linux/kernel/futex.c

4066 lines
110 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Fast Userspace Mutexes (which I call "Futexes!").
* (C) Rusty Russell, IBM 2002
*
* Generalized futexes, futex requeueing, misc fixes by Ingo Molnar
* (C) Copyright 2003 Red Hat Inc, All Rights Reserved
*
* Removed page pinning, fix privately mapped COW pages and other cleanups
* (C) Copyright 2003, 2004 Jamie Lokier
*
* Robust futex support started by Ingo Molnar
* (C) Copyright 2006 Red Hat Inc, All Rights Reserved
* Thanks to Thomas Gleixner for suggestions, analysis and fixes.
*
* PI-futex support started by Ingo Molnar and Thomas Gleixner
* Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
* Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
*
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
* PRIVATE futexes by Eric Dumazet
* Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com>
*
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
* Requeue-PI support by Darren Hart <dvhltc@us.ibm.com>
* Copyright (C) IBM Corporation, 2009
* Thanks to Thomas Gleixner for conceptual design and careful reviews.
*
* Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly
* enough at me, Linus for the original (flawed) idea, Matthew
* Kirkwood for proof-of-concept implementation.
*
* "The futexes are also cursed."
* "But they come in a choice of three flavours!"
*/
#include <linux/compat.h>
#include <linux/jhash.h>
#include <linux/pagemap.h>
#include <linux/syscalls.h>
futex: Take hugepages into account when generating futex_key The futex_keys of process shared futexes are generated from the page offset, the mapping host and the mapping index of the futex user space address. This should result in an unique identifier for each futex. Though this is not true when futexes are located in different subpages of an hugepage. The reason is, that the mapping index for all those futexes evaluates to the index of the base page of the hugetlbfs mapping. So a futex at offset 0 of the hugepage mapping and another one at offset PAGE_SIZE of the same hugepage mapping have identical futex_keys. This happens because the futex code blindly uses page->index. Steps to reproduce the bug: 1. Map a file from hugetlbfs. Initialize pthread_mutex1 at offset 0 and pthread_mutex2 at offset PAGE_SIZE of the hugetlbfs mapping. The mutexes must be initialized as PTHREAD_PROCESS_SHARED because PTHREAD_PROCESS_PRIVATE mutexes are not affected by this issue as their keys solely depend on the user space address. 2. Lock mutex1 and mutex2 3. Create thread1 and in the thread function lock mutex1, which results in thread1 blocking on the locked mutex1. 4. Create thread2 and in the thread function lock mutex2, which results in thread2 blocking on the locked mutex2. 5. Unlock mutex2. Despite the fact that mutex2 got unlocked, thread2 still blocks on mutex2 because the futex_key points to mutex1. To solve this issue we need to take the normal page index of the page which contains the futex into account, if the futex is in an hugetlbfs mapping. In other words, we calculate the normal page mapping index of the subpage in the hugetlbfs mapping. Mappings which are not based on hugetlbfs are not affected and still use page->index. Thanks to Mel Gorman who provided a patch for adding proper evaluation functions to the hugetlbfs code to avoid exposing hugetlbfs specific details to the futex code. [ tglx: Massaged changelog ] Signed-off-by: Zhang Yi <zhang.yi20@zte.com.cn> Reviewed-by: Jiang Biao <jiang.biao2@zte.com.cn> Tested-by: Ma Chenggong <ma.chenggong@zte.com.cn> Reviewed-by: 'Mel Gorman' <mgorman@suse.de> Acked-by: 'Darren Hart' <dvhart@linux.intel.com> Cc: 'Peter Zijlstra' <peterz@infradead.org> Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/000101ce71a6%24a83c5880%24f8b50980%24@com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2013-06-25 13:19:31 +00:00
#include <linux/hugetlb.h>
#include <linux/freezer.h>
mm: remove include/linux/bootmem.h Move remaining definitions and declarations from include/linux/bootmem.h into include/linux/memblock.h and remove the redundant header. The includes were replaced with the semantic patch below and then semi-automated removal of duplicated '#include <linux/memblock.h> @@ @@ - #include <linux/bootmem.h> + #include <linux/memblock.h> [sfr@canb.auug.org.au: dma-direct: fix up for the removal of linux/bootmem.h] Link: http://lkml.kernel.org/r/20181002185342.133d1680@canb.auug.org.au [sfr@canb.auug.org.au: powerpc: fix up for removal of linux/bootmem.h] Link: http://lkml.kernel.org/r/20181005161406.73ef8727@canb.auug.org.au [sfr@canb.auug.org.au: x86/kaslr, ACPI/NUMA: fix for linux/bootmem.h removal] Link: http://lkml.kernel.org/r/20181008190341.5e396491@canb.auug.org.au Link: http://lkml.kernel.org/r/1536927045-23536-30-git-send-email-rppt@linux.vnet.ibm.com Signed-off-by: Mike Rapoport <rppt@linux.vnet.ibm.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Acked-by: Michal Hocko <mhocko@suse.com> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Chris Zankel <chris@zankel.net> Cc: "David S. Miller" <davem@davemloft.net> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Greentime Hu <green.hu@gmail.com> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Guan Xuetao <gxt@pku.edu.cn> Cc: Ingo Molnar <mingo@redhat.com> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Ley Foon Tan <lftan@altera.com> Cc: Mark Salter <msalter@redhat.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: Michael Ellerman <mpe@ellerman.id.au> Cc: Michal Simek <monstr@monstr.eu> Cc: Palmer Dabbelt <palmer@sifive.com> Cc: Paul Burton <paul.burton@mips.com> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Richard Weinberger <richard@nod.at> Cc: Rich Felker <dalias@libc.org> Cc: Russell King <linux@armlinux.org.uk> Cc: Serge Semin <fancer.lancer@gmail.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Tony Luck <tony.luck@intel.com> Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-10-30 22:09:49 +00:00
#include <linux/memblock.h>
#include <linux/fault-inject.h>
#include <linux/time_namespace.h>
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
#include <asm/futex.h>
#include "locking/rtmutex_common.h"
/*
* READ this before attempting to hack on futexes!
*
* Basic futex operation and ordering guarantees
* =============================================
*
* The waiter reads the futex value in user space and calls
* futex_wait(). This function computes the hash bucket and acquires
* the hash bucket lock. After that it reads the futex user space value
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* again and verifies that the data has not changed. If it has not changed
* it enqueues itself into the hash bucket, releases the hash bucket lock
* and schedules.
*
* The waker side modifies the user space value of the futex and calls
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* futex_wake(). This function computes the hash bucket and acquires the
* hash bucket lock. Then it looks for waiters on that futex in the hash
* bucket and wakes them.
*
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* In futex wake up scenarios where no tasks are blocked on a futex, taking
* the hb spinlock can be avoided and simply return. In order for this
* optimization to work, ordering guarantees must exist so that the waiter
* being added to the list is acknowledged when the list is concurrently being
* checked by the waker, avoiding scenarios like the following:
*
* CPU 0 CPU 1
* val = *futex;
* sys_futex(WAIT, futex, val);
* futex_wait(futex, val);
* uval = *futex;
* *futex = newval;
* sys_futex(WAKE, futex);
* futex_wake(futex);
* if (queue_empty())
* return;
* if (uval == val)
* lock(hash_bucket(futex));
* queue();
* unlock(hash_bucket(futex));
* schedule();
*
* This would cause the waiter on CPU 0 to wait forever because it
* missed the transition of the user space value from val to newval
* and the waker did not find the waiter in the hash bucket queue.
*
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* The correct serialization ensures that a waiter either observes
* the changed user space value before blocking or is woken by a
* concurrent waker:
*
* CPU 0 CPU 1
* val = *futex;
* sys_futex(WAIT, futex, val);
* futex_wait(futex, val);
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
*
* waiters++; (a)
* smp_mb(); (A) <-- paired with -.
* |
* lock(hash_bucket(futex)); |
* |
* uval = *futex; |
* | *futex = newval;
* | sys_futex(WAKE, futex);
* | futex_wake(futex);
* |
* `--------> smp_mb(); (B)
* if (uval == val)
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* queue();
* unlock(hash_bucket(futex));
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
* schedule(); if (waiters)
* lock(hash_bucket(futex));
* else wake_waiters(futex);
* waiters--; (b) unlock(hash_bucket(futex));
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
*
* Where (A) orders the waiters increment and the futex value read through
* atomic operations (see hb_waiters_inc) and where (B) orders the write
* to futex and the waiters read (see hb_waiters_pending()).
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
*
* This yields the following case (where X:=waiters, Y:=futex):
*
* X = Y = 0
*
* w[X]=1 w[Y]=1
* MB MB
* r[Y]=y r[X]=x
*
* Which guarantees that x==0 && y==0 is impossible; which translates back into
* the guarantee that we cannot both miss the futex variable change and the
* enqueue.
*
* Note that a new waiter is accounted for in (a) even when it is possible that
* the wait call can return error, in which case we backtrack from it in (b).
* Refer to the comment in queue_lock().
*
* Similarly, in order to account for waiters being requeued on another
* address we always increment the waiters for the destination bucket before
* acquiring the lock. It then decrements them again after releasing it -
* the code that actually moves the futex(es) between hash buckets (requeue_futex)
* will do the additional required waiter count housekeeping. This is done for
* double_lock_hb() and double_unlock_hb(), respectively.
*/
#ifdef CONFIG_HAVE_FUTEX_CMPXCHG
#define futex_cmpxchg_enabled 1
#else
static int __read_mostly futex_cmpxchg_enabled;
#endif
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
/*
* Futex flags used to encode options to functions and preserve them across
* restarts.
*/
#ifdef CONFIG_MMU
# define FLAGS_SHARED 0x01
#else
/*
* NOMMU does not have per process address space. Let the compiler optimize
* code away.
*/
# define FLAGS_SHARED 0x00
#endif
#define FLAGS_CLOCKRT 0x02
#define FLAGS_HAS_TIMEOUT 0x04
/*
* Priority Inheritance state:
*/
struct futex_pi_state {
/*
* list of 'owned' pi_state instances - these have to be
* cleaned up in do_exit() if the task exits prematurely:
*/
struct list_head list;
/*
* The PI object:
*/
struct rt_mutex pi_mutex;
struct task_struct *owner;
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
refcount_t refcount;
union futex_key key;
} __randomize_layout;
/**
* struct futex_q - The hashed futex queue entry, one per waiting task
* @list: priority-sorted list of tasks waiting on this futex
* @task: the task waiting on the futex
* @lock_ptr: the hash bucket lock
* @key: the key the futex is hashed on
* @pi_state: optional priority inheritance state
* @rt_waiter: rt_waiter storage for use with requeue_pi
* @requeue_pi_key: the requeue_pi target futex key
* @bitset: bitset for the optional bitmasked wakeup
*
* We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so
* we can wake only the relevant ones (hashed queues may be shared).
*
* A futex_q has a woken state, just like tasks have TASK_RUNNING.
* It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
* The order of wakeup is always to make the first condition true, then
* the second.
*
* PI futexes are typically woken before they are removed from the hash list via
* the rt_mutex code. See unqueue_me_pi().
*/
struct futex_q {
struct plist_node list;
struct task_struct *task;
spinlock_t *lock_ptr;
union futex_key key;
struct futex_pi_state *pi_state;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
struct rt_mutex_waiter *rt_waiter;
union futex_key *requeue_pi_key;
u32 bitset;
} __randomize_layout;
static const struct futex_q futex_q_init = {
/* list gets initialized in queue_me()*/
.key = FUTEX_KEY_INIT,
.bitset = FUTEX_BITSET_MATCH_ANY
};
/*
* Hash buckets are shared by all the futex_keys that hash to the same
* location. Each key may have multiple futex_q structures, one for each task
* waiting on a futex.
*/
struct futex_hash_bucket {
atomic_t waiters;
spinlock_t lock;
struct plist_head chain;
futexes: Increase hash table size for better performance Currently, the futex global hash table suffers from its fixed, smallish (for today's standards) size of 256 entries, as well as its lack of NUMA awareness. Large systems, using many futexes, can be prone to high amounts of collisions; where these futexes hash to the same bucket and lead to extra contention on the same hb->lock. Furthermore, cacheline bouncing is a reality when we have multiple hb->locks residing on the same cacheline and different futexes hash to adjacent buckets. This patch keeps the current static size of 16 entries for small systems, or otherwise, 256 * ncpus (or larger as we need to round the number to a power of 2). Note that this number of CPUs accounts for all CPUs that can ever be available in the system, taking into consideration things like hotpluging. While we do impose extra overhead at bootup by making the hash table larger, this is a one time thing, and does not shadow the benefits of this patch. Furthermore, as suggested by tglx, by cache aligning the hash buckets we can avoid access across cacheline boundaries and also avoid massive cache line bouncing if multiple cpus are hammering away at different hash buckets which happen to reside in the same cache line. Also, similar to other core kernel components (pid, dcache, tcp), by using alloc_large_system_hash() we benefit from its NUMA awareness and thus the table is distributed among the nodes instead of in a single one. For a custom microbenchmark that pounds on the uaddr hashing -- making the wait path fail at futex_wait_setup() returning -EWOULDBLOCK for large amounts of futexes, we can see the following benefits on a 80-core, 8-socket 1Tb server: +---------+--------------------+------------------------+-----------------------+-------------------------------+ | threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) | +---------+--------------------+------------------------+-----------------------+-------------------------------+ |     512 |              32426 | 50531  (+55.8%)        | 255274  (+687.2%)     | 292553  (+802.2%)             | |     256 |              65360 | 99588  (+52.3%)        | 443563  (+578.6%)     | 508088  (+677.3%)             | |     128 |             125635 | 200075 (+59.2%)        | 742613  (+491.1%)     | 835452  (+564.9%)             | |      80 |             193559 | 323425 (+67.1%)        | 1028147 (+431.1%)     | 1130304 (+483.9%)             | |      64 |             247667 | 443740 (+79.1%)        | 997300  (+302.6%)     | 1145494 (+362.5%)             | |      32 |             628412 | 721401 (+14.7%)        | 965996  (+53.7%)      | 1122115 (+78.5%)              | +---------+--------------------+------------------------+-----------------------+-------------------------------+ Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Waiman Long <Waiman.Long@hp.com> Reviewed-and-tested-by: Jason Low <jason.low2@hp.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:23 +00:00
} ____cacheline_aligned_in_smp;
/*
* The base of the bucket array and its size are always used together
* (after initialization only in hash_futex()), so ensure that they
* reside in the same cacheline.
*/
static struct {
struct futex_hash_bucket *queues;
unsigned long hashsize;
} __futex_data __read_mostly __aligned(2*sizeof(long));
#define futex_queues (__futex_data.queues)
#define futex_hashsize (__futex_data.hashsize)
futexes: Increase hash table size for better performance Currently, the futex global hash table suffers from its fixed, smallish (for today's standards) size of 256 entries, as well as its lack of NUMA awareness. Large systems, using many futexes, can be prone to high amounts of collisions; where these futexes hash to the same bucket and lead to extra contention on the same hb->lock. Furthermore, cacheline bouncing is a reality when we have multiple hb->locks residing on the same cacheline and different futexes hash to adjacent buckets. This patch keeps the current static size of 16 entries for small systems, or otherwise, 256 * ncpus (or larger as we need to round the number to a power of 2). Note that this number of CPUs accounts for all CPUs that can ever be available in the system, taking into consideration things like hotpluging. While we do impose extra overhead at bootup by making the hash table larger, this is a one time thing, and does not shadow the benefits of this patch. Furthermore, as suggested by tglx, by cache aligning the hash buckets we can avoid access across cacheline boundaries and also avoid massive cache line bouncing if multiple cpus are hammering away at different hash buckets which happen to reside in the same cache line. Also, similar to other core kernel components (pid, dcache, tcp), by using alloc_large_system_hash() we benefit from its NUMA awareness and thus the table is distributed among the nodes instead of in a single one. For a custom microbenchmark that pounds on the uaddr hashing -- making the wait path fail at futex_wait_setup() returning -EWOULDBLOCK for large amounts of futexes, we can see the following benefits on a 80-core, 8-socket 1Tb server: +---------+--------------------+------------------------+-----------------------+-------------------------------+ | threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) | +---------+--------------------+------------------------+-----------------------+-------------------------------+ |     512 |              32426 | 50531  (+55.8%)        | 255274  (+687.2%)     | 292553  (+802.2%)             | |     256 |              65360 | 99588  (+52.3%)        | 443563  (+578.6%)     | 508088  (+677.3%)             | |     128 |             125635 | 200075 (+59.2%)        | 742613  (+491.1%)     | 835452  (+564.9%)             | |      80 |             193559 | 323425 (+67.1%)        | 1028147 (+431.1%)     | 1130304 (+483.9%)             | |      64 |             247667 | 443740 (+79.1%)        | 997300  (+302.6%)     | 1145494 (+362.5%)             | |      32 |             628412 | 721401 (+14.7%)        | 965996  (+53.7%)      | 1122115 (+78.5%)              | +---------+--------------------+------------------------+-----------------------+-------------------------------+ Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Waiman Long <Waiman.Long@hp.com> Reviewed-and-tested-by: Jason Low <jason.low2@hp.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:23 +00:00
/*
* Fault injections for futexes.
*/
#ifdef CONFIG_FAIL_FUTEX
static struct {
struct fault_attr attr;
bool ignore_private;
} fail_futex = {
.attr = FAULT_ATTR_INITIALIZER,
.ignore_private = false,
};
static int __init setup_fail_futex(char *str)
{
return setup_fault_attr(&fail_futex.attr, str);
}
__setup("fail_futex=", setup_fail_futex);
static bool should_fail_futex(bool fshared)
{
if (fail_futex.ignore_private && !fshared)
return false;
return should_fail(&fail_futex.attr, 1);
}
#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
static int __init fail_futex_debugfs(void)
{
umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
struct dentry *dir;
dir = fault_create_debugfs_attr("fail_futex", NULL,
&fail_futex.attr);
if (IS_ERR(dir))
return PTR_ERR(dir);
debugfs_create_bool("ignore-private", mode, dir,
&fail_futex.ignore_private);
return 0;
}
late_initcall(fail_futex_debugfs);
#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
#else
static inline bool should_fail_futex(bool fshared)
{
return false;
}
#endif /* CONFIG_FAIL_FUTEX */
#ifdef CONFIG_COMPAT
static void compat_exit_robust_list(struct task_struct *curr);
#endif
/*
* Reflects a new waiter being added to the waitqueue.
*/
static inline void hb_waiters_inc(struct futex_hash_bucket *hb)
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
{
#ifdef CONFIG_SMP
atomic_inc(&hb->waiters);
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
/*
* Full barrier (A), see the ordering comment above.
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
*/
smp_mb__after_atomic();
#endif
}
/*
* Reflects a waiter being removed from the waitqueue by wakeup
* paths.
*/
static inline void hb_waiters_dec(struct futex_hash_bucket *hb)
{
#ifdef CONFIG_SMP
atomic_dec(&hb->waiters);
#endif
}
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
static inline int hb_waiters_pending(struct futex_hash_bucket *hb)
{
#ifdef CONFIG_SMP
/*
* Full barrier (B), see the ordering comment above.
*/
smp_mb();
return atomic_read(&hb->waiters);
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
#else
return 1;
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
#endif
}
/**
* hash_futex - Return the hash bucket in the global hash
* @key: Pointer to the futex key for which the hash is calculated
*
* We hash on the keys returned from get_futex_key (see below) and return the
* corresponding hash bucket in the global hash.
*/
static struct futex_hash_bucket *hash_futex(union futex_key *key)
{
u32 hash = jhash2((u32 *)key, offsetof(typeof(*key), both.offset) / 4,
key->both.offset);
futexes: Increase hash table size for better performance Currently, the futex global hash table suffers from its fixed, smallish (for today's standards) size of 256 entries, as well as its lack of NUMA awareness. Large systems, using many futexes, can be prone to high amounts of collisions; where these futexes hash to the same bucket and lead to extra contention on the same hb->lock. Furthermore, cacheline bouncing is a reality when we have multiple hb->locks residing on the same cacheline and different futexes hash to adjacent buckets. This patch keeps the current static size of 16 entries for small systems, or otherwise, 256 * ncpus (or larger as we need to round the number to a power of 2). Note that this number of CPUs accounts for all CPUs that can ever be available in the system, taking into consideration things like hotpluging. While we do impose extra overhead at bootup by making the hash table larger, this is a one time thing, and does not shadow the benefits of this patch. Furthermore, as suggested by tglx, by cache aligning the hash buckets we can avoid access across cacheline boundaries and also avoid massive cache line bouncing if multiple cpus are hammering away at different hash buckets which happen to reside in the same cache line. Also, similar to other core kernel components (pid, dcache, tcp), by using alloc_large_system_hash() we benefit from its NUMA awareness and thus the table is distributed among the nodes instead of in a single one. For a custom microbenchmark that pounds on the uaddr hashing -- making the wait path fail at futex_wait_setup() returning -EWOULDBLOCK for large amounts of futexes, we can see the following benefits on a 80-core, 8-socket 1Tb server: +---------+--------------------+------------------------+-----------------------+-------------------------------+ | threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) | +---------+--------------------+------------------------+-----------------------+-------------------------------+ |     512 |              32426 | 50531  (+55.8%)        | 255274  (+687.2%)     | 292553  (+802.2%)             | |     256 |              65360 | 99588  (+52.3%)        | 443563  (+578.6%)     | 508088  (+677.3%)             | |     128 |             125635 | 200075 (+59.2%)        | 742613  (+491.1%)     | 835452  (+564.9%)             | |      80 |             193559 | 323425 (+67.1%)        | 1028147 (+431.1%)     | 1130304 (+483.9%)             | |      64 |             247667 | 443740 (+79.1%)        | 997300  (+302.6%)     | 1145494 (+362.5%)             | |      32 |             628412 | 721401 (+14.7%)        | 965996  (+53.7%)      | 1122115 (+78.5%)              | +---------+--------------------+------------------------+-----------------------+-------------------------------+ Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Waiman Long <Waiman.Long@hp.com> Reviewed-and-tested-by: Jason Low <jason.low2@hp.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:23 +00:00
return &futex_queues[hash & (futex_hashsize - 1)];
}
/**
* match_futex - Check whether two futex keys are equal
* @key1: Pointer to key1
* @key2: Pointer to key2
*
* Return 1 if two futex_keys are equal, 0 otherwise.
*/
static inline int match_futex(union futex_key *key1, union futex_key *key2)
{
return (key1 && key2
&& key1->both.word == key2->both.word
&& key1->both.ptr == key2->both.ptr
&& key1->both.offset == key2->both.offset);
}
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
enum futex_access {
FUTEX_READ,
FUTEX_WRITE
};
/**
* futex_setup_timer - set up the sleeping hrtimer.
* @time: ptr to the given timeout value
* @timeout: the hrtimer_sleeper structure to be set up
* @flags: futex flags
* @range_ns: optional range in ns
*
* Return: Initialized hrtimer_sleeper structure or NULL if no timeout
* value given
*/
static inline struct hrtimer_sleeper *
futex_setup_timer(ktime_t *time, struct hrtimer_sleeper *timeout,
int flags, u64 range_ns)
{
if (!time)
return NULL;
hrtimer_init_sleeper_on_stack(timeout, (flags & FLAGS_CLOCKRT) ?
CLOCK_REALTIME : CLOCK_MONOTONIC,
HRTIMER_MODE_ABS);
/*
* If range_ns is 0, calling hrtimer_set_expires_range_ns() is
* effectively the same as calling hrtimer_set_expires().
*/
hrtimer_set_expires_range_ns(&timeout->timer, *time, range_ns);
return timeout;
}
/*
* Generate a machine wide unique identifier for this inode.
*
* This relies on u64 not wrapping in the life-time of the machine; which with
* 1ns resolution means almost 585 years.
*
* This further relies on the fact that a well formed program will not unmap
* the file while it has a (shared) futex waiting on it. This mapping will have
* a file reference which pins the mount and inode.
*
* If for some reason an inode gets evicted and read back in again, it will get
* a new sequence number and will _NOT_ match, even though it is the exact same
* file.
*
* It is important that match_futex() will never have a false-positive, esp.
* for PI futexes that can mess up the state. The above argues that false-negatives
* are only possible for malformed programs.
*/
static u64 get_inode_sequence_number(struct inode *inode)
{
static atomic64_t i_seq;
u64 old;
/* Does the inode already have a sequence number? */
old = atomic64_read(&inode->i_sequence);
if (likely(old))
return old;
for (;;) {
u64 new = atomic64_add_return(1, &i_seq);
if (WARN_ON_ONCE(!new))
continue;
old = atomic64_cmpxchg_relaxed(&inode->i_sequence, 0, new);
if (old)
return old;
return new;
}
}
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
/**
* get_futex_key() - Get parameters which are the keys for a futex
* @uaddr: virtual address of the futex
* @fshared: false for a PROCESS_PRIVATE futex, true for PROCESS_SHARED
* @key: address where result is stored.
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
* @rw: mapping needs to be read/write (values: FUTEX_READ,
* FUTEX_WRITE)
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
*
* Return: a negative error code or 0
*
* The key words are stored in @key on success.
*
* For shared mappings (when @fshared), the key is:
*
* ( inode->i_sequence, page->index, offset_within_page )
*
* [ also see get_inode_sequence_number() ]
*
* For private mappings (or when !@fshared), the key is:
*
* ( current->mm, address, 0 )
*
* This allows (cross process, where applicable) identification of the futex
* without keeping the page pinned for the duration of the FUTEX_WAIT.
*
* lock_page() might sleep, the caller should not hold a spinlock.
*/
static int get_futex_key(u32 __user *uaddr, bool fshared, union futex_key *key,
enum futex_access rw)
{
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
unsigned long address = (unsigned long)uaddr;
struct mm_struct *mm = current->mm;
futex: Calculate the futex key based on a tail page for file-based futexes Mike Galbraith reported that the LTP test case futex_wake04 was broken by commit 65d8fc777f6d ("futex: Remove requirement for lock_page() in get_futex_key()"). This test case uses futexes backed by hugetlbfs pages and so there is an associated inode with a futex stored on such pages. The problem is that the key is being calculated based on the head page index of the hugetlbfs page and not the tail page. Prior to the optimisation, the page lock was used to stabilise mappings and pin the inode is file-backed which is overkill. If the page was a compound page, the head page was automatically looked up as part of the page lock operation but the tail page index was used to calculate the futex key. After the optimisation, the compound head is looked up early and the page lock is only relied upon to identify truncated pages, special pages or a shmem page moving to swapcache. The head page is looked up because without the page lock, special care has to be taken to pin the inode correctly. However, the tail page is still required to calculate the futex key so this patch records the tail page. On vanilla 4.6, the output of the test case is; futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TFAIL : futex_wake04.c:126: Bug: wait_thread2 did not wake after 30 secs. With the patch applied futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TPASS : Hi hydra, thread2 awake! Fixes: 65d8fc777f6d "futex: Remove requirement for lock_page() in get_futex_key()" Reported-and-tested-by: Mike Galbraith <umgwanakikbuti@gmail.com> Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/20160608132522.GM2469@suse.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-06-08 13:25:22 +00:00
struct page *page, *tail;
struct address_space *mapping;
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
int err, ro = 0;
/*
* The futex address must be "naturally" aligned.
*/
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
key->both.offset = address % PAGE_SIZE;
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
if (unlikely((address % sizeof(u32)) != 0))
return -EINVAL;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
address -= key->both.offset;
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
if (unlikely(!access_ok(uaddr, sizeof(u32))))
return -EFAULT;
if (unlikely(should_fail_futex(fshared)))
return -EFAULT;
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
/*
* PROCESS_PRIVATE futexes are fast.
* As the mm cannot disappear under us and the 'key' only needs
* virtual address, we dont even have to find the underlying vma.
* Note : We do have to check 'uaddr' is a valid user address,
* but access_ok() should be faster than find_vma()
*/
if (!fshared) {
key->private.mm = mm;
key->private.address = address;
return 0;
}
again:
/* Ignore any VERIFY_READ mapping (futex common case) */
if (unlikely(should_fail_futex(true)))
return -EFAULT;
mm/gup: change GUP fast to use flags rather than a write 'bool' To facilitate additional options to get_user_pages_fast() change the singular write parameter to be gup_flags. This patch does not change any functionality. New functionality will follow in subsequent patches. Some of the get_user_pages_fast() call sites were unchanged because they already passed FOLL_WRITE or 0 for the write parameter. NOTE: It was suggested to change the ordering of the get_user_pages_fast() arguments to ensure that callers were converted. This breaks the current GUP call site convention of having the returned pages be the final parameter. So the suggestion was rejected. Link: http://lkml.kernel.org/r/20190328084422.29911-4-ira.weiny@intel.com Link: http://lkml.kernel.org/r/20190317183438.2057-4-ira.weiny@intel.com Signed-off-by: Ira Weiny <ira.weiny@intel.com> Reviewed-by: Mike Marshall <hubcap@omnibond.com> Cc: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Dan Williams <dan.j.williams@intel.com> Cc: "David S. Miller" <davem@davemloft.net> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Hogan <jhogan@kernel.org> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: John Hubbard <jhubbard@nvidia.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Paul Mackerras <paulus@samba.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Rich Felker <dalias@libc.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-05-14 00:17:11 +00:00
err = get_user_pages_fast(address, 1, FOLL_WRITE, &page);
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
/*
* If write access is not required (eg. FUTEX_WAIT), try
* and get read-only access.
*/
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
if (err == -EFAULT && rw == FUTEX_READ) {
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
err = get_user_pages_fast(address, 1, 0, &page);
ro = 1;
}
if (err < 0)
return err;
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
else
err = 0;
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
/*
* The treatment of mapping from this point on is critical. The page
* lock protects many things but in this context the page lock
* stabilizes mapping, prevents inode freeing in the shared
* file-backed region case and guards against movement to swap cache.
*
* Strictly speaking the page lock is not needed in all cases being
* considered here and page lock forces unnecessarily serialization
* From this point on, mapping will be re-verified if necessary and
* page lock will be acquired only if it is unavoidable
futex: Calculate the futex key based on a tail page for file-based futexes Mike Galbraith reported that the LTP test case futex_wake04 was broken by commit 65d8fc777f6d ("futex: Remove requirement for lock_page() in get_futex_key()"). This test case uses futexes backed by hugetlbfs pages and so there is an associated inode with a futex stored on such pages. The problem is that the key is being calculated based on the head page index of the hugetlbfs page and not the tail page. Prior to the optimisation, the page lock was used to stabilise mappings and pin the inode is file-backed which is overkill. If the page was a compound page, the head page was automatically looked up as part of the page lock operation but the tail page index was used to calculate the futex key. After the optimisation, the compound head is looked up early and the page lock is only relied upon to identify truncated pages, special pages or a shmem page moving to swapcache. The head page is looked up because without the page lock, special care has to be taken to pin the inode correctly. However, the tail page is still required to calculate the futex key so this patch records the tail page. On vanilla 4.6, the output of the test case is; futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TFAIL : futex_wake04.c:126: Bug: wait_thread2 did not wake after 30 secs. With the patch applied futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TPASS : Hi hydra, thread2 awake! Fixes: 65d8fc777f6d "futex: Remove requirement for lock_page() in get_futex_key()" Reported-and-tested-by: Mike Galbraith <umgwanakikbuti@gmail.com> Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/20160608132522.GM2469@suse.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-06-08 13:25:22 +00:00
*
* Mapping checks require the head page for any compound page so the
* head page and mapping is looked up now. For anonymous pages, it
* does not matter if the page splits in the future as the key is
* based on the address. For filesystem-backed pages, the tail is
* required as the index of the page determines the key. For
* base pages, there is no tail page and tail == page.
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
*/
futex: Calculate the futex key based on a tail page for file-based futexes Mike Galbraith reported that the LTP test case futex_wake04 was broken by commit 65d8fc777f6d ("futex: Remove requirement for lock_page() in get_futex_key()"). This test case uses futexes backed by hugetlbfs pages and so there is an associated inode with a futex stored on such pages. The problem is that the key is being calculated based on the head page index of the hugetlbfs page and not the tail page. Prior to the optimisation, the page lock was used to stabilise mappings and pin the inode is file-backed which is overkill. If the page was a compound page, the head page was automatically looked up as part of the page lock operation but the tail page index was used to calculate the futex key. After the optimisation, the compound head is looked up early and the page lock is only relied upon to identify truncated pages, special pages or a shmem page moving to swapcache. The head page is looked up because without the page lock, special care has to be taken to pin the inode correctly. However, the tail page is still required to calculate the futex key so this patch records the tail page. On vanilla 4.6, the output of the test case is; futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TFAIL : futex_wake04.c:126: Bug: wait_thread2 did not wake after 30 secs. With the patch applied futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TPASS : Hi hydra, thread2 awake! Fixes: 65d8fc777f6d "futex: Remove requirement for lock_page() in get_futex_key()" Reported-and-tested-by: Mike Galbraith <umgwanakikbuti@gmail.com> Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/20160608132522.GM2469@suse.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-06-08 13:25:22 +00:00
tail = page;
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
page = compound_head(page);
mapping = READ_ONCE(page->mapping);
/*
* If page->mapping is NULL, then it cannot be a PageAnon
* page; but it might be the ZERO_PAGE or in the gate area or
* in a special mapping (all cases which we are happy to fail);
* or it may have been a good file page when get_user_pages_fast
* found it, but truncated or holepunched or subjected to
* invalidate_complete_page2 before we got the page lock (also
* cases which we are happy to fail). And we hold a reference,
* so refcount care in invalidate_complete_page's remove_mapping
* prevents drop_caches from setting mapping to NULL beneath us.
*
* The case we do have to guard against is when memory pressure made
* shmem_writepage move it from filecache to swapcache beneath us:
* an unlikely race, but we do need to retry for page->mapping.
*/
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
if (unlikely(!mapping)) {
int shmem_swizzled;
/*
* Page lock is required to identify which special case above
* applies. If this is really a shmem page then the page lock
* will prevent unexpected transitions.
*/
lock_page(page);
shmem_swizzled = PageSwapCache(page) || page->mapping;
unlock_page(page);
put_page(page);
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
if (shmem_swizzled)
goto again;
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
return -EFAULT;
}
/*
* Private mappings are handled in a simple way.
*
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
* If the futex key is stored on an anonymous page, then the associated
* object is the mm which is implicitly pinned by the calling process.
*
* NOTE: When userspace waits on a MAP_SHARED mapping, even if
* it's a read-only handle, it's expected that futexes attach to
* the object not the particular process.
*/
if (PageAnon(page)) {
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
/*
* A RO anonymous page will never change and thus doesn't make
* sense for futex operations.
*/
if (unlikely(should_fail_futex(true)) || ro) {
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
err = -EFAULT;
goto out;
}
key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
key->private.mm = mm;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
key->private.address = address;
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
} else {
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
struct inode *inode;
/*
* The associated futex object in this case is the inode and
* the page->mapping must be traversed. Ordinarily this should
* be stabilised under page lock but it's not strictly
* necessary in this case as we just want to pin the inode, not
* update the radix tree or anything like that.
*
* The RCU read lock is taken as the inode is finally freed
* under RCU. If the mapping still matches expectations then the
* mapping->host can be safely accessed as being a valid inode.
*/
rcu_read_lock();
if (READ_ONCE(page->mapping) != mapping) {
rcu_read_unlock();
put_page(page);
goto again;
}
inode = READ_ONCE(mapping->host);
if (!inode) {
rcu_read_unlock();
put_page(page);
goto again;
}
key->both.offset |= FUT_OFF_INODE; /* inode-based key */
key->shared.i_seq = get_inode_sequence_number(inode);
futex: Calculate the futex key based on a tail page for file-based futexes Mike Galbraith reported that the LTP test case futex_wake04 was broken by commit 65d8fc777f6d ("futex: Remove requirement for lock_page() in get_futex_key()"). This test case uses futexes backed by hugetlbfs pages and so there is an associated inode with a futex stored on such pages. The problem is that the key is being calculated based on the head page index of the hugetlbfs page and not the tail page. Prior to the optimisation, the page lock was used to stabilise mappings and pin the inode is file-backed which is overkill. If the page was a compound page, the head page was automatically looked up as part of the page lock operation but the tail page index was used to calculate the futex key. After the optimisation, the compound head is looked up early and the page lock is only relied upon to identify truncated pages, special pages or a shmem page moving to swapcache. The head page is looked up because without the page lock, special care has to be taken to pin the inode correctly. However, the tail page is still required to calculate the futex key so this patch records the tail page. On vanilla 4.6, the output of the test case is; futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TFAIL : futex_wake04.c:126: Bug: wait_thread2 did not wake after 30 secs. With the patch applied futex_wake04 0 TINFO : Hugepagesize 2097152 futex_wake04 1 TPASS : Hi hydra, thread2 awake! Fixes: 65d8fc777f6d "futex: Remove requirement for lock_page() in get_futex_key()" Reported-and-tested-by: Mike Galbraith <umgwanakikbuti@gmail.com> Signed-off-by: Mel Gorman <mgorman@techsingularity.net> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Davidlohr Bueso <dave@stgolabs.net> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: http://lkml.kernel.org/r/20160608132522.GM2469@suse.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2016-06-08 13:25:22 +00:00
key->shared.pgoff = basepage_index(tail);
futex: Remove requirement for lock_page() in get_futex_key() When dealing with key handling for shared futexes, we can drastically reduce the usage/need of the page lock. 1) For anonymous pages, the associated futex object is the mm_struct which does not require the page lock. 2) For inode based, keys, we can check under RCU read lock if the page mapping is still valid and take reference to the inode. This just leaves one rare race that requires the page lock in the slow path when examining the swapcache. Additionally realtime users currently have a problem with the page lock being contended for unbounded periods of time during futex operations. Task A get_futex_key() lock_page() ---> preempted Now any other task trying to lock that page will have to wait until task A gets scheduled back in, which is an unbound time. With this patch, we pretty much have a lockless futex_get_key(). Experiments show that this patch can boost/speedup the hashing of shared futexes with the perf futex benchmarks (which is good for measuring such change) by up to 45% when there are high (> 100) thread counts on a 60 core Westmere. Lower counts are pretty much in the noise range or less than 10%, but mid range can be seen at over 30% overall throughput (hash ops/sec). This makes anon-mem shared futexes much closer to its private counterpart. Signed-off-by: Mel Gorman <mgorman@suse.de> [ Ported on top of thp refcount rework, changelog, comments, fixes. ] Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Chris Mason <clm@fb.com> Cc: Darren Hart <dvhart@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: dave@stgolabs.net Link: http://lkml.kernel.org/r/1455045314-8305-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-02-09 19:15:14 +00:00
rcu_read_unlock();
}
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
out:
put_page(page);
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
return err;
}
/**
* fault_in_user_writeable() - Fault in user address and verify RW access
* @uaddr: pointer to faulting user space address
*
* Slow path to fixup the fault we just took in the atomic write
* access to @uaddr.
*
* We have no generic implementation of a non-destructive write to the
* user address. We know that we faulted in the atomic pagefault
* disabled section so we can as well avoid the #PF overhead by
* calling get_user_pages() right away.
*/
static int fault_in_user_writeable(u32 __user *uaddr)
{
struct mm_struct *mm = current->mm;
int ret;
mmap locking API: use coccinelle to convert mmap_sem rwsem call sites This change converts the existing mmap_sem rwsem calls to use the new mmap locking API instead. The change is generated using coccinelle with the following rule: // spatch --sp-file mmap_lock_api.cocci --in-place --include-headers --dir . @@ expression mm; @@ ( -init_rwsem +mmap_init_lock | -down_write +mmap_write_lock | -down_write_killable +mmap_write_lock_killable | -down_write_trylock +mmap_write_trylock | -up_write +mmap_write_unlock | -downgrade_write +mmap_write_downgrade | -down_read +mmap_read_lock | -down_read_killable +mmap_read_lock_killable | -down_read_trylock +mmap_read_trylock | -up_read +mmap_read_unlock ) -(&mm->mmap_sem) +(mm) Signed-off-by: Michel Lespinasse <walken@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com> Reviewed-by: Laurent Dufour <ldufour@linux.ibm.com> Reviewed-by: Vlastimil Babka <vbabka@suse.cz> Cc: Davidlohr Bueso <dbueso@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Hugh Dickins <hughd@google.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Hubbard <jhubbard@nvidia.com> Cc: Liam Howlett <Liam.Howlett@oracle.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ying Han <yinghan@google.com> Link: http://lkml.kernel.org/r/20200520052908.204642-5-walken@google.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-09 04:33:25 +00:00
mmap_read_lock(mm);
ret = fixup_user_fault(mm, (unsigned long)uaddr,
mm: bring in additional flag for fixup_user_fault to signal unlock During Jason's work with postcopy migration support for s390 a problem regarding gmap faults was discovered. The gmap code will call fixup_user_fault which will end up always in handle_mm_fault. Till now we never cared about retries, but as the userfaultfd code kind of relies on it. this needs some fix. This patchset does not take care of the futex code. I will now look closer at this. This patch (of 2): With the introduction of userfaultfd, kvm on s390 needs fixup_user_fault to pass in FAULT_FLAG_ALLOW_RETRY and give feedback if during the faulting we ever unlocked mmap_sem. This patch brings in the logic to handle retries as well as it cleans up the current documentation. fixup_user_fault was not having the same semantics as filemap_fault. It never indicated if a retry happened and so a caller wasn't able to handle that case. So we now changed the behaviour to always retry a locked mmap_sem. Signed-off-by: Dominik Dingel <dingel@linux.vnet.ibm.com> Reviewed-by: Andrea Arcangeli <aarcange@redhat.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Christian Borntraeger <borntraeger@de.ibm.com> Cc: "Jason J. Herne" <jjherne@linux.vnet.ibm.com> Cc: David Rientjes <rientjes@google.com> Cc: Eric B Munson <emunson@akamai.com> Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com> Cc: Mel Gorman <mgorman@suse.de> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Dominik Dingel <dingel@linux.vnet.ibm.com> Cc: Paolo Bonzini <pbonzini@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-16 00:57:04 +00:00
FAULT_FLAG_WRITE, NULL);
mmap locking API: use coccinelle to convert mmap_sem rwsem call sites This change converts the existing mmap_sem rwsem calls to use the new mmap locking API instead. The change is generated using coccinelle with the following rule: // spatch --sp-file mmap_lock_api.cocci --in-place --include-headers --dir . @@ expression mm; @@ ( -init_rwsem +mmap_init_lock | -down_write +mmap_write_lock | -down_write_killable +mmap_write_lock_killable | -down_write_trylock +mmap_write_trylock | -up_write +mmap_write_unlock | -downgrade_write +mmap_write_downgrade | -down_read +mmap_read_lock | -down_read_killable +mmap_read_lock_killable | -down_read_trylock +mmap_read_trylock | -up_read +mmap_read_unlock ) -(&mm->mmap_sem) +(mm) Signed-off-by: Michel Lespinasse <walken@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Reviewed-by: Daniel Jordan <daniel.m.jordan@oracle.com> Reviewed-by: Laurent Dufour <ldufour@linux.ibm.com> Reviewed-by: Vlastimil Babka <vbabka@suse.cz> Cc: Davidlohr Bueso <dbueso@suse.de> Cc: David Rientjes <rientjes@google.com> Cc: Hugh Dickins <hughd@google.com> Cc: Jason Gunthorpe <jgg@ziepe.ca> Cc: Jerome Glisse <jglisse@redhat.com> Cc: John Hubbard <jhubbard@nvidia.com> Cc: Liam Howlett <Liam.Howlett@oracle.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ying Han <yinghan@google.com> Link: http://lkml.kernel.org/r/20200520052908.204642-5-walken@google.com Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-06-09 04:33:25 +00:00
mmap_read_unlock(mm);
return ret < 0 ? ret : 0;
}
/**
* futex_top_waiter() - Return the highest priority waiter on a futex
* @hb: the hash bucket the futex_q's reside in
* @key: the futex key (to distinguish it from other futex futex_q's)
*
* Must be called with the hb lock held.
*/
static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb,
union futex_key *key)
{
struct futex_q *this;
plist_for_each_entry(this, &hb->chain, list) {
if (match_futex(&this->key, key))
return this;
}
return NULL;
}
static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr,
u32 uval, u32 newval)
{
int ret;
pagefault_disable();
ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval);
pagefault_enable();
return ret;
}
static int get_futex_value_locked(u32 *dest, u32 __user *from)
{
int ret;
pagefault_disable();
ret = __get_user(*dest, from);
pagefault_enable();
return ret ? -EFAULT : 0;
}
/*
* PI code:
*/
static int refill_pi_state_cache(void)
{
struct futex_pi_state *pi_state;
if (likely(current->pi_state_cache))
return 0;
pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL);
if (!pi_state)
return -ENOMEM;
INIT_LIST_HEAD(&pi_state->list);
/* pi_mutex gets initialized later */
pi_state->owner = NULL;
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
refcount_set(&pi_state->refcount, 1);
pi_state->key = FUTEX_KEY_INIT;
current->pi_state_cache = pi_state;
return 0;
}
static struct futex_pi_state *alloc_pi_state(void)
{
struct futex_pi_state *pi_state = current->pi_state_cache;
WARN_ON(!pi_state);
current->pi_state_cache = NULL;
return pi_state;
}
static void get_pi_state(struct futex_pi_state *pi_state)
{
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
WARN_ON_ONCE(!refcount_inc_not_zero(&pi_state->refcount));
}
/*
* Drops a reference to the pi_state object and frees or caches it
* when the last reference is gone.
*/
static void put_pi_state(struct futex_pi_state *pi_state)
{
if (!pi_state)
return;
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
if (!refcount_dec_and_test(&pi_state->refcount))
return;
/*
* If pi_state->owner is NULL, the owner is most probably dying
* and has cleaned up the pi_state already
*/
if (pi_state->owner) {
struct task_struct *owner;
unsigned long flags;
raw_spin_lock_irqsave(&pi_state->pi_mutex.wait_lock, flags);
owner = pi_state->owner;
if (owner) {
raw_spin_lock(&owner->pi_lock);
list_del_init(&pi_state->list);
raw_spin_unlock(&owner->pi_lock);
}
rt_mutex_proxy_unlock(&pi_state->pi_mutex, owner);
raw_spin_unlock_irqrestore(&pi_state->pi_mutex.wait_lock, flags);
}
if (current->pi_state_cache) {
kfree(pi_state);
} else {
/*
* pi_state->list is already empty.
* clear pi_state->owner.
* refcount is at 0 - put it back to 1.
*/
pi_state->owner = NULL;
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
refcount_set(&pi_state->refcount, 1);
current->pi_state_cache = pi_state;
}
}
#ifdef CONFIG_FUTEX_PI
/*
* This task is holding PI mutexes at exit time => bad.
* Kernel cleans up PI-state, but userspace is likely hosed.
* (Robust-futex cleanup is separate and might save the day for userspace.)
*/
static void exit_pi_state_list(struct task_struct *curr)
{
struct list_head *next, *head = &curr->pi_state_list;
struct futex_pi_state *pi_state;
struct futex_hash_bucket *hb;
union futex_key key = FUTEX_KEY_INIT;
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
if (!futex_cmpxchg_enabled)
return;
/*
* We are a ZOMBIE and nobody can enqueue itself on
* pi_state_list anymore, but we have to be careful
* versus waiters unqueueing themselves:
*/
raw_spin_lock_irq(&curr->pi_lock);
while (!list_empty(head)) {
next = head->next;
pi_state = list_entry(next, struct futex_pi_state, list);
key = pi_state->key;
hb = hash_futex(&key);
futex: Fix more put_pi_state() vs. exit_pi_state_list() races Dmitry (through syzbot) reported being able to trigger the WARN in get_pi_state() and a use-after-free on: raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); Both are due to this race: exit_pi_state_list() put_pi_state() lock(&curr->pi_lock) while() { pi_state = list_first_entry(head); hb = hash_futex(&pi_state->key); unlock(&curr->pi_lock); dec_and_test(&pi_state->refcount); lock(&hb->lock) lock(&pi_state->pi_mutex.wait_lock) // uaf if pi_state free'd lock(&curr->pi_lock); .... unlock(&curr->pi_lock); get_pi_state(); // WARN; refcount==0 The problem is we take the reference count too late, and don't allow it being 0. Fix it by using inc_not_zero() and simply retrying the loop when we fail to get a refcount. In that case put_pi_state() should remove the entry from the list. Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Gratian Crisan <gratian.crisan@ni.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: dvhart@infradead.org Cc: syzbot <bot+2af19c9e1ffe4d4ee1d16c56ae7580feaee75765@syzkaller.appspotmail.com> Cc: syzkaller-bugs@googlegroups.com Cc: <stable@vger.kernel.org> Fixes: c74aef2d06a9 ("futex: Fix pi_state->owner serialization") Link: http://lkml.kernel.org/r/20171031101853.xpfh72y643kdfhjs@hirez.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-31 10:18:53 +00:00
/*
* We can race against put_pi_state() removing itself from the
* list (a waiter going away). put_pi_state() will first
* decrement the reference count and then modify the list, so
* its possible to see the list entry but fail this reference
* acquire.
*
* In that case; drop the locks to let put_pi_state() make
* progress and retry the loop.
*/
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
if (!refcount_inc_not_zero(&pi_state->refcount)) {
futex: Fix more put_pi_state() vs. exit_pi_state_list() races Dmitry (through syzbot) reported being able to trigger the WARN in get_pi_state() and a use-after-free on: raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); Both are due to this race: exit_pi_state_list() put_pi_state() lock(&curr->pi_lock) while() { pi_state = list_first_entry(head); hb = hash_futex(&pi_state->key); unlock(&curr->pi_lock); dec_and_test(&pi_state->refcount); lock(&hb->lock) lock(&pi_state->pi_mutex.wait_lock) // uaf if pi_state free'd lock(&curr->pi_lock); .... unlock(&curr->pi_lock); get_pi_state(); // WARN; refcount==0 The problem is we take the reference count too late, and don't allow it being 0. Fix it by using inc_not_zero() and simply retrying the loop when we fail to get a refcount. In that case put_pi_state() should remove the entry from the list. Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Gratian Crisan <gratian.crisan@ni.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: dvhart@infradead.org Cc: syzbot <bot+2af19c9e1ffe4d4ee1d16c56ae7580feaee75765@syzkaller.appspotmail.com> Cc: syzkaller-bugs@googlegroups.com Cc: <stable@vger.kernel.org> Fixes: c74aef2d06a9 ("futex: Fix pi_state->owner serialization") Link: http://lkml.kernel.org/r/20171031101853.xpfh72y643kdfhjs@hirez.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-31 10:18:53 +00:00
raw_spin_unlock_irq(&curr->pi_lock);
cpu_relax();
raw_spin_lock_irq(&curr->pi_lock);
continue;
}
raw_spin_unlock_irq(&curr->pi_lock);
spin_lock(&hb->lock);
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
raw_spin_lock(&curr->pi_lock);
/*
* We dropped the pi-lock, so re-check whether this
* task still owns the PI-state:
*/
if (head->next != next) {
futex: Fix more put_pi_state() vs. exit_pi_state_list() races Dmitry (through syzbot) reported being able to trigger the WARN in get_pi_state() and a use-after-free on: raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); Both are due to this race: exit_pi_state_list() put_pi_state() lock(&curr->pi_lock) while() { pi_state = list_first_entry(head); hb = hash_futex(&pi_state->key); unlock(&curr->pi_lock); dec_and_test(&pi_state->refcount); lock(&hb->lock) lock(&pi_state->pi_mutex.wait_lock) // uaf if pi_state free'd lock(&curr->pi_lock); .... unlock(&curr->pi_lock); get_pi_state(); // WARN; refcount==0 The problem is we take the reference count too late, and don't allow it being 0. Fix it by using inc_not_zero() and simply retrying the loop when we fail to get a refcount. In that case put_pi_state() should remove the entry from the list. Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Gratian Crisan <gratian.crisan@ni.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: dvhart@infradead.org Cc: syzbot <bot+2af19c9e1ffe4d4ee1d16c56ae7580feaee75765@syzkaller.appspotmail.com> Cc: syzkaller-bugs@googlegroups.com Cc: <stable@vger.kernel.org> Fixes: c74aef2d06a9 ("futex: Fix pi_state->owner serialization") Link: http://lkml.kernel.org/r/20171031101853.xpfh72y643kdfhjs@hirez.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-31 10:18:53 +00:00
/* retain curr->pi_lock for the loop invariant */
raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
spin_unlock(&hb->lock);
futex: Fix more put_pi_state() vs. exit_pi_state_list() races Dmitry (through syzbot) reported being able to trigger the WARN in get_pi_state() and a use-after-free on: raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); Both are due to this race: exit_pi_state_list() put_pi_state() lock(&curr->pi_lock) while() { pi_state = list_first_entry(head); hb = hash_futex(&pi_state->key); unlock(&curr->pi_lock); dec_and_test(&pi_state->refcount); lock(&hb->lock) lock(&pi_state->pi_mutex.wait_lock) // uaf if pi_state free'd lock(&curr->pi_lock); .... unlock(&curr->pi_lock); get_pi_state(); // WARN; refcount==0 The problem is we take the reference count too late, and don't allow it being 0. Fix it by using inc_not_zero() and simply retrying the loop when we fail to get a refcount. In that case put_pi_state() should remove the entry from the list. Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Gratian Crisan <gratian.crisan@ni.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: dvhart@infradead.org Cc: syzbot <bot+2af19c9e1ffe4d4ee1d16c56ae7580feaee75765@syzkaller.appspotmail.com> Cc: syzkaller-bugs@googlegroups.com Cc: <stable@vger.kernel.org> Fixes: c74aef2d06a9 ("futex: Fix pi_state->owner serialization") Link: http://lkml.kernel.org/r/20171031101853.xpfh72y643kdfhjs@hirez.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-31 10:18:53 +00:00
put_pi_state(pi_state);
continue;
}
WARN_ON(pi_state->owner != curr);
WARN_ON(list_empty(&pi_state->list));
list_del_init(&pi_state->list);
pi_state->owner = NULL;
futex: Fix more put_pi_state() vs. exit_pi_state_list() races Dmitry (through syzbot) reported being able to trigger the WARN in get_pi_state() and a use-after-free on: raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); Both are due to this race: exit_pi_state_list() put_pi_state() lock(&curr->pi_lock) while() { pi_state = list_first_entry(head); hb = hash_futex(&pi_state->key); unlock(&curr->pi_lock); dec_and_test(&pi_state->refcount); lock(&hb->lock) lock(&pi_state->pi_mutex.wait_lock) // uaf if pi_state free'd lock(&curr->pi_lock); .... unlock(&curr->pi_lock); get_pi_state(); // WARN; refcount==0 The problem is we take the reference count too late, and don't allow it being 0. Fix it by using inc_not_zero() and simply retrying the loop when we fail to get a refcount. In that case put_pi_state() should remove the entry from the list. Reported-by: Dmitry Vyukov <dvyukov@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Cc: Gratian Crisan <gratian.crisan@ni.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: dvhart@infradead.org Cc: syzbot <bot+2af19c9e1ffe4d4ee1d16c56ae7580feaee75765@syzkaller.appspotmail.com> Cc: syzkaller-bugs@googlegroups.com Cc: <stable@vger.kernel.org> Fixes: c74aef2d06a9 ("futex: Fix pi_state->owner serialization") Link: http://lkml.kernel.org/r/20171031101853.xpfh72y643kdfhjs@hirez.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2017-10-31 10:18:53 +00:00
raw_spin_unlock(&curr->pi_lock);
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
spin_unlock(&hb->lock);
rt_mutex_futex_unlock(&pi_state->pi_mutex);
put_pi_state(pi_state);
raw_spin_lock_irq(&curr->pi_lock);
}
raw_spin_unlock_irq(&curr->pi_lock);
}
#else
static inline void exit_pi_state_list(struct task_struct *curr) { }
#endif
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
/*
* We need to check the following states:
*
* Waiter | pi_state | pi->owner | uTID | uODIED | ?
*
* [1] NULL | --- | --- | 0 | 0/1 | Valid
* [2] NULL | --- | --- | >0 | 0/1 | Valid
*
* [3] Found | NULL | -- | Any | 0/1 | Invalid
*
* [4] Found | Found | NULL | 0 | 1 | Valid
* [5] Found | Found | NULL | >0 | 1 | Invalid
*
* [6] Found | Found | task | 0 | 1 | Valid
*
* [7] Found | Found | NULL | Any | 0 | Invalid
*
* [8] Found | Found | task | ==taskTID | 0/1 | Valid
* [9] Found | Found | task | 0 | 0 | Invalid
* [10] Found | Found | task | !=taskTID | 0/1 | Invalid
*
* [1] Indicates that the kernel can acquire the futex atomically. We
* came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit.
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
*
* [2] Valid, if TID does not belong to a kernel thread. If no matching
* thread is found then it indicates that the owner TID has died.
*
* [3] Invalid. The waiter is queued on a non PI futex
*
* [4] Valid state after exit_robust_list(), which sets the user space
* value to FUTEX_WAITERS | FUTEX_OWNER_DIED.
*
* [5] The user space value got manipulated between exit_robust_list()
* and exit_pi_state_list()
*
* [6] Valid state after exit_pi_state_list() which sets the new owner in
* the pi_state but cannot access the user space value.
*
* [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set.
*
* [8] Owner and user space value match
*
* [9] There is no transient state which sets the user space TID to 0
* except exit_robust_list(), but this is indicated by the
* FUTEX_OWNER_DIED bit. See [4]
*
* [10] There is no transient state which leaves owner and user space
* TID out of sync.
*
*
* Serialization and lifetime rules:
*
* hb->lock:
*
* hb -> futex_q, relation
* futex_q -> pi_state, relation
*
* (cannot be raw because hb can contain arbitrary amount
* of futex_q's)
*
* pi_mutex->wait_lock:
*
* {uval, pi_state}
*
* (and pi_mutex 'obviously')
*
* p->pi_lock:
*
* p->pi_state_list -> pi_state->list, relation
*
* pi_state->refcount:
*
* pi_state lifetime
*
*
* Lock order:
*
* hb->lock
* pi_mutex->wait_lock
* p->pi_lock
*
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
*/
/*
* Validate that the existing waiter has a pi_state and sanity check
* the pi_state against the user space value. If correct, attach to
* it.
*/
static int attach_to_pi_state(u32 __user *uaddr, u32 uval,
struct futex_pi_state *pi_state,
struct futex_pi_state **ps)
{
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
pid_t pid = uval & FUTEX_TID_MASK;
u32 uval2;
int ret;
/*
* Userspace might have messed up non-PI and PI futexes [3]
*/
if (unlikely(!pi_state))
return -EINVAL;
/*
* We get here with hb->lock held, and having found a
* futex_top_waiter(). This means that futex_lock_pi() of said futex_q
* has dropped the hb->lock in between queue_me() and unqueue_me_pi(),
* which in turn means that futex_lock_pi() still has a reference on
* our pi_state.
*
* The waiter holding a reference on @pi_state also protects against
* the unlocked put_pi_state() in futex_unlock_pi(), futex_lock_pi()
* and futex_wait_requeue_pi() as it cannot go to 0 and consequently
* free pi_state before we can take a reference ourselves.
*/
futex: Convert futex_pi_state.refcount to refcount_t atomic_t variables are currently used to implement reference counters with the following properties: - counter is initialized to 1 using atomic_set() - a resource is freed upon counter reaching zero - once counter reaches zero, its further increments aren't allowed - counter schema uses basic atomic operations (set, inc, inc_not_zero, dec_and_test, etc.) Such atomic variables should be converted to a newly provided refcount_t type and API that prevents accidental counter overflows and underflows. This is important since overflows and underflows can lead to use-after-free situation and be exploitable. The variable futex_pi_state.refcount is used as pure reference counter. Convert it to refcount_t and fix up the operations. **Important note for maintainers: Some functions from refcount_t API defined in lib/refcount.c have different memory ordering guarantees than their atomic counterparts. Please check Documentation/core-api/refcount-vs-atomic.rst for more information. Normally the differences should not matter since refcount_t provides enough guarantees to satisfy the refcounting use cases, but in some rare cases it might matter. Please double check that you don't have some undocumented memory guarantees for this variable usage. For the futex_pi_state.refcount it might make a difference in following places: - get_pi_state() and exit_pi_state_list(): increment in refcount_inc_not_zero() only guarantees control dependency on success vs. fully ordered atomic counterpart - put_pi_state(): decrement in refcount_dec_and_test() provides RELEASE ordering and ACQUIRE ordering on success vs. fully ordered atomic counterpart Suggested-by: Kees Cook <keescook@chromium.org> Signed-off-by: Elena Reshetova <elena.reshetova@intel.com> Reviewed-by: David Windsor <dwindsor@gmail.com> Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Will Deacon <will.deacon@arm.com> Cc: dvhart@infradead.org Link: http://lkml.kernel.org/r/1549369467-3505-1-git-send-email-elena.reshetova@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2019-02-05 12:24:27 +00:00
WARN_ON(!refcount_read(&pi_state->refcount));
/*
* Now that we have a pi_state, we can acquire wait_lock
* and do the state validation.
*/
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
/*
* Since {uval, pi_state} is serialized by wait_lock, and our current
* uval was read without holding it, it can have changed. Verify it
* still is what we expect it to be, otherwise retry the entire
* operation.
*/
if (get_futex_value_locked(&uval2, uaddr))
goto out_efault;
if (uval != uval2)
goto out_eagain;
/*
* Handle the owner died case:
*/
if (uval & FUTEX_OWNER_DIED) {
/*
* exit_pi_state_list sets owner to NULL and wakes the
* topmost waiter. The task which acquires the
* pi_state->rt_mutex will fixup owner.
*/
if (!pi_state->owner) {
/*
* No pi state owner, but the user space TID
* is not 0. Inconsistent state. [5]
*/
if (pid)
goto out_einval;
/*
* Take a ref on the state and return success. [4]
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
*/
goto out_attach;
}
/*
* If TID is 0, then either the dying owner has not
* yet executed exit_pi_state_list() or some waiter
* acquired the rtmutex in the pi state, but did not
* yet fixup the TID in user space.
*
* Take a ref on the state and return success. [6]
*/
if (!pid)
goto out_attach;
} else {
/*
* If the owner died bit is not set, then the pi_state
* must have an owner. [7]
*/
if (!pi_state->owner)
goto out_einval;
}
/*
* Bail out if user space manipulated the futex value. If pi
* state exists then the owner TID must be the same as the
* user space TID. [9/10]
*/
if (pid != task_pid_vnr(pi_state->owner))
goto out_einval;
out_attach:
get_pi_state(pi_state);
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
*ps = pi_state;
return 0;
out_einval:
ret = -EINVAL;
goto out_error;
out_eagain:
ret = -EAGAIN;
goto out_error;
out_efault:
ret = -EFAULT;
goto out_error;
out_error:
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
return ret;
}
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
/**
* wait_for_owner_exiting - Block until the owner has exited
* @ret: owner's current futex lock status
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
* @exiting: Pointer to the exiting task
*
* Caller must hold a refcount on @exiting.
*/
static void wait_for_owner_exiting(int ret, struct task_struct *exiting)
{
if (ret != -EBUSY) {
WARN_ON_ONCE(exiting);
return;
}
if (WARN_ON_ONCE(ret == -EBUSY && !exiting))
return;
mutex_lock(&exiting->futex_exit_mutex);
/*
* No point in doing state checking here. If the waiter got here
* while the task was in exec()->exec_futex_release() then it can
* have any FUTEX_STATE_* value when the waiter has acquired the
* mutex. OK, if running, EXITING or DEAD if it reached exit()
* already. Highly unlikely and not a problem. Just one more round
* through the futex maze.
*/
mutex_unlock(&exiting->futex_exit_mutex);
put_task_struct(exiting);
}
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
static int handle_exit_race(u32 __user *uaddr, u32 uval,
struct task_struct *tsk)
{
u32 uval2;
/*
* If the futex exit state is not yet FUTEX_STATE_DEAD, tell the
* caller that the alleged owner is busy.
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
*/
if (tsk && tsk->futex_state != FUTEX_STATE_DEAD)
return -EBUSY;
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
/*
* Reread the user space value to handle the following situation:
*
* CPU0 CPU1
*
* sys_exit() sys_futex()
* do_exit() futex_lock_pi()
* futex_lock_pi_atomic()
* exit_signals(tsk) No waiters:
* tsk->flags |= PF_EXITING; *uaddr == 0x00000PID
* mm_release(tsk) Set waiter bit
* exit_robust_list(tsk) { *uaddr = 0x80000PID;
* Set owner died attach_to_pi_owner() {
* *uaddr = 0xC0000000; tsk = get_task(PID);
* } if (!tsk->flags & PF_EXITING) {
* ... attach();
* tsk->futex_state = } else {
* FUTEX_STATE_DEAD; if (tsk->futex_state !=
* FUTEX_STATE_DEAD)
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
* return -EAGAIN;
* return -ESRCH; <--- FAIL
* }
*
* Returning ESRCH unconditionally is wrong here because the
* user space value has been changed by the exiting task.
*
* The same logic applies to the case where the exiting task is
* already gone.
*/
if (get_futex_value_locked(&uval2, uaddr))
return -EFAULT;
/* If the user space value has changed, try again. */
if (uval2 != uval)
return -EAGAIN;
/*
* The exiting task did not have a robust list, the robust list was
* corrupted or the user space value in *uaddr is simply bogus.
* Give up and tell user space.
*/
return -ESRCH;
}
/*
* Lookup the task for the TID provided from user space and attach to
* it after doing proper sanity checks.
*/
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
static int attach_to_pi_owner(u32 __user *uaddr, u32 uval, union futex_key *key,
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
struct futex_pi_state **ps,
struct task_struct **exiting)
{
pid_t pid = uval & FUTEX_TID_MASK;
struct futex_pi_state *pi_state;
struct task_struct *p;
/*
* We are the first waiter - try to look up the real owner and attach
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
* the new pi_state to it, but bail out when TID = 0 [1]
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
*
* The !pid check is paranoid. None of the call sites should end up
* with pid == 0, but better safe than sorry. Let the caller retry
*/
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
if (!pid)
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
return -EAGAIN;
p = find_get_task_by_vpid(pid);
futex: futex_find_get_task remove credentails check futex_find_get_task is currently used (through lookup_pi_state) from two contexts, futex_requeue and futex_lock_pi_atomic. None of the paths looks it needs the credentials check, though. Different (e)uids shouldn't matter at all because the only thing that is important for shared futex is the accessibility of the shared memory. The credentail check results in glibc assert failure or process hang (if glibc is compiled without assert support) for shared robust pthread mutex with priority inheritance if a process tries to lock already held lock owned by a process with a different euid: pthread_mutex_lock.c:312: __pthread_mutex_lock_full: Assertion `(-(e)) != 3 || !robust' failed. The problem is that futex_lock_pi_atomic which is called when we try to lock already held lock checks the current holder (tid is stored in the futex value) to get the PI state. It uses lookup_pi_state which in turn gets task struct from futex_find_get_task. ESRCH is returned either when the task is not found or if credentials check fails. futex_lock_pi_atomic simply returns if it gets ESRCH. glibc code, however, doesn't expect that robust lock returns with ESRCH because it should get either success or owner died. Signed-off-by: Michal Hocko <mhocko@suse.cz> Acked-by: Darren Hart <dvhltc@us.ibm.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Nick Piggin <npiggin@suse.de> Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru> Cc: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-06-30 07:51:19 +00:00
if (!p)
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
return handle_exit_race(uaddr, uval, NULL);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
if (unlikely(p->flags & PF_KTHREAD)) {
put_task_struct(p);
return -EPERM;
}
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
/*
* We need to look at the task state to figure out, whether the
* task is exiting. To protect against the change of the task state
* in futex_exit_release(), we do this protected by p->pi_lock:
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
*/
raw_spin_lock_irq(&p->pi_lock);
if (unlikely(p->futex_state != FUTEX_STATE_OK)) {
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
/*
* The task is on the way out. When the futex state is
* FUTEX_STATE_DEAD, we know that the task has finished
* the cleanup:
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
*/
futex: Cure exit race Stefan reported, that the glibc tst-robustpi4 test case fails occasionally. That case creates the following race between sys_exit() and sys_futex_lock_pi(): CPU0 CPU1 sys_exit() sys_futex() do_exit() futex_lock_pi() exit_signals(tsk) No waiters: tsk->flags |= PF_EXITING; *uaddr == 0x00000PID mm_release(tsk) Set waiter bit exit_robust_list(tsk) { *uaddr = 0x80000PID; Set owner died attach_to_pi_owner() { *uaddr = 0xC0000000; tsk = get_task(PID); } if (!tsk->flags & PF_EXITING) { ... attach(); tsk->flags |= PF_EXITPIDONE; } else { if (!(tsk->flags & PF_EXITPIDONE)) return -EAGAIN; return -ESRCH; <--- FAIL } ESRCH is returned all the way to user space, which triggers the glibc test case assert. Returning ESRCH unconditionally is wrong here because the user space value has been changed by the exiting task to 0xC0000000, i.e. the FUTEX_OWNER_DIED bit is set and the futex PID value has been cleared. This is a valid state and the kernel has to handle it, i.e. taking the futex. Cure it by rereading the user space value when PF_EXITING and PF_EXITPIDONE is set in the task which 'owns' the futex. If the value has changed, let the kernel retry the operation, which includes all regular sanity checks and correctly handles the FUTEX_OWNER_DIED case. If it hasn't changed, then return ESRCH as there is no way to distinguish this case from malfunctioning user space. This happens when the exiting task did not have a robust list, the robust list was corrupted or the user space value in the futex was simply bogus. Reported-by: Stefan Liebler <stli@linux.ibm.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Peter Zijlstra <peterz@infradead.org> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Darren Hart <dvhart@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sasha Levin <sashal@kernel.org> Cc: stable@vger.kernel.org Link: https://bugzilla.kernel.org/show_bug.cgi?id=200467 Link: https://lkml.kernel.org/r/20181210152311.986181245@linutronix.de
2018-12-10 13:35:14 +00:00
int ret = handle_exit_race(uaddr, uval, p);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
raw_spin_unlock_irq(&p->pi_lock);
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
/*
* If the owner task is between FUTEX_STATE_EXITING and
* FUTEX_STATE_DEAD then store the task pointer and keep
* the reference on the task struct. The calling code will
* drop all locks, wait for the task to reach
* FUTEX_STATE_DEAD and then drop the refcount. This is
* required to prevent a live lock when the current task
* preempted the exiting task between the two states.
*/
if (ret == -EBUSY)
*exiting = p;
else
put_task_struct(p);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
return ret;
}
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
/*
* No existing pi state. First waiter. [2]
*
* This creates pi_state, we have hb->lock held, this means nothing can
* observe this state, wait_lock is irrelevant.
futex: Make lookup_pi_state more robust The current implementation of lookup_pi_state has ambigous handling of the TID value 0 in the user space futex. We can get into the kernel even if the TID value is 0, because either there is a stale waiters bit or the owner died bit is set or we are called from the requeue_pi path or from user space just for fun. The current code avoids an explicit sanity check for pid = 0 in case that kernel internal state (waiters) are found for the user space address. This can lead to state leakage and worse under some circumstances. Handle the cases explicit: Waiter | pi_state | pi->owner | uTID | uODIED | ? [1] NULL | --- | --- | 0 | 0/1 | Valid [2] NULL | --- | --- | >0 | 0/1 | Valid [3] Found | NULL | -- | Any | 0/1 | Invalid [4] Found | Found | NULL | 0 | 1 | Valid [5] Found | Found | NULL | >0 | 1 | Invalid [6] Found | Found | task | 0 | 1 | Valid [7] Found | Found | NULL | Any | 0 | Invalid [8] Found | Found | task | ==taskTID | 0/1 | Valid [9] Found | Found | task | 0 | 0 | Invalid [10] Found | Found | task | !=taskTID | 0/1 | Invalid [1] Indicates that the kernel can acquire the futex atomically. We came came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. [2] Valid, if TID does not belong to a kernel thread. If no matching thread is found then it indicates that the owner TID has died. [3] Invalid. The waiter is queued on a non PI futex [4] Valid state after exit_robust_list(), which sets the user space value to FUTEX_WAITERS | FUTEX_OWNER_DIED. [5] The user space value got manipulated between exit_robust_list() and exit_pi_state_list() [6] Valid state after exit_pi_state_list() which sets the new owner in the pi_state but cannot access the user space value. [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. [8] Owner and user space value match [9] There is no transient state which sets the user space TID to 0 except exit_robust_list(), but this is indicated by the FUTEX_OWNER_DIED bit. See [4] [10] There is no transient state which leaves owner and user space TID out of sync. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Kees Cook <keescook@chromium.org> Cc: Will Drewry <wad@chromium.org> Cc: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-03 12:27:08 +00:00
*/
pi_state = alloc_pi_state();
/*
* Initialize the pi_mutex in locked state and make @p
* the owner of it:
*/
rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p);
/* Store the key for possible exit cleanups: */
pi_state->key = *key;
WARN_ON(!list_empty(&pi_state->list));
list_add(&pi_state->list, &p->pi_state_list);
/*
* Assignment without holding pi_state->pi_mutex.wait_lock is safe
* because there is no concurrency as the object is not published yet.
*/
pi_state->owner = p;
raw_spin_unlock_irq(&p->pi_lock);
put_task_struct(p);
*ps = pi_state;
return 0;
}
static int lookup_pi_state(u32 __user *uaddr, u32 uval,
struct futex_hash_bucket *hb,
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
union futex_key *key, struct futex_pi_state **ps,
struct task_struct **exiting)
{
struct futex_q *top_waiter = futex_top_waiter(hb, key);
/*
* If there is a waiter on that futex, validate it and
* attach to the pi_state when the validation succeeds.
*/
if (top_waiter)
return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps);
/*
* We are the first waiter - try to look up the owner based on
* @uval and attach to it.
*/
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
return attach_to_pi_owner(uaddr, uval, key, ps, exiting);
}
static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval)
{
int err;
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 curval;
if (unlikely(should_fail_futex(true)))
return -EFAULT;
err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval);
if (unlikely(err))
return err;
/* If user space value changed, let the caller retry */
return curval != uval ? -EAGAIN : 0;
}
/**
* futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex
* @uaddr: the pi futex user address
* @hb: the pi futex hash bucket
* @key: the futex key associated with uaddr and hb
* @ps: the pi_state pointer where we store the result of the
* lookup
* @task: the task to perform the atomic lock work for. This will
* be "current" except in the case of requeue pi.
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
* @exiting: Pointer to store the task pointer of the owner task
* which is in the middle of exiting
* @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
*
* Return:
* - 0 - ready to wait;
* - 1 - acquired the lock;
* - <0 - error
*
* The hb->lock and futex_key refs shall be held by the caller.
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
*
* @exiting is only set when the return value is -EBUSY. If so, this holds
* a refcount on the exiting task on return and the caller needs to drop it
* after waiting for the exit to complete.
*/
static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb,
union futex_key *key,
struct futex_pi_state **ps,
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
struct task_struct *task,
struct task_struct **exiting,
int set_waiters)
{
u32 uval, newval, vpid = task_pid_vnr(task);
struct futex_q *top_waiter;
int ret;
/*
* Read the user space value first so we can validate a few
* things before proceeding further.
*/
if (get_futex_value_locked(&uval, uaddr))
return -EFAULT;
if (unlikely(should_fail_futex(true)))
return -EFAULT;
/*
* Detect deadlocks.
*/
if ((unlikely((uval & FUTEX_TID_MASK) == vpid)))
return -EDEADLK;
if ((unlikely(should_fail_futex(true))))
return -EDEADLK;
/*
* Lookup existing state first. If it exists, try to attach to
* its pi_state.
*/
top_waiter = futex_top_waiter(hb, key);
if (top_waiter)
return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps);
/*
* No waiter and user TID is 0. We are here because the
* waiters or the owner died bit is set or called from
* requeue_cmp_pi or for whatever reason something took the
* syscall.
*/
if (!(uval & FUTEX_TID_MASK)) {
futex: Handle futex_pi OWNER_DIED take over correctly Siddhesh analyzed a failure in the take over of pi futexes in case the owner died and provided a workaround. See: http://sourceware.org/bugzilla/show_bug.cgi?id=14076 The detailed problem analysis shows: Futex F is initialized with PTHREAD_PRIO_INHERIT and PTHREAD_MUTEX_ROBUST_NP attributes. T1 lock_futex_pi(F); T2 lock_futex_pi(F); --> T2 blocks on the futex and creates pi_state which is associated to T1. T1 exits --> exit_robust_list() runs --> Futex F userspace value TID field is set to 0 and FUTEX_OWNER_DIED bit is set. T3 lock_futex_pi(F); --> Succeeds due to the check for F's userspace TID field == 0 --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space T1 --> exit_pi_state_list() --> Transfers pi_state to waiter T2 and wakes T2 via rt_mutex_unlock(&pi_state->mutex) T2 --> acquires pi_state->mutex and gains real ownership of the pi_state --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space T3 --> observes inconsistent state This problem is independent of UP/SMP, preemptible/non preemptible kernels, or process shared vs. private. The only difference is that certain configurations are more likely to expose it. So as Siddhesh correctly analyzed the following check in futex_lock_pi_atomic() is the culprit: if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { We check the userspace value for a TID value of 0 and take over the futex unconditionally if that's true. AFAICT this check is there as it is correct for a different corner case of futexes: the WAITERS bit became stale. Now the proposed change - if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { + if (unlikely(ownerdied || + !(curval & (FUTEX_TID_MASK | FUTEX_WAITERS)))) { solves the problem, but it's not obvious why and it wreckages the "stale WAITERS bit" case. What happens is, that due to the WAITERS bit being set (T2 is blocked on that futex) it enforces T3 to go through lookup_pi_state(), which in the above case returns an existing pi_state and therefor forces T3 to legitimately fight with T2 over the ownership of the pi_state (via pi_state->mutex). Probelm solved! Though that does not work for the "WAITERS bit is stale" problem because if lookup_pi_state() does not find existing pi_state it returns -ERSCH (due to TID == 0) which causes futex_lock_pi() to return -ESRCH to user space because the OWNER_DIED bit is not set. Now there is a different solution to that problem. Do not look at the user space value at all and enforce a lookup of possibly available pi_state. If pi_state can be found, then the new incoming locker T3 blocks on that pi_state and legitimately races with T2 to acquire the rt_mutex and the pi_state and therefor the proper ownership of the user space futex. lookup_pi_state() has the correct order of checks. It first tries to find a pi_state associated with the user space futex and only if that fails it checks for futex TID value = 0. If no pi_state is available nothing can create new state at that point because this happens with the hash bucket lock held. So the above scenario changes to: T1 lock_futex_pi(F); T2 lock_futex_pi(F); --> T2 blocks on the futex and creates pi_state which is associated to T1. T1 exits --> exit_robust_list() runs --> Futex F userspace value TID field is set to 0 and FUTEX_OWNER_DIED bit is set. T3 lock_futex_pi(F); --> Finds pi_state and blocks on pi_state->rt_mutex T1 --> exit_pi_state_list() --> Transfers pi_state to waiter T2 and wakes it via rt_mutex_unlock(&pi_state->mutex) T2 --> acquires pi_state->mutex and gains ownership of the pi_state --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space This covers all gazillion points on which T3 might come in between T1's exit_robust_list() clearing the TID field and T2 fixing it up. It also solves the "WAITERS bit stale" problem by forcing the take over. Another benefit of changing the code this way is that it makes it less dependent on untrusted user space values and therefor minimizes the possible wreckage which might be inflicted. As usual after staring for too long at the futex code my brain hurts so much that I really want to ditch that whole optimization of avoiding the syscall for the non contended case for PI futexes and rip out the maze of corner case handling code. Unfortunately we can't as user space relies on that existing behaviour, but at least thinking about it helps me to preserve my mental sanity. Maybe we should nevertheless :) Reported-and-tested-by: Siddhesh Poyarekar <siddhesh.poyarekar@gmail.com> Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1210232138540.2756@ionos Acked-by: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2012-10-23 20:29:38 +00:00
/*
* We take over the futex. No other waiters and the user space
* TID is 0. We preserve the owner died bit.
futex: Handle futex_pi OWNER_DIED take over correctly Siddhesh analyzed a failure in the take over of pi futexes in case the owner died and provided a workaround. See: http://sourceware.org/bugzilla/show_bug.cgi?id=14076 The detailed problem analysis shows: Futex F is initialized with PTHREAD_PRIO_INHERIT and PTHREAD_MUTEX_ROBUST_NP attributes. T1 lock_futex_pi(F); T2 lock_futex_pi(F); --> T2 blocks on the futex and creates pi_state which is associated to T1. T1 exits --> exit_robust_list() runs --> Futex F userspace value TID field is set to 0 and FUTEX_OWNER_DIED bit is set. T3 lock_futex_pi(F); --> Succeeds due to the check for F's userspace TID field == 0 --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space T1 --> exit_pi_state_list() --> Transfers pi_state to waiter T2 and wakes T2 via rt_mutex_unlock(&pi_state->mutex) T2 --> acquires pi_state->mutex and gains real ownership of the pi_state --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space T3 --> observes inconsistent state This problem is independent of UP/SMP, preemptible/non preemptible kernels, or process shared vs. private. The only difference is that certain configurations are more likely to expose it. So as Siddhesh correctly analyzed the following check in futex_lock_pi_atomic() is the culprit: if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { We check the userspace value for a TID value of 0 and take over the futex unconditionally if that's true. AFAICT this check is there as it is correct for a different corner case of futexes: the WAITERS bit became stale. Now the proposed change - if (unlikely(ownerdied || !(curval & FUTEX_TID_MASK))) { + if (unlikely(ownerdied || + !(curval & (FUTEX_TID_MASK | FUTEX_WAITERS)))) { solves the problem, but it's not obvious why and it wreckages the "stale WAITERS bit" case. What happens is, that due to the WAITERS bit being set (T2 is blocked on that futex) it enforces T3 to go through lookup_pi_state(), which in the above case returns an existing pi_state and therefor forces T3 to legitimately fight with T2 over the ownership of the pi_state (via pi_state->mutex). Probelm solved! Though that does not work for the "WAITERS bit is stale" problem because if lookup_pi_state() does not find existing pi_state it returns -ERSCH (due to TID == 0) which causes futex_lock_pi() to return -ESRCH to user space because the OWNER_DIED bit is not set. Now there is a different solution to that problem. Do not look at the user space value at all and enforce a lookup of possibly available pi_state. If pi_state can be found, then the new incoming locker T3 blocks on that pi_state and legitimately races with T2 to acquire the rt_mutex and the pi_state and therefor the proper ownership of the user space futex. lookup_pi_state() has the correct order of checks. It first tries to find a pi_state associated with the user space futex and only if that fails it checks for futex TID value = 0. If no pi_state is available nothing can create new state at that point because this happens with the hash bucket lock held. So the above scenario changes to: T1 lock_futex_pi(F); T2 lock_futex_pi(F); --> T2 blocks on the futex and creates pi_state which is associated to T1. T1 exits --> exit_robust_list() runs --> Futex F userspace value TID field is set to 0 and FUTEX_OWNER_DIED bit is set. T3 lock_futex_pi(F); --> Finds pi_state and blocks on pi_state->rt_mutex T1 --> exit_pi_state_list() --> Transfers pi_state to waiter T2 and wakes it via rt_mutex_unlock(&pi_state->mutex) T2 --> acquires pi_state->mutex and gains ownership of the pi_state --> Claims ownership of the futex and sets its own TID into the userspace TID field of futex F --> returns to user space This covers all gazillion points on which T3 might come in between T1's exit_robust_list() clearing the TID field and T2 fixing it up. It also solves the "WAITERS bit stale" problem by forcing the take over. Another benefit of changing the code this way is that it makes it less dependent on untrusted user space values and therefor minimizes the possible wreckage which might be inflicted. As usual after staring for too long at the futex code my brain hurts so much that I really want to ditch that whole optimization of avoiding the syscall for the non contended case for PI futexes and rip out the maze of corner case handling code. Unfortunately we can't as user space relies on that existing behaviour, but at least thinking about it helps me to preserve my mental sanity. Maybe we should nevertheless :) Reported-and-tested-by: Siddhesh Poyarekar <siddhesh.poyarekar@gmail.com> Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1210232138540.2756@ionos Acked-by: Darren Hart <dvhart@linux.intel.com> Cc: stable@vger.kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2012-10-23 20:29:38 +00:00
*/
newval = uval & FUTEX_OWNER_DIED;
newval |= vpid;
/* The futex requeue_pi code can enforce the waiters bit */
if (set_waiters)
newval |= FUTEX_WAITERS;
ret = lock_pi_update_atomic(uaddr, uval, newval);
/* If the take over worked, return 1 */
return ret < 0 ? ret : 1;
}
/*
* First waiter. Set the waiters bit before attaching ourself to
* the owner. If owner tries to unlock, it will be forced into
* the kernel and blocked on hb->lock.
*/
newval = uval | FUTEX_WAITERS;
ret = lock_pi_update_atomic(uaddr, uval, newval);
if (ret)
return ret;
/*
* If the update of the user space value succeeded, we try to
* attach to the owner. If that fails, no harm done, we only
* set the FUTEX_WAITERS bit in the user space variable.
*/
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
return attach_to_pi_owner(uaddr, newval, key, ps, exiting);
}
/**
* __unqueue_futex() - Remove the futex_q from its futex_hash_bucket
* @q: The futex_q to unqueue
*
* The q->lock_ptr must not be NULL and must be held by the caller.
*/
static void __unqueue_futex(struct futex_q *q)
{
struct futex_hash_bucket *hb;
if (WARN_ON_SMP(!q->lock_ptr) || WARN_ON(plist_node_empty(&q->list)))
return;
lockdep_assert_held(q->lock_ptr);
hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock);
plist_del(&q->list, &hb->chain);
hb_waiters_dec(hb);
}
/*
* The hash bucket lock must be held when this is called.
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
* Afterwards, the futex_q must not be accessed. Callers
* must ensure to later call wake_up_q() for the actual
* wakeups to occur.
*/
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q)
{
struct task_struct *p = q->task;
if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n"))
return;
get_task_struct(p);
__unqueue_futex(q);
/*
* The waiting task can free the futex_q as soon as q->lock_ptr = NULL
* is written, without taking any locks. This is possible in the event
* of a spurious wakeup, for example. A memory barrier is required here
* to prevent the following store to lock_ptr from getting ahead of the
* plist_del in __unqueue_futex().
*/
smp_store_release(&q->lock_ptr, NULL);
/*
* Queue the task for later wakeup for after we've released
* the hb->lock.
*/
sched/wake_q: Reduce reference counting for special users Some users, specifically futexes and rwsems, required fixes that allowed the callers to be safe when wakeups occur before they are expected by wake_up_q(). Such scenarios also play games and rely on reference counting, and until now were pivoting on wake_q doing it. With the wake_q_add() call being moved down, this can no longer be the case. As such we end up with a a double task refcounting overhead; and these callers care enough about this (being rather core-ish). This patch introduces a wake_q_add_safe() call that serves for callers that have already done refcounting and therefore the task is 'safe' from wake_q point of view (int that it requires reference throughout the entire queue/>wakeup cycle). In the one case it has internal reference counting, in the other case it consumes the reference counting. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Waiman Long <longman@redhat.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Xie Yongji <xieyongji@baidu.com> Cc: Yongji Xie <elohimes@gmail.com> Cc: andrea.parri@amarulasolutions.com Cc: lilin24@baidu.com Cc: liuqi16@baidu.com Cc: nixun@baidu.com Cc: yuanlinsi01@baidu.com Cc: zhangyu31@baidu.com Link: https://lkml.kernel.org/r/20181218195352.7orq3upiwfdbrdne@linux-r8p5 Signed-off-by: Ingo Molnar <mingo@kernel.org>
2018-12-18 19:53:52 +00:00
wake_q_add_safe(wake_q, p);
}
/*
* Caller must hold a reference on @pi_state.
*/
static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state)
{
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 curval, newval;
struct task_struct *new_owner;
bool postunlock = false;
DEFINE_WAKE_Q(wake_q);
int ret = 0;
new_owner = rt_mutex_next_owner(&pi_state->pi_mutex);
if (WARN_ON_ONCE(!new_owner)) {
/*
* As per the comment in futex_unlock_pi() this should not happen.
*
* When this happens, give up our locks and try again, giving
* the futex_lock_pi() instance time to complete, either by
* waiting on the rtmutex or removing itself from the futex
* queue.
*/
ret = -EAGAIN;
goto out_unlock;
futex: Rework inconsistent rt_mutex/futex_q state There is a weird state in the futex_unlock_pi() path when it interleaves with a concurrent futex_lock_pi() at the point where it drops hb->lock. In this case, it can happen that the rt_mutex wait_list and the futex_q disagree on pending waiters, in particular rt_mutex will find no pending waiters where futex_q thinks there are. In this case the rt_mutex unlock code cannot assign an owner. The futex side fixup code has to cleanup the inconsistencies with quite a bunch of interesting corner cases. Simplify all this by changing wake_futex_pi() to return -EAGAIN when this situation occurs. This then gives the futex_lock_pi() code the opportunity to continue and the retried futex_unlock_pi() will now observe a coherent state. The only problem is that this breaks RT timeliness guarantees. That is, consider the following scenario: T1 and T2 are both pinned to CPU0. prio(T2) > prio(T1) CPU0 T1 lock_pi() queue_me() <- Waiter is visible preemption T2 unlock_pi() loops with -EAGAIN forever Which is undesirable for PI primitives. Future patches will rectify this. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104151.850383690@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:54 +00:00
}
/*
* We pass it to the next owner. The WAITERS bit is always kept
* enabled while there is PI state around. We cleanup the owner
* died bit, because we are the owner.
*/
newval = FUTEX_WAITERS | task_pid_vnr(new_owner);
if (unlikely(should_fail_futex(true))) {
ret = -EFAULT;
goto out_unlock;
}
ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval);
if (!ret && (curval != uval)) {
/*
* If a unconditional UNLOCK_PI operation (user space did not
* try the TID->0 transition) raced with a waiter setting the
* FUTEX_WAITERS flag between get_user() and locking the hash
* bucket lock, retry the operation.
*/
if ((FUTEX_TID_MASK & curval) == uval)
ret = -EAGAIN;
else
ret = -EINVAL;
}
if (ret)
goto out_unlock;
/*
* This is a point of no return; once we modify the uval there is no
* going back and subsequent operations must not fail.
*/
raw_spin_lock(&pi_state->owner->pi_lock);
WARN_ON(list_empty(&pi_state->list));
list_del_init(&pi_state->list);
raw_spin_unlock(&pi_state->owner->pi_lock);
raw_spin_lock(&new_owner->pi_lock);
WARN_ON(!list_empty(&pi_state->list));
list_add(&pi_state->list, &new_owner->pi_state_list);
pi_state->owner = new_owner;
raw_spin_unlock(&new_owner->pi_lock);
postunlock = __rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q);
out_unlock:
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
if (postunlock)
rt_mutex_postunlock(&wake_q);
return ret;
}
/*
* Express the locking dependencies for lockdep:
*/
static inline void
double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
{
if (hb1 <= hb2) {
spin_lock(&hb1->lock);
if (hb1 < hb2)
spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING);
} else { /* hb1 > hb2 */
spin_lock(&hb2->lock);
spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING);
}
}
static inline void
double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
{
spin_unlock(&hb1->lock);
if (hb1 != hb2)
spin_unlock(&hb2->lock);
}
/*
* Wake up waiters matching bitset queued on this futex (uaddr).
*/
static int
futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset)
{
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
struct futex_hash_bucket *hb;
struct futex_q *this, *next;
union futex_key key = FUTEX_KEY_INIT;
int ret;
DEFINE_WAKE_Q(wake_q);
if (!bitset)
return -EINVAL;
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, FUTEX_READ);
if (unlikely(ret != 0))
return ret;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
hb = hash_futex(&key);
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
/* Make sure we really have tasks to wakeup */
if (!hb_waiters_pending(hb))
return ret;
futexes: Avoid taking the hb->lock if there's nothing to wake up In futex_wake() there is clearly no point in taking the hb->lock if we know beforehand that there are no tasks to be woken. While the hash bucket's plist head is a cheap way of knowing this, we cannot rely 100% on it as there is a racy window between the futex_wait call and when the task is actually added to the plist. To this end, we couple it with the spinlock check as tasks trying to enter the critical region are most likely potential waiters that will be added to the plist, thus preventing tasks sleeping forever if wakers don't acknowledge all possible waiters. Furthermore, the futex ordering guarantees are preserved, ensuring that waiters either observe the changed user space value before blocking or is woken by a concurrent waker. For wakers, this is done by relying on the barriers in get_futex_key_refs() -- for archs that do not have implicit mb in atomic_inc(), we explicitly add them through a new futex_get_mm function. For waiters we rely on the fact that spin_lock calls already update the head counter, so spinners are visible even if the lock hasn't been acquired yet. For more details please refer to the updated comments in the code and related discussion: https://lkml.org/lkml/2013/11/26/556 Special thanks to tglx for careful review and feedback. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Cc: Waiman Long <Waiman.Long@hp.com> Cc: Jason Low <jason.low2@hp.com> Cc: Andrew Morton <akpm@linux-foundation.org> Link: http://lkml.kernel.org/r/1389569486-25487-5-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:25 +00:00
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
spin_lock(&hb->lock);
plist_for_each_entry_safe(this, next, &hb->chain, list) {
if (match_futex (&this->key, &key)) {
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (this->pi_state || this->rt_waiter) {
ret = -EINVAL;
break;
}
/* Check if one of the bits is set in both bitsets */
if (!(this->bitset & bitset))
continue;
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
mark_wake_futex(&wake_q, this);
if (++ret >= nr_wake)
break;
}
}
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
spin_unlock(&hb->lock);
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
wake_up_q(&wake_q);
return ret;
}
futex: Remove duplicated code and fix undefined behaviour There is code duplicated over all architecture's headers for futex_atomic_op_inuser. Namely op decoding, access_ok check for uaddr, and comparison of the result. Remove this duplication and leave up to the arches only the needed assembly which is now in arch_futex_atomic_op_inuser. This effectively distributes the Will Deacon's arm64 fix for undefined behaviour reported by UBSAN to all architectures. The fix was done in commit 5f16a046f8e1 (arm64: futex: Fix undefined behaviour with FUTEX_OP_OPARG_SHIFT usage). Look there for an example dump. And as suggested by Thomas, check for negative oparg too, because it was also reported to cause undefined behaviour report. Note that s390 removed access_ok check in d12a29703 ("s390/uaccess: remove pointless access_ok() checks") as access_ok there returns true. We introduce it back to the helper for the sake of simplicity (it gets optimized away anyway). Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com> [s390] Acked-by: Chris Metcalf <cmetcalf@mellanox.com> [for tile] Reviewed-by: Darren Hart (VMware) <dvhart@infradead.org> Reviewed-by: Will Deacon <will.deacon@arm.com> [core/arm64] Cc: linux-mips@linux-mips.org Cc: Rich Felker <dalias@libc.org> Cc: linux-ia64@vger.kernel.org Cc: linux-sh@vger.kernel.org Cc: peterz@infradead.org Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: sparclinux@vger.kernel.org Cc: Jonas Bonn <jonas@southpole.se> Cc: linux-s390@vger.kernel.org Cc: linux-arch@vger.kernel.org Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: linux-hexagon@vger.kernel.org Cc: Helge Deller <deller@gmx.de> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: linux-snps-arc@lists.infradead.org Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: linux-xtensa@linux-xtensa.org Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi> Cc: openrisc@lists.librecores.org Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Stafford Horne <shorne@gmail.com> Cc: linux-arm-kernel@lists.infradead.org Cc: Richard Henderson <rth@twiddle.net> Cc: Chris Zankel <chris@zankel.net> Cc: Michal Simek <monstr@monstr.eu> Cc: Tony Luck <tony.luck@intel.com> Cc: linux-parisc@vger.kernel.org Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: linux-alpha@vger.kernel.org Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Cc: "David S. Miller" <davem@davemloft.net> Link: http://lkml.kernel.org/r/20170824073105.3901-1-jslaby@suse.cz
2017-08-24 07:31:05 +00:00
static int futex_atomic_op_inuser(unsigned int encoded_op, u32 __user *uaddr)
{
unsigned int op = (encoded_op & 0x70000000) >> 28;
unsigned int cmp = (encoded_op & 0x0f000000) >> 24;
int oparg = sign_extend32((encoded_op & 0x00fff000) >> 12, 11);
int cmparg = sign_extend32(encoded_op & 0x00000fff, 11);
futex: Remove duplicated code and fix undefined behaviour There is code duplicated over all architecture's headers for futex_atomic_op_inuser. Namely op decoding, access_ok check for uaddr, and comparison of the result. Remove this duplication and leave up to the arches only the needed assembly which is now in arch_futex_atomic_op_inuser. This effectively distributes the Will Deacon's arm64 fix for undefined behaviour reported by UBSAN to all architectures. The fix was done in commit 5f16a046f8e1 (arm64: futex: Fix undefined behaviour with FUTEX_OP_OPARG_SHIFT usage). Look there for an example dump. And as suggested by Thomas, check for negative oparg too, because it was also reported to cause undefined behaviour report. Note that s390 removed access_ok check in d12a29703 ("s390/uaccess: remove pointless access_ok() checks") as access_ok there returns true. We introduce it back to the helper for the sake of simplicity (it gets optimized away anyway). Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com> [s390] Acked-by: Chris Metcalf <cmetcalf@mellanox.com> [for tile] Reviewed-by: Darren Hart (VMware) <dvhart@infradead.org> Reviewed-by: Will Deacon <will.deacon@arm.com> [core/arm64] Cc: linux-mips@linux-mips.org Cc: Rich Felker <dalias@libc.org> Cc: linux-ia64@vger.kernel.org Cc: linux-sh@vger.kernel.org Cc: peterz@infradead.org Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: sparclinux@vger.kernel.org Cc: Jonas Bonn <jonas@southpole.se> Cc: linux-s390@vger.kernel.org Cc: linux-arch@vger.kernel.org Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: linux-hexagon@vger.kernel.org Cc: Helge Deller <deller@gmx.de> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: linux-snps-arc@lists.infradead.org Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: linux-xtensa@linux-xtensa.org Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi> Cc: openrisc@lists.librecores.org Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Stafford Horne <shorne@gmail.com> Cc: linux-arm-kernel@lists.infradead.org Cc: Richard Henderson <rth@twiddle.net> Cc: Chris Zankel <chris@zankel.net> Cc: Michal Simek <monstr@monstr.eu> Cc: Tony Luck <tony.luck@intel.com> Cc: linux-parisc@vger.kernel.org Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: linux-alpha@vger.kernel.org Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Cc: "David S. Miller" <davem@davemloft.net> Link: http://lkml.kernel.org/r/20170824073105.3901-1-jslaby@suse.cz
2017-08-24 07:31:05 +00:00
int oldval, ret;
if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) {
futex: futex_wake_op, do not fail on invalid op In commit 30d6e0a4190d ("futex: Remove duplicated code and fix undefined behaviour"), I let FUTEX_WAKE_OP to fail on invalid op. Namely when op should be considered as shift and the shift is out of range (< 0 or > 31). But strace's test suite does this madness: futex(0x7fabd78bcffc, 0x5, 0xfacefeed, 0xb, 0x7fabd78bcffc, 0xa0caffee); futex(0x7fabd78bcffc, 0x5, 0xfacefeed, 0xb, 0x7fabd78bcffc, 0xbadfaced); futex(0x7fabd78bcffc, 0x5, 0xfacefeed, 0xb, 0x7fabd78bcffc, 0xffffffff); When I pick the first 0xa0caffee, it decodes as: 0x80000000 & 0xa0caffee: oparg is shift 0x70000000 & 0xa0caffee: op is FUTEX_OP_OR 0x0f000000 & 0xa0caffee: cmp is FUTEX_OP_CMP_EQ 0x00fff000 & 0xa0caffee: oparg is sign-extended 0xcaf = -849 0x00000fff & 0xa0caffee: cmparg is sign-extended 0xfee = -18 That means the op tries to do this: (futex |= (1 << (-849))) == -18 which is completely bogus. The new check of op in the code is: if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) { if (oparg < 0 || oparg > 31) return -EINVAL; oparg = 1 << oparg; } which results obviously in the "Invalid argument" errno: FAIL: futex =========== futex(0x7fabd78bcffc, 0x5, 0xfacefeed, 0xb, 0x7fabd78bcffc, 0xa0caffee) = -1: Invalid argument futex.test: failed test: ../futex failed with code 1 So let us soften the failure to print only a (ratelimited) message, crop the value and continue as if it were right. When userspace keeps up, we can switch this to return -EINVAL again. [v2] Do not return 0 immediatelly, proceed with the cropped value. Fixes: 30d6e0a4190d ("futex: Remove duplicated code and fix undefined behaviour") Signed-off-by: Jiri Slaby <jslaby@suse.cz> Cc: Ingo Molnar <mingo@redhat.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <dvhart@infradead.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-10-23 11:41:51 +00:00
if (oparg < 0 || oparg > 31) {
char comm[sizeof(current->comm)];
/*
* kill this print and return -EINVAL when userspace
* is sane again
*/
pr_info_ratelimited("futex_wake_op: %s tries to shift op by %d; fix this program\n",
get_task_comm(comm, current), oparg);
oparg &= 31;
}
futex: Remove duplicated code and fix undefined behaviour There is code duplicated over all architecture's headers for futex_atomic_op_inuser. Namely op decoding, access_ok check for uaddr, and comparison of the result. Remove this duplication and leave up to the arches only the needed assembly which is now in arch_futex_atomic_op_inuser. This effectively distributes the Will Deacon's arm64 fix for undefined behaviour reported by UBSAN to all architectures. The fix was done in commit 5f16a046f8e1 (arm64: futex: Fix undefined behaviour with FUTEX_OP_OPARG_SHIFT usage). Look there for an example dump. And as suggested by Thomas, check for negative oparg too, because it was also reported to cause undefined behaviour report. Note that s390 removed access_ok check in d12a29703 ("s390/uaccess: remove pointless access_ok() checks") as access_ok there returns true. We introduce it back to the helper for the sake of simplicity (it gets optimized away anyway). Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com> [s390] Acked-by: Chris Metcalf <cmetcalf@mellanox.com> [for tile] Reviewed-by: Darren Hart (VMware) <dvhart@infradead.org> Reviewed-by: Will Deacon <will.deacon@arm.com> [core/arm64] Cc: linux-mips@linux-mips.org Cc: Rich Felker <dalias@libc.org> Cc: linux-ia64@vger.kernel.org Cc: linux-sh@vger.kernel.org Cc: peterz@infradead.org Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: sparclinux@vger.kernel.org Cc: Jonas Bonn <jonas@southpole.se> Cc: linux-s390@vger.kernel.org Cc: linux-arch@vger.kernel.org Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: linux-hexagon@vger.kernel.org Cc: Helge Deller <deller@gmx.de> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: linux-snps-arc@lists.infradead.org Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: linux-xtensa@linux-xtensa.org Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi> Cc: openrisc@lists.librecores.org Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Stafford Horne <shorne@gmail.com> Cc: linux-arm-kernel@lists.infradead.org Cc: Richard Henderson <rth@twiddle.net> Cc: Chris Zankel <chris@zankel.net> Cc: Michal Simek <monstr@monstr.eu> Cc: Tony Luck <tony.luck@intel.com> Cc: linux-parisc@vger.kernel.org Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: linux-alpha@vger.kernel.org Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Cc: "David S. Miller" <davem@davemloft.net> Link: http://lkml.kernel.org/r/20170824073105.3901-1-jslaby@suse.cz
2017-08-24 07:31:05 +00:00
oparg = 1 << oparg;
}
pagefault_disable();
futex: Remove duplicated code and fix undefined behaviour There is code duplicated over all architecture's headers for futex_atomic_op_inuser. Namely op decoding, access_ok check for uaddr, and comparison of the result. Remove this duplication and leave up to the arches only the needed assembly which is now in arch_futex_atomic_op_inuser. This effectively distributes the Will Deacon's arm64 fix for undefined behaviour reported by UBSAN to all architectures. The fix was done in commit 5f16a046f8e1 (arm64: futex: Fix undefined behaviour with FUTEX_OP_OPARG_SHIFT usage). Look there for an example dump. And as suggested by Thomas, check for negative oparg too, because it was also reported to cause undefined behaviour report. Note that s390 removed access_ok check in d12a29703 ("s390/uaccess: remove pointless access_ok() checks") as access_ok there returns true. We introduce it back to the helper for the sake of simplicity (it gets optimized away anyway). Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com> [s390] Acked-by: Chris Metcalf <cmetcalf@mellanox.com> [for tile] Reviewed-by: Darren Hart (VMware) <dvhart@infradead.org> Reviewed-by: Will Deacon <will.deacon@arm.com> [core/arm64] Cc: linux-mips@linux-mips.org Cc: Rich Felker <dalias@libc.org> Cc: linux-ia64@vger.kernel.org Cc: linux-sh@vger.kernel.org Cc: peterz@infradead.org Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: sparclinux@vger.kernel.org Cc: Jonas Bonn <jonas@southpole.se> Cc: linux-s390@vger.kernel.org Cc: linux-arch@vger.kernel.org Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: linux-hexagon@vger.kernel.org Cc: Helge Deller <deller@gmx.de> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: linux-snps-arc@lists.infradead.org Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: linux-xtensa@linux-xtensa.org Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi> Cc: openrisc@lists.librecores.org Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Stafford Horne <shorne@gmail.com> Cc: linux-arm-kernel@lists.infradead.org Cc: Richard Henderson <rth@twiddle.net> Cc: Chris Zankel <chris@zankel.net> Cc: Michal Simek <monstr@monstr.eu> Cc: Tony Luck <tony.luck@intel.com> Cc: linux-parisc@vger.kernel.org Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: linux-alpha@vger.kernel.org Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Cc: "David S. Miller" <davem@davemloft.net> Link: http://lkml.kernel.org/r/20170824073105.3901-1-jslaby@suse.cz
2017-08-24 07:31:05 +00:00
ret = arch_futex_atomic_op_inuser(op, oparg, &oldval, uaddr);
pagefault_enable();
futex: Remove duplicated code and fix undefined behaviour There is code duplicated over all architecture's headers for futex_atomic_op_inuser. Namely op decoding, access_ok check for uaddr, and comparison of the result. Remove this duplication and leave up to the arches only the needed assembly which is now in arch_futex_atomic_op_inuser. This effectively distributes the Will Deacon's arm64 fix for undefined behaviour reported by UBSAN to all architectures. The fix was done in commit 5f16a046f8e1 (arm64: futex: Fix undefined behaviour with FUTEX_OP_OPARG_SHIFT usage). Look there for an example dump. And as suggested by Thomas, check for negative oparg too, because it was also reported to cause undefined behaviour report. Note that s390 removed access_ok check in d12a29703 ("s390/uaccess: remove pointless access_ok() checks") as access_ok there returns true. We introduce it back to the helper for the sake of simplicity (it gets optimized away anyway). Signed-off-by: Jiri Slaby <jslaby@suse.cz> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Russell King <rmk+kernel@armlinux.org.uk> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com> [s390] Acked-by: Chris Metcalf <cmetcalf@mellanox.com> [for tile] Reviewed-by: Darren Hart (VMware) <dvhart@infradead.org> Reviewed-by: Will Deacon <will.deacon@arm.com> [core/arm64] Cc: linux-mips@linux-mips.org Cc: Rich Felker <dalias@libc.org> Cc: linux-ia64@vger.kernel.org Cc: linux-sh@vger.kernel.org Cc: peterz@infradead.org Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Paul Mackerras <paulus@samba.org> Cc: sparclinux@vger.kernel.org Cc: Jonas Bonn <jonas@southpole.se> Cc: linux-s390@vger.kernel.org Cc: linux-arch@vger.kernel.org Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: linux-hexagon@vger.kernel.org Cc: Helge Deller <deller@gmx.de> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Matt Turner <mattst88@gmail.com> Cc: linux-snps-arc@lists.infradead.org Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: Arnd Bergmann <arnd@arndb.de> Cc: linux-xtensa@linux-xtensa.org Cc: Stefan Kristiansson <stefan.kristiansson@saunalahti.fi> Cc: openrisc@lists.librecores.org Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Stafford Horne <shorne@gmail.com> Cc: linux-arm-kernel@lists.infradead.org Cc: Richard Henderson <rth@twiddle.net> Cc: Chris Zankel <chris@zankel.net> Cc: Michal Simek <monstr@monstr.eu> Cc: Tony Luck <tony.luck@intel.com> Cc: linux-parisc@vger.kernel.org Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: linux-alpha@vger.kernel.org Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linuxppc-dev@lists.ozlabs.org Cc: "David S. Miller" <davem@davemloft.net> Link: http://lkml.kernel.org/r/20170824073105.3901-1-jslaby@suse.cz
2017-08-24 07:31:05 +00:00
if (ret)
return ret;
switch (cmp) {
case FUTEX_OP_CMP_EQ:
return oldval == cmparg;
case FUTEX_OP_CMP_NE:
return oldval != cmparg;
case FUTEX_OP_CMP_LT:
return oldval < cmparg;
case FUTEX_OP_CMP_GE:
return oldval >= cmparg;
case FUTEX_OP_CMP_LE:
return oldval <= cmparg;
case FUTEX_OP_CMP_GT:
return oldval > cmparg;
default:
return -ENOSYS;
}
}
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
/*
* Wake up all waiters hashed on the physical page that is mapped
* to this virtual address:
*/
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
static int
futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2,
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
int nr_wake, int nr_wake2, int op)
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
{
union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
struct futex_hash_bucket *hb1, *hb2;
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
struct futex_q *this, *next;
int ret, op_ret;
DEFINE_WAKE_Q(wake_q);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
retry:
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, FUTEX_READ);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (unlikely(ret != 0))
return ret;
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, FUTEX_WRITE);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (unlikely(ret != 0))
return ret;
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
hb1 = hash_futex(&key1);
hb2 = hash_futex(&key2);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
retry_private:
double_lock_hb(hb1, hb2);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
op_ret = futex_atomic_op_inuser(op, uaddr2);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (unlikely(op_ret < 0)) {
double_unlock_hb(hb1, hb2);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (!IS_ENABLED(CONFIG_MMU) ||
unlikely(op_ret != -EFAULT && op_ret != -EAGAIN)) {
/*
* we don't get EFAULT from MMU faults if we don't have
* an MMU, but we might get them from range checking
*/
ret = op_ret;
return ret;
}
if (op_ret == -EFAULT) {
ret = fault_in_user_writeable(uaddr2);
if (ret)
return ret;
}
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (!(flags & FLAGS_SHARED)) {
cond_resched();
goto retry_private;
}
cond_resched();
goto retry;
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
}
plist_for_each_entry_safe(this, next, &hb1->chain, list) {
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (match_futex (&this->key, &key1)) {
if (this->pi_state || this->rt_waiter) {
ret = -EINVAL;
goto out_unlock;
}
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
mark_wake_futex(&wake_q, this);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (++ret >= nr_wake)
break;
}
}
if (op_ret > 0) {
op_ret = 0;
plist_for_each_entry_safe(this, next, &hb2->chain, list) {
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (match_futex (&this->key, &key2)) {
if (this->pi_state || this->rt_waiter) {
ret = -EINVAL;
goto out_unlock;
}
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
mark_wake_futex(&wake_q, this);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
if (++op_ret >= nr_wake2)
break;
}
}
ret += op_ret;
}
out_unlock:
double_unlock_hb(hb1, hb2);
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
wake_up_q(&wake_q);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
return ret;
}
/**
* requeue_futex() - Requeue a futex_q from one hb to another
* @q: the futex_q to requeue
* @hb1: the source hash_bucket
* @hb2: the target hash_bucket
* @key2: the new key for the requeued futex_q
*/
static inline
void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
struct futex_hash_bucket *hb2, union futex_key *key2)
{
/*
* If key1 and key2 hash to the same bucket, no need to
* requeue.
*/
if (likely(&hb1->chain != &hb2->chain)) {
plist_del(&q->list, &hb1->chain);
hb_waiters_dec(hb1);
hb_waiters_inc(hb2);
plist_add(&q->list, &hb2->chain);
q->lock_ptr = &hb2->lock;
}
q->key = *key2;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/**
* requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
* @q: the futex_q
* @key: the key of the requeue target futex
* @hb: the hash_bucket of the requeue target futex
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* During futex_requeue, with requeue_pi=1, it is possible to acquire the
* target futex if it is uncontended or via a lock steal. Set the futex_q key
* to the requeue target futex so the waiter can detect the wakeup on the right
* futex, but remove it from the hb and NULL the rt_waiter so it can detect
futex: Update futex_q lock_ptr on requeue proxy lock futex_requeue() can acquire the lock on behalf of a waiter early on or during the requeue loop if it is uncontended or in the event of a lock steal or owner died. On wakeup, the waiter (in futex_wait_requeue_pi()) cleans up the pi_state owner using the lock_ptr to protect against concurrent access to the pi_state. The pi_state is hung off futex_q's on the requeue target futex hash bucket so the lock_ptr needs to be updated accordingly. The problem manifested by triggering the WARN_ON in lookup_pi_state() about the pid != pi_state->owner->pid. With this patch, the pi_state is properly guarded against concurrent access via the requeue target hb lock. The astute reviewer may notice that there is a window of time between when futex_requeue() unlocks the hb locks and when futex_wait_requeue_pi() will acquire hb2->lock. During this time the pi_state and uval are not in sync with the underlying rtmutex owner (but the uval does indicate there are waiters, so no atomic changes will occur in userspace). However, this is not a problem. Should a contending thread enter lookup_pi_state() and acquire hb2->lock before the ownership is fixed up, it will find the pi_state hung off a waiter's (possibly the pending owner's) futex_q and block on the rtmutex. Once futex_wait_requeue_pi() fixes up the owner, it will also move the pi_state from the old owner's task->pi_state_list to its own. v3: Fix plist lock name for application to mainline (rather than -rt) Compile tested against tip/v2.6.31-rc5. Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Dinakar Guniguntala <dino@in.ibm.com> Cc: John Stultz <johnstul@linux.vnet.ibm.com> LKML-Reference: <4A7F4EFF.6090903@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-09 22:34:39 +00:00
* atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock
* to protect access to the pi_state to fixup the owner later. Must be called
* with both q->lock_ptr and hb->lock held.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
static inline
futex: Update futex_q lock_ptr on requeue proxy lock futex_requeue() can acquire the lock on behalf of a waiter early on or during the requeue loop if it is uncontended or in the event of a lock steal or owner died. On wakeup, the waiter (in futex_wait_requeue_pi()) cleans up the pi_state owner using the lock_ptr to protect against concurrent access to the pi_state. The pi_state is hung off futex_q's on the requeue target futex hash bucket so the lock_ptr needs to be updated accordingly. The problem manifested by triggering the WARN_ON in lookup_pi_state() about the pid != pi_state->owner->pid. With this patch, the pi_state is properly guarded against concurrent access via the requeue target hb lock. The astute reviewer may notice that there is a window of time between when futex_requeue() unlocks the hb locks and when futex_wait_requeue_pi() will acquire hb2->lock. During this time the pi_state and uval are not in sync with the underlying rtmutex owner (but the uval does indicate there are waiters, so no atomic changes will occur in userspace). However, this is not a problem. Should a contending thread enter lookup_pi_state() and acquire hb2->lock before the ownership is fixed up, it will find the pi_state hung off a waiter's (possibly the pending owner's) futex_q and block on the rtmutex. Once futex_wait_requeue_pi() fixes up the owner, it will also move the pi_state from the old owner's task->pi_state_list to its own. v3: Fix plist lock name for application to mainline (rather than -rt) Compile tested against tip/v2.6.31-rc5. Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Dinakar Guniguntala <dino@in.ibm.com> Cc: John Stultz <johnstul@linux.vnet.ibm.com> LKML-Reference: <4A7F4EFF.6090903@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-09 22:34:39 +00:00
void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
struct futex_hash_bucket *hb)
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
{
q->key = *key;
__unqueue_futex(q);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
WARN_ON(!q->rt_waiter);
q->rt_waiter = NULL;
futex: Update futex_q lock_ptr on requeue proxy lock futex_requeue() can acquire the lock on behalf of a waiter early on or during the requeue loop if it is uncontended or in the event of a lock steal or owner died. On wakeup, the waiter (in futex_wait_requeue_pi()) cleans up the pi_state owner using the lock_ptr to protect against concurrent access to the pi_state. The pi_state is hung off futex_q's on the requeue target futex hash bucket so the lock_ptr needs to be updated accordingly. The problem manifested by triggering the WARN_ON in lookup_pi_state() about the pid != pi_state->owner->pid. With this patch, the pi_state is properly guarded against concurrent access via the requeue target hb lock. The astute reviewer may notice that there is a window of time between when futex_requeue() unlocks the hb locks and when futex_wait_requeue_pi() will acquire hb2->lock. During this time the pi_state and uval are not in sync with the underlying rtmutex owner (but the uval does indicate there are waiters, so no atomic changes will occur in userspace). However, this is not a problem. Should a contending thread enter lookup_pi_state() and acquire hb2->lock before the ownership is fixed up, it will find the pi_state hung off a waiter's (possibly the pending owner's) futex_q and block on the rtmutex. Once futex_wait_requeue_pi() fixes up the owner, it will also move the pi_state from the old owner's task->pi_state_list to its own. v3: Fix plist lock name for application to mainline (rather than -rt) Compile tested against tip/v2.6.31-rc5. Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Dinakar Guniguntala <dino@in.ibm.com> Cc: John Stultz <johnstul@linux.vnet.ibm.com> LKML-Reference: <4A7F4EFF.6090903@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-09 22:34:39 +00:00
q->lock_ptr = &hb->lock;
wake_up_state(q->task, TASK_NORMAL);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
/**
* futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter
* @pifutex: the user address of the to futex
* @hb1: the from futex hash bucket, must be locked by the caller
* @hb2: the to futex hash bucket, must be locked by the caller
* @key1: the from futex key
* @key2: the to futex key
* @ps: address to store the pi_state pointer
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
* @exiting: Pointer to store the task pointer of the owner task
* which is in the middle of exiting
* @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0)
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* Try and get the lock on behalf of the top waiter if we can do it atomically.
* Wake the top waiter if we succeed. If the caller specified set_waiters,
* then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit.
* hb1 and hb2 must be held by the caller.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
* @exiting is only set when the return value is -EBUSY. If so, this holds
* a refcount on the exiting task on return and the caller needs to drop it
* after waiting for the exit to complete.
*
* Return:
* - 0 - failed to acquire the lock atomically;
* - >0 - acquired the lock, return value is vpid of the top_waiter
* - <0 - error
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
static int
futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1,
struct futex_hash_bucket *hb2, union futex_key *key1,
union futex_key *key2, struct futex_pi_state **ps,
struct task_struct **exiting, int set_waiters)
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
{
struct futex_q *top_waiter = NULL;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
u32 curval;
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
int ret, vpid;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (get_futex_value_locked(&curval, pifutex))
return -EFAULT;
if (unlikely(should_fail_futex(true)))
return -EFAULT;
/*
* Find the top_waiter and determine if there are additional waiters.
* If the caller intends to requeue more than 1 waiter to pifutex,
* force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now,
* as we have means to handle the possible fault. If not, don't set
* the bit unecessarily as it will force the subsequent unlock to enter
* the kernel.
*/
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
top_waiter = futex_top_waiter(hb1, key1);
/* There are no waiters, nothing for us to do. */
if (!top_waiter)
return 0;
/* Ensure we requeue to the expected futex. */
if (!match_futex(top_waiter->requeue_pi_key, key2))
return -EINVAL;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in
* the contended case or if set_waiters is 1. The pi_state is returned
* in ps in contended cases.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
vpid = task_pid_vnr(top_waiter->task);
ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
exiting, set_waiters);
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
if (ret == 1) {
futex: Update futex_q lock_ptr on requeue proxy lock futex_requeue() can acquire the lock on behalf of a waiter early on or during the requeue loop if it is uncontended or in the event of a lock steal or owner died. On wakeup, the waiter (in futex_wait_requeue_pi()) cleans up the pi_state owner using the lock_ptr to protect against concurrent access to the pi_state. The pi_state is hung off futex_q's on the requeue target futex hash bucket so the lock_ptr needs to be updated accordingly. The problem manifested by triggering the WARN_ON in lookup_pi_state() about the pid != pi_state->owner->pid. With this patch, the pi_state is properly guarded against concurrent access via the requeue target hb lock. The astute reviewer may notice that there is a window of time between when futex_requeue() unlocks the hb locks and when futex_wait_requeue_pi() will acquire hb2->lock. During this time the pi_state and uval are not in sync with the underlying rtmutex owner (but the uval does indicate there are waiters, so no atomic changes will occur in userspace). However, this is not a problem. Should a contending thread enter lookup_pi_state() and acquire hb2->lock before the ownership is fixed up, it will find the pi_state hung off a waiter's (possibly the pending owner's) futex_q and block on the rtmutex. Once futex_wait_requeue_pi() fixes up the owner, it will also move the pi_state from the old owner's task->pi_state_list to its own. v3: Fix plist lock name for application to mainline (rather than -rt) Compile tested against tip/v2.6.31-rc5. Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Dinakar Guniguntala <dino@in.ibm.com> Cc: John Stultz <johnstul@linux.vnet.ibm.com> LKML-Reference: <4A7F4EFF.6090903@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-09 22:34:39 +00:00
requeue_pi_wake_futex(top_waiter, key2, hb2);
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
return vpid;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
return ret;
}
/**
* futex_requeue() - Requeue waiters from uaddr1 to uaddr2
* @uaddr1: source futex user address
* @flags: futex flags (FLAGS_SHARED, etc.)
* @uaddr2: target futex user address
* @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
* @nr_requeue: number of waiters to requeue (0-INT_MAX)
* @cmpval: @uaddr1 expected value (or %NULL)
* @requeue_pi: if we are attempting to requeue from a non-pi futex to a
* pi futex (pi to pi requeue is not supported)
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
* uaddr2 atomically on behalf of the top waiter.
*
* Return:
* - >=0 - on success, the number of tasks requeued or woken;
* - <0 - on error
*/
static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
u32 __user *uaddr2, int nr_wake, int nr_requeue,
u32 *cmpval, int requeue_pi)
{
union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT;
int task_count = 0, ret;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
struct futex_pi_state *pi_state = NULL;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
struct futex_hash_bucket *hb1, *hb2;
struct futex_q *this, *next;
DEFINE_WAKE_Q(wake_q);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (nr_wake < 0 || nr_requeue < 0)
return -EINVAL;
/*
* When PI not supported: return -ENOSYS if requeue_pi is true,
* consequently the compiler knows requeue_pi is always false past
* this point which will optimize away all the conditional code
* further down.
*/
if (!IS_ENABLED(CONFIG_FUTEX_PI) && requeue_pi)
return -ENOSYS;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (requeue_pi) {
/*
* Requeue PI only works on two distinct uaddrs. This
* check is only valid for private futexes. See below.
*/
if (uaddr1 == uaddr2)
return -EINVAL;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* requeue_pi requires a pi_state, try to allocate it now
* without any locks in case it fails.
*/
if (refill_pi_state_cache())
return -ENOMEM;
/*
* requeue_pi must wake as many tasks as it can, up to nr_wake
* + nr_requeue, since it acquires the rt_mutex prior to
* returning to userspace, so as to not leave the rt_mutex with
* waiters and no owner. However, second and third wake-ups
* cannot be predicted as they involve race conditions with the
* first wake and a fault while looking up the pi_state. Both
* pthread_cond_signal() and pthread_cond_broadcast() should
* use nr_wake=1.
*/
if (nr_wake != 1)
return -EINVAL;
}
retry:
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, FUTEX_READ);
if (unlikely(ret != 0))
return ret;
futex: Fix regression with read only mappings commit 7485d0d3758e8e6491a5c9468114e74dc050785d (futexes: Remove rw parameter from get_futex_key()) in 2.6.33 fixed two problems: First, It prevented a loop when encountering a ZERO_PAGE. Second, it fixed RW MAP_PRIVATE futex operations by forcing the COW to occur by unconditionally performing a write access get_user_pages_fast() to get the page. The commit also introduced a user-mode regression in that it broke futex operations on read-only memory maps. For example, this breaks workloads that have one or more reader processes doing a FUTEX_WAIT on a futex within a read only shared file mapping, and a writer processes that has a writable mapping issuing the FUTEX_WAKE. This fixes the regression for valid futex operations on RO mappings by trying a RO get_user_pages_fast() when the RW get_user_pages_fast() fails. This change makes it necessary to also check for invalid use cases, such as anonymous RO mappings (which can never change) and the ZERO_PAGE which the commit referenced above was written to address. This patch does restore the original behavior with RO MAP_PRIVATE mappings, which have inherent user-mode usage problems and don't really make sense. With this patch performing a FUTEX_WAIT within a RO MAP_PRIVATE mapping will be successfully woken provided another process updates the region of the underlying mapped file. However, the mmap() man page states that for a MAP_PRIVATE mapping: It is unspecified whether changes made to the file after the mmap() call are visible in the mapped region. So user-mode users attempting to use futex operations on RO MAP_PRIVATE mappings are depending on unspecified behavior. Additionally a RO MAP_PRIVATE mapping could fail to wake up in the following case. Thread-A: call futex(FUTEX_WAIT, memory-region-A). get_futex_key() return inode based key. sleep on the key Thread-B: call mprotect(PROT_READ|PROT_WRITE, memory-region-A) Thread-B: write memory-region-A. COW happen. This process's memory-region-A become related to new COWed private (ie PageAnon=1) page. Thread-B: call futex(FUETX_WAKE, memory-region-A). get_futex_key() return mm based key. IOW, we fail to wake up Thread-A. Once again doing something like this is just silly and users who do something like this get what they deserve. While RO MAP_PRIVATE mappings are nonsensical, checking for a private mapping requires walking the vmas and was deemed too costly to avoid a userspace hang. This Patch is based on Peter Zijlstra's initial patch with modifications to only allow RO mappings for futex operations that need VERIFY_READ access. Reported-by: David Oliver <david@rgmadvisors.com> Signed-off-by: Shawn Bohrer <sbohrer@rgmadvisors.com> Acked-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Darren Hart <dvhart@linux.intel.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: peterz@infradead.org Cc: eric.dumazet@gmail.com Cc: zvonler@rgmadvisors.com Cc: hughd@google.com Link: http://lkml.kernel.org/r/1309450892-30676-1-git-send-email-sbohrer@rgmadvisors.com Cc: stable@kernel.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-06-30 16:21:32 +00:00
ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2,
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
requeue_pi ? FUTEX_WRITE : FUTEX_READ);
if (unlikely(ret != 0))
return ret;
/*
* The check above which compares uaddrs is not sufficient for
* shared futexes. We need to compare the keys:
*/
if (requeue_pi && match_futex(&key1, &key2))
return -EINVAL;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
hb1 = hash_futex(&key1);
hb2 = hash_futex(&key2);
retry_private:
futex: avoid race between requeue and wake Jan Stancek reported: "pthread_cond_broadcast/4-1.c testcase from openposix testsuite (LTP) occasionally fails, because some threads fail to wake up. Testcase creates 5 threads, which are all waiting on same condition. Main thread then calls pthread_cond_broadcast() without holding mutex, which calls: futex(uaddr1, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, uaddr2, ..) This immediately wakes up single thread A, which unlocks mutex and tries to wake up another thread: futex(uaddr2, FUTEX_WAKE_PRIVATE, 1) If thread A manages to call futex_wake() before any waiters are requeued for uaddr2, no other thread is woken up" The ordering constraints for the hash bucket waiter counting are that the waiter counts have to be incremented _before_ getting the spinlock (because the spinlock acts as part of the memory barrier), but the "requeue" operation didn't honor those rules, and nobody had even thought about that case. This fairly simple patch just increments the waiter count for the target hash bucket (hb2) when requeing a futex before taking the locks. It then decrements them again after releasing the lock - the code that actually moves the futex(es) between hash buckets will do the additional required waiter count housekeeping. Reported-and-tested-by: Jan Stancek <jstancek@redhat.com> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org # 3.14 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-08 22:30:07 +00:00
hb_waiters_inc(hb2);
double_lock_hb(hb1, hb2);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
if (likely(cmpval != NULL)) {
u32 curval;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
ret = get_futex_value_locked(&curval, uaddr1);
if (unlikely(ret)) {
double_unlock_hb(hb1, hb2);
futex: avoid race between requeue and wake Jan Stancek reported: "pthread_cond_broadcast/4-1.c testcase from openposix testsuite (LTP) occasionally fails, because some threads fail to wake up. Testcase creates 5 threads, which are all waiting on same condition. Main thread then calls pthread_cond_broadcast() without holding mutex, which calls: futex(uaddr1, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, uaddr2, ..) This immediately wakes up single thread A, which unlocks mutex and tries to wake up another thread: futex(uaddr2, FUTEX_WAKE_PRIVATE, 1) If thread A manages to call futex_wake() before any waiters are requeued for uaddr2, no other thread is woken up" The ordering constraints for the hash bucket waiter counting are that the waiter counts have to be incremented _before_ getting the spinlock (because the spinlock acts as part of the memory barrier), but the "requeue" operation didn't honor those rules, and nobody had even thought about that case. This fairly simple patch just increments the waiter count for the target hash bucket (hb2) when requeing a futex before taking the locks. It then decrements them again after releasing the lock - the code that actually moves the futex(es) between hash buckets will do the additional required waiter count housekeeping. Reported-and-tested-by: Jan Stancek <jstancek@redhat.com> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org # 3.14 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-08 22:30:07 +00:00
hb_waiters_dec(hb2);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
ret = get_user(curval, uaddr1);
if (ret)
return ret;
if (!(flags & FLAGS_SHARED))
goto retry_private;
goto retry;
}
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
if (curval != *cmpval) {
ret = -EAGAIN;
goto out_unlock;
}
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
struct task_struct *exiting = NULL;
/*
* Attempt to acquire uaddr2 and wake the top waiter. If we
* intend to requeue waiters, force setting the FUTEX_WAITERS
* bit. We force this here where we are able to easily handle
* faults rather in the requeue loop below.
*/
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
&key2, &pi_state,
&exiting, nr_requeue);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* At this point the top_waiter has either taken uaddr2 or is
* waiting on it. If the former, then the pi_state will not
* exist yet, look it up one more time to ensure we have a
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
* reference to it. If the lock was taken, ret contains the
* vpid of the top waiter task.
* If the lock was not taken, we have pi_state and an initial
* refcount on it. In case of an error we have nothing.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
if (ret > 0) {
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
WARN_ON(pi_state);
task_count++;
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
/*
* If we acquired the lock, then the user space value
* of uaddr2 should be vpid. It cannot be changed by
* the top waiter as it is blocked on hb2 lock if it
* tries to do so. If something fiddled with it behind
* our back the pi state lookup might unearth it. So
* we rather use the known value than rereading and
* handing potential crap to lookup_pi_state.
*
* If that call succeeds then we have pi_state and an
* initial refcount on it.
futex: Add another early deadlock detection check Dave Jones trinity syscall fuzzer exposed an issue in the deadlock detection code of rtmutex: http://lkml.kernel.org/r/20140429151655.GA14277@redhat.com That underlying issue has been fixed with a patch to the rtmutex code, but the futex code must not call into rtmutex in that case because - it can detect that issue early - it avoids a different and more complex fixup for backing out If the user space variable got manipulated to 0x80000000 which means no lock holder, but the waiters bit set and an active pi_state in the kernel is found we can figure out the recursive locking issue by looking at the pi_state owner. If that is the current task, then we can safely return -EDEADLK. The check should have been added in commit 59fa62451 (futex: Handle futex_pi OWNER_DIED take over correctly) already, but I did not see the above issue caused by user space manipulation back then. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Dave Jones <davej@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Darren Hart <darren@dvhart.com> Cc: Davidlohr Bueso <davidlohr@hp.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Clark Williams <williams@redhat.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Lai Jiangshan <laijs@cn.fujitsu.com> Cc: Roland McGrath <roland@hack.frob.com> Cc: Carlos ODonell <carlos@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Link: http://lkml.kernel.org/r/20140512201701.097349971@linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org
2014-05-12 20:45:34 +00:00
*/
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
ret = lookup_pi_state(uaddr2, ret, hb2, &key2,
&pi_state, &exiting);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
switch (ret) {
case 0:
/* We hold a reference on the pi state. */
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
break;
/* If the above failed, then pi_state is NULL */
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
case -EFAULT:
double_unlock_hb(hb1, hb2);
futex: avoid race between requeue and wake Jan Stancek reported: "pthread_cond_broadcast/4-1.c testcase from openposix testsuite (LTP) occasionally fails, because some threads fail to wake up. Testcase creates 5 threads, which are all waiting on same condition. Main thread then calls pthread_cond_broadcast() without holding mutex, which calls: futex(uaddr1, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, uaddr2, ..) This immediately wakes up single thread A, which unlocks mutex and tries to wake up another thread: futex(uaddr2, FUTEX_WAKE_PRIVATE, 1) If thread A manages to call futex_wake() before any waiters are requeued for uaddr2, no other thread is woken up" The ordering constraints for the hash bucket waiter counting are that the waiter counts have to be incremented _before_ getting the spinlock (because the spinlock acts as part of the memory barrier), but the "requeue" operation didn't honor those rules, and nobody had even thought about that case. This fairly simple patch just increments the waiter count for the target hash bucket (hb2) when requeing a futex before taking the locks. It then decrements them again after releasing the lock - the code that actually moves the futex(es) between hash buckets will do the additional required waiter count housekeeping. Reported-and-tested-by: Jan Stancek <jstancek@redhat.com> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org # 3.14 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-08 22:30:07 +00:00
hb_waiters_dec(hb2);
ret = fault_in_user_writeable(uaddr2);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (!ret)
goto retry;
return ret;
case -EBUSY:
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
case -EAGAIN:
/*
* Two reasons for this:
* - EBUSY: Owner is exiting and we just wait for the
* exit to complete.
* - EAGAIN: The user space value changed.
*/
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
double_unlock_hb(hb1, hb2);
futex: avoid race between requeue and wake Jan Stancek reported: "pthread_cond_broadcast/4-1.c testcase from openposix testsuite (LTP) occasionally fails, because some threads fail to wake up. Testcase creates 5 threads, which are all waiting on same condition. Main thread then calls pthread_cond_broadcast() without holding mutex, which calls: futex(uaddr1, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, uaddr2, ..) This immediately wakes up single thread A, which unlocks mutex and tries to wake up another thread: futex(uaddr2, FUTEX_WAKE_PRIVATE, 1) If thread A manages to call futex_wake() before any waiters are requeued for uaddr2, no other thread is woken up" The ordering constraints for the hash bucket waiter counting are that the waiter counts have to be incremented _before_ getting the spinlock (because the spinlock acts as part of the memory barrier), but the "requeue" operation didn't honor those rules, and nobody had even thought about that case. This fairly simple patch just increments the waiter count for the target hash bucket (hb2) when requeing a futex before taking the locks. It then decrements them again after releasing the lock - the code that actually moves the futex(es) between hash buckets will do the additional required waiter count housekeeping. Reported-and-tested-by: Jan Stancek <jstancek@redhat.com> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org # 3.14 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-08 22:30:07 +00:00
hb_waiters_dec(hb2);
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
/*
* Handle the case where the owner is in the middle of
* exiting. Wait for the exit to complete otherwise
* this task might loop forever, aka. live lock.
*/
wait_for_owner_exiting(ret, exiting);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
cond_resched();
goto retry;
default:
goto out_unlock;
}
}
plist_for_each_entry_safe(this, next, &hb1->chain, list) {
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (task_count - nr_wake >= nr_requeue)
break;
if (!match_futex(&this->key, &key1))
continue;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* FUTEX_WAIT_REQEUE_PI and FUTEX_CMP_REQUEUE_PI should always
* be paired with each other and no other futex ops.
*
* We should never be requeueing a futex_q with a pi_state,
* which is awaiting a futex_unlock_pi().
*/
if ((requeue_pi && !this->rt_waiter) ||
(!requeue_pi && this->rt_waiter) ||
this->pi_state) {
ret = -EINVAL;
break;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* Wake nr_wake waiters. For requeue_pi, if we acquired the
* lock, we already woke the top_waiter. If not, it will be
* woken by futex_unlock_pi().
*/
if (++task_count <= nr_wake && !requeue_pi) {
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
mark_wake_futex(&wake_q, this);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
continue;
}
/* Ensure we requeue to the expected futex for requeue_pi. */
if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
ret = -EINVAL;
break;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* Requeue nr_requeue waiters and possibly one more in the case
* of requeue_pi if we couldn't acquire the lock atomically.
*/
if (requeue_pi) {
/*
* Prepare the waiter to take the rt_mutex. Take a
* refcount on the pi_state and store the pointer in
* the futex_q object of the waiter.
*/
get_pi_state(pi_state);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
this->pi_state = pi_state;
ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
this->rt_waiter,
this->task);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (ret == 1) {
/*
* We got the lock. We do neither drop the
* refcount on pi_state nor clear
* this->pi_state because the waiter needs the
* pi_state for cleaning up the user space
* value. It will drop the refcount after
* doing so.
*/
futex: Update futex_q lock_ptr on requeue proxy lock futex_requeue() can acquire the lock on behalf of a waiter early on or during the requeue loop if it is uncontended or in the event of a lock steal or owner died. On wakeup, the waiter (in futex_wait_requeue_pi()) cleans up the pi_state owner using the lock_ptr to protect against concurrent access to the pi_state. The pi_state is hung off futex_q's on the requeue target futex hash bucket so the lock_ptr needs to be updated accordingly. The problem manifested by triggering the WARN_ON in lookup_pi_state() about the pid != pi_state->owner->pid. With this patch, the pi_state is properly guarded against concurrent access via the requeue target hb lock. The astute reviewer may notice that there is a window of time between when futex_requeue() unlocks the hb locks and when futex_wait_requeue_pi() will acquire hb2->lock. During this time the pi_state and uval are not in sync with the underlying rtmutex owner (but the uval does indicate there are waiters, so no atomic changes will occur in userspace). However, this is not a problem. Should a contending thread enter lookup_pi_state() and acquire hb2->lock before the ownership is fixed up, it will find the pi_state hung off a waiter's (possibly the pending owner's) futex_q and block on the rtmutex. Once futex_wait_requeue_pi() fixes up the owner, it will also move the pi_state from the old owner's task->pi_state_list to its own. v3: Fix plist lock name for application to mainline (rather than -rt) Compile tested against tip/v2.6.31-rc5. Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Dinakar Guniguntala <dino@in.ibm.com> Cc: John Stultz <johnstul@linux.vnet.ibm.com> LKML-Reference: <4A7F4EFF.6090903@us.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2009-08-09 22:34:39 +00:00
requeue_pi_wake_futex(this, &key2, hb2);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
continue;
} else if (ret) {
/*
* rt_mutex_start_proxy_lock() detected a
* potential deadlock when we tried to queue
* that waiter. Drop the pi_state reference
* which we took above and remove the pointer
* to the state from the waiters futex_q
* object.
*/
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
this->pi_state = NULL;
put_pi_state(pi_state);
/*
* We stop queueing more waiters and let user
* space deal with the mess.
*/
break;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
requeue_futex(this, hb1, hb2, &key2);
}
/*
* We took an extra initial reference to the pi_state either
* in futex_proxy_trylock_atomic() or in lookup_pi_state(). We
* need to drop it here again.
*/
put_pi_state(pi_state);
out_unlock:
double_unlock_hb(hb1, hb2);
futex: Implement lockless wakeups Given the overall futex architecture, any chance of reducing hb->lock contention is welcome. In this particular case, using wake-queues to enable lockless wakeups addresses very much real world performance concerns, even cases of soft-lockups in cases of large amounts of blocked tasks (which is not hard to find in large boxes, using but just a handful of futex). At the lowest level, this patch can reduce latency of a single thread attempting to acquire hb->lock in highly contended scenarios by a up to 2x. At lower counts of nr_wake there are no regressions, confirming, of course, that the wake_q handling overhead is practically non existent. For instance, while a fair amount of variation, the extended pef-bench wakeup benchmark shows for a 20 core machine the following avg per-thread time to wakeup its share of tasks: nr_thr ms-before ms-after 16 0.0590 0.0215 32 0.0396 0.0220 48 0.0417 0.0182 64 0.0536 0.0236 80 0.0414 0.0097 96 0.0672 0.0152 Naturally, this can cause spurious wakeups. However there is no core code that cannot handle them afaict, and furthermore tglx does have the point that other events can already trigger them anyway. Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Acked-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Chris Mason <clm@fb.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: George Spelvin <linux@horizon.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Steven Rostedt <rostedt@goodmis.org> Link: http://lkml.kernel.org/r/1430494072-30283-3-git-send-email-dave@stgolabs.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2015-05-01 15:27:51 +00:00
wake_up_q(&wake_q);
futex: avoid race between requeue and wake Jan Stancek reported: "pthread_cond_broadcast/4-1.c testcase from openposix testsuite (LTP) occasionally fails, because some threads fail to wake up. Testcase creates 5 threads, which are all waiting on same condition. Main thread then calls pthread_cond_broadcast() without holding mutex, which calls: futex(uaddr1, FUTEX_CMP_REQUEUE_PRIVATE, 1, 2147483647, uaddr2, ..) This immediately wakes up single thread A, which unlocks mutex and tries to wake up another thread: futex(uaddr2, FUTEX_WAKE_PRIVATE, 1) If thread A manages to call futex_wake() before any waiters are requeued for uaddr2, no other thread is woken up" The ordering constraints for the hash bucket waiter counting are that the waiter counts have to be incremented _before_ getting the spinlock (because the spinlock acts as part of the memory barrier), but the "requeue" operation didn't honor those rules, and nobody had even thought about that case. This fairly simple patch just increments the waiter count for the target hash bucket (hb2) when requeing a futex before taking the locks. It then decrements them again after releasing the lock - the code that actually moves the futex(es) between hash buckets will do the additional required waiter count housekeeping. Reported-and-tested-by: Jan Stancek <jstancek@redhat.com> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org # 3.14 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-04-08 22:30:07 +00:00
hb_waiters_dec(hb2);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
return ret ? ret : task_count;
}
/* The key must be already stored in q->key. */
static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
__acquires(&hb->lock)
{
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
struct futex_hash_bucket *hb;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
hb = hash_futex(&q->key);
/*
* Increment the counter before taking the lock so that
* a potential waker won't miss a to-be-slept task that is
* waiting for the spinlock. This is safe as all queue_lock()
* users end up calling queue_me(). Similarly, for housekeeping,
* decrement the counter at queue_unlock() when some error has
* occurred and we don't end up adding the task to the list.
*/
hb_waiters_inc(hb); /* implies smp_mb(); (A) */
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
q->lock_ptr = &hb->lock;
spin_lock(&hb->lock);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
return hb;
}
static inline void
queue_unlock(struct futex_hash_bucket *hb)
__releases(&hb->lock)
{
spin_unlock(&hb->lock);
hb_waiters_dec(hb);
}
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
{
int prio;
/*
* The priority used to register this element is
* - either the real thread-priority for the real-time threads
* (i.e. threads with a priority lower than MAX_RT_PRIO)
* - or MAX_RT_PRIO for non-RT threads.
* Thus, all RT-threads are woken first in priority order, and
* the others are woken last, in FIFO order.
*/
prio = min(current->normal_prio, MAX_RT_PRIO);
plist_node_init(&q->list, prio);
plist_add(&q->list, &hb->chain);
q->task = current;
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
}
/**
* queue_me() - Enqueue the futex_q on the futex_hash_bucket
* @q: The futex_q to enqueue
* @hb: The destination hash bucket
*
* The hb->lock must be held by the caller, and is released here. A call to
* queue_me() is typically paired with exactly one call to unqueue_me(). The
* exceptions involve the PI related operations, which may use unqueue_me_pi()
* or nothing if the unqueue is done as part of the wake process and the unqueue
* state is implicit in the state of woken task (see futex_wait_requeue_pi() for
* an example).
*/
static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
__releases(&hb->lock)
{
__queue_me(q, hb);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
spin_unlock(&hb->lock);
}
/**
* unqueue_me() - Remove the futex_q from its futex_hash_bucket
* @q: The futex_q to unqueue
*
* The q->lock_ptr must not be held by the caller. A call to unqueue_me() must
* be paired with exactly one earlier call to queue_me().
*
* Return:
* - 1 - if the futex_q was still queued (and we removed unqueued it);
* - 0 - if the futex_q was already removed by the waking thread
*/
static int unqueue_me(struct futex_q *q)
{
spinlock_t *lock_ptr;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
int ret = 0;
/* In the common case we don't take the spinlock, which is nice. */
retry:
/*
* q->lock_ptr can change between this read and the following spin_lock.
* Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and
* optimizing lock_ptr out of the logic below.
*/
lock_ptr = READ_ONCE(q->lock_ptr);
if (lock_ptr != NULL) {
spin_lock(lock_ptr);
/*
* q->lock_ptr can change between reading it and
* spin_lock(), causing us to take the wrong lock. This
* corrects the race condition.
*
* Reasoning goes like this: if we have the wrong lock,
* q->lock_ptr must have changed (maybe several times)
* between reading it and the spin_lock(). It can
* change again after the spin_lock() but only if it was
* already changed before the spin_lock(). It cannot,
* however, change back to the original value. Therefore
* we can detect whether we acquired the correct lock.
*/
if (unlikely(lock_ptr != q->lock_ptr)) {
spin_unlock(lock_ptr);
goto retry;
}
__unqueue_futex(q);
BUG_ON(q->pi_state);
spin_unlock(lock_ptr);
ret = 1;
}
return ret;
}
/*
* PI futexes can not be requeued and must remove themself from the
* hash bucket. The hash bucket lock (i.e. lock_ptr) is held on entry
* and dropped here.
*/
static void unqueue_me_pi(struct futex_q *q)
__releases(q->lock_ptr)
{
__unqueue_futex(q);
BUG_ON(!q->pi_state);
put_pi_state(q->pi_state);
q->pi_state = NULL;
spin_unlock(q->lock_ptr);
}
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
struct task_struct *argowner)
{
struct futex_pi_state *pi_state = q->pi_state;
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 uval, curval, newval;
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
struct task_struct *oldowner, *newowner;
u32 newtid;
int ret, err = 0;
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
lockdep_assert_held(q->lock_ptr);
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
oldowner = pi_state->owner;
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
/*
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
* We are here because either:
*
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
* - we stole the lock and pi_state->owner needs updating to reflect
* that (@argowner == current),
*
* or:
*
* - someone stole our lock and we need to fix things to point to the
* new owner (@argowner == NULL).
*
* Either way, we have to replace the TID in the user space variable.
rtmutex: Simplify PI algorithm and make highest prio task get lock In current rtmutex, the pending owner may be boosted by the tasks in the rtmutex's waitlist when the pending owner is deboosted or a task in the waitlist is boosted. This boosting is unrelated, because the pending owner does not really take the rtmutex. It is not reasonable. Example. time1: A(high prio) onwers the rtmutex. B(mid prio) and C (low prio) in the waitlist. time2 A release the lock, B becomes the pending owner A(or other high prio task) continues to run. B's prio is lower than A, so B is just queued at the runqueue. time3 A or other high prio task sleeps, but we have passed some time The B and C's prio are changed in the period (time2 ~ time3) due to boosting or deboosting. Now C has the priority higher than B. ***Is it reasonable that C has to boost B and help B to get the rtmutex? NO!! I think, it is unrelated/unneed boosting before B really owns the rtmutex. We should give C a chance to beat B and win the rtmutex. This is the motivation of this patch. This patch *ensures* only the top waiter or higher priority task can take the lock. How? 1) we don't dequeue the top waiter when unlock, if the top waiter is changed, the old top waiter will fail and go to sleep again. 2) when requiring lock, it will get the lock when the lock is not taken and: there is no waiter OR higher priority than waiters OR it is top waiter. 3) In any time, the top waiter is changed, the top waiter will be woken up. The algorithm is much simpler than before, no pending owner, no boosting for pending owner. Other advantage of this patch: 1) The states of a rtmutex are reduced a half, easier to read the code. 2) the codes become shorter. 3) top waiter is not dequeued until it really take the lock: they will retain FIFO when it is stolen. Not advantage nor disadvantage 1) Even we may wakeup multiple waiters(any time when top waiter changed), we hardly cause "thundering herd", the number of wokenup task is likely 1 or very little. 2) two APIs are changed. rt_mutex_owner() will not return pending owner, it will return NULL when the top waiter is going to take the lock. rt_mutex_next_owner() always return the top waiter. will not return NULL if we have waiters because the top waiter is not dequeued. I have fixed the code that use these APIs. need updated after this patch is accepted 1) Document/* 2) the testcase scripts/rt-tester/t4-l2-pi-deboost.tst Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com> LKML-Reference: <4D3012D5.4060709@cn.fujitsu.com> Reviewed-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2011-01-14 09:09:41 +00:00
* This must be atomic as we have to preserve the owner died bit here.
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
*
* Note: We write the user space value _before_ changing the pi_state
* because we can fault here. Imagine swapped out pages or a fork
* that marked all the anonymous memory readonly for cow.
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
*
* Modifying pi_state _before_ the user space value would leave the
* pi_state in an inconsistent state when we fault here, because we
* need to drop the locks to handle the fault. This might be observed
* in the PID check in lookup_pi_state.
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
*/
retry:
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
if (!argowner) {
if (oldowner != current) {
/*
* We raced against a concurrent self; things are
* already fixed up. Nothing to do.
*/
ret = 0;
goto out_unlock;
}
if (__rt_mutex_futex_trylock(&pi_state->pi_mutex)) {
/* We got the lock. pi_state is correct. Tell caller. */
ret = 1;
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
goto out_unlock;
}
/*
futex: Handle transient "ownerless" rtmutex state correctly Gratian managed to trigger the BUG_ON(!newowner) in fixup_pi_state_owner(). This is one possible chain of events leading to this: Task Prio Operation T1 120 lock(F) T2 120 lock(F) -> blocks (top waiter) T3 50 (RT) lock(F) -> boosts T1 and blocks (new top waiter) XX timeout/ -> wakes T2 signal T1 50 unlock(F) -> wakes T3 (rtmutex->owner == NULL, waiter bit is set) T2 120 cleanup -> try_to_take_mutex() fails because T3 is the top waiter and the lower priority T2 cannot steal the lock. -> fixup_pi_state_owner() sees newowner == NULL -> BUG_ON() The comment states that this is invalid and rt_mutex_real_owner() must return a non NULL owner when the trylock failed, but in case of a queued and woken up waiter rt_mutex_real_owner() == NULL is a valid transient state. The higher priority waiter has simply not yet managed to take over the rtmutex. The BUG_ON() is therefore wrong and this is just another retry condition in fixup_pi_state_owner(). Drop the locks, so that T3 can make progress, and then try the fixup again. Gratian provided a great analysis, traces and a reproducer. The analysis is to the point, but it confused the hell out of that tglx dude who had to page in all the futex horrors again. Condensed version is above. [ tglx: Wrote comment and changelog ] Fixes: c1e2f0eaf015 ("futex: Avoid violating the 10th rule of futex") Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/87a6w6x7bb.fsf@ni.com Link: https://lore.kernel.org/r/87sg9pkvf7.fsf@nanos.tec.linutronix.de
2020-11-04 15:12:44 +00:00
* The trylock just failed, so either there is an owner or
* there is a higher priority waiter than this one.
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
*/
newowner = rt_mutex_owner(&pi_state->pi_mutex);
futex: Handle transient "ownerless" rtmutex state correctly Gratian managed to trigger the BUG_ON(!newowner) in fixup_pi_state_owner(). This is one possible chain of events leading to this: Task Prio Operation T1 120 lock(F) T2 120 lock(F) -> blocks (top waiter) T3 50 (RT) lock(F) -> boosts T1 and blocks (new top waiter) XX timeout/ -> wakes T2 signal T1 50 unlock(F) -> wakes T3 (rtmutex->owner == NULL, waiter bit is set) T2 120 cleanup -> try_to_take_mutex() fails because T3 is the top waiter and the lower priority T2 cannot steal the lock. -> fixup_pi_state_owner() sees newowner == NULL -> BUG_ON() The comment states that this is invalid and rt_mutex_real_owner() must return a non NULL owner when the trylock failed, but in case of a queued and woken up waiter rt_mutex_real_owner() == NULL is a valid transient state. The higher priority waiter has simply not yet managed to take over the rtmutex. The BUG_ON() is therefore wrong and this is just another retry condition in fixup_pi_state_owner(). Drop the locks, so that T3 can make progress, and then try the fixup again. Gratian provided a great analysis, traces and a reproducer. The analysis is to the point, but it confused the hell out of that tglx dude who had to page in all the futex horrors again. Condensed version is above. [ tglx: Wrote comment and changelog ] Fixes: c1e2f0eaf015 ("futex: Avoid violating the 10th rule of futex") Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Mike Galbraith <efault@gmx.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: stable@vger.kernel.org Link: https://lore.kernel.org/r/87a6w6x7bb.fsf@ni.com Link: https://lore.kernel.org/r/87sg9pkvf7.fsf@nanos.tec.linutronix.de
2020-11-04 15:12:44 +00:00
/*
* If the higher priority waiter has not yet taken over the
* rtmutex then newowner is NULL. We can't return here with
* that state because it's inconsistent vs. the user space
* state. So drop the locks and try again. It's a valid
* situation and not any different from the other retry
* conditions.
*/
if (unlikely(!newowner)) {
err = -EAGAIN;
goto handle_err;
}
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
} else {
WARN_ON_ONCE(argowner != current);
if (oldowner == current) {
/*
* We raced against a concurrent self; things are
* already fixed up. Nothing to do.
*/
ret = 1;
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
goto out_unlock;
}
newowner = argowner;
}
newtid = task_pid_vnr(newowner) | FUTEX_WAITERS;
/* Owner died? */
if (!pi_state->owner)
newtid |= FUTEX_OWNER_DIED;
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
err = get_futex_value_locked(&uval, uaddr);
if (err)
goto handle_err;
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
for (;;) {
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
newval = (uval & FUTEX_OWNER_DIED) | newtid;
err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval);
if (err)
goto handle_err;
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
if (curval == uval)
break;
uval = curval;
}
/*
* We fixed up user space. Now we need to fix the pi_state
* itself.
*/
if (pi_state->owner != NULL) {
raw_spin_lock(&pi_state->owner->pi_lock);
WARN_ON(list_empty(&pi_state->list));
list_del_init(&pi_state->list);
raw_spin_unlock(&pi_state->owner->pi_lock);
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
}
futex: Prevent stale futex owner when interrupted/timeout Roland Westrelin did a great analysis of a long standing thinko in the return path of futex_lock_pi. While we fixed the lock steal case long ago, which was easy to trigger, we never had a test case which exposed this problem and stupidly never thought about the reverse lock stealing scenario and the return to user space with a stale state. When a blocked tasks returns from rt_mutex_timed_locked without holding the rt_mutex (due to a signal or timeout) and at the same time the task holding the futex is releasing the futex and assigning the ownership of the futex to the returning task, then it might happen that a third task acquires the rt_mutex before the final rt_mutex_trylock() of the returning task happens under the futex hash bucket lock. The returning task returns to user space with ETIMEOUT or EINTR, but the user space futex value is assigned to this task. The task which acquired the rt_mutex fixes the user space futex value right after the hash bucket lock has been released by the returning task, but for a short period of time the user space value is wrong. Detailed description is available at: https://bugzilla.redhat.com/show_bug.cgi?id=400541 The fix for this is the same as we do when the rt_mutex was acquired by a higher priority task via lock stealing from the designated new owner. In that case we already fix the user space value and the internal pi_state up before we return. This mechanism can be used to fixup the above corner case as well. When the returning task, which failed to acquire the rt_mutex, notices that it is the designated owner of the futex, then it fixes up the stale user space value and the pi_state, before returning to user space. This happens with the futex hash bucket lock held, so the task which acquired the rt_mutex is guaranteed to be blocked on the hash bucket lock. We can access the rt_mutex owner, which gives us the pid of the new owner, safely here as the owner is not able to modify (release) it while waiting on the hash bucket lock. Rename the "curr" argument of fixup_pi_state_owner() to "newowner" to avoid confusion with current and add the check for the stale state into the failure path of rt_mutex_trylock() in the return path of unlock_futex_pi(). If the situation is detected use fixup_pi_state_owner() to assign everything to the owner of the rt_mutex. Pointed-out-and-tested-by: Roland Westrelin <roland.westrelin@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-01-08 18:47:38 +00:00
pi_state->owner = newowner;
raw_spin_lock(&newowner->pi_lock);
WARN_ON(!list_empty(&pi_state->list));
futex: Prevent stale futex owner when interrupted/timeout Roland Westrelin did a great analysis of a long standing thinko in the return path of futex_lock_pi. While we fixed the lock steal case long ago, which was easy to trigger, we never had a test case which exposed this problem and stupidly never thought about the reverse lock stealing scenario and the return to user space with a stale state. When a blocked tasks returns from rt_mutex_timed_locked without holding the rt_mutex (due to a signal or timeout) and at the same time the task holding the futex is releasing the futex and assigning the ownership of the futex to the returning task, then it might happen that a third task acquires the rt_mutex before the final rt_mutex_trylock() of the returning task happens under the futex hash bucket lock. The returning task returns to user space with ETIMEOUT or EINTR, but the user space futex value is assigned to this task. The task which acquired the rt_mutex fixes the user space futex value right after the hash bucket lock has been released by the returning task, but for a short period of time the user space value is wrong. Detailed description is available at: https://bugzilla.redhat.com/show_bug.cgi?id=400541 The fix for this is the same as we do when the rt_mutex was acquired by a higher priority task via lock stealing from the designated new owner. In that case we already fix the user space value and the internal pi_state up before we return. This mechanism can be used to fixup the above corner case as well. When the returning task, which failed to acquire the rt_mutex, notices that it is the designated owner of the futex, then it fixes up the stale user space value and the pi_state, before returning to user space. This happens with the futex hash bucket lock held, so the task which acquired the rt_mutex is guaranteed to be blocked on the hash bucket lock. We can access the rt_mutex owner, which gives us the pid of the new owner, safely here as the owner is not able to modify (release) it while waiting on the hash bucket lock. Rename the "curr" argument of fixup_pi_state_owner() to "newowner" to avoid confusion with current and add the check for the stale state into the failure path of rt_mutex_trylock() in the return path of unlock_futex_pi(). If the situation is detected use fixup_pi_state_owner() to assign everything to the owner of the rt_mutex. Pointed-out-and-tested-by: Roland Westrelin <roland.westrelin@sun.com> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-01-08 18:47:38 +00:00
list_add(&pi_state->list, &newowner->pi_state_list);
raw_spin_unlock(&newowner->pi_lock);
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
return argowner == current;
/*
* In order to reschedule or handle a page fault, we need to drop the
* locks here. In the case of a fault, this gives the other task
* (either the highest priority waiter itself or the task which stole
* the rtmutex) the chance to try the fixup of the pi_state. So once we
* are back from handling the fault we need to check the pi_state after
* reacquiring the locks and before trying to do another fixup. When
* the fixup has been done already we simply return.
*
* Note: we hold both hb->lock and pi_mutex->wait_lock. We can safely
* drop hb->lock since the caller owns the hb -> futex_q relation.
* Dropping the pi_mutex->wait_lock requires the state revalidate.
*/
handle_err:
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
spin_unlock(q->lock_ptr);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
switch (err) {
case -EFAULT:
ret = fault_in_user_writeable(uaddr);
break;
case -EAGAIN:
cond_resched();
ret = 0;
break;
default:
WARN_ON_ONCE(1);
ret = err;
break;
}
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
spin_lock(q->lock_ptr);
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
/*
* Check if someone else fixed it for us:
*/
if (pi_state->owner != oldowner) {
ret = argowner == current;
goto out_unlock;
}
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
if (ret)
goto out_unlock;
futexes: fix fault handling in futex_lock_pi This patch addresses a very sporadic pi-futex related failure in highly threaded java apps on large SMP systems. David Holmes reported that the pi_state consistency check in lookup_pi_state triggered with his test application. This means that the kernel internal pi_state and the user space futex variable are out of sync. First we assumed that this is a user space data corruption, but deeper investigation revieled that the problem happend because the pi-futex code is not handling a fault in the futex_lock_pi path when the user space variable needs to be fixed up. The fault happens when a fork mapped the anon memory which contains the futex readonly for COW or the page got swapped out exactly between the unlock of the futex and the return of either the new futex owner or the task which was the expected owner but failed to acquire the kernel internal rtmutex. The current futex_lock_pi() code drops out with an inconsistent in case it faults and returns -EFAULT to user space. User space has no way to fixup that state. When we wrote this code we thought that we could not drop the hash bucket lock at this point to handle the fault. After analysing the code again it turned out to be wrong because there are only two tasks involved which might modify the pi_state and the user space variable: - the task which acquired the rtmutex - the pending owner of the pi_state which did not get the rtmutex Both tasks drop into the fixup_pi_state() function before returning to user space. The first task which acquired the hash bucket lock faults in the fixup of the user space variable, drops the spinlock and calls futex_handle_fault() to fault in the page. Now the second task could acquire the hash bucket lock and tries to fixup the user space variable as well. It either faults as well or it succeeds because the first task already faulted the page in. One caveat is to avoid a double fixup. After returning from the fault handling we reacquire the hash bucket lock and check whether the pi_state owner has been modified already. Reported-by: David Holmes <david.holmes@sun.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: David Holmes <david.holmes@sun.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: <stable@kernel.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> kernel/futex.c | 93 ++++++++++++++++++++++++++++++++++++++++++++------------- 1 file changed, 73 insertions(+), 20 deletions(-)
2008-06-23 09:21:58 +00:00
goto retry;
out_unlock:
raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock);
return ret;
}
static long futex_wait_restart(struct restart_block *restart);
/**
* fixup_owner() - Post lock pi_state and corner case management
* @uaddr: user address of the futex
* @q: futex_q (contains pi_state and access to the rt_mutex)
* @locked: if the attempt to take the rt_mutex succeeded (1) or not (0)
*
* After attempting to lock an rt_mutex, this function is called to cleanup
* the pi_state owner as well as handle race conditions that may allow us to
* acquire the lock. Must be called with the hb lock held.
*
* Return:
* - 1 - success, lock taken;
* - 0 - success, lock not taken;
* - <0 - on error (-EFAULT)
*/
static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked)
{
if (locked) {
/*
* Got the lock. We might not be the anticipated owner if we
* did a lock-steal - fix up the PI-state in that case:
*
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
* Speculative pi_state->owner read (we don't hold wait_lock);
* since we own the lock pi_state->owner == current is the
* stable state, anything else needs more attention.
*/
if (q->pi_state->owner != current)
return fixup_pi_state_owner(uaddr, q, current);
return 1;
}
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
/*
* If we didn't get the lock; check if anybody stole it from us. In
* that case, we need to fix up the uval to point to them instead of
* us, otherwise bad things happen. [10]
*
* Another speculative read; pi_state->owner == current is unstable
* but needs our attention.
*/
if (q->pi_state->owner == current)
return fixup_pi_state_owner(uaddr, q, NULL);
futex: Avoid violating the 10th rule of futex Julia reported futex state corruption in the following scenario: waiter waker stealer (prio > waiter) futex(WAIT_REQUEUE_PI, uaddr, uaddr2, timeout=[N ms]) futex_wait_requeue_pi() futex_wait_queue_me() freezable_schedule() <scheduled out> futex(LOCK_PI, uaddr2) futex(CMP_REQUEUE_PI, uaddr, uaddr2, 1, 0) /* requeues waiter to uaddr2 */ futex(UNLOCK_PI, uaddr2) wake_futex_pi() cmp_futex_value_locked(uaddr2, waiter) wake_up_q() <woken by waker> <hrtimer_wakeup() fires, clears sleeper->task> futex(LOCK_PI, uaddr2) __rt_mutex_start_proxy_lock() try_to_take_rt_mutex() /* steals lock */ rt_mutex_set_owner(lock, stealer) <preempted> <scheduled in> rt_mutex_wait_proxy_lock() __rt_mutex_slowlock() try_to_take_rt_mutex() /* fails, lock held by stealer */ if (timeout && !timeout->task) return -ETIMEDOUT; fixup_owner() /* lock wasn't acquired, so, fixup_pi_state_owner skipped */ return -ETIMEDOUT; /* At this point, we've returned -ETIMEDOUT to userspace, but the * futex word shows waiter to be the owner, and the pi_mutex has * stealer as the owner */ futex_lock(LOCK_PI, uaddr2) -> bails with EDEADLK, futex word says we're owner. And suggested that what commit: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") removes from fixup_owner() looks to be just what is needed. And indeed it is -- I completely missed that requeue_pi could also result in this case. So we need to restore that, except that subsequent patches, like commit: 16ffa12d7425 ("futex: Pull rt_mutex_futex_unlock() out from under hb->lock") changed all the locking rules. Even without that, the sequence: - if (rt_mutex_futex_trylock(&q->pi_state->pi_mutex)) { - locked = 1; - goto out; - } - raw_spin_lock_irq(&q->pi_state->pi_mutex.wait_lock); - owner = rt_mutex_owner(&q->pi_state->pi_mutex); - if (!owner) - owner = rt_mutex_next_owner(&q->pi_state->pi_mutex); - raw_spin_unlock_irq(&q->pi_state->pi_mutex.wait_lock); - ret = fixup_pi_state_owner(uaddr, q, owner); already suggests there were races; otherwise we'd never have to look at next_owner. So instead of doing 3 consecutive wait_lock sections with who knows what races, we do it all in a single section. Additionally, the usage of pi_state->owner in fixup_owner() was only safe because only the rt_mutex owner would modify it, which this additional case wrecks. Luckily the values can only change away and not to the value we're testing, this means we can do a speculative test and double check once we have the wait_lock. Fixes: 73d786bd043e ("futex: Rework inconsistent rt_mutex/futex_q state") Reported-by: Julia Cartwright <julia@ni.com> Reported-by: Gratian Crisan <gratian.crisan@ni.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Julia Cartwright <julia@ni.com> Tested-by: Gratian Crisan <gratian.crisan@ni.com> Cc: Darren Hart <dvhart@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/20171208124939.7livp7no2ov65rrc@hirez.programming.kicks-ass.net
2017-12-08 12:49:39 +00:00
/*
* Paranoia check. If we did not take the lock, then we should not be
* the owner of the rt_mutex. Warn and establish consistent state.
*/
if (WARN_ON_ONCE(rt_mutex_owner(&q->pi_state->pi_mutex) == current))
return fixup_pi_state_owner(uaddr, q, current);
return 0;
}
/**
* futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal
* @hb: the futex hash bucket, must be locked by the caller
* @q: the futex_q to queue up on
* @timeout: the prepared hrtimer_sleeper, or null for no timeout
*/
static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
struct hrtimer_sleeper *timeout)
{
/*
* The task state is guaranteed to be set before another task can
* wake it. set_current_state() is implemented using smp_store_mb() and
* queue_me() calls spin_unlock() upon completion, both serializing
* access to the hash list and forcing another memory barrier.
*/
set_current_state(TASK_INTERRUPTIBLE);
2009-09-22 05:30:38 +00:00
queue_me(q, hb);
/* Arm the timer */
if (timeout)
hrtimer_sleeper_start_expires(timeout, HRTIMER_MODE_ABS);
/*
2009-09-22 05:30:38 +00:00
* If we have been removed from the hash list, then another task
* has tried to wake us, and we can skip the call to schedule().
*/
if (likely(!plist_node_empty(&q->list))) {
/*
* If the timer has already expired, current will already be
* flagged for rescheduling. Only call schedule if there
* is no timeout, or if it has yet to expire.
*/
if (!timeout || timeout->task)
freezable_schedule();
}
__set_current_state(TASK_RUNNING);
}
/**
* futex_wait_setup() - Prepare to wait on a futex
* @uaddr: the futex userspace address
* @val: the expected value
* @flags: futex flags (FLAGS_SHARED, etc.)
* @q: the associated futex_q
* @hb: storage for hash_bucket pointer to be returned to caller
*
* Setup the futex_q and locate the hash_bucket. Get the futex value and
* compare it with the expected value. Handle atomic faults internally.
* Return with the hb lock held and a q.key reference on success, and unlocked
* with no q.key reference on failure.
*
* Return:
* - 0 - uaddr contains val and hb has been locked;
* - <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked
*/
static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags,
struct futex_q *q, struct futex_hash_bucket **hb)
{
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
u32 uval;
int ret;
/*
* Access the page AFTER the hash-bucket is locked.
* Order is important:
*
* Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
* Userspace waker: if (cond(var)) { var = new; futex_wake(&var); }
*
* The basic logical guarantee of a futex is that it blocks ONLY
* if cond(var) is known to be true at the time of blocking, for
* any cond. If we locked the hash-bucket after testing *uaddr, that
* would open a race condition where we could block indefinitely with
* cond(var) false, which would violate the guarantee.
*
* On the other hand, we insert q and release the hash-bucket only
* after testing *uaddr. This guarantees that futex_wait() will NOT
* absorb a wakeup if *uaddr does not match the desired values
* while the syscall executes.
*/
retry:
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, FUTEX_READ);
if (unlikely(ret != 0))
return ret;
retry_private:
*hb = queue_lock(q);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
ret = get_futex_value_locked(&uval, uaddr);
if (ret) {
queue_unlock(*hb);
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
ret = get_user(uval, uaddr);
if (ret)
return ret;
if (!(flags & FLAGS_SHARED))
goto retry_private;
goto retry;
}
if (uval != val) {
queue_unlock(*hb);
ret = -EWOULDBLOCK;
}
return ret;
}
static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
ktime_t *abs_time, u32 bitset)
{
struct hrtimer_sleeper timeout, *to;
struct restart_block *restart;
struct futex_hash_bucket *hb;
struct futex_q q = futex_q_init;
int ret;
if (!bitset)
return -EINVAL;
q.bitset = bitset;
to = futex_setup_timer(abs_time, &timeout, flags,
current->timer_slack_ns);
retry:
/*
* Prepare to wait on uaddr. On success, holds hb lock and increments
* q.key refs.
*/
ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
if (ret)
goto out;
/* queue_me and wait for wakeup, timeout, or a signal. */
futex_wait_queue_me(hb, &q, to);
/* If we were woken (and unqueued), we succeeded, whatever. */
ret = 0;
/* unqueue_me() drops q.key ref */
if (!unqueue_me(&q))
goto out;
ret = -ETIMEDOUT;
if (to && !to->task)
goto out;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
/*
* We expect signal_pending(current), but we might be the
* victim of a spurious wakeup as well.
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
*/
if (!signal_pending(current))
goto retry;
ret = -ERESTARTSYS;
if (!abs_time)
goto out;
all arches, signal: move restart_block to struct task_struct If an attacker can cause a controlled kernel stack overflow, overwriting the restart block is a very juicy exploit target. This is because the restart_block is held in the same memory allocation as the kernel stack. Moving the restart block to struct task_struct prevents this exploit by making the restart_block harder to locate. Note that there are other fields in thread_info that are also easy targets, at least on some architectures. It's also a decent simplification, since the restart code is more or less identical on all architectures. [james.hogan@imgtec.com: metag: align thread_info::supervisor_stack] Signed-off-by: Andy Lutomirski <luto@amacapital.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Kees Cook <keescook@chromium.org> Cc: David Miller <davem@davemloft.net> Acked-by: Richard Weinberger <richard@nod.at> Cc: Richard Henderson <rth@twiddle.net> Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru> Cc: Matt Turner <mattst88@gmail.com> Cc: Vineet Gupta <vgupta@synopsys.com> Cc: Russell King <rmk@arm.linux.org.uk> Cc: Catalin Marinas <catalin.marinas@arm.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Hans-Christian Egtvedt <egtvedt@samfundet.no> Cc: Steven Miao <realmz6@gmail.com> Cc: Mark Salter <msalter@redhat.com> Cc: Aurelien Jacquiot <a-jacquiot@ti.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Jesper Nilsson <jesper.nilsson@axis.com> Cc: David Howells <dhowells@redhat.com> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Geert Uytterhoeven <geert@linux-m68k.org> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@samba.org> Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Tested-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc) Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Chen Liqin <liqin.linux@gmail.com> Cc: Lennox Wu <lennox.wu@gmail.com> Cc: Chris Metcalf <cmetcalf@ezchip.com> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn> Cc: Chris Zankel <chris@zankel.net> Cc: Max Filippov <jcmvbkbc@gmail.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Guenter Roeck <linux@roeck-us.net> Signed-off-by: James Hogan <james.hogan@imgtec.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-02-12 23:01:14 +00:00
restart = &current->restart_block;
restart->fn = futex_wait_restart;
restart->futex.uaddr = uaddr;
restart->futex.val = val;
restart->futex.time = *abs_time;
restart->futex.bitset = bitset;
restart->futex.flags = flags | FLAGS_HAS_TIMEOUT;
ret = -ERESTART_RESTARTBLOCK;
out:
if (to) {
hrtimer_cancel(&to->timer);
destroy_hrtimer_on_stack(&to->timer);
}
return ret;
}
static long futex_wait_restart(struct restart_block *restart)
{
u32 __user *uaddr = restart->futex.uaddr;
ktime_t t, *tp = NULL;
if (restart->futex.flags & FLAGS_HAS_TIMEOUT) {
t = restart->futex.time;
tp = &t;
}
restart->fn = do_no_restart_syscall;
return (long)futex_wait(uaddr, restart->futex.flags,
restart->futex.val, tp, restart->futex.bitset);
}
/*
* Userspace tried a 0 -> TID atomic transition of the futex value
* and failed. The kernel side here does the whole locking operation:
* if there are waiters then it will block as a consequence of relying
* on rt-mutexes, it does PI, etc. (Due to races the kernel might see
* a 0 value of the futex too.).
*
* Also serves as futex trylock_pi()'ing, and due semantics.
*/
futex: Fix argument handling in futex_lock_pi() calls This patch fixes two separate buglets in calls to futex_lock_pi(): * Eliminate unused 'detect' argument * Change unused 'timeout' argument of FUTEX_TRYLOCK_PI to NULL The 'detect' argument of futex_lock_pi() seems never to have been used (when it was included with the initial PI mutex implementation in Linux 2.6.18, all checks against its value were disabled by ANDing against 0 (i.e., if (detect... && 0)), and with commit 778e9a9c3e7193ea9f434f382947155ffb59c755, any mention of this argument in futex_lock_pi() went way altogether. Its presence now serves only to confuse readers of the code, by giving the impression that the futex() FUTEX_LOCK_PI operation actually does use the 'val' argument. This patch removes the argument. The futex_lock_pi() call that corresponds to FUTEX_TRYLOCK_PI includes 'timeout' as one of its arguments. This misleads the reader into thinking that the FUTEX_TRYLOCK_PI operation does employ timeouts for some sensible purpose; but it does not. Indeed, it cannot, because the checks at the start of sys_futex() exclude FUTEX_TRYLOCK_PI from the set of operations that do copy_from_user() on the timeout argument. So, in the FUTEX_TRYLOCK_PI futex_lock_pi() call it would be simplest to change 'timeout' to 'NULL'. This patch does that. Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Darren Hart <darren@dvhart.com> Link: http://lkml.kernel.org/r/54B96646.8010200@gmail.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2015-01-16 19:28:06 +00:00
static int futex_lock_pi(u32 __user *uaddr, unsigned int flags,
ktime_t *time, int trylock)
{
struct hrtimer_sleeper timeout, *to;
struct futex_pi_state *pi_state = NULL;
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
struct task_struct *exiting = NULL;
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
struct rt_mutex_waiter rt_waiter;
struct futex_hash_bucket *hb;
struct futex_q q = futex_q_init;
int res, ret;
if (!IS_ENABLED(CONFIG_FUTEX_PI))
return -ENOSYS;
if (refill_pi_state_cache())
return -ENOMEM;
to = futex_setup_timer(time, &timeout, FLAGS_CLOCKRT, 0);
retry:
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, FUTEX_WRITE);
if (unlikely(ret != 0))
goto out;
retry_private:
hb = queue_lock(&q);
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current,
&exiting, 0);
if (unlikely(ret)) {
/*
* Atomic work succeeded and we got the lock,
* or failed. Either way, we do _not_ block.
*/
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
switch (ret) {
case 1:
/* We got the lock. */
ret = 0;
goto out_unlock_put_key;
case -EFAULT:
goto uaddr_faulted;
case -EBUSY:
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
case -EAGAIN:
/*
* Two reasons for this:
* - EBUSY: Task is exiting and we just wait for the
* exit to complete.
* - EAGAIN: The user space value changed.
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
*/
queue_unlock(hb);
futex: Prevent exit livelock Oleg provided the following test case: int main(void) { struct sched_param sp = {}; sp.sched_priority = 2; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); int lock = vfork(); if (!lock) { sp.sched_priority = 1; assert(sched_setscheduler(0, SCHED_FIFO, &sp) == 0); _exit(0); } syscall(__NR_futex, &lock, FUTEX_LOCK_PI, 0,0,0); return 0; } This creates an unkillable RT process spinning in futex_lock_pi() on a UP machine or if the process is affine to a single CPU. The reason is: parent child set FIFO prio 2 vfork() -> set FIFO prio 1 implies wait_for_child() sched_setscheduler(...) exit() do_exit() .... mm_release() tsk->futex_state = FUTEX_STATE_EXITING; exit_futex(); (NOOP in this case) complete() --> wakes parent sys_futex() loop infinite because tsk->futex_state == FUTEX_STATE_EXITING The same problem can happen just by regular preemption as well: task holds futex ... do_exit() tsk->futex_state = FUTEX_STATE_EXITING; --> preemption (unrelated wakeup of some other higher prio task, e.g. timer) switch_to(other_task) return to user sys_futex() loop infinite as above Just for the fun of it the futex exit cleanup could trigger the wakeup itself before the task sets its futex state to DEAD. To cure this, the handling of the exiting owner is changed so: - A refcount is held on the task - The task pointer is stored in a caller visible location - The caller drops all locks (hash bucket, mmap_sem) and blocks on task::futex_exit_mutex. When the mutex is acquired then the exiting task has completed the cleanup and the state is consistent and can be reevaluated. This is not a pretty solution, but there is no choice other than returning an error code to user space, which would break the state consistency guarantee and open another can of problems including regressions. For stable backports the preparatory commits ac31c7ff8624 .. ba31c1a48538 are required as well, but for anything older than 5.3.y the backports are going to be provided when this hits mainline as the other dependencies for those kernels are definitely not stable material. Fixes: 778e9a9c3e71 ("pi-futex: fix exit races and locking problems") Reported-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: Stable Team <stable@vger.kernel.org> Link: https://lkml.kernel.org/r/20191106224557.041676471@linutronix.de
2019-11-06 21:55:46 +00:00
/*
* Handle the case where the owner is in the middle of
* exiting. Wait for the exit to complete otherwise
* this task might loop forever, aka. live lock.
*/
wait_for_owner_exiting(ret, exiting);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
cond_resched();
goto retry;
default:
goto out_unlock_put_key;
}
}
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
WARN_ON(!q.pi_state);
/*
* Only actually queue now that the atomic ops are done:
*/
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
__queue_me(&q, hb);
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
if (trylock) {
ret = rt_mutex_futex_trylock(&q.pi_state->pi_mutex);
/* Fixup the trylock return value: */
ret = ret ? 0 : -EWOULDBLOCK;
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
goto no_block;
}
futex: Drop hb->lock before enqueueing on the rtmutex When PREEMPT_RT_FULL does the spinlock -> rt_mutex substitution the PI chain code will (falsely) report a deadlock and BUG. The problem is that it hold hb->lock (now an rt_mutex) while doing task_blocks_on_rt_mutex on the futex's pi_state::rtmutex. This, when interleaved just right with futex_unlock_pi() leads it to believe to see an AB-BA deadlock. Task1 (holds rt_mutex, Task2 (does FUTEX_LOCK_PI) does FUTEX_UNLOCK_PI) lock hb->lock lock rt_mutex (as per start_proxy) lock hb->lock Which is a trivial AB-BA. It is not an actual deadlock, because it won't be holding hb->lock by the time it actually blocks on the rt_mutex, but the chainwalk code doesn't know that and it would be a nightmare to handle this gracefully. To avoid this problem, do the same as in futex_unlock_pi() and drop hb->lock after acquiring wait_lock. This still fully serializes against futex_unlock_pi(), since adding to the wait_list does the very same lock dance, and removing it holds both locks. Aside of solving the RT problem this makes the lock and unlock mechanism symetric and reduces the hb->lock held time. Reported-and-tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Suggested-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.161341537@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:36:00 +00:00
rt_mutex_init_waiter(&rt_waiter);
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
/*
futex: Drop hb->lock before enqueueing on the rtmutex When PREEMPT_RT_FULL does the spinlock -> rt_mutex substitution the PI chain code will (falsely) report a deadlock and BUG. The problem is that it hold hb->lock (now an rt_mutex) while doing task_blocks_on_rt_mutex on the futex's pi_state::rtmutex. This, when interleaved just right with futex_unlock_pi() leads it to believe to see an AB-BA deadlock. Task1 (holds rt_mutex, Task2 (does FUTEX_LOCK_PI) does FUTEX_UNLOCK_PI) lock hb->lock lock rt_mutex (as per start_proxy) lock hb->lock Which is a trivial AB-BA. It is not an actual deadlock, because it won't be holding hb->lock by the time it actually blocks on the rt_mutex, but the chainwalk code doesn't know that and it would be a nightmare to handle this gracefully. To avoid this problem, do the same as in futex_unlock_pi() and drop hb->lock after acquiring wait_lock. This still fully serializes against futex_unlock_pi(), since adding to the wait_list does the very same lock dance, and removing it holds both locks. Aside of solving the RT problem this makes the lock and unlock mechanism symetric and reduces the hb->lock held time. Reported-and-tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Suggested-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.161341537@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:36:00 +00:00
* On PREEMPT_RT_FULL, when hb->lock becomes an rt_mutex, we must not
* hold it while doing rt_mutex_start_proxy(), because then it will
* include hb->lock in the blocking chain, even through we'll not in
* fact hold it while blocking. This will lead it to report -EDEADLK
* and BUG when futex_unlock_pi() interleaves with this.
*
* Therefore acquire wait_lock while holding hb->lock, but drop the
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
* latter before calling __rt_mutex_start_proxy_lock(). This
* interleaves with futex_unlock_pi() -- which does a similar lock
* handoff -- such that the latter can observe the futex_q::pi_state
* before __rt_mutex_start_proxy_lock() is done.
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
*/
futex: Drop hb->lock before enqueueing on the rtmutex When PREEMPT_RT_FULL does the spinlock -> rt_mutex substitution the PI chain code will (falsely) report a deadlock and BUG. The problem is that it hold hb->lock (now an rt_mutex) while doing task_blocks_on_rt_mutex on the futex's pi_state::rtmutex. This, when interleaved just right with futex_unlock_pi() leads it to believe to see an AB-BA deadlock. Task1 (holds rt_mutex, Task2 (does FUTEX_LOCK_PI) does FUTEX_UNLOCK_PI) lock hb->lock lock rt_mutex (as per start_proxy) lock hb->lock Which is a trivial AB-BA. It is not an actual deadlock, because it won't be holding hb->lock by the time it actually blocks on the rt_mutex, but the chainwalk code doesn't know that and it would be a nightmare to handle this gracefully. To avoid this problem, do the same as in futex_unlock_pi() and drop hb->lock after acquiring wait_lock. This still fully serializes against futex_unlock_pi(), since adding to the wait_list does the very same lock dance, and removing it holds both locks. Aside of solving the RT problem this makes the lock and unlock mechanism symetric and reduces the hb->lock held time. Reported-and-tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Suggested-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.161341537@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:36:00 +00:00
raw_spin_lock_irq(&q.pi_state->pi_mutex.wait_lock);
spin_unlock(q.lock_ptr);
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
/*
* __rt_mutex_start_proxy_lock() unconditionally enqueues the @rt_waiter
* such that futex_unlock_pi() is guaranteed to observe the waiter when
* it sees the futex_q::pi_state.
*/
futex: Drop hb->lock before enqueueing on the rtmutex When PREEMPT_RT_FULL does the spinlock -> rt_mutex substitution the PI chain code will (falsely) report a deadlock and BUG. The problem is that it hold hb->lock (now an rt_mutex) while doing task_blocks_on_rt_mutex on the futex's pi_state::rtmutex. This, when interleaved just right with futex_unlock_pi() leads it to believe to see an AB-BA deadlock. Task1 (holds rt_mutex, Task2 (does FUTEX_LOCK_PI) does FUTEX_UNLOCK_PI) lock hb->lock lock rt_mutex (as per start_proxy) lock hb->lock Which is a trivial AB-BA. It is not an actual deadlock, because it won't be holding hb->lock by the time it actually blocks on the rt_mutex, but the chainwalk code doesn't know that and it would be a nightmare to handle this gracefully. To avoid this problem, do the same as in futex_unlock_pi() and drop hb->lock after acquiring wait_lock. This still fully serializes against futex_unlock_pi(), since adding to the wait_list does the very same lock dance, and removing it holds both locks. Aside of solving the RT problem this makes the lock and unlock mechanism symetric and reduces the hb->lock held time. Reported-and-tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Suggested-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.161341537@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:36:00 +00:00
ret = __rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current);
raw_spin_unlock_irq(&q.pi_state->pi_mutex.wait_lock);
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
if (ret) {
if (ret == 1)
ret = 0;
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
goto cleanup;
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
}
if (unlikely(to))
hrtimer_sleeper_start_expires(to, HRTIMER_MODE_ABS);
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
cleanup:
spin_lock(q.lock_ptr);
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
/*
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
* If we failed to acquire the lock (deadlock/signal/timeout), we must
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
* first acquire the hb->lock before removing the lock from the
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
* rt_mutex waitqueue, such that we can keep the hb and rt_mutex wait
* lists consistent.
futex: Drop hb->lock before enqueueing on the rtmutex When PREEMPT_RT_FULL does the spinlock -> rt_mutex substitution the PI chain code will (falsely) report a deadlock and BUG. The problem is that it hold hb->lock (now an rt_mutex) while doing task_blocks_on_rt_mutex on the futex's pi_state::rtmutex. This, when interleaved just right with futex_unlock_pi() leads it to believe to see an AB-BA deadlock. Task1 (holds rt_mutex, Task2 (does FUTEX_LOCK_PI) does FUTEX_UNLOCK_PI) lock hb->lock lock rt_mutex (as per start_proxy) lock hb->lock Which is a trivial AB-BA. It is not an actual deadlock, because it won't be holding hb->lock by the time it actually blocks on the rt_mutex, but the chainwalk code doesn't know that and it would be a nightmare to handle this gracefully. To avoid this problem, do the same as in futex_unlock_pi() and drop hb->lock after acquiring wait_lock. This still fully serializes against futex_unlock_pi(), since adding to the wait_list does the very same lock dance, and removing it holds both locks. Aside of solving the RT problem this makes the lock and unlock mechanism symetric and reduces the hb->lock held time. Reported-and-tested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Suggested-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.161341537@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:36:00 +00:00
*
* In particular; it is important that futex_unlock_pi() can not
* observe this inconsistency.
futex: Rework futex_lock_pi() to use rt_mutex_*_proxy_lock() By changing futex_lock_pi() to use rt_mutex_*_proxy_lock() all wait_list modifications are done under both hb->lock and wait_lock. This closes the obvious interleave pattern between futex_lock_pi() and futex_unlock_pi(), but not entirely so. See below: Before: futex_lock_pi() futex_unlock_pi() unlock hb->lock lock hb->lock unlock hb->lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock schedule() lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock <idem> -EAGAIN lock hb->lock After: futex_lock_pi() futex_unlock_pi() lock hb->lock lock rt_mutex->wait_lock list_add unlock rt_mutex->wait_lock unlock hb->lock schedule() lock hb->lock unlock hb->lock lock hb->lock lock rt_mutex->wait_lock list_del unlock rt_mutex->wait_lock lock rt_mutex->wait_lock unlock rt_mutex_wait_lock -EAGAIN unlock hb->lock It does however solve the earlier starvation/live-lock scenario which got introduced with the -EAGAIN since unlike the before scenario; where the -EAGAIN happens while futex_unlock_pi() doesn't hold any locks; in the after scenario it happens while futex_unlock_pi() actually holds a lock, and then it is serialized on that lock. Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: juri.lelli@arm.com Cc: bigeasy@linutronix.de Cc: xlpang@redhat.com Cc: rostedt@goodmis.org Cc: mathieu.desnoyers@efficios.com Cc: jdesfossez@efficios.com Cc: dvhart@infradead.org Cc: bristot@redhat.com Link: http://lkml.kernel.org/r/20170322104152.062785528@infradead.org Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2017-03-22 10:35:58 +00:00
*/
if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter))
ret = 0;
no_block:
/*
* Fixup the pi_state owner and possibly acquire the lock if we
* haven't already.
*/
res = fixup_owner(uaddr, &q, !ret);
/*
* If fixup_owner() returned an error, proprogate that. If it acquired
* the lock, clear our -ETIMEDOUT or -EINTR.
*/
if (res)
ret = (res < 0) ? res : 0;
/*
* If fixup_owner() faulted and was unable to handle the fault, unlock
* it and return the fault to userspace.
*/
if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current)) {
pi_state = q.pi_state;
get_pi_state(pi_state);
}
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
/* Unqueue and drop the lock */
unqueue_me_pi(&q);
if (pi_state) {
rt_mutex_futex_unlock(&pi_state->pi_mutex);
put_pi_state(pi_state);
}
goto out;
out_unlock_put_key:
queue_unlock(hb);
out:
if (to) {
hrtimer_cancel(&to->timer);
destroy_hrtimer_on_stack(&to->timer);
}
return ret != -EINTR ? ret : -ERESTARTNOINTR;
uaddr_faulted:
queue_unlock(hb);
pi-futex: fix exit races and locking problems 1. New entries can be added to tsk->pi_state_list after task completed exit_pi_state_list(). The result is memory leakage and deadlocks. 2. handle_mm_fault() is called under spinlock. The result is obvious. 3. results in self-inflicted deadlock inside glibc. Sometimes futex_lock_pi returns -ESRCH, when it is not expected and glibc enters to for(;;) sleep() to simulate deadlock. This problem is quite obvious and I think the patch is right. Though it looks like each "if" in futex_lock_pi() got some stupid special case "else if". :-) 4. sometimes futex_lock_pi() returns -EDEADLK, when nobody has the lock. The reason is also obvious (see comment in the patch), but correct fix is far beyond my comprehension. I guess someone already saw this, the chunk: if (rt_mutex_trylock(&q.pi_state->pi_mutex)) ret = 0; is obviously from the same opera. But it does not work, because the rtmutex is really taken at this point: wake_futex_pi() of previous owner reassigned it to us. My fix works. But it looks very stupid. I would think about removal of shift of ownership in wake_futex_pi() and making all the work in context of process taking lock. From: Thomas Gleixner <tglx@linutronix.de> Fix 1) Avoid the tasklist lock variant of the exit race fix by adding an additional state transition to the exit code. This fixes also the issue, when a task with recursive segfaults is not able to release the futexes. Fix 2) Cleanup the lookup_pi_state() failure path and solve the -ESRCH problem finally. Fix 3) Solve the fixup_pi_state_owner() problem which needs to do the fixup in the lock protected section by using the in_atomic userspace access functions. This removes also the ugly lock drop / unqueue inside of fixup_pi_state() Fix 4) Fix a stale lock in the error path of futex_wake_pi() Added some error checks for verification. The -EDEADLK problem is solved by the rtmutex fixups. Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Eric Dumazet <dada1@cosmosbay.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-08 20:47:00 +00:00
ret = fault_in_user_writeable(uaddr);
if (ret)
goto out;
if (!(flags & FLAGS_SHARED))
goto retry_private;
goto retry;
}
/*
* Userspace attempted a TID -> 0 atomic transition, and failed.
* This is the in-kernel slowpath: we look up the PI state (if any),
* and do the rt-mutex unlock.
*/
static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags)
{
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 curval, uval, vpid = task_pid_vnr(current);
union futex_key key = FUTEX_KEY_INIT;
struct futex_hash_bucket *hb;
struct futex_q *top_waiter;
int ret;
if (!IS_ENABLED(CONFIG_FUTEX_PI))
return -ENOSYS;
retry:
if (get_user(uval, uaddr))
return -EFAULT;
/*
* We release only a lock we actually own:
*/
if ((uval & FUTEX_TID_MASK) != vpid)
return -EPERM;
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, FUTEX_WRITE);
if (ret)
return ret;
hb = hash_futex(&key);
spin_lock(&hb->lock);
/*
* Check waiters first. We do not trust user space values at
* all and we at least want to know if user space fiddled
* with the futex value instead of blindly unlocking.
*/
top_waiter = futex_top_waiter(hb, &key);
if (top_waiter) {
struct futex_pi_state *pi_state = top_waiter->pi_state;
ret = -EINVAL;
if (!pi_state)
goto out_unlock;
/*
* If current does not own the pi_state then the futex is
* inconsistent and user space fiddled with the futex value.
*/
if (pi_state->owner != current)
goto out_unlock;
get_pi_state(pi_state);
/*
* By taking wait_lock while still holding hb->lock, we ensure
* there is no point where we hold neither; and therefore
* wake_futex_pi() must observe a state consistent with what we
* observed.
futex: Handle early deadlock return correctly commit 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") changed the locking rules in the futex code so that the hash bucket lock is not longer held while the waiter is enqueued into the rtmutex wait list. This made the lock and the unlock path symmetric, but unfortunately the possible early exit from __rt_mutex_proxy_start() due to a detected deadlock was not updated accordingly. That allows a concurrent unlocker to observe inconsitent state which triggers the warning in the unlock path. futex_lock_pi() futex_unlock_pi() lock(hb->lock) queue(hb_waiter) lock(hb->lock) lock(rtmutex->wait_lock) unlock(hb->lock) // acquired hb->lock hb_waiter = futex_top_waiter() lock(rtmutex->wait_lock) __rt_mutex_proxy_start() ---> fail remove(rtmutex_waiter); ---> returns -EDEADLOCK unlock(rtmutex->wait_lock) // acquired wait_lock wake_futex_pi() rt_mutex_next_owner() --> returns NULL --> WARN lock(hb->lock) unqueue(hb_waiter) The problem is caused by the remove(rtmutex_waiter) in the failure case of __rt_mutex_proxy_start() as this lets the unlocker observe a waiter in the hash bucket but no waiter on the rtmutex, i.e. inconsistent state. The original commit handles this correctly for the other early return cases (timeout, signal) by delaying the removal of the rtmutex waiter until the returning task reacquired the hash bucket lock. Treat the failure case of __rt_mutex_proxy_start() in the same way and let the existing cleanup code handle the eventual handover of the rtmutex gracefully. The regular rt_mutex_proxy_start() gains the rtmutex waiter removal for the failure case, so that the other callsites are still operating correctly. Add proper comments to the code so all these details are fully documented. Thanks to Peter for helping with the analysis and writing the really valuable code comments. Fixes: 56222b212e8e ("futex: Drop hb->lock before enqueueing on the rtmutex") Reported-by: Heiko Carstens <heiko.carstens@de.ibm.com> Co-developed-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Peter Zijlstra <peterz@infradead.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: linux-s390@vger.kernel.org Cc: Stefan Liebler <stli@linux.ibm.com> Cc: Sebastian Sewior <bigeasy@linutronix.de> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/alpine.DEB.2.21.1901292311410.1950@nanos.tec.linutronix.de
2019-01-29 22:15:12 +00:00
*
* In particular; this forces __rt_mutex_start_proxy() to
* complete such that we're guaranteed to observe the
* rt_waiter. Also see the WARN in wake_futex_pi().
*/
raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock);
spin_unlock(&hb->lock);
/* drops pi_state->pi_mutex.wait_lock */
ret = wake_futex_pi(uaddr, uval, pi_state);
put_pi_state(pi_state);
/*
* Success, we're done! No tricky corner cases.
*/
if (!ret)
goto out_putkey;
/*
* The atomic access to the futex value generated a
* pagefault, so retry the user-access and the wakeup:
*/
if (ret == -EFAULT)
goto pi_faulted;
/*
* A unconditional UNLOCK_PI op raced against a waiter
* setting the FUTEX_WAITERS bit. Try again.
*/
if (ret == -EAGAIN)
goto pi_retry;
/*
* wake_futex_pi has detected invalid state. Tell user
* space.
*/
goto out_putkey;
}
/*
* We have no kernel internal state, i.e. no waiters in the
* kernel. Waiters which are about to queue themselves are stuck
* on hb->lock. So we can safely ignore them. We do neither
* preserve the WAITERS bit not the OWNER_DIED one. We are the
* owner.
*/
if ((ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))) {
spin_unlock(&hb->lock);
switch (ret) {
case -EFAULT:
goto pi_faulted;
case -EAGAIN:
goto pi_retry;
default:
WARN_ON_ONCE(1);
goto out_putkey;
}
}
/*
* If uval has changed, let user space handle it.
*/
ret = (curval == uval) ? 0 : -EAGAIN;
out_unlock:
spin_unlock(&hb->lock);
out_putkey:
return ret;
pi_retry:
cond_resched();
goto retry;
pi_faulted:
ret = fault_in_user_writeable(uaddr);
if (!ret)
goto retry;
return ret;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/**
* handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
* @hb: the hash_bucket futex_q was original enqueued on
* @q: the futex_q woken while waiting to be requeued
* @key2: the futex_key of the requeue target futex
* @timeout: the timeout associated with the wait (NULL if none)
*
* Detect if the task was woken on the initial futex as opposed to the requeue
* target futex. If so, determine if it was a timeout or a signal that caused
* the wakeup and return the appropriate error code to the caller. Must be
* called with the hb lock held.
*
* Return:
* - 0 = no early wakeup detected;
* - <0 = -ETIMEDOUT or -ERESTARTNOINTR
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
static inline
int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
struct futex_q *q, union futex_key *key2,
struct hrtimer_sleeper *timeout)
{
int ret = 0;
/*
* With the hb lock held, we avoid races while we process the wakeup.
* We only need to hold hb (and not hb2) to ensure atomicity as the
* wakeup code can't change q.key from uaddr to uaddr2 if we hold hb.
* It can't be requeued from uaddr2 to something else since we don't
* support a PI aware source futex for requeue.
*/
if (!match_futex(&q->key, key2)) {
WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
/*
* We were woken prior to requeue by a timeout or a signal.
* Unqueue the futex_q and determine which it was.
*/
plist_del(&q->list, &hb->chain);
hb_waiters_dec(hb);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/* Handle spurious wakeups gracefully */
ret = -EWOULDBLOCK;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (timeout && !timeout->task)
ret = -ETIMEDOUT;
else if (signal_pending(current))
ret = -ERESTARTNOINTR;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
return ret;
}
/**
* futex_wait_requeue_pi() - Wait on uaddr and take uaddr2
* @uaddr: the futex we initially wait on (non-pi)
* @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be
* the same type, no requeueing from private to shared, etc.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
* @val: the expected value of uaddr
* @abs_time: absolute timeout
* @bitset: 32 bit wakeup bitset set by userspace, defaults to all
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
* @uaddr2: the pi futex we will take prior to returning to user-space
*
* The caller will wait on uaddr and will be requeued by futex_requeue() to
* uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake
* on uaddr2 and complete the acquisition of the rt_mutex prior to returning to
* userspace. This ensures the rt_mutex maintains an owner when it has waiters;
* without one, the pi logic would not know which task to boost/deboost, if
* there was a need to.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* We call schedule in futex_wait_queue_me() when we enqueue and return there
* via the following--
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
* 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue()
* 2) wakeup on uaddr2 after a requeue
* 3) signal
* 4) timeout
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* If 3, cleanup and return -ERESTARTNOINTR.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* If 2, we may then block on trying to take the rt_mutex and return via:
* 5) successful lock
* 6) signal
* 7) timeout
* 8) other lock acquisition failure
*
* If 6, return -EWOULDBLOCK (restarting the syscall would do the same).
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*
* If 4 or 7, we cleanup and return with -ETIMEDOUT.
*
* Return:
* - 0 - On success;
* - <0 - On error
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
u32 val, ktime_t *abs_time, u32 bitset,
u32 __user *uaddr2)
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
{
struct hrtimer_sleeper timeout, *to;
struct futex_pi_state *pi_state = NULL;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
struct rt_mutex_waiter rt_waiter;
struct futex_hash_bucket *hb;
union futex_key key2 = FUTEX_KEY_INIT;
struct futex_q q = futex_q_init;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
int res, ret;
if (!IS_ENABLED(CONFIG_FUTEX_PI))
return -ENOSYS;
if (uaddr == uaddr2)
return -EINVAL;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (!bitset)
return -EINVAL;
to = futex_setup_timer(abs_time, &timeout, flags,
current->timer_slack_ns);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* The waiter is allocated on our stack, manipulated by the requeue
* code while we sleep on uaddr.
*/
rt_mutex_init_waiter(&rt_waiter);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
Remove 'type' argument from access_ok() function Nobody has actually used the type (VERIFY_READ vs VERIFY_WRITE) argument of the user address range verification function since we got rid of the old racy i386-only code to walk page tables by hand. It existed because the original 80386 would not honor the write protect bit when in kernel mode, so you had to do COW by hand before doing any user access. But we haven't supported that in a long time, and these days the 'type' argument is a purely historical artifact. A discussion about extending 'user_access_begin()' to do the range checking resulted this patch, because there is no way we're going to move the old VERIFY_xyz interface to that model. And it's best done at the end of the merge window when I've done most of my merges, so let's just get this done once and for all. This patch was mostly done with a sed-script, with manual fix-ups for the cases that weren't of the trivial 'access_ok(VERIFY_xyz' form. There were a couple of notable cases: - csky still had the old "verify_area()" name as an alias. - the iter_iov code had magical hardcoded knowledge of the actual values of VERIFY_{READ,WRITE} (not that they mattered, since nothing really used it) - microblaze used the type argument for a debug printout but other than those oddities this should be a total no-op patch. I tried to fix up all architectures, did fairly extensive grepping for access_ok() uses, and the changes are trivial, but I may have missed something. Any missed conversion should be trivially fixable, though. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-01-04 02:57:57 +00:00
ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, FUTEX_WRITE);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
if (unlikely(ret != 0))
goto out;
q.bitset = bitset;
q.rt_waiter = &rt_waiter;
q.requeue_pi_key = &key2;
/*
* Prepare to wait on uaddr. On success, increments q.key (key1) ref
* count.
*/
ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
if (ret)
goto out;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* The check above which compares uaddrs is not sufficient for
* shared futexes. We need to compare the keys:
*/
if (match_futex(&q.key, &key2)) {
queue_unlock(hb);
ret = -EINVAL;
goto out;
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/* Queue the futex_q, drop the hb lock, wait for wakeup. */
futex_wait_queue_me(hb, &q, to);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
spin_lock(&hb->lock);
ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
spin_unlock(&hb->lock);
if (ret)
goto out;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* In order for us to be here, we know our q.key == key2, and since
* we took the hb->lock above, we also know that futex_requeue() has
* completed and we no longer have to concern ourselves with a wakeup
* race with the atomic proxy lock acquisition by the requeue code. The
* futex_requeue dropped our key1 reference and incremented our key2
* reference count.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
/* Check if the requeue code acquired the second futex for us. */
if (!q.rt_waiter) {
/*
* Got the lock. We might not be the anticipated owner if we
* did a lock-steal - fix up the PI-state in that case.
*/
if (q.pi_state && (q.pi_state->owner != current)) {
spin_lock(q.lock_ptr);
ret = fixup_pi_state_owner(uaddr2, &q, current);
if (ret < 0 && rt_mutex_owner(&q.pi_state->pi_mutex) == current) {
pi_state = q.pi_state;
get_pi_state(pi_state);
}
/*
* Drop the reference to the pi state which
* the requeue_pi() code acquired for us.
*/
put_pi_state(q.pi_state);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
spin_unlock(q.lock_ptr);
/*
* Adjust the return value. It's either -EFAULT or
* success (1) but the caller expects 0 for success.
*/
ret = ret < 0 ? ret : 0;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
} else {
struct rt_mutex *pi_mutex;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* We have been woken up by futex_unlock_pi(), a timeout, or a
* signal. futex_unlock_pi() will not destroy the lock_ptr nor
* the pi_state.
*/
WARN_ON(!q.pi_state);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
pi_mutex = &q.pi_state->pi_mutex;
ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
spin_lock(q.lock_ptr);
if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter))
ret = 0;
debug_rt_mutex_free_waiter(&rt_waiter);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* Fixup the pi_state owner and possibly acquire the lock if we
* haven't already.
*/
res = fixup_owner(uaddr2, &q, !ret);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* If fixup_owner() returned an error, proprogate that. If it
* acquired the lock, clear -ETIMEDOUT or -EINTR.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
if (res)
ret = (res < 0) ? res : 0;
/*
* If fixup_pi_state_owner() faulted and was unable to handle
* the fault, unlock the rt_mutex and return the fault to
* userspace.
*/
if (ret && rt_mutex_owner(&q.pi_state->pi_mutex) == current) {
pi_state = q.pi_state;
get_pi_state(pi_state);
}
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/* Unqueue and drop the lock. */
unqueue_me_pi(&q);
}
if (pi_state) {
rt_mutex_futex_unlock(&pi_state->pi_mutex);
put_pi_state(pi_state);
}
if (ret == -EINTR) {
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
/*
* We've already been requeued, but cannot restart by calling
* futex_lock_pi() directly. We could restart this syscall, but
* it would detect that the user space "val" changed and return
* -EWOULDBLOCK. Save the overhead of the restart and return
* -EWOULDBLOCK directly.
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
*/
ret = -EWOULDBLOCK;
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
}
out:
if (to) {
hrtimer_cancel(&to->timer);
destroy_hrtimer_on_stack(&to->timer);
}
return ret;
}
/*
* Support for robust futexes: the kernel cleans up held futexes at
* thread exit time.
*
* Implementation: user-space maintains a per-thread list of locks it
* is holding. Upon do_exit(), the kernel carefully walks this list,
* and marks all locks that are owned by this thread with the
* FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
* always manipulated with the lock held, so the list is private and
* per-thread. Userspace also maintains a per-thread 'list_op_pending'
* field, to allow the kernel to clean up if the thread dies after
* acquiring the lock, but just before it could have added itself to
* the list. There can only be one such pending lock.
*/
/**
* sys_set_robust_list() - Set the robust-futex list head of a task
* @head: pointer to the list-head
* @len: length of the list-head, as userspace expects
*/
SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head,
size_t, len)
{
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
if (!futex_cmpxchg_enabled)
return -ENOSYS;
/*
* The kernel knows only one size for now:
*/
if (unlikely(len != sizeof(*head)))
return -EINVAL;
current->robust_list = head;
return 0;
}
/**
* sys_get_robust_list() - Get the robust-futex list head of a task
* @pid: pid of the process [zero for current task]
* @head_ptr: pointer to a list-head pointer, the kernel fills it in
* @len_ptr: pointer to a length field, the kernel fills in the header size
*/
SYSCALL_DEFINE3(get_robust_list, int, pid,
struct robust_list_head __user * __user *, head_ptr,
size_t __user *, len_ptr)
{
struct robust_list_head __user *head;
unsigned long ret;
struct task_struct *p;
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
if (!futex_cmpxchg_enabled)
return -ENOSYS;
rcu_read_lock();
ret = -ESRCH;
if (!pid)
p = current;
else {
p = find_task_by_vpid(pid);
if (!p)
goto err_unlock;
}
ret = -EPERM;
ptrace: use fsuid, fsgid, effective creds for fs access checks By checking the effective credentials instead of the real UID / permitted capabilities, ensure that the calling process actually intended to use its credentials. To ensure that all ptrace checks use the correct caller credentials (e.g. in case out-of-tree code or newly added code omits the PTRACE_MODE_*CREDS flag), use two new flags and require one of them to be set. The problem was that when a privileged task had temporarily dropped its privileges, e.g. by calling setreuid(0, user_uid), with the intent to perform following syscalls with the credentials of a user, it still passed ptrace access checks that the user would not be able to pass. While an attacker should not be able to convince the privileged task to perform a ptrace() syscall, this is a problem because the ptrace access check is reused for things in procfs. In particular, the following somewhat interesting procfs entries only rely on ptrace access checks: /proc/$pid/stat - uses the check for determining whether pointers should be visible, useful for bypassing ASLR /proc/$pid/maps - also useful for bypassing ASLR /proc/$pid/cwd - useful for gaining access to restricted directories that contain files with lax permissions, e.g. in this scenario: lrwxrwxrwx root root /proc/13020/cwd -> /root/foobar drwx------ root root /root drwxr-xr-x root root /root/foobar -rw-r--r-- root root /root/foobar/secret Therefore, on a system where a root-owned mode 6755 binary changes its effective credentials as described and then dumps a user-specified file, this could be used by an attacker to reveal the memory layout of root's processes or reveal the contents of files he is not allowed to access (through /proc/$pid/cwd). [akpm@linux-foundation.org: fix warning] Signed-off-by: Jann Horn <jann@thejh.net> Acked-by: Kees Cook <keescook@chromium.org> Cc: Casey Schaufler <casey@schaufler-ca.com> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: "Serge E. Hallyn" <serge.hallyn@ubuntu.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Andy Lutomirski <luto@kernel.org> Cc: Al Viro <viro@zeniv.linux.org.uk> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Willy Tarreau <w@1wt.eu> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-20 23:00:04 +00:00
if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS))
goto err_unlock;
head = p->robust_list;
rcu_read_unlock();
if (put_user(sizeof(*head), len_ptr))
return -EFAULT;
return put_user(head, head_ptr);
err_unlock:
rcu_read_unlock();
return ret;
}
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
/* Constants for the pending_op argument of handle_futex_death */
#define HANDLE_DEATH_PENDING true
#define HANDLE_DEATH_LIST false
/*
* Process a futex-list entry, check whether it's owned by the
* dying task, and do notification if so:
*/
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
static int handle_futex_death(u32 __user *uaddr, struct task_struct *curr,
bool pi, bool pending_op)
{
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
u32 uval, nval, mval;
int err;
/* Futex address must be 32bit aligned */
if ((((unsigned long)uaddr) % sizeof(*uaddr)) != 0)
return -1;
retry:
if (get_user(uval, uaddr))
return -1;
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
/*
* Special case for regular (non PI) futexes. The unlock path in
* user space has two race scenarios:
*
* 1. The unlock path releases the user space futex value and
* before it can execute the futex() syscall to wake up
* waiters it is killed.
*
* 2. A woken up waiter is killed before it can acquire the
* futex in user space.
*
* In both cases the TID validation below prevents a wakeup of
* potential waiters which can cause these waiters to block
* forever.
*
* In both cases the following conditions are met:
*
* 1) task->robust_list->list_op_pending != NULL
* @pending_op == true
* 2) User space futex value == 0
* 3) Regular futex: @pi == false
*
* If these conditions are met, it is safe to attempt waking up a
* potential waiter without touching the user space futex value and
* trying to set the OWNER_DIED bit. The user space futex value is
* uncontended and the rest of the user space mutex state is
* consistent, so a woken waiter will just take over the
* uncontended futex. Setting the OWNER_DIED bit would create
* inconsistent state and malfunction of the user space owner died
* handling.
*/
if (pending_op && !pi && !uval) {
futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
return 0;
}
if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr))
return 0;
/*
* Ok, this dying thread is truly holding a futex
* of interest. Set the OWNER_DIED bit atomically
* via cmpxchg, and if the value had FUTEX_WAITERS
* set, wake up a waiter (if any). (We have to do a
* futex_wake() even if OWNER_DIED is already set -
* to handle the rare but possible case of recursive
* thread-death.) The rest of the cleanup is done in
* userspace.
*/
mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED;
/*
* We are not holding a lock here, but we want to have
* the pagefault_disable/enable() protection because
* we want to handle the fault gracefully. If the
* access fails we try to fault in the futex with R/W
* verification via get_user_pages. get_user() above
* does not guarantee R/W access. If that fails we
* give up and leave the futex locked.
*/
if ((err = cmpxchg_futex_value_locked(&nval, uaddr, uval, mval))) {
switch (err) {
case -EFAULT:
futex: Deobfuscate handle_futex_death() handle_futex_death() uses futex_atomic_cmpxchg_inatomic() without disabling page faults. That's ok, but totally non obvious. We don't hold locks so we actually can and want to fault here, because the get_user() before futex_atomic_cmpxchg_inatomic() does not guarantee a R/W mapping. We could just add a big fat comment to explain this, but actually changing the code so that the functionality is entirely clear is better. Use the helper function which disables page faults around the futex_atomic_cmpxchg_inatomic() and handle a fault with a call to fault_in_user_writeable() as all other places in the futex code do as well. Pointed-out-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Darren Hart <darren@dvhart.com> Cc: Michel Lespinasse <walken@google.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Matt Turner <mattst88@gmail.com> Cc: Russell King <linux@arm.linux.org.uk> Cc: David Howells <dhowells@redhat.com> Cc: Tony Luck <tony.luck@intel.com> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Paul Mundt <lethal@linux-sh.org> Cc: "David S. Miller" <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> LKML-Reference: <alpine.LFD.2.00.1103141126590.2787@localhost6.localdomain6> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2011-03-14 09:34:35 +00:00
if (fault_in_user_writeable(uaddr))
return -1;
goto retry;
case -EAGAIN:
cond_resched();
goto retry;
default:
WARN_ON_ONCE(1);
return err;
}
}
if (nval != uval)
goto retry;
/*
* Wake robust non-PI futexes here. The wakeup of
* PI futexes happens in exit_pi_state():
*/
if (!pi && (uval & FUTEX_WAITERS))
futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY);
return 0;
}
/*
* Fetch a robust-list pointer. Bit 0 signals PI futexes:
*/
static inline int fetch_robust_entry(struct robust_list __user **entry,
struct robust_list __user * __user *head,
unsigned int *pi)
{
unsigned long uentry;
if (get_user(uentry, (unsigned long __user *)head))
return -EFAULT;
*entry = (void __user *)(uentry & ~1UL);
*pi = uentry & 1;
return 0;
}
/*
* Walk curr->robust_list (very carefully, it's a userspace list!)
* and mark any locks found there dead, and notify any waiters.
*
* We silently return on any sign of list-walking problem.
*/
static void exit_robust_list(struct task_struct *curr)
{
struct robust_list_head __user *head = curr->robust_list;
struct robust_list __user *entry, *next_entry, *pending;
unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
unsigned int next_pi;
unsigned long futex_offset;
int rc;
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
if (!futex_cmpxchg_enabled)
return;
/*
* Fetch the list head (which was registered earlier, via
* sys_set_robust_list()):
*/
if (fetch_robust_entry(&entry, &head->list.next, &pi))
return;
/*
* Fetch the relative futex offset:
*/
if (get_user(futex_offset, &head->futex_offset))
return;
/*
* Fetch any possibly pending lock-add first, and handle it
* if it exists:
*/
if (fetch_robust_entry(&pending, &head->list_op_pending, &pip))
return;
next_entry = NULL; /* avoid warning with gcc */
while (entry != &head->list) {
/*
* Fetch the next entry in the list before calling
* handle_futex_death:
*/
rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi);
/*
* A pending lock might already be on the list, so
* don't process it twice:
*/
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
if (entry != pending) {
if (handle_futex_death((void __user *)entry + futex_offset,
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
curr, pi, HANDLE_DEATH_LIST))
return;
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
}
if (rc)
return;
entry = next_entry;
pi = next_pi;
/*
* Avoid excessively long or circular lists:
*/
if (!--limit)
break;
cond_resched();
}
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
if (pending) {
handle_futex_death((void __user *)pending + futex_offset,
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
curr, pip, HANDLE_DEATH_PENDING);
}
}
static void futex_cleanup(struct task_struct *tsk)
{
if (unlikely(tsk->robust_list)) {
exit_robust_list(tsk);
tsk->robust_list = NULL;
}
#ifdef CONFIG_COMPAT
if (unlikely(tsk->compat_robust_list)) {
compat_exit_robust_list(tsk);
tsk->compat_robust_list = NULL;
}
#endif
if (unlikely(!list_empty(&tsk->pi_state_list)))
exit_pi_state_list(tsk);
}
/**
* futex_exit_recursive - Set the tasks futex state to FUTEX_STATE_DEAD
* @tsk: task to set the state on
*
* Set the futex exit state of the task lockless. The futex waiter code
* observes that state when a task is exiting and loops until the task has
* actually finished the futex cleanup. The worst case for this is that the
* waiter runs through the wait loop until the state becomes visible.
*
* This is called from the recursive fault handling path in do_exit().
*
* This is best effort. Either the futex exit code has run already or
* not. If the OWNER_DIED bit has been set on the futex then the waiter can
* take it over. If not, the problem is pushed back to user space. If the
* futex exit code did not run yet, then an already queued waiter might
* block forever, but there is nothing which can be done about that.
*/
void futex_exit_recursive(struct task_struct *tsk)
{
/* If the state is FUTEX_STATE_EXITING then futex_exit_mutex is held */
if (tsk->futex_state == FUTEX_STATE_EXITING)
mutex_unlock(&tsk->futex_exit_mutex);
tsk->futex_state = FUTEX_STATE_DEAD;
}
static void futex_cleanup_begin(struct task_struct *tsk)
{
/*
* Prevent various race issues against a concurrent incoming waiter
* including live locks by forcing the waiter to block on
* tsk->futex_exit_mutex when it observes FUTEX_STATE_EXITING in
* attach_to_pi_owner().
*/
mutex_lock(&tsk->futex_exit_mutex);
/*
* Switch the state to FUTEX_STATE_EXITING under tsk->pi_lock.
*
* This ensures that all subsequent checks of tsk->futex_state in
* attach_to_pi_owner() must observe FUTEX_STATE_EXITING with
* tsk->pi_lock held.
*
* It guarantees also that a pi_state which was queued right before
* the state change under tsk->pi_lock by a concurrent waiter must
* be observed in exit_pi_state_list().
*/
raw_spin_lock_irq(&tsk->pi_lock);
tsk->futex_state = FUTEX_STATE_EXITING;
raw_spin_unlock_irq(&tsk->pi_lock);
}
static void futex_cleanup_end(struct task_struct *tsk, int state)
{
/*
* Lockless store. The only side effect is that an observer might
* take another loop until it becomes visible.
*/
tsk->futex_state = state;
/*
* Drop the exit protection. This unblocks waiters which observed
* FUTEX_STATE_EXITING to reevaluate the state.
*/
mutex_unlock(&tsk->futex_exit_mutex);
}
void futex_exec_release(struct task_struct *tsk)
{
/*
* The state handling is done for consistency, but in the case of
* exec() there is no way to prevent futher damage as the PID stays
* the same. But for the unlikely and arguably buggy case that a
* futex is held on exec(), this provides at least as much state
* consistency protection which is possible.
*/
futex_cleanup_begin(tsk);
futex_cleanup(tsk);
/*
* Reset the state to FUTEX_STATE_OK. The task is alive and about
* exec a new binary.
*/
futex_cleanup_end(tsk, FUTEX_STATE_OK);
}
void futex_exit_release(struct task_struct *tsk)
{
futex_cleanup_begin(tsk);
futex_cleanup(tsk);
futex_cleanup_end(tsk, FUTEX_STATE_DEAD);
}
long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
u32 __user *uaddr2, u32 val2, u32 val3)
{
int cmd = op & FUTEX_CMD_MASK;
unsigned int flags = 0;
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
if (!(op & FUTEX_PRIVATE_FLAG))
flags |= FLAGS_SHARED;
if (op & FUTEX_CLOCK_REALTIME) {
flags |= FLAGS_CLOCKRT;
if (cmd != FUTEX_WAIT && cmd != FUTEX_WAIT_BITSET && \
cmd != FUTEX_WAIT_REQUEUE_PI)
return -ENOSYS;
}
switch (cmd) {
case FUTEX_LOCK_PI:
case FUTEX_UNLOCK_PI:
case FUTEX_TRYLOCK_PI:
case FUTEX_WAIT_REQUEUE_PI:
case FUTEX_CMP_REQUEUE_PI:
if (!futex_cmpxchg_enabled)
return -ENOSYS;
}
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
switch (cmd) {
case FUTEX_WAIT:
val3 = FUTEX_BITSET_MATCH_ANY;
fallthrough;
case FUTEX_WAIT_BITSET:
return futex_wait(uaddr, flags, val, timeout, val3);
case FUTEX_WAKE:
val3 = FUTEX_BITSET_MATCH_ANY;
fallthrough;
case FUTEX_WAKE_BITSET:
return futex_wake(uaddr, flags, val, val3);
case FUTEX_REQUEUE:
return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0);
case FUTEX_CMP_REQUEUE:
return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0);
[PATCH] FUTEX_WAKE_OP: pthread_cond_signal() speedup ATM pthread_cond_signal is unnecessarily slow, because it wakes one waiter (which at least on UP usually means an immediate context switch to one of the waiter threads). This waiter wakes up and after a few instructions it attempts to acquire the cv internal lock, but that lock is still held by the thread calling pthread_cond_signal. So it goes to sleep and eventually the signalling thread is scheduled in, unlocks the internal lock and wakes the waiter again. Now, before 2003-09-21 NPTL was using FUTEX_REQUEUE in pthread_cond_signal to avoid this performance issue, but it was removed when locks were redesigned to the 3 state scheme (unlocked, locked uncontended, locked contended). Following scenario shows why simply using FUTEX_REQUEUE in pthread_cond_signal together with using lll_mutex_unlock_force in place of lll_mutex_unlock is not enough and probably why it has been disabled at that time: The number is value in cv->__data.__lock. thr1 thr2 thr3 0 pthread_cond_wait 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) 0 lll_futex_wait (&cv->__data.__futex, futexval) 0 pthread_cond_signal 1 lll_mutex_lock (cv->__data.__lock) 1 pthread_cond_signal 2 lll_mutex_lock (cv->__data.__lock) 2 lll_futex_wait (&cv->__data.__lock, 2) 2 lll_futex_requeue (&cv->__data.__futex, 0, 1, &cv->__data.__lock) # FUTEX_REQUEUE, not FUTEX_CMP_REQUEUE 2 lll_mutex_unlock_force (cv->__data.__lock) 0 cv->__data.__lock = 0 0 lll_futex_wake (&cv->__data.__lock, 1) 1 lll_mutex_lock (cv->__data.__lock) 0 lll_mutex_unlock (cv->__data.__lock) # Here, lll_mutex_unlock doesn't know there are threads waiting # on the internal cv's lock Now, I believe it is possible to use FUTEX_REQUEUE in pthread_cond_signal, but it will cost us not one, but 2 extra syscalls and, what's worse, one of these extra syscalls will be done for every single waiting loop in pthread_cond_*wait. We would need to use lll_mutex_unlock_force in pthread_cond_signal after requeue and lll_mutex_cond_lock in pthread_cond_*wait after lll_futex_wait. Another alternative is to do the unlocking pthread_cond_signal needs to do (the lock can't be unlocked before lll_futex_wake, as that is racy) in the kernel. I have implemented both variants, futex-requeue-glibc.patch is the first one and futex-wake_op{,-glibc}.patch is the unlocking inside of the kernel. The kernel interface allows userland to specify how exactly an unlocking operation should look like (some atomic arithmetic operation with optional constant argument and comparison of the previous futex value with another constant). It has been implemented just for ppc*, x86_64 and i?86, for other architectures I'm including just a stub header which can be used as a starting point by maintainers to write support for their arches and ATM will just return -ENOSYS for FUTEX_WAKE_OP. The requeue patch has been (lightly) tested just on x86_64, the wake_op patch on ppc64 kernel running 32-bit and 64-bit NPTL and x86_64 kernel running 32-bit and 64-bit NPTL. With the following benchmark on UP x86-64 I get: for i in nptl-orig nptl-requeue nptl-wake_op; do echo time elf/ld.so --library-path .:$i /tmp/bench; \ for j in 1 2; do echo ( time elf/ld.so --library-path .:$i /tmp/bench ) 2>&1; done; done time elf/ld.so --library-path .:nptl-orig /tmp/bench real 0m0.655s user 0m0.253s sys 0m0.403s real 0m0.657s user 0m0.269s sys 0m0.388s time elf/ld.so --library-path .:nptl-requeue /tmp/bench real 0m0.496s user 0m0.225s sys 0m0.271s real 0m0.531s user 0m0.242s sys 0m0.288s time elf/ld.so --library-path .:nptl-wake_op /tmp/bench real 0m0.380s user 0m0.176s sys 0m0.204s real 0m0.382s user 0m0.175s sys 0m0.207s The benchmark is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00001.txt Older futex-requeue-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00002.txt Older futex-wake_op-glibc.patch version is at: http://sourceware.org/ml/libc-alpha/2005-03/txt00003.txt Will post a new version (just x86-64 fixes so that the patch applies against pthread_cond_signal.S) to libc-hacker ml soon. Attached is the kernel FUTEX_WAKE_OP patch as well as a simple-minded testcase that will not test the atomicity of the operation, but at least check if the threads that should have been woken up are woken up and whether the arithmetic operation in the kernel gave the expected results. Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jamie Lokier <jamie@shareable.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Yoichi Yuasa <yuasa@hh.iij4u.or.jp> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-09-06 22:16:25 +00:00
case FUTEX_WAKE_OP:
return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
case FUTEX_LOCK_PI:
futex: Fix argument handling in futex_lock_pi() calls This patch fixes two separate buglets in calls to futex_lock_pi(): * Eliminate unused 'detect' argument * Change unused 'timeout' argument of FUTEX_TRYLOCK_PI to NULL The 'detect' argument of futex_lock_pi() seems never to have been used (when it was included with the initial PI mutex implementation in Linux 2.6.18, all checks against its value were disabled by ANDing against 0 (i.e., if (detect... && 0)), and with commit 778e9a9c3e7193ea9f434f382947155ffb59c755, any mention of this argument in futex_lock_pi() went way altogether. Its presence now serves only to confuse readers of the code, by giving the impression that the futex() FUTEX_LOCK_PI operation actually does use the 'val' argument. This patch removes the argument. The futex_lock_pi() call that corresponds to FUTEX_TRYLOCK_PI includes 'timeout' as one of its arguments. This misleads the reader into thinking that the FUTEX_TRYLOCK_PI operation does employ timeouts for some sensible purpose; but it does not. Indeed, it cannot, because the checks at the start of sys_futex() exclude FUTEX_TRYLOCK_PI from the set of operations that do copy_from_user() on the timeout argument. So, in the FUTEX_TRYLOCK_PI futex_lock_pi() call it would be simplest to change 'timeout' to 'NULL'. This patch does that. Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Darren Hart <darren@dvhart.com> Link: http://lkml.kernel.org/r/54B96646.8010200@gmail.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2015-01-16 19:28:06 +00:00
return futex_lock_pi(uaddr, flags, timeout, 0);
case FUTEX_UNLOCK_PI:
return futex_unlock_pi(uaddr, flags);
case FUTEX_TRYLOCK_PI:
futex: Fix argument handling in futex_lock_pi() calls This patch fixes two separate buglets in calls to futex_lock_pi(): * Eliminate unused 'detect' argument * Change unused 'timeout' argument of FUTEX_TRYLOCK_PI to NULL The 'detect' argument of futex_lock_pi() seems never to have been used (when it was included with the initial PI mutex implementation in Linux 2.6.18, all checks against its value were disabled by ANDing against 0 (i.e., if (detect... && 0)), and with commit 778e9a9c3e7193ea9f434f382947155ffb59c755, any mention of this argument in futex_lock_pi() went way altogether. Its presence now serves only to confuse readers of the code, by giving the impression that the futex() FUTEX_LOCK_PI operation actually does use the 'val' argument. This patch removes the argument. The futex_lock_pi() call that corresponds to FUTEX_TRYLOCK_PI includes 'timeout' as one of its arguments. This misleads the reader into thinking that the FUTEX_TRYLOCK_PI operation does employ timeouts for some sensible purpose; but it does not. Indeed, it cannot, because the checks at the start of sys_futex() exclude FUTEX_TRYLOCK_PI from the set of operations that do copy_from_user() on the timeout argument. So, in the FUTEX_TRYLOCK_PI futex_lock_pi() call it would be simplest to change 'timeout' to 'NULL'. This patch does that. Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Darren Hart <darren@dvhart.com> Link: http://lkml.kernel.org/r/54B96646.8010200@gmail.com Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2015-01-16 19:28:06 +00:00
return futex_lock_pi(uaddr, flags, NULL, 1);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
case FUTEX_WAIT_REQUEUE_PI:
val3 = FUTEX_BITSET_MATCH_ANY;
return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
uaddr2);
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
case FUTEX_CMP_REQUEUE_PI:
return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1);
}
return -ENOSYS;
}
SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
struct __kernel_timespec __user *, utime, u32 __user *, uaddr2,
u32, val3)
{
struct timespec64 ts;
ktime_t t, *tp = NULL;
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
u32 val2 = 0;
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
int cmd = op & FUTEX_CMD_MASK;
if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
cmd == FUTEX_WAIT_BITSET ||
cmd == FUTEX_WAIT_REQUEUE_PI)) {
if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG))))
return -EFAULT;
if (get_timespec64(&ts, utime))
return -EFAULT;
if (!timespec64_valid(&ts))
return -EINVAL;
t = timespec64_to_ktime(ts);
FUTEX: new PRIVATE futexes Analysis of current linux futex code : -------------------------------------- A central hash table futex_queues[] holds all contexts (futex_q) of waiting threads. Each futex_wait()/futex_wait() has to obtain a spinlock on a hash slot to perform lookups or insert/deletion of a futex_q. When a futex_wait() is done, calling thread has to : 1) - Obtain a read lock on mmap_sem to be able to validate the user pointer (calling find_vma()). This validation tells us if the futex uses an inode based store (mapped file), or mm based store (anonymous mem) 2) - compute a hash key 3) - Atomic increment of reference counter on an inode or a mm_struct 4) - lock part of futex_queues[] hash table 5) - perform the test on value of futex. (rollback is value != expected_value, returns EWOULDBLOCK) (various loops if test triggers mm faults) 6) queue the context into hash table, release the lock got in 4) 7) - release the read_lock on mmap_sem <block> 8) Eventually unqueue the context (but rarely, as this part  may be done by the futex_wake()) Futexes were designed to improve scalability but current implementation has various problems : - Central hashtable : This means scalability problems if many processes/threads want to use futexes at the same time. This means NUMA unbalance because this hashtable is located on one node. - Using mmap_sem on every futex() syscall : Even if mmap_sem is a rw_semaphore, up_read()/down_read() are doing atomic ops on mmap_sem, dirtying cache line : - lot of cache line ping pongs on SMP configurations. mmap_sem is also extensively used by mm code (page faults, mmap()/munmap()) Highly threaded processes might suffer from mmap_sem contention. mmap_sem is also used by oprofile code. Enabling oprofile hurts threaded programs because of contention on the mmap_sem cache line. - Using an atomic_inc()/atomic_dec() on inode ref counter or mm ref counter: It's also a cache line ping pong on SMP. It also increases mmap_sem hold time because of cache misses. Most of these scalability problems come from the fact that futexes are in one global namespace. As we use a central hash table, we must make sure they are all using the same reference (given by the mm subsystem). We chose to force all futexes be 'shared'. This has a cost. But fact is POSIX defined PRIVATE and SHARED, allowing clear separation, and optimal performance if carefuly implemented. Time has come for linux to have better threading performance. The goal is to permit new futex commands to avoid : - Taking the mmap_sem semaphore, conflicting with other subsystems. - Modifying a ref_count on mm or an inode, still conflicting with mm or fs. This is possible because, for one process using PTHREAD_PROCESS_PRIVATE futexes, we only need to distinguish futexes by their virtual address, no matter the underlying mm storage is. If glibc wants to exploit this new infrastructure, it should use new _PRIVATE futex subcommands for PTHREAD_PROCESS_PRIVATE futexes. And be prepared to fallback on old subcommands for old kernels. Using one global variable with the FUTEX_PRIVATE_FLAG or 0 value should be OK. PTHREAD_PROCESS_SHARED futexes should still use the old subcommands. Compatibility with old applications is preserved, they still hit the scalability problems, but new applications can fly :) Note : the same SHARED futex (mapped on a file) can be used by old binaries *and* new binaries, because both binaries will use the old subcommands. Note : Vast majority of futexes should be using PROCESS_PRIVATE semantic, as this is the default semantic. Almost all applications should benefit of this changes (new kernel and updated libc) Some bench results on a Pentium M 1.6 GHz (SMP kernel on a UP machine) /* calling futex_wait(addr, value) with value != *addr */ 433 cycles per futex(FUTEX_WAIT) call (mixing 2 futexes) 424 cycles per futex(FUTEX_WAIT) call (using one futex) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (mixing 2 futexes) 334 cycles per futex(FUTEX_WAIT_PRIVATE) call (using one futex) For reference : 187 cycles per getppid() call 188 cycles per umask() call 181 cycles per ni_syscall() call Signed-off-by: Eric Dumazet <dada1@cosmosbay.com> Pierre Peiffer <pierre.peiffer@bull.net> Cc: "Ulrich Drepper" <drepper@gmail.com> Cc: "Nick Piggin" <nickpiggin@yahoo.com.au> Cc: "Ingo Molnar" <mingo@elte.hu> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-09 09:35:04 +00:00
if (cmd == FUTEX_WAIT)
t = ktime_add_safe(ktime_get(), t);
else if (!(op & FUTEX_CLOCK_REALTIME))
t = timens_ktime_to_host(CLOCK_MONOTONIC, t);
tp = &t;
}
/*
futex: add requeue_pi functionality PI Futexes and their underlying rt_mutex cannot be left ownerless if there are pending waiters as this will break the PI boosting logic, so the standard requeue commands aren't sufficient. The new commands properly manage pi futex ownership by ensuring a futex with waiters has an owner at all times. This will allow glibc to properly handle pi mutexes with pthread_condvars. The approach taken here is to create two new futex op codes: FUTEX_WAIT_REQUEUE_PI: Tasks will use this op code to wait on a futex (such as a non-pi waitqueue) and wake after they have been requeued to a pi futex. Prior to returning to userspace, they will acquire this pi futex (and the underlying rt_mutex). futex_wait_requeue_pi() is the result of a high speed collision between futex_wait() and futex_lock_pi() (with the first part of futex_lock_pi() being done by futex_proxy_trylock_atomic() on behalf of the top_waiter). FUTEX_REQUEUE_PI (and FUTEX_CMP_REQUEUE_PI): This call must be used to wake tasks waiting with FUTEX_WAIT_REQUEUE_PI, regardless of how many tasks the caller intends to wake or requeue. pthread_cond_broadcast() should call this with nr_wake=1 and nr_requeue=INT_MAX. pthread_cond_signal() should call this with nr_wake=1 and nr_requeue=0. The reason being we need both callers to get the benefit of the futex_proxy_trylock_atomic() routine. futex_requeue() also enqueues the top_waiter on the rt_mutex via rt_mutex_start_proxy_lock(). Signed-off-by: Darren Hart <dvhltc@us.ibm.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2009-04-03 20:40:49 +00:00
* requeue parameter in 'utime' if cmd == FUTEX_*_REQUEUE_*.
* number of waiters to wake in 'utime' if cmd == FUTEX_WAKE_OP.
*/
if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
[PATCH] pi-futex: futex code cleanups We are pleased to announce "lightweight userspace priority inheritance" (PI) support for futexes. The following patchset and glibc patch implements it, ontop of the robust-futexes patchset which is included in 2.6.16-mm1. We are calling it lightweight for 3 reasons: - in the user-space fastpath a PI-enabled futex involves no kernel work (or any other PI complexity) at all. No registration, no extra kernel calls - just pure fast atomic ops in userspace. - in the slowpath (in the lock-contention case), the system call and scheduling pattern is in fact better than that of normal futexes, due to the 'integrated' nature of FUTEX_LOCK_PI. [more about that further down] - the in-kernel PI implementation is streamlined around the mutex abstraction, with strict rules that keep the implementation relatively simple: only a single owner may own a lock (i.e. no read-write lock support), only the owner may unlock a lock, no recursive locking, etc. Priority Inheritance - why, oh why??? ------------------------------------- Many of you heard the horror stories about the evil PI code circling Linux for years, which makes no real sense at all and is only used by buggy applications and which has horrible overhead. Some of you have dreaded this very moment, when someone actually submits working PI code ;-) So why would we like to see PI support for futexes? We'd like to see it done purely for technological reasons. We dont think it's a buggy concept, we think it's useful functionality to offer to applications, which functionality cannot be achieved in other ways. We also think it's the right thing to do, and we think we've got the right arguments and the right numbers to prove that. We also believe that we can address all the counter-arguments as well. For these reasons (and the reasons outlined below) we are submitting this patch-set for upstream kernel inclusion. What are the benefits of PI? The short reply: ---------------- User-space PI helps achieving/improving determinism for user-space applications. In the best-case, it can help achieve determinism and well-bound latencies. Even in the worst-case, PI will improve the statistical distribution of locking related application delays. The longer reply: ----------------- Firstly, sharing locks between multiple tasks is a common programming technique that often cannot be replaced with lockless algorithms. As we can see it in the kernel [which is a quite complex program in itself], lockless structures are rather the exception than the norm - the current ratio of lockless vs. locky code for shared data structures is somewhere between 1:10 and 1:100. Lockless is hard, and the complexity of lockless algorithms often endangers to ability to do robust reviews of said code. I.e. critical RT apps often choose lock structures to protect critical data structures, instead of lockless algorithms. Furthermore, there are cases (like shared hardware, or other resource limits) where lockless access is mathematically impossible. Media players (such as Jack) are an example of reasonable application design with multiple tasks (with multiple priority levels) sharing short-held locks: for example, a highprio audio playback thread is combined with medium-prio construct-audio-data threads and low-prio display-colory-stuff threads. Add video and decoding to the mix and we've got even more priority levels. So once we accept that synchronization objects (locks) are an unavoidable fact of life, and once we accept that multi-task userspace apps have a very fair expectation of being able to use locks, we've got to think about how to offer the option of a deterministic locking implementation to user-space. Most of the technical counter-arguments against doing priority inheritance only apply to kernel-space locks. But user-space locks are different, there we cannot disable interrupts or make the task non-preemptible in a critical section, so the 'use spinlocks' argument does not apply (user-space spinlocks have the same priority inversion problems as other user-space locking constructs). Fact is, pretty much the only technique that currently enables good determinism for userspace locks (such as futex-based pthread mutexes) is priority inheritance: Currently (without PI), if a high-prio and a low-prio task shares a lock [this is a quite common scenario for most non-trivial RT applications], even if all critical sections are coded carefully to be deterministic (i.e. all critical sections are short in duration and only execute a limited number of instructions), the kernel cannot guarantee any deterministic execution of the high-prio task: any medium-priority task could preempt the low-prio task while it holds the shared lock and executes the critical section, and could delay it indefinitely. Implementation: --------------- As mentioned before, the userspace fastpath of PI-enabled pthread mutexes involves no kernel work at all - they behave quite similarly to normal futex-based locks: a 0 value means unlocked, and a value==TID means locked. (This is the same method as used by list-based robust futexes.) Userspace uses atomic ops to lock/unlock these mutexes without entering the kernel. To handle the slowpath, we have added two new futex ops: FUTEX_LOCK_PI FUTEX_UNLOCK_PI If the lock-acquire fastpath fails, [i.e. an atomic transition from 0 to TID fails], then FUTEX_LOCK_PI is called. The kernel does all the remaining work: if there is no futex-queue attached to the futex address yet then the code looks up the task that owns the futex [it has put its own TID into the futex value], and attaches a 'PI state' structure to the futex-queue. The pi_state includes an rt-mutex, which is a PI-aware, kernel-based synchronization object. The 'other' task is made the owner of the rt-mutex, and the FUTEX_WAITERS bit is atomically set in the futex value. Then this task tries to lock the rt-mutex, on which it blocks. Once it returns, it has the mutex acquired, and it sets the futex value to its own TID and returns. Userspace has no other work to perform - it now owns the lock, and futex value contains FUTEX_WAITERS|TID. If the unlock side fastpath succeeds, [i.e. userspace manages to do a TID -> 0 atomic transition of the futex value], then no kernel work is triggered. If the unlock fastpath fails (because the FUTEX_WAITERS bit is set), then FUTEX_UNLOCK_PI is called, and the kernel unlocks the futex on the behalf of userspace - and it also unlocks the attached pi_state->rt_mutex and thus wakes up any potential waiters. Note that under this approach, contrary to other PI-futex approaches, there is no prior 'registration' of a PI-futex. [which is not quite possible anyway, due to existing ABI properties of pthread mutexes.] Also, under this scheme, 'robustness' and 'PI' are two orthogonal properties of futexes, and all four combinations are possible: futex, robust-futex, PI-futex, robust+PI-futex. glibc support: -------------- Ulrich Drepper and Jakub Jelinek have written glibc support for PI-futexes (and robust futexes), enabling robust and PI (PTHREAD_PRIO_INHERIT) POSIX mutexes. (PTHREAD_PRIO_PROTECT support will be added later on too, no additional kernel changes are needed for that). [NOTE: The glibc patch is obviously inofficial and unsupported without matching upstream kernel functionality.] the patch-queue and the glibc patch can also be downloaded from: http://redhat.com/~mingo/PI-futex-patches/ Many thanks go to the people who helped us create this kernel feature: Steven Rostedt, Esben Nielsen, Benedikt Spranger, Daniel Walker, John Cooper, Arjan van de Ven, Oleg Nesterov and others. Credits for related prior projects goes to Dirk Grambow, Inaky Perez-Gonzalez, Bill Huey and many others. Clean up the futex code, before adding more features to it: - use u32 as the futex field type - that's the ABI - use __user and pointers to u32 instead of unsigned long - code style / comment style cleanups - rename hash-bucket name from 'bh' to 'hb'. I checked the pre and post futex.o object files to make sure this patch has no code effects. Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Arjan van de Ven <arjan@linux.intel.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-06-27 09:54:47 +00:00
val2 = (u32) (unsigned long) utime;
return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
}
#ifdef CONFIG_COMPAT
/*
* Fetch a robust-list pointer. Bit 0 signals PI futexes:
*/
static inline int
compat_fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry,
compat_uptr_t __user *head, unsigned int *pi)
{
if (get_user(*uentry, head))
return -EFAULT;
*entry = compat_ptr((*uentry) & ~1);
*pi = (unsigned int)(*uentry) & 1;
return 0;
}
static void __user *futex_uaddr(struct robust_list __user *entry,
compat_long_t futex_offset)
{
compat_uptr_t base = ptr_to_compat(entry);
void __user *uaddr = compat_ptr(base + futex_offset);
return uaddr;
}
/*
* Walk curr->robust_list (very carefully, it's a userspace list!)
* and mark any locks found there dead, and notify any waiters.
*
* We silently return on any sign of list-walking problem.
*/
static void compat_exit_robust_list(struct task_struct *curr)
{
struct compat_robust_list_head __user *head = curr->compat_robust_list;
struct robust_list __user *entry, *next_entry, *pending;
unsigned int limit = ROBUST_LIST_LIMIT, pi, pip;
treewide: Remove uninitialized_var() usage Using uninitialized_var() is dangerous as it papers over real bugs[1] (or can in the future), and suppresses unrelated compiler warnings (e.g. "unused variable"). If the compiler thinks it is uninitialized, either simply initialize the variable or make compiler changes. In preparation for removing[2] the[3] macro[4], remove all remaining needless uses with the following script: git grep '\buninitialized_var\b' | cut -d: -f1 | sort -u | \ xargs perl -pi -e \ 's/\buninitialized_var\(([^\)]+)\)/\1/g; s:\s*/\* (GCC be quiet|to make compiler happy) \*/$::g;' drivers/video/fbdev/riva/riva_hw.c was manually tweaked to avoid pathological white-space. No outstanding warnings were found building allmodconfig with GCC 9.3.0 for x86_64, i386, arm64, arm, powerpc, powerpc64le, s390x, mips, sparc64, alpha, and m68k. [1] https://lore.kernel.org/lkml/20200603174714.192027-1-glider@google.com/ [2] https://lore.kernel.org/lkml/CA+55aFw+Vbj0i=1TGqCR5vQkCzWJ0QxK6CernOU6eedsudAixw@mail.gmail.com/ [3] https://lore.kernel.org/lkml/CA+55aFwgbgqhbp1fkxvRKEpzyR5J8n1vKT1VZdz9knmPuXhOeg@mail.gmail.com/ [4] https://lore.kernel.org/lkml/CA+55aFz2500WfbKXAx8s67wrm9=yVJu65TpLgN_ybYNv0VEOKA@mail.gmail.com/ Reviewed-by: Leon Romanovsky <leonro@mellanox.com> # drivers/infiniband and mlx4/mlx5 Acked-by: Jason Gunthorpe <jgg@mellanox.com> # IB Acked-by: Kalle Valo <kvalo@codeaurora.org> # wireless drivers Reviewed-by: Chao Yu <yuchao0@huawei.com> # erofs Signed-off-by: Kees Cook <keescook@chromium.org>
2020-06-03 20:09:38 +00:00
unsigned int next_pi;
compat_uptr_t uentry, next_uentry, upending;
compat_long_t futex_offset;
int rc;
if (!futex_cmpxchg_enabled)
return;
/*
* Fetch the list head (which was registered earlier, via
* sys_set_robust_list()):
*/
if (compat_fetch_robust_entry(&uentry, &entry, &head->list.next, &pi))
return;
/*
* Fetch the relative futex offset:
*/
if (get_user(futex_offset, &head->futex_offset))
return;
/*
* Fetch any possibly pending lock-add first, and handle it
* if it exists:
*/
if (compat_fetch_robust_entry(&upending, &pending,
&head->list_op_pending, &pip))
return;
next_entry = NULL; /* avoid warning with gcc */
while (entry != (struct robust_list __user *) &head->list) {
/*
* Fetch the next entry in the list before calling
* handle_futex_death:
*/
rc = compat_fetch_robust_entry(&next_uentry, &next_entry,
(compat_uptr_t __user *)&entry->next, &next_pi);
/*
* A pending lock might already be on the list, so
* dont process it twice:
*/
if (entry != pending) {
void __user *uaddr = futex_uaddr(entry, futex_offset);
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
if (handle_futex_death(uaddr, curr, pi,
HANDLE_DEATH_LIST))
return;
}
if (rc)
return;
uentry = next_uentry;
entry = next_entry;
pi = next_pi;
/*
* Avoid excessively long or circular lists:
*/
if (!--limit)
break;
cond_resched();
}
if (pending) {
void __user *uaddr = futex_uaddr(pending, futex_offset);
futex: Prevent robust futex exit race Robust futexes utilize the robust_list mechanism to allow the kernel to release futexes which are held when a task exits. The exit can be voluntary or caused by a signal or fault. This prevents that waiters block forever. The futex operations in user space store a pointer to the futex they are either locking or unlocking in the op_pending member of the per task robust list. After a lock operation has succeeded the futex is queued in the robust list linked list and the op_pending pointer is cleared. After an unlock operation has succeeded the futex is removed from the robust list linked list and the op_pending pointer is cleared. The robust list exit code checks for the pending operation and any futex which is queued in the linked list. It carefully checks whether the futex value is the TID of the exiting task. If so, it sets the OWNER_DIED bit and tries to wake up a potential waiter. This is race free for the lock operation but unlock has two race scenarios where waiters might not be woken up. These issues can be observed with regular robust pthread mutexes. PI aware pthread mutexes are not affected. (1) Unlocking task is killed after unlocking the futex value in user space before being able to wake a waiter. pthread_mutex_unlock() | V atomic_exchange_rel (&mutex->__data.__lock, 0) <------------------------killed lll_futex_wake () | | |(__lock = 0) |(enter kernel) | V do_exit() exit_mm() mm_release() exit_robust_list() handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters which in consequence block infinitely. (2) Waiting task is killed after a wakeup and before it can acquire the futex in user space. OWNER WAITER futex_wait() pthread_mutex_unlock() | | | |(__lock = 0) | | | V | futex_wake() ------------> wakeup() | |(return to userspace) |(__lock = 0) | V oldval = mutex->__data.__lock <-----------------killed atomic_compare_and_exchange_val_acq (&mutex->__data.__lock, | id | assume_other_futex_waiters, 0) | | | (enter kernel)| | V do_exit() | | V handle_futex_death() | |(__lock = 0) |(uval = 0) | V if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) return 0; The sanity check which ensures that the user space futex is owned by the exiting task prevents the wakeup of waiters, which seems to be correct as the exiting task does not own the futex value, but the consequence is that other waiters wont be woken up and block infinitely. In both scenarios the following conditions are true: - task->robust_list->list_op_pending != NULL - user space futex value == 0 - Regular futex (not PI) If these conditions are met then it is reasonably safe to wake up a potential waiter in order to prevent the above problems. As this might be a false positive it can cause spurious wakeups, but the waiter side has to handle other types of unrelated wakeups, e.g. signals gracefully anyway. So such a spurious wakeup will not affect the correctness of these operations. This workaround must not touch the user space futex value and cannot set the OWNER_DIED bit because the lock value is 0, i.e. uncontended. Setting OWNER_DIED in this case would result in inconsistent state and subsequently in malfunction of the owner died handling in user space. The rest of the user space state is still consistent as no other task can observe the list_op_pending entry in the exiting tasks robust list. The eventually woken up waiter will observe the uncontended lock value and take it over. [ tglx: Massaged changelog and comment. Made the return explicit and not depend on the subsequent check and added constants to hand into handle_futex_death() instead of plain numbers. Fixed a few coding style issues. ] Fixes: 0771dfefc9e5 ("[PATCH] lightweight robust futexes: core") Signed-off-by: Yang Tao <yang.tao172@zte.com.cn> Signed-off-by: Yi Wang <wang.yi59@zte.com.cn> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Reviewed-by: Ingo Molnar <mingo@kernel.org> Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: stable@vger.kernel.org Link: https://lkml.kernel.org/r/1573010582-35297-1-git-send-email-wang.yi59@zte.com.cn Link: https://lkml.kernel.org/r/20191106224555.943191378@linutronix.de
2019-11-06 21:55:35 +00:00
handle_futex_death(uaddr, curr, pip, HANDLE_DEATH_PENDING);
}
}
COMPAT_SYSCALL_DEFINE2(set_robust_list,
struct compat_robust_list_head __user *, head,
compat_size_t, len)
{
if (!futex_cmpxchg_enabled)
return -ENOSYS;
if (unlikely(len != sizeof(*head)))
return -EINVAL;
current->compat_robust_list = head;
return 0;
}
COMPAT_SYSCALL_DEFINE3(get_robust_list, int, pid,
compat_uptr_t __user *, head_ptr,
compat_size_t __user *, len_ptr)
{
struct compat_robust_list_head __user *head;
unsigned long ret;
struct task_struct *p;
if (!futex_cmpxchg_enabled)
return -ENOSYS;
rcu_read_lock();
ret = -ESRCH;
if (!pid)
p = current;
else {
p = find_task_by_vpid(pid);
if (!p)
goto err_unlock;
}
ret = -EPERM;
if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS))
goto err_unlock;
head = p->compat_robust_list;
rcu_read_unlock();
if (put_user(sizeof(*head), len_ptr))
return -EFAULT;
return put_user(ptr_to_compat(head), head_ptr);
err_unlock:
rcu_read_unlock();
return ret;
}
#endif /* CONFIG_COMPAT */
#ifdef CONFIG_COMPAT_32BIT_TIME
SYSCALL_DEFINE6(futex_time32, u32 __user *, uaddr, int, op, u32, val,
struct old_timespec32 __user *, utime, u32 __user *, uaddr2,
u32, val3)
{
struct timespec64 ts;
ktime_t t, *tp = NULL;
int val2 = 0;
int cmd = op & FUTEX_CMD_MASK;
if (utime && (cmd == FUTEX_WAIT || cmd == FUTEX_LOCK_PI ||
cmd == FUTEX_WAIT_BITSET ||
cmd == FUTEX_WAIT_REQUEUE_PI)) {
if (get_old_timespec32(&ts, utime))
return -EFAULT;
if (!timespec64_valid(&ts))
return -EINVAL;
t = timespec64_to_ktime(ts);
if (cmd == FUTEX_WAIT)
t = ktime_add_safe(ktime_get(), t);
else if (!(op & FUTEX_CLOCK_REALTIME))
t = timens_ktime_to_host(CLOCK_MONOTONIC, t);
tp = &t;
}
if (cmd == FUTEX_REQUEUE || cmd == FUTEX_CMP_REQUEUE ||
cmd == FUTEX_CMP_REQUEUE_PI || cmd == FUTEX_WAKE_OP)
val2 = (int) (unsigned long) utime;
return do_futex(uaddr, op, val, tp, uaddr2, val2, val3);
}
#endif /* CONFIG_COMPAT_32BIT_TIME */
static void __init futex_detect_cmpxchg(void)
{
#ifndef CONFIG_HAVE_FUTEX_CMPXCHG
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
u32 curval;
/*
* This will fail and we want it. Some arch implementations do
* runtime detection of the futex_atomic_cmpxchg_inatomic()
* functionality. We want to know that before we call in any
* of the complex code paths. Also we want to prevent
* registration of robust lists in that case. NULL is
* guaranteed to fault and we get -EFAULT on functional
* implementation, the non-functional ones will return
* -ENOSYS.
*/
if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT)
futex_cmpxchg_enabled = 1;
#endif
}
static int __init futex_init(void)
{
unsigned int futex_shift;
futexes: Increase hash table size for better performance Currently, the futex global hash table suffers from its fixed, smallish (for today's standards) size of 256 entries, as well as its lack of NUMA awareness. Large systems, using many futexes, can be prone to high amounts of collisions; where these futexes hash to the same bucket and lead to extra contention on the same hb->lock. Furthermore, cacheline bouncing is a reality when we have multiple hb->locks residing on the same cacheline and different futexes hash to adjacent buckets. This patch keeps the current static size of 16 entries for small systems, or otherwise, 256 * ncpus (or larger as we need to round the number to a power of 2). Note that this number of CPUs accounts for all CPUs that can ever be available in the system, taking into consideration things like hotpluging. While we do impose extra overhead at bootup by making the hash table larger, this is a one time thing, and does not shadow the benefits of this patch. Furthermore, as suggested by tglx, by cache aligning the hash buckets we can avoid access across cacheline boundaries and also avoid massive cache line bouncing if multiple cpus are hammering away at different hash buckets which happen to reside in the same cache line. Also, similar to other core kernel components (pid, dcache, tcp), by using alloc_large_system_hash() we benefit from its NUMA awareness and thus the table is distributed among the nodes instead of in a single one. For a custom microbenchmark that pounds on the uaddr hashing -- making the wait path fail at futex_wait_setup() returning -EWOULDBLOCK for large amounts of futexes, we can see the following benefits on a 80-core, 8-socket 1Tb server: +---------+--------------------+------------------------+-----------------------+-------------------------------+ | threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) | +---------+--------------------+------------------------+-----------------------+-------------------------------+ |     512 |              32426 | 50531  (+55.8%)        | 255274  (+687.2%)     | 292553  (+802.2%)             | |     256 |              65360 | 99588  (+52.3%)        | 443563  (+578.6%)     | 508088  (+677.3%)             | |     128 |             125635 | 200075 (+59.2%)        | 742613  (+491.1%)     | 835452  (+564.9%)             | |      80 |             193559 | 323425 (+67.1%)        | 1028147 (+431.1%)     | 1130304 (+483.9%)             | |      64 |             247667 | 443740 (+79.1%)        | 997300  (+302.6%)     | 1145494 (+362.5%)             | |      32 |             628412 | 721401 (+14.7%)        | 965996  (+53.7%)      | 1122115 (+78.5%)              | +---------+--------------------+------------------------+-----------------------+-------------------------------+ Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Waiman Long <Waiman.Long@hp.com> Reviewed-and-tested-by: Jason Low <jason.low2@hp.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:23 +00:00
unsigned long i;
#if CONFIG_BASE_SMALL
futex_hashsize = 16;
#else
futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus());
#endif
futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues),
futex_hashsize, 0,
futex_hashsize < 256 ? HASH_SMALL : 0,
&futex_shift, NULL,
futex_hashsize, futex_hashsize);
futex_hashsize = 1UL << futex_shift;
futex_detect_cmpxchg();
futex: runtime enable pi and robust functionality Not all architectures implement futex_atomic_cmpxchg_inatomic(). The default implementation returns -ENOSYS, which is currently not handled inside of the futex guts. Futex PI calls and robust list exits with a held futex result in an endless loop in the futex code on architectures which have no support. Fixing up every place where futex_atomic_cmpxchg_inatomic() is called would add a fair amount of extra if/else constructs to the already complex code. It is also not possible to disable the robust feature before user space tries to register robust lists. Compile time disabling is not a good idea either, as there are already architectures with runtime detection of futex_atomic_cmpxchg_inatomic support. Detect the functionality at runtime instead by calling cmpxchg_futex_value_locked() with a NULL pointer from the futex initialization code. This is guaranteed to fail, but the call of futex_atomic_cmpxchg_inatomic() happens with pagefaults disabled. On architectures, which use the asm-generic implementation or have a runtime CPU feature detection, a -ENOSYS return value disables the PI/robust features. On architectures with a working implementation the call returns -EFAULT and the PI/robust features are enabled. The relevant syscalls return -ENOSYS and the robust list exit code is blocked, when the detection fails. Fixes http://lkml.org/lkml/2008/2/11/149 Originally reported by: Lennart Buytenhek Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Acked-by: Ingo Molnar <mingo@elte.hu> Cc: Lennert Buytenhek <buytenh@wantstofly.org> Cc: Riku Voipio <riku.voipio@movial.fi> Cc: <stable@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-23 23:23:57 +00:00
futexes: Increase hash table size for better performance Currently, the futex global hash table suffers from its fixed, smallish (for today's standards) size of 256 entries, as well as its lack of NUMA awareness. Large systems, using many futexes, can be prone to high amounts of collisions; where these futexes hash to the same bucket and lead to extra contention on the same hb->lock. Furthermore, cacheline bouncing is a reality when we have multiple hb->locks residing on the same cacheline and different futexes hash to adjacent buckets. This patch keeps the current static size of 16 entries for small systems, or otherwise, 256 * ncpus (or larger as we need to round the number to a power of 2). Note that this number of CPUs accounts for all CPUs that can ever be available in the system, taking into consideration things like hotpluging. While we do impose extra overhead at bootup by making the hash table larger, this is a one time thing, and does not shadow the benefits of this patch. Furthermore, as suggested by tglx, by cache aligning the hash buckets we can avoid access across cacheline boundaries and also avoid massive cache line bouncing if multiple cpus are hammering away at different hash buckets which happen to reside in the same cache line. Also, similar to other core kernel components (pid, dcache, tcp), by using alloc_large_system_hash() we benefit from its NUMA awareness and thus the table is distributed among the nodes instead of in a single one. For a custom microbenchmark that pounds on the uaddr hashing -- making the wait path fail at futex_wait_setup() returning -EWOULDBLOCK for large amounts of futexes, we can see the following benefits on a 80-core, 8-socket 1Tb server: +---------+--------------------+------------------------+-----------------------+-------------------------------+ | threads | baseline (ops/sec) | aligned-only (ops/sec) | large table (ops/sec) | large table+aligned (ops/sec) | +---------+--------------------+------------------------+-----------------------+-------------------------------+ |     512 |              32426 | 50531  (+55.8%)        | 255274  (+687.2%)     | 292553  (+802.2%)             | |     256 |              65360 | 99588  (+52.3%)        | 443563  (+578.6%)     | 508088  (+677.3%)             | |     128 |             125635 | 200075 (+59.2%)        | 742613  (+491.1%)     | 835452  (+564.9%)             | |      80 |             193559 | 323425 (+67.1%)        | 1028147 (+431.1%)     | 1130304 (+483.9%)             | |      64 |             247667 | 443740 (+79.1%)        | 997300  (+302.6%)     | 1145494 (+362.5%)             | |      32 |             628412 | 721401 (+14.7%)        | 965996  (+53.7%)      | 1122115 (+78.5%)              | +---------+--------------------+------------------------+-----------------------+-------------------------------+ Reviewed-by: Darren Hart <dvhart@linux.intel.com> Reviewed-by: Peter Zijlstra <peterz@infradead.org> Reviewed-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Waiman Long <Waiman.Long@hp.com> Reviewed-and-tested-by: Jason Low <jason.low2@hp.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Jeff Mahoney <jeffm@suse.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Scott Norton <scott.norton@hp.com> Cc: Tom Vaden <tom.vaden@hp.com> Cc: Aswin Chandramouleeswaran <aswin@hp.com> Link: http://lkml.kernel.org/r/1389569486-25487-3-git-send-email-davidlohr@hp.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2014-01-12 23:31:23 +00:00
for (i = 0; i < futex_hashsize; i++) {
atomic_set(&futex_queues[i].waiters, 0);
plist_head_init(&futex_queues[i].chain);
spin_lock_init(&futex_queues[i].lock);
}
return 0;
}
core_initcall(futex_init);