linux/ipc/sem.c

2400 lines
60 KiB
C
Raw Normal View History

License cleanup: add SPDX GPL-2.0 license identifier to files with no license Many source files in the tree are missing licensing information, which makes it harder for compliance tools to determine the correct license. By default all files without license information are under the default license of the kernel, which is GPL version 2. Update the files which contain no license information with the 'GPL-2.0' SPDX license identifier. The SPDX identifier is a legally binding shorthand, which can be used instead of the full boiler plate text. This patch is based on work done by Thomas Gleixner and Kate Stewart and Philippe Ombredanne. How this work was done: Patches were generated and checked against linux-4.14-rc6 for a subset of the use cases: - file had no licensing information it it. - file was a */uapi/* one with no licensing information in it, - file was a */uapi/* one with existing licensing information, Further patches will be generated in subsequent months to fix up cases where non-standard license headers were used, and references to license had to be inferred by heuristics based on keywords. The analysis to determine which SPDX License Identifier to be applied to a file was done in a spreadsheet of side by side results from of the output of two independent scanners (ScanCode & Windriver) producing SPDX tag:value files created by Philippe Ombredanne. Philippe prepared the base worksheet, and did an initial spot review of a few 1000 files. The 4.13 kernel was the starting point of the analysis with 60,537 files assessed. Kate Stewart did a file by file comparison of the scanner results in the spreadsheet to determine which SPDX license identifier(s) to be applied to the file. She confirmed any determination that was not immediately clear with lawyers working with the Linux Foundation. Criteria used to select files for SPDX license identifier tagging was: - Files considered eligible had to be source code files. - Make and config files were included as candidates if they contained >5 lines of source - File already had some variant of a license header in it (even if <5 lines). All documentation files were explicitly excluded. The following heuristics were used to determine which SPDX license identifiers to apply. - when both scanners couldn't find any license traces, file was considered to have no license information in it, and the top level COPYING file license applied. For non */uapi/* files that summary was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 11139 and resulted in the first patch in this series. If that file was a */uapi/* path one, it was "GPL-2.0 WITH Linux-syscall-note" otherwise it was "GPL-2.0". Results of that was: SPDX license identifier # files ---------------------------------------------------|------- GPL-2.0 WITH Linux-syscall-note 930 and resulted in the second patch in this series. - if a file had some form of licensing information in it, and was one of the */uapi/* ones, it was denoted with the Linux-syscall-note if any GPL family license was found in the file or had no licensing in it (per prior point). Results summary: SPDX license identifier # files ---------------------------------------------------|------ GPL-2.0 WITH Linux-syscall-note 270 GPL-2.0+ WITH Linux-syscall-note 169 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) 21 ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) 17 LGPL-2.1+ WITH Linux-syscall-note 15 GPL-1.0+ WITH Linux-syscall-note 14 ((GPL-2.0+ WITH Linux-syscall-note) OR BSD-3-Clause) 5 LGPL-2.0+ WITH Linux-syscall-note 4 LGPL-2.1 WITH Linux-syscall-note 3 ((GPL-2.0 WITH Linux-syscall-note) OR MIT) 3 ((GPL-2.0 WITH Linux-syscall-note) AND MIT) 1 and that resulted in the third patch in this series. - when the two scanners agreed on the detected license(s), that became the concluded license(s). - when there was disagreement between the two scanners (one detected a license but the other didn't, or they both detected different licenses) a manual inspection of the file occurred. - In most cases a manual inspection of the information in the file resulted in a clear resolution of the license that should apply (and which scanner probably needed to revisit its heuristics). - When it was not immediately clear, the license identifier was confirmed with lawyers working with the Linux Foundation. - If there was any question as to the appropriate license identifier, the file was flagged for further research and to be revisited later in time. In total, over 70 hours of logged manual review was done on the spreadsheet to determine the SPDX license identifiers to apply to the source files by Kate, Philippe, Thomas and, in some cases, confirmation by lawyers working with the Linux Foundation. Kate also obtained a third independent scan of the 4.13 code base from FOSSology, and compared selected files where the other two scanners disagreed against that SPDX file, to see if there was new insights. The Windriver scanner is based on an older version of FOSSology in part, so they are related. Thomas did random spot checks in about 500 files from the spreadsheets for the uapi headers and agreed with SPDX license identifier in the files he inspected. For the non-uapi files Thomas did random spot checks in about 15000 files. In initial set of patches against 4.14-rc6, 3 files were found to have copy/paste license identifier errors, and have been fixed to reflect the correct identifier. Additionally Philippe spent 10 hours this week doing a detailed manual inspection and review of the 12,461 patched files from the initial patch version early this week with: - a full scancode scan run, collecting the matched texts, detected license ids and scores - reviewing anything where there was a license detected (about 500+ files) to ensure that the applied SPDX license was correct - reviewing anything where there was no detection but the patch license was not GPL-2.0 WITH Linux-syscall-note to ensure that the applied SPDX license was correct This produced a worksheet with 20 files needing minor correction. This worksheet was then exported into 3 different .csv files for the different types of files to be modified. These .csv files were then reviewed by Greg. Thomas wrote a script to parse the csv files and add the proper SPDX tag to the file, in the format that the file expected. This script was further refined by Greg based on the output to detect more types of files automatically and to distinguish between header and source .c files (which need different comment types.) Finally Greg ran the script using the .csv files to generate the patches. Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org> Reviewed-by: Philippe Ombredanne <pombredanne@nexb.com> Reviewed-by: Thomas Gleixner <tglx@linutronix.de> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
2017-11-01 14:07:57 +00:00
// SPDX-License-Identifier: GPL-2.0
/*
* linux/ipc/sem.c
* Copyright (C) 1992 Krishna Balasubramanian
* Copyright (C) 1995 Eric Schenk, Bruno Haible
*
* /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
*
* SMP-threaded, sysctl's added
* (c) 1999 Manfred Spraul <manfred@colorfullife.com>
* Enforced range limit on SEM_UNDO
* (c) 2001 Red Hat Inc
* Lockless wakeup
* (c) 2003 Manfred Spraul <manfred@colorfullife.com>
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
* Further wakeup optimizations, documentation
* (c) 2010 Manfred Spraul <manfred@colorfullife.com>
[PATCH] Rework of IPC auditing 1) The audit_ipc_perms() function has been split into two different functions: - audit_ipc_obj() - audit_ipc_set_perm() There's a key shift here... The audit_ipc_obj() collects the uid, gid, mode, and SElinux context label of the current ipc object. This audit_ipc_obj() hook is now found in several places. Most notably, it is hooked in ipcperms(), which is called in various places around the ipc code permforming a MAC check. Additionally there are several places where *checkid() is used to validate that an operation is being performed on a valid object while not necessarily having a nearby ipcperms() call. In these locations, audit_ipc_obj() is called to ensure that the information is captured by the audit system. The audit_set_new_perm() function is called any time the permissions on the ipc object changes. In this case, the NEW permissions are recorded (and note that an audit_ipc_obj() call exists just a few lines before each instance). 2) Support for an AUDIT_IPC_SET_PERM audit message type. This allows for separate auxiliary audit records for normal operations on an IPC object and permissions changes. Note that the same struct audit_aux_data_ipcctl is used and populated, however there are separate audit_log_format statements based on the type of the message. Finally, the AUDIT_IPC block of code in audit_free_aux() was extended to handle aux messages of this new type. No more mem leaks I hope ;-) Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-02 21:07:33 +00:00
*
* support for audit of ipc object properties and permission changes
* Dustin Kirkland <dustin.kirkland@us.ibm.com>
*
* namespaces support
* OpenVZ, SWsoft Inc.
* Pavel Emelianov <xemul@openvz.org>
*
* Implementation notes: (May 2010)
* This file implements System V semaphores.
*
* User space visible behavior:
* - FIFO ordering for semop() operations (just FIFO, not starvation
* protection)
* - multiple semaphore operations that alter the same semaphore in
* one semop() are handled.
* - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
* SETALL calls.
* - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
* - undo adjustments at process exit are limited to 0..SEMVMX.
* - namespace are supported.
* - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
* to /proc/sys/kernel/sem.
* - statistics about the usage are reported in /proc/sysvipc/sem.
*
* Internals:
* - scalability:
* - all global variables are read-mostly.
* - semop() calls and semctl(RMID) are synchronized by RCU.
* - most operations do write operations (actually: spin_lock calls) to
* the per-semaphore array structure.
* Thus: Perfect SMP scaling between independent semaphore arrays.
* If multiple semaphores in one array are used, then cache line
* trashing on the semaphore array spinlock will limit the scaling.
* - semncnt and semzcnt are calculated on demand in count_semcnt()
* - the task that performs a successful semop() scans the list of all
* sleeping tasks and completes any pending operations that can be fulfilled.
* Semaphores are actively given to waiting tasks (necessary for FIFO).
* (see update_queue())
* - To improve the scalability, the actual wake-up calls are performed after
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* dropping all locks. (see wake_up_sem_queue_prepare())
* - All work is done by the waker, the woken up task does not have to do
* anything - not even acquiring a lock or dropping a refcount.
* - A woken up task may not even touch the semaphore array anymore, it may
* have been destroyed already by a semctl(RMID).
* - UNDO values are stored in an array (one per process and per
* semaphore array, lazily allocated). For backwards compatibility, multiple
* modes for the UNDO variables are supported (per process, per thread)
* (see copy_semundo, CLONE_SYSVSEM)
* - There are two lists of the pending operations: a per-array list
* and per-semaphore list (stored in the array). This allows to achieve FIFO
* ordering without always scanning all pending operations.
* The worst-case behavior is nevertheless O(N^2) for N wakeups.
*/
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/init.h>
#include <linux/proc_fs.h>
#include <linux/time.h>
#include <linux/security.h>
#include <linux/syscalls.h>
#include <linux/audit.h>
#include <linux/capability.h>
#include <linux/seq_file.h>
#include <linux/rwsem.h>
#include <linux/nsproxy.h>
namespaces: move the IPC namespace under IPC_NS option Currently the IPC namespace management code is spread over the ipc/*.c files. I moved this code into ipc/namespace.c file which is compiled out when needed. The linux/ipc_namespace.h file is used to store the prototypes of the functions in namespace.c and the stubs for NAMESPACES=n case. This is done so, because the stub for copy_ipc_namespace requires the knowledge of the CLONE_NEWIPC flag, which is in sched.h. But the linux/ipc.h file itself in included into many many .c files via the sys.h->sem.h sequence so adding the sched.h into it will make all these .c depend on sched.h which is not that good. On the other hand the knowledge about the namespaces stuff is required in 4 .c files only. Besides, this patch compiles out some auxiliary functions from ipc/sem.c, msg.c and shm.c files. It turned out that moving these functions into namespaces.c is not that easy because they use many other calls and macros from the original file. Moving them would make this patch complicated. On the other hand all these functions can be consolidated, so I will send a separate patch doing this a bit later. Signed-off-by: Pavel Emelyanov <xemul@openvz.org> Acked-by: Serge Hallyn <serue@us.ibm.com> Cc: Cedric Le Goater <clg@fr.ibm.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: Kirill Korotaev <dev@sw.ru> Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:18:22 +00:00
#include <linux/ipc_namespace.h>
#include <linux/sched/wake_q.h>
#include <linux/uaccess.h>
#include "util.h"
/* One semaphore structure for each semaphore in the system. */
struct sem {
int semval; /* current value */
/*
* PID of the process that last modified the semaphore. For
* Linux, specifically these are:
* - semop
* - semctl, via SETVAL and SETALL.
* - at task exit when performing undo adjustments (see exit_sem).
*/
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
struct pid *sempid;
spinlock_t lock; /* spinlock for fine-grained semtimedop */
struct list_head pending_alter; /* pending single-sop operations */
/* that alter the semaphore */
struct list_head pending_const; /* pending single-sop operations */
/* that do not alter the semaphore*/
time_t sem_otime; /* candidate for sem_otime */
} ____cacheline_aligned_in_smp;
/* One sem_array data structure for each set of semaphores in the system. */
struct sem_array {
struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */
time64_t sem_ctime; /* create/last semctl() time */
struct list_head pending_alter; /* pending operations */
/* that alter the array */
struct list_head pending_const; /* pending complex operations */
/* that do not alter semvals */
struct list_head list_id; /* undo requests on this array */
int sem_nsems; /* no. of semaphores in array */
int complex_count; /* pending complex operations */
unsigned int use_global_lock;/* >0: global lock required */
struct sem sems[];
} __randomize_layout;
/* One queue for each sleeping process in the system. */
struct sem_queue {
struct list_head list; /* queue of pending operations */
struct task_struct *sleeper; /* this process */
struct sem_undo *undo; /* undo structure */
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
struct pid *pid; /* process id of requesting process */
int status; /* completion status of operation */
struct sembuf *sops; /* array of pending operations */
struct sembuf *blocking; /* the operation that blocked */
int nsops; /* number of operations */
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
bool alter; /* does *sops alter the array? */
bool dupsop; /* sops on more than one sem_num */
};
/* Each task has a list of undo requests. They are executed automatically
* when the process exits.
*/
struct sem_undo {
struct list_head list_proc; /* per-process list: *
* all undos from one process
* rcu protected */
struct rcu_head rcu; /* rcu struct for sem_undo */
struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
struct list_head list_id; /* per semaphore array list:
* all undos for one array */
int semid; /* semaphore set identifier */
short *semadj; /* array of adjustments */
/* one per semaphore */
};
/* sem_undo_list controls shared access to the list of sem_undo structures
* that may be shared among all a CLONE_SYSVSEM task group.
*/
struct sem_undo_list {
refcount_t refcnt;
spinlock_t lock;
struct list_head list_proc;
};
#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
static int newary(struct ipc_namespace *, struct ipc_params *);
static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
#ifdef CONFIG_PROC_FS
static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
#endif
#define SEMMSL_FAST 256 /* 512 bytes on stack */
#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
/*
* Switching from the mode suitable for simple ops
* to the mode for complex ops is costly. Therefore:
* use some hysteresis
*/
#define USE_GLOBAL_LOCK_HYSTERESIS 10
/*
* Locking:
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* a) global sem_lock() for read/write
* sem_undo.id_next,
* sem_array.complex_count,
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* sem_array.pending{_alter,_const},
* sem_array.sem_undo
*
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* b) global or semaphore sem_lock() for read/write:
* sem_array.sems[i].pending_{const,alter}:
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
*
* c) special:
* sem_undo_list.list_proc:
* * undo_list->lock for write
* * rcu for read
* use_global_lock:
* * global sem_lock() for write
* * either local or global sem_lock() for read.
*
* Memory ordering:
* Most ordering is enforced by using spin_lock() and spin_unlock().
* The special case is use_global_lock:
* Setting it from non-zero to 0 is a RELEASE, this is ensured by
* using smp_store_release().
* Testing if it is non-zero is an ACQUIRE, this is ensured by using
* smp_load_acquire().
* Setting it from 0 to non-zero must be ordered with regards to
* this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
* is inside a spin_lock() and after a write from 0 to non-zero a
* spin_lock()+spin_unlock() is done.
*/
#define sc_semmsl sem_ctls[0]
#define sc_semmns sem_ctls[1]
#define sc_semopm sem_ctls[2]
#define sc_semmni sem_ctls[3]
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
int sem_init_ns(struct ipc_namespace *ns)
{
ns->sc_semmsl = SEMMSL;
ns->sc_semmns = SEMMNS;
ns->sc_semopm = SEMOPM;
ns->sc_semmni = SEMMNI;
ns->used_sems = 0;
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
return ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
}
namespaces: move the IPC namespace under IPC_NS option Currently the IPC namespace management code is spread over the ipc/*.c files. I moved this code into ipc/namespace.c file which is compiled out when needed. The linux/ipc_namespace.h file is used to store the prototypes of the functions in namespace.c and the stubs for NAMESPACES=n case. This is done so, because the stub for copy_ipc_namespace requires the knowledge of the CLONE_NEWIPC flag, which is in sched.h. But the linux/ipc.h file itself in included into many many .c files via the sys.h->sem.h sequence so adding the sched.h into it will make all these .c depend on sched.h which is not that good. On the other hand the knowledge about the namespaces stuff is required in 4 .c files only. Besides, this patch compiles out some auxiliary functions from ipc/sem.c, msg.c and shm.c files. It turned out that moving these functions into namespaces.c is not that easy because they use many other calls and macros from the original file. Moving them would make this patch complicated. On the other hand all these functions can be consolidated, so I will send a separate patch doing this a bit later. Signed-off-by: Pavel Emelyanov <xemul@openvz.org> Acked-by: Serge Hallyn <serue@us.ibm.com> Cc: Cedric Le Goater <clg@fr.ibm.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: Kirill Korotaev <dev@sw.ru> Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:18:22 +00:00
#ifdef CONFIG_IPC_NS
void sem_exit_ns(struct ipc_namespace *ns)
{
free_ipcs(ns, &sem_ids(ns), freeary);
idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
}
namespaces: move the IPC namespace under IPC_NS option Currently the IPC namespace management code is spread over the ipc/*.c files. I moved this code into ipc/namespace.c file which is compiled out when needed. The linux/ipc_namespace.h file is used to store the prototypes of the functions in namespace.c and the stubs for NAMESPACES=n case. This is done so, because the stub for copy_ipc_namespace requires the knowledge of the CLONE_NEWIPC flag, which is in sched.h. But the linux/ipc.h file itself in included into many many .c files via the sys.h->sem.h sequence so adding the sched.h into it will make all these .c depend on sched.h which is not that good. On the other hand the knowledge about the namespaces stuff is required in 4 .c files only. Besides, this patch compiles out some auxiliary functions from ipc/sem.c, msg.c and shm.c files. It turned out that moving these functions into namespaces.c is not that easy because they use many other calls and macros from the original file. Moving them would make this patch complicated. On the other hand all these functions can be consolidated, so I will send a separate patch doing this a bit later. Signed-off-by: Pavel Emelyanov <xemul@openvz.org> Acked-by: Serge Hallyn <serue@us.ibm.com> Cc: Cedric Le Goater <clg@fr.ibm.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Herbert Poetzl <herbert@13thfloor.at> Cc: Kirill Korotaev <dev@sw.ru> Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-02-08 12:18:22 +00:00
#endif
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
int __init sem_init(void)
{
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
const int err = sem_init_ns(&init_ipc_ns);
ipc_init_proc_interface("sysvipc/sem",
" key semid perms nsems uid gid cuid cgid otime ctime\n",
IPC_SEM_IDS, sysvipc_sem_proc_show);
ipc: optimize semget/shmget/msgget for lots of keys ipc_findkey() used to scan all objects to look for the wanted key. This is slow when using a high number of keys. This change adds an rhashtable of kern_ipc_perm objects in ipc_ids, so that one lookup cease to be O(n). This change gives a 865% improvement of benchmark reaim.jobs_per_min on a 56 threads Intel(R) Xeon(R) CPU E5-2695 v3 @ 2.30GHz with 256G memory [1] Other (more micro) benchmark results, by the author: On an i5 laptop, the following loop executed right after a reboot took, without and with this change: for (int i = 0, k=0x424242; i < KEYS; ++i) semget(k++, 1, IPC_CREAT | 0600); total total max single max single KEYS without with call without call with 1 3.5 4.9 µs 3.5 4.9 10 7.6 8.6 µs 3.7 4.7 32 16.2 15.9 µs 4.3 5.3 100 72.9 41.8 µs 3.7 4.7 1000 5,630.0 502.0 µs * * 10000 1,340,000.0 7,240.0 µs * * 31900 17,600,000.0 22,200.0 µs * * *: unreliable measure: high variance The duration for a lookup-only usage was obtained by the same loop once the keys are present: total total max single max single KEYS without with call without call with 1 2.1 2.5 µs 2.1 2.5 10 4.5 4.8 µs 2.2 2.3 32 13.0 10.8 µs 2.3 2.8 100 82.9 25.1 µs * 2.3 1000 5,780.0 217.0 µs * * 10000 1,470,000.0 2,520.0 µs * * 31900 17,400,000.0 7,810.0 µs * * Finally, executing each semget() in a new process gave, when still summing only the durations of these syscalls: creation: total total KEYS without with 1 3.7 5.0 µs 10 32.9 36.7 µs 32 125.0 109.0 µs 100 523.0 353.0 µs 1000 20,300.0 3,280.0 µs 10000 2,470,000.0 46,700.0 µs 31900 27,800,000.0 219,000.0 µs lookup-only: total total KEYS without with 1 2.5 2.7 µs 10 25.4 24.4 µs 32 106.0 72.6 µs 100 591.0 352.0 µs 1000 22,400.0 2,250.0 µs 10000 2,510,000.0 25,700.0 µs 31900 28,200,000.0 115,000.0 µs [1] http://lkml.kernel.org/r/20170814060507.GE23258@yexl-desktop Link: http://lkml.kernel.org/r/20170815194954.ck32ta2z35yuzpwp@debix Signed-off-by: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Reviewed-by: Marc Pardo <marc.pardo@supersonicimagine.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: "Peter Zijlstra (Intel)" <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Serge Hallyn <serge@hallyn.com> Cc: Andrey Vagin <avagin@openvz.org> Cc: Guillaume Knispel <guillaume.knispel@supersonicimagine.com> Cc: Marc Pardo <marc.pardo@supersonicimagine.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-09-08 23:17:55 +00:00
return err;
}
/**
* unmerge_queues - unmerge queues, if possible.
* @sma: semaphore array
*
* The function unmerges the wait queues if complex_count is 0.
* It must be called prior to dropping the global semaphore array lock.
*/
static void unmerge_queues(struct sem_array *sma)
{
struct sem_queue *q, *tq;
/* complex operations still around? */
if (sma->complex_count)
return;
/*
* We will switch back to simple mode.
* Move all pending operation back into the per-semaphore
* queues.
*/
list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
struct sem *curr;
curr = &sma->sems[q->sops[0].sem_num];
list_add_tail(&q->list, &curr->pending_alter);
}
INIT_LIST_HEAD(&sma->pending_alter);
}
/**
* merge_queues - merge single semop queues into global queue
* @sma: semaphore array
*
* This function merges all per-semaphore queues into the global queue.
* It is necessary to achieve FIFO ordering for the pending single-sop
* operations when a multi-semop operation must sleep.
* Only the alter operations must be moved, the const operations can stay.
*/
static void merge_queues(struct sem_array *sma)
{
int i;
for (i = 0; i < sma->sem_nsems; i++) {
struct sem *sem = &sma->sems[i];
list_splice_init(&sem->pending_alter, &sma->pending_alter);
}
}
ipc: fix race with LSMs Currently, IPC mechanisms do security and auditing related checks under RCU. However, since security modules can free the security structure, for example, through selinux_[sem,msg_queue,shm]_free_security(), we can race if the structure is freed before other tasks are done with it, creating a use-after-free condition. Manfred illustrates this nicely, for instance with shared mem and selinux: -> do_shmat calls rcu_read_lock() -> do_shmat calls shm_object_check(). Checks that the object is still valid - but doesn't acquire any locks. Then it returns. -> do_shmat calls security_shm_shmat (e.g. selinux_shm_shmat) -> selinux_shm_shmat calls ipc_has_perm() -> ipc_has_perm accesses ipc_perms->security shm_close() -> shm_close acquires rw_mutex & shm_lock -> shm_close calls shm_destroy -> shm_destroy calls security_shm_free (e.g. selinux_shm_free_security) -> selinux_shm_free_security calls ipc_free_security(&shp->shm_perm) -> ipc_free_security calls kfree(ipc_perms->security) This patch delays the freeing of the security structures after all RCU readers are done. Furthermore it aligns the security life cycle with that of the rest of IPC - freeing them based on the reference counter. For situations where we need not free security, the current behavior is kept. Linus states: "... the old behavior was suspect for another reason too: having the security blob go away from under a user sounds like it could cause various other problems anyway, so I think the old code was at least _prone_ to bugs even if it didn't have catastrophic behavior." I have tested this patch with IPC testcases from LTP on both my quad-core laptop and on a 64 core NUMA server. In both cases selinux is enabled, and tests pass for both voluntary and forced preemption models. While the mentioned races are theoretical (at least no one as reported them), I wanted to make sure that this new logic doesn't break anything we weren't aware of. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-24 00:04:45 +00:00
static void sem_rcu_free(struct rcu_head *head)
{
struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
ipc: fix race with LSMs Currently, IPC mechanisms do security and auditing related checks under RCU. However, since security modules can free the security structure, for example, through selinux_[sem,msg_queue,shm]_free_security(), we can race if the structure is freed before other tasks are done with it, creating a use-after-free condition. Manfred illustrates this nicely, for instance with shared mem and selinux: -> do_shmat calls rcu_read_lock() -> do_shmat calls shm_object_check(). Checks that the object is still valid - but doesn't acquire any locks. Then it returns. -> do_shmat calls security_shm_shmat (e.g. selinux_shm_shmat) -> selinux_shm_shmat calls ipc_has_perm() -> ipc_has_perm accesses ipc_perms->security shm_close() -> shm_close acquires rw_mutex & shm_lock -> shm_close calls shm_destroy -> shm_destroy calls security_shm_free (e.g. selinux_shm_free_security) -> selinux_shm_free_security calls ipc_free_security(&shp->shm_perm) -> ipc_free_security calls kfree(ipc_perms->security) This patch delays the freeing of the security structures after all RCU readers are done. Furthermore it aligns the security life cycle with that of the rest of IPC - freeing them based on the reference counter. For situations where we need not free security, the current behavior is kept. Linus states: "... the old behavior was suspect for another reason too: having the security blob go away from under a user sounds like it could cause various other problems anyway, so I think the old code was at least _prone_ to bugs even if it didn't have catastrophic behavior." I have tested this patch with IPC testcases from LTP on both my quad-core laptop and on a 64 core NUMA server. In both cases selinux is enabled, and tests pass for both voluntary and forced preemption models. While the mentioned races are theoretical (at least no one as reported them), I wanted to make sure that this new logic doesn't break anything we weren't aware of. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-24 00:04:45 +00:00
security_sem_free(&sma->sem_perm);
kvfree(sma);
ipc: fix race with LSMs Currently, IPC mechanisms do security and auditing related checks under RCU. However, since security modules can free the security structure, for example, through selinux_[sem,msg_queue,shm]_free_security(), we can race if the structure is freed before other tasks are done with it, creating a use-after-free condition. Manfred illustrates this nicely, for instance with shared mem and selinux: -> do_shmat calls rcu_read_lock() -> do_shmat calls shm_object_check(). Checks that the object is still valid - but doesn't acquire any locks. Then it returns. -> do_shmat calls security_shm_shmat (e.g. selinux_shm_shmat) -> selinux_shm_shmat calls ipc_has_perm() -> ipc_has_perm accesses ipc_perms->security shm_close() -> shm_close acquires rw_mutex & shm_lock -> shm_close calls shm_destroy -> shm_destroy calls security_shm_free (e.g. selinux_shm_free_security) -> selinux_shm_free_security calls ipc_free_security(&shp->shm_perm) -> ipc_free_security calls kfree(ipc_perms->security) This patch delays the freeing of the security structures after all RCU readers are done. Furthermore it aligns the security life cycle with that of the rest of IPC - freeing them based on the reference counter. For situations where we need not free security, the current behavior is kept. Linus states: "... the old behavior was suspect for another reason too: having the security blob go away from under a user sounds like it could cause various other problems anyway, so I think the old code was at least _prone_ to bugs even if it didn't have catastrophic behavior." I have tested this patch with IPC testcases from LTP on both my quad-core laptop and on a 64 core NUMA server. In both cases selinux is enabled, and tests pass for both voluntary and forced preemption models. While the mentioned races are theoretical (at least no one as reported them), I wanted to make sure that this new logic doesn't break anything we weren't aware of. Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-24 00:04:45 +00:00
}
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
/*
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* Enter the mode suitable for non-simple operations:
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
* Caller must own sem_perm.lock.
*/
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
static void complexmode_enter(struct sem_array *sma)
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
{
int i;
struct sem *sem;
if (sma->use_global_lock > 0) {
/*
* We are already in global lock mode.
* Nothing to do, just reset the
* counter until we return to simple mode.
*/
sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
return;
}
sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
for (i = 0; i < sma->sem_nsems; i++) {
sem = &sma->sems[i];
spin_lock(&sem->lock);
spin_unlock(&sem->lock);
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
}
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
}
/*
* Try to leave the mode that disallows simple operations:
* Caller must own sem_perm.lock.
*/
static void complexmode_tryleave(struct sem_array *sma)
{
if (sma->complex_count) {
/* Complex ops are sleeping.
* We must stay in complex mode
*/
return;
}
if (sma->use_global_lock == 1) {
/*
* Immediately after setting use_global_lock to 0,
* a simple op can start. Thus: all memory writes
* performed by the current operation must be visible
* before we set use_global_lock to 0.
*/
smp_store_release(&sma->use_global_lock, 0);
} else {
sma->use_global_lock--;
}
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
}
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
#define SEM_GLOBAL_LOCK (-1)
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
/*
* If the request contains only one semaphore operation, and there are
* no complex transactions pending, lock only the semaphore involved.
* Otherwise, lock the entire semaphore array, since we either have
* multiple semaphores in our own semops, or we need to look at
* semaphores from other pending complex operations.
*/
static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
int nsops)
{
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
struct sem *sem;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
if (nsops != 1) {
/* Complex operation - acquire a full lock */
ipc_lock_object(&sma->sem_perm);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
/* Prevent parallel simple ops */
complexmode_enter(sma);
return SEM_GLOBAL_LOCK;
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
}
/*
* Only one semaphore affected - try to optimize locking.
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* Optimized locking is possible if no complex operation
* is either enqueued or processed right now.
*
* Both facts are tracked by use_global_mode.
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
*/
sem = &sma->sems[sops->sem_num];
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
/*
* Initial check for use_global_lock. Just an optimization,
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* no locking, no memory barrier.
*/
if (!sma->use_global_lock) {
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
/*
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
* It appears that no complex operation is around.
* Acquire the per-semaphore lock.
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
*/
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
spin_lock(&sem->lock);
/* pairs with smp_store_release() */
if (!smp_load_acquire(&sma->use_global_lock)) {
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
/* fast path successful! */
return sops->sem_num;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
spin_unlock(&sem->lock);
}
/* slow path: acquire the full lock */
ipc_lock_object(&sma->sem_perm);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (sma->use_global_lock == 0) {
/*
* The use_global_lock mode ended while we waited for
* sma->sem_perm.lock. Thus we must switch to locking
* with sem->lock.
* Unlike in the fast path, there is no need to recheck
* sma->use_global_lock after we have acquired sem->lock:
* We own sma->sem_perm.lock, thus use_global_lock cannot
* change.
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
*/
spin_lock(&sem->lock);
ipc/sem.c: fix race in sem_lock() The exclusion of complex operations in sem_lock() is insufficient: after acquiring the per-semaphore lock, a simple op must first check that sem_perm.lock is not locked and only after that test check complex_count. The current code does it the other way around - and that creates a race. Details are below. The patch is a complete rewrite of sem_lock(), based in part on the code from Mike Galbraith. It removes all gotos and all loops and thus the risk of livelocks. I have tested the patch (together with the next one) on my i3 laptop and it didn't cause any problems. The bug is probably also present in 3.10 and 3.11, but for these kernels it might be simpler just to move the test of sma->complex_count after the spin_is_locked() test. Details of the bug: Assume: - sma->complex_count = 0. - Thread 1: semtimedop(complex op that must sleep) - Thread 2: semtimedop(simple op). Pseudo-Trace: Thread 1: sem_lock(): acquire sem_perm.lock Thread 1: sem_lock(): check for ongoing simple ops Nothing ongoing, thread 2 is still before sem_lock(). Thread 1: try_atomic_semop() <<< preempted. Thread 2: sem_lock(): static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, int nsops) { int locknum; again: if (nsops == 1 && !sma->complex_count) { struct sem *sem = sma->sem_base + sops->sem_num; /* Lock just the semaphore we are interested in. */ spin_lock(&sem->lock); /* * If sma->complex_count was set while we were spinning, * we may need to look at things we did not lock here. */ if (unlikely(sma->complex_count)) { spin_unlock(&sem->lock); goto lock_array; } <<<<<<<<< <<< complex_count is still 0. <<< <<< Here it is preempted <<<<<<<<< Thread 1: try_atomic_semop() returns, notices that it must sleep. Thread 1: increases sma->complex_count. Thread 1: drops sem_perm.lock Thread 2: /* * Another process is holding the global lock on the * sem_array; we cannot enter our critical section, * but have to wait for the global lock to be released. */ if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { spin_unlock(&sem->lock); spin_unlock_wait(&sma->sem_perm.lock); goto again; } <<< sem_perm.lock already dropped, thus no "goto again;" locknum = sops->sem_num; Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Mike Galbraith <bitbucket@online.de> Cc: Rik van Riel <riel@redhat.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-09-30 20:45:04 +00:00
ipc_unlock_object(&sma->sem_perm);
return sops->sem_num;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
} else {
/*
* Not a false alarm, thus continue to use the global lock
* mode. No need for complexmode_enter(), this was done by
* the caller that has set use_global_mode to non-zero.
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
*/
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
return SEM_GLOBAL_LOCK;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
}
static inline void sem_unlock(struct sem_array *sma, int locknum)
{
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
if (locknum == SEM_GLOBAL_LOCK) {
unmerge_queues(sma);
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
complexmode_tryleave(sma);
ipc_unlock_object(&sma->sem_perm);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
} else {
struct sem *sem = &sma->sems[locknum];
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
spin_unlock(&sem->lock);
}
}
/*
* sem_lock_(check_) routines are called in the paths where the rwsem
* is not held.
*
* The caller holds the RCU read lock.
*/
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
{
struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
if (IS_ERR(ipcp))
return ERR_CAST(ipcp);
return container_of(ipcp, struct sem_array, sem_perm);
}
static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
int id)
{
struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
if (IS_ERR(ipcp))
return ERR_CAST(ipcp);
return container_of(ipcp, struct sem_array, sem_perm);
}
static inline void sem_lock_and_putref(struct sem_array *sma)
{
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
}
static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
{
ipc_rmid(&sem_ids(ns), &s->sem_perm);
}
static struct sem_array *sem_alloc(size_t nsems)
{
struct sem_array *sma;
size_t size;
if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
return NULL;
size = sizeof(*sma) + nsems * sizeof(sma->sems[0]);
sma = kvmalloc(size, GFP_KERNEL);
if (unlikely(!sma))
return NULL;
memset(sma, 0, size);
return sma;
}
/**
* newary - Create a new semaphore set
* @ns: namespace
* @params: ptr to the structure that contains key, semflg and nsems
*
* Called with sem_ids.rwsem held (as a writer)
*/
static int newary(struct ipc_namespace *ns, struct ipc_params *params)
{
int retval;
struct sem_array *sma;
key_t key = params->key;
int nsems = params->u.nsems;
int semflg = params->flg;
int i;
if (!nsems)
return -EINVAL;
if (ns->used_sems + nsems > ns->sc_semmns)
return -ENOSPC;
sma = sem_alloc(nsems);
if (!sma)
return -ENOMEM;
sma->sem_perm.mode = (semflg & S_IRWXUGO);
sma->sem_perm.key = key;
sma->sem_perm.security = NULL;
retval = security_sem_alloc(&sma->sem_perm);
if (retval) {
kvfree(sma);
return retval;
}
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
for (i = 0; i < nsems; i++) {
INIT_LIST_HEAD(&sma->sems[i].pending_alter);
INIT_LIST_HEAD(&sma->sems[i].pending_const);
spin_lock_init(&sma->sems[i].lock);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
sma->complex_count = 0;
sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
INIT_LIST_HEAD(&sma->pending_alter);
INIT_LIST_HEAD(&sma->pending_const);
INIT_LIST_HEAD(&sma->list_id);
sma->sem_nsems = nsems;
sma->sem_ctime = ktime_get_real_seconds();
/* ipc_addid() locks sma upon success. */
retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
if (retval < 0) {
call_rcu(&sma->sem_perm.rcu, sem_rcu_free);
return retval;
}
ns->used_sems += nsems;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
return sma->sem_perm.id;
}
/*
* Called with sem_ids.rwsem and ipcp locked.
*/
static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
struct ipc_params *params)
{
struct sem_array *sma;
sma = container_of(ipcp, struct sem_array, sem_perm);
if (params->u.nsems > sma->sem_nsems)
return -EINVAL;
return 0;
}
long ksys_semget(key_t key, int nsems, int semflg)
{
struct ipc_namespace *ns;
static const struct ipc_ops sem_ops = {
.getnew = newary,
.associate = security_sem_associate,
.more_checks = sem_more_checks,
};
struct ipc_params sem_params;
ns = current->nsproxy->ipc_ns;
if (nsems < 0 || nsems > ns->sc_semmsl)
return -EINVAL;
sem_params.key = key;
sem_params.flg = semflg;
sem_params.u.nsems = nsems;
return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
}
SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
{
return ksys_semget(key, nsems, semflg);
}
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
/**
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
* perform_atomic_semop[_slow] - Attempt to perform semaphore
* operations on a given array.
* @sma: semaphore array
* @q: struct sem_queue that describes the operation
*
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
* Caller blocking are as follows, based the value
* indicated by the semaphore operation (sem_op):
*
* (1) >0 never blocks.
* (2) 0 (wait-for-zero operation): semval is non-zero.
* (3) <0 attempting to decrement semval to a value smaller than zero.
*
* Returns 0 if the operation was possible.
* Returns 1 if the operation is impossible, the caller must sleep.
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
* Returns <0 for error codes.
*/
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
{
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
int result, sem_op, nsops;
struct pid *pid;
struct sembuf *sop;
struct sem *curr;
struct sembuf *sops;
struct sem_undo *un;
sops = q->sops;
nsops = q->nsops;
un = q->undo;
for (sop = sops; sop < sops + nsops; sop++) {
curr = &sma->sems[sop->sem_num];
sem_op = sop->sem_op;
result = curr->semval;
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
if (!sem_op && result)
goto would_block;
result += sem_op;
if (result < 0)
goto would_block;
if (result > SEMVMX)
goto out_of_range;
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
if (sop->sem_flg & SEM_UNDO) {
int undo = un->semadj[sop->sem_num] - sem_op;
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
/* Exceeding the undo range is an error. */
if (undo < (-SEMAEM - 1) || undo > SEMAEM)
goto out_of_range;
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
un->semadj[sop->sem_num] = undo;
}
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
curr->semval = result;
}
sop--;
pid = q->pid;
while (sop >= sops) {
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
sop--;
}
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
return 0;
out_of_range:
result = -ERANGE;
goto undo;
would_block:
q->blocking = sop;
if (sop->sem_flg & IPC_NOWAIT)
result = -EAGAIN;
else
result = 1;
undo:
sop--;
while (sop >= sops) {
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
sem_op = sop->sem_op;
sma->sems[sop->sem_num].semval -= sem_op;
ipc/sem.c: avoid overflow of semop undo (semadj) value When trying to understand semop code, I found a small mistake in the check for semadj (undo) value overflow. The new undo value is not stored immediately and next potential checks are done against the old value. The failing scenario is not much practical. One semop call has to do more operations on the same semaphore. Also semval and semadj must have different values, so there has to be some operations without SEM_UNDO flag. For example: struct sembuf depositor_op[1]; struct sembuf collector_op[2]; depositor_op[0].sem_num = 0; depositor_op[0].sem_op = 20000; depositor_op[0].sem_flg = 0; collector_op[0].sem_num = 0; collector_op[0].sem_op = -10000; collector_op[0].sem_flg = SEM_UNDO; collector_op[1].sem_num = 0; collector_op[1].sem_op = -10000; collector_op[1].sem_flg = SEM_UNDO; if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 1st deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 1st collect"); return 1; } if (semop(semid, depositor_op, 1) == -1) { perror("Failed to do 2nd deposit"); return 1; } if (semop(semid, collector_op, 2) == -1) { perror("Failed to do 2nd collect"); return 1; } return 0; It passes without error now but the semadj value has overflown in the 2nd collector operation. [akpm@linux-foundation.org: restore lessened scope of local `undo'] [davidlohr@hp.com: correct header comment for perform_atomic_semop] Signed-off-by: Petr Mladek <pmladek@suse.cz> Acked-by: Davidlohr Bueso <davidlohr@hp.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Jiri Kosina <jkosina@suse.cz> Signed-off-by: Davidlohr Bueso <davidlohr@hp.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-01-28 01:07:00 +00:00
if (sop->sem_flg & SEM_UNDO)
un->semadj[sop->sem_num] += sem_op;
sop--;
}
return result;
}
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
{
int result, sem_op, nsops;
struct sembuf *sop;
struct sem *curr;
struct sembuf *sops;
struct sem_undo *un;
sops = q->sops;
nsops = q->nsops;
un = q->undo;
if (unlikely(q->dupsop))
return perform_atomic_semop_slow(sma, q);
/*
* We scan the semaphore set twice, first to ensure that the entire
* operation can succeed, therefore avoiding any pointless writes
* to shared memory and having to undo such changes in order to block
* until the operations can go through.
*/
for (sop = sops; sop < sops + nsops; sop++) {
curr = &sma->sems[sop->sem_num];
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
sem_op = sop->sem_op;
result = curr->semval;
if (!sem_op && result)
goto would_block; /* wait-for-zero */
result += sem_op;
if (result < 0)
goto would_block;
if (result > SEMVMX)
return -ERANGE;
if (sop->sem_flg & SEM_UNDO) {
int undo = un->semadj[sop->sem_num] - sem_op;
/* Exceeding the undo range is an error. */
if (undo < (-SEMAEM - 1) || undo > SEMAEM)
return -ERANGE;
}
}
for (sop = sops; sop < sops + nsops; sop++) {
curr = &sma->sems[sop->sem_num];
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
sem_op = sop->sem_op;
result = curr->semval;
if (sop->sem_flg & SEM_UNDO) {
int undo = un->semadj[sop->sem_num] - sem_op;
un->semadj[sop->sem_num] = undo;
}
curr->semval += sem_op;
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&curr->sempid, q->pid);
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
}
return 0;
would_block:
q->blocking = sop;
return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
}
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
struct wake_q_head *wake_q)
{
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_q_add(wake_q, q->sleeper);
/*
* Rely on the above implicit barrier, such that we can
* ensure that we hold reference to the task before setting
* q->status. Otherwise we could race with do_exit if the
* task is awoken by an external event before calling
* wake_up_process().
*/
WRITE_ONCE(q->status, error);
}
static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
{
list_del(&q->list);
if (q->nsops > 1)
sma->complex_count--;
}
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
/** check_restart(sma, q)
* @sma: semaphore array
* @q: the operation that just completed
*
* update_queue is O(N^2) when it restarts scanning the whole queue of
* waiting operations. Therefore this function checks if the restart is
* really necessary. It is called after a previously waiting operation
* modified the array.
* Note that wait-for-zero operations are handled without restart.
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
*/
static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
{
/* pending complex alter operations are too difficult to analyse */
if (!list_empty(&sma->pending_alter))
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
return 1;
/* we were a sleeping complex operation. Too difficult */
if (q->nsops > 1)
return 1;
/* It is impossible that someone waits for the new value:
* - complex operations always restart.
* - wait-for-zero are handled seperately.
* - q is a previously sleeping simple operation that
* altered the array. It must be a decrement, because
* simple increments never sleep.
* - If there are older (higher priority) decrements
* in the queue, then they have observed the original
* semval value and couldn't proceed. The operation
* decremented to value - thus they won't proceed either.
*/
return 0;
}
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
/**
* wake_const_ops - wake up non-alter tasks
* @sma: semaphore array.
* @semnum: semaphore that was modified.
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* @wake_q: lockless wake-queue head.
*
* wake_const_ops must be called after a semaphore in a semaphore array
* was set to 0. If complex const operations are pending, wake_const_ops must
* be called with semnum = -1, as well as with the number of each modified
* semaphore.
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* The tasks that must be woken up are added to @wake_q. The return code
* is stored in q->pid.
* The function returns 1 if at least one operation was completed successfully.
*/
static int wake_const_ops(struct sem_array *sma, int semnum,
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
struct wake_q_head *wake_q)
{
struct sem_queue *q, *tmp;
struct list_head *pending_list;
int semop_completed = 0;
if (semnum == -1)
pending_list = &sma->pending_const;
else
pending_list = &sma->sems[semnum].pending_const;
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
list_for_each_entry_safe(q, tmp, pending_list, list) {
int error = perform_atomic_semop(sma, q);
if (error > 0)
continue;
/* operation completed, remove from queue & wakeup */
unlink_queue(sma, q);
wake_up_sem_queue_prepare(q, error, wake_q);
if (error == 0)
semop_completed = 1;
}
return semop_completed;
}
/**
* do_smart_wakeup_zero - wakeup all wait for zero tasks
* @sma: semaphore array
* @sops: operations that were performed
* @nsops: number of operations
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* @wake_q: lockless wake-queue head
*
* Checks all required queue for wait-for-zero operations, based
* on the actual changes that were performed on the semaphore array.
* The function returns 1 if at least one operation was completed successfully.
*/
static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
int nsops, struct wake_q_head *wake_q)
{
int i;
int semop_completed = 0;
int got_zero = 0;
/* first: the per-semaphore queues, if known */
if (sops) {
for (i = 0; i < nsops; i++) {
int num = sops[i].sem_num;
if (sma->sems[num].semval == 0) {
got_zero = 1;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
semop_completed |= wake_const_ops(sma, num, wake_q);
}
}
} else {
/*
* No sops means modified semaphores not known.
* Assume all were changed.
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
*/
for (i = 0; i < sma->sem_nsems; i++) {
if (sma->sems[i].semval == 0) {
got_zero = 1;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
semop_completed |= wake_const_ops(sma, i, wake_q);
}
}
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
}
/*
* If one of the modified semaphores got 0,
* then check the global queue, too.
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
*/
if (got_zero)
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
semop_completed |= wake_const_ops(sma, -1, wake_q);
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
return semop_completed;
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
}
/**
* update_queue - look for tasks that can be completed.
* @sma: semaphore array.
* @semnum: semaphore that was modified.
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* @wake_q: lockless wake-queue head.
*
* update_queue must be called after a semaphore in a semaphore array
* was modified. If multiple semaphores were modified, update_queue must
* be called with semnum = -1, as well as with the number of each modified
* semaphore.
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* The tasks that must be woken up are added to @wake_q. The return code
* is stored in q->pid.
* The function internally checks if const operations can now succeed.
*
* The function return 1 if at least one semop was completed successfully.
*/
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
{
struct sem_queue *q, *tmp;
struct list_head *pending_list;
int semop_completed = 0;
if (semnum == -1)
pending_list = &sma->pending_alter;
else
pending_list = &sma->sems[semnum].pending_alter;
again:
list_for_each_entry_safe(q, tmp, pending_list, list) {
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
int error, restart;
/* If we are scanning the single sop, per-semaphore list of
* one semaphore and that semaphore is 0, then it is not
* necessary to scan further: simple increments
* that affect only one entry succeed immediately and cannot
* be in the per semaphore pending queue, and decrements
* cannot be successful if the value is already 0.
*/
if (semnum != -1 && sma->sems[semnum].semval == 0)
break;
error = perform_atomic_semop(sma, q);
/* Does q->sleeper still need to sleep? */
if (error > 0)
continue;
unlink_queue(sma, q);
if (error) {
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
restart = 0;
} else {
semop_completed = 1;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
restart = check_restart(sma, q);
}
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_sem_queue_prepare(q, error, wake_q);
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
if (restart)
goto again;
}
return semop_completed;
}
/**
* set_semotime - set sem_otime
* @sma: semaphore array
* @sops: operations that modified the array, may be NULL
*
* sem_otime is replicated to avoid cache line trashing.
* This function sets one instance to the current time.
*/
static void set_semotime(struct sem_array *sma, struct sembuf *sops)
{
if (sops == NULL) {
sma->sems[0].sem_otime = get_seconds();
} else {
sma->sems[sops[0].sem_num].sem_otime =
get_seconds();
}
}
/**
* do_smart_update - optimized update_queue
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
* @sma: semaphore array
* @sops: operations that were performed
* @nsops: number of operations
* @otime: force setting otime
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* @wake_q: lockless wake-queue head
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
*
* do_smart_update() does the required calls to update_queue and wakeup_zero,
* based on the actual changes that were performed on the semaphore array.
* Note that the function does not do the actual wake-up: the caller is
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
* responsible for calling wake_up_q().
* It is safe to perform this call after dropping all locks.
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
*/
static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
int otime, struct wake_q_head *wake_q)
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
{
int i;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
if (!list_empty(&sma->pending_alter)) {
/* semaphore array uses the global queue - just process it. */
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
otime |= update_queue(sma, -1, wake_q);
} else {
if (!sops) {
/*
* No sops, thus the modified semaphores are not
* known. Check all.
*/
for (i = 0; i < sma->sem_nsems; i++)
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
otime |= update_queue(sma, i, wake_q);
} else {
/*
* Check the semaphores that were increased:
* - No complex ops, thus all sleeping ops are
* decrease.
* - if we decreased the value, then any sleeping
* semaphore ops wont be able to run: If the
* previous value was too small, then the new
* value will be too small, too.
*/
for (i = 0; i < nsops; i++) {
if (sops[i].sem_op > 0) {
otime |= update_queue(sma,
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
sops[i].sem_num, wake_q);
}
}
}
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
}
if (otime)
set_semotime(sma, sops);
ipc/sem.c: optimize update_queue() for bulk wakeup calls The following series of patches tries to fix the spinlock contention reported by Chris Mason - his benchmark exposes problems of the current code: - In the worst case, the algorithm used by update_queue() is O(N^2). Bulk wake-up calls can enter this worst case. The patch series fix that. Note that the benchmark app doesn't expose the problem, it just should be fixed: Real world apps might do the wake-ups in another order than perfect FIFO. - The part of the code that runs within the semaphore array spinlock is significantly larger than necessary. The patch series fixes that. This change is responsible for the main improvement. - The cacheline with the spinlock is also used for a variable that is read in the hot path (sem_base) and for a variable that is unnecessarily written to multiple times (sem_otime). The last step of the series cacheline-aligns the spinlock. This patch: The SysV semaphore code allows to perform multiple operations on all semaphores in the array as atomic operations. After a modification, update_queue() checks which of the waiting tasks can complete. The algorithm that is used to identify the tasks is O(N^2) in the worst case. For some cases, it is simple to avoid the O(N^2). The patch adds a detection logic for some cases, especially for the case of an array where all sleeping tasks are single sembuf operations and a multi-sembuf operation is used to wake up multiple tasks. A big database application uses that approach. The patch fixes wakeup due to semctl(,,SETALL,) - the initial version of the patch breaks that. [akpm@linux-foundation.org: make do_smart_update() static] Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Chris Mason <chris.mason@oracle.com> Cc: Zach Brown <zach.brown@oracle.com> Cc: Jens Axboe <jens.axboe@oracle.com> Cc: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-05-26 21:43:40 +00:00
}
/*
ipc/sem.c: make semctl(,,{GETNCNT,GETZCNT}) standard compliant SUSv4 clearly defines how semncnt and semzcnt must be calculated: A task waits on exactly one semaphore: The semaphore from the first operation in the sop array that cannot proceed. The Linux implementation never followed the standard, it tried to count all semaphores that might be the reason why a task sleeps. This patch fixes that. Note: a) The implementation assumes that GETNCNT and GETZCNT are rare operations, therefore the code counts them only on demand. (If they wouldn't be rare, then the non-compliance would have been found earlier) b) compared to the initial version of the patch, the BUG_ONs were removed and it was clarified that the new behavior conforms to SUS. Back-compatibility concerns: Manfred: : - there is no application in Fedora that uses GETNCNT or GETZCNT. : : - application that use only single-sop semop() are also safe, the : difference only affects complex apps. : : - portable application are also safe, the new behavior is standard : compliant. : : But that's it. The old behavior existed in Linux from 0.99.something : until now. Michael: : * These operations seem to be very little used. Grepping the public : source that is contained Fedora 20 source DVD, there appear to be no : uses. Of course, this says nothing about uses in private / : non-mainstream FOSS code, but it seems likely that the same pattern : is followed there. : : * The existing behavior is hard enough to understand that I suspect : that no one understood it well enough to rely on it anyway : (especially as that behavior contradicted both man page and POSIX). : : So, there's a chance of breakage, but I estimate that it's minute. Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-06 21:37:51 +00:00
* check_qop: Test if a queued operation sleeps on the semaphore semnum
*/
static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
bool count_zero)
{
ipc/sem.c: make semctl(,,{GETNCNT,GETZCNT}) standard compliant SUSv4 clearly defines how semncnt and semzcnt must be calculated: A task waits on exactly one semaphore: The semaphore from the first operation in the sop array that cannot proceed. The Linux implementation never followed the standard, it tried to count all semaphores that might be the reason why a task sleeps. This patch fixes that. Note: a) The implementation assumes that GETNCNT and GETZCNT are rare operations, therefore the code counts them only on demand. (If they wouldn't be rare, then the non-compliance would have been found earlier) b) compared to the initial version of the patch, the BUG_ONs were removed and it was clarified that the new behavior conforms to SUS. Back-compatibility concerns: Manfred: : - there is no application in Fedora that uses GETNCNT or GETZCNT. : : - application that use only single-sop semop() are also safe, the : difference only affects complex apps. : : - portable application are also safe, the new behavior is standard : compliant. : : But that's it. The old behavior existed in Linux from 0.99.something : until now. Michael: : * These operations seem to be very little used. Grepping the public : source that is contained Fedora 20 source DVD, there appear to be no : uses. Of course, this says nothing about uses in private / : non-mainstream FOSS code, but it seems likely that the same pattern : is followed there. : : * The existing behavior is hard enough to understand that I suspect : that no one understood it well enough to rely on it anyway : (especially as that behavior contradicted both man page and POSIX). : : So, there's a chance of breakage, but I estimate that it's minute. Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-06 21:37:51 +00:00
struct sembuf *sop = q->blocking;
/*
* Linux always (since 0.99.10) reported a task as sleeping on all
* semaphores. This violates SUS, therefore it was changed to the
* standard compliant behavior.
* Give the administrators a chance to notice that an application
* might misbehave because it relies on the Linux behavior.
*/
pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
"The task %s (%d) triggered the difference, watch for misbehavior.\n",
current->comm, task_pid_nr(current));
ipc/sem.c: make semctl(,,{GETNCNT,GETZCNT}) standard compliant SUSv4 clearly defines how semncnt and semzcnt must be calculated: A task waits on exactly one semaphore: The semaphore from the first operation in the sop array that cannot proceed. The Linux implementation never followed the standard, it tried to count all semaphores that might be the reason why a task sleeps. This patch fixes that. Note: a) The implementation assumes that GETNCNT and GETZCNT are rare operations, therefore the code counts them only on demand. (If they wouldn't be rare, then the non-compliance would have been found earlier) b) compared to the initial version of the patch, the BUG_ONs were removed and it was clarified that the new behavior conforms to SUS. Back-compatibility concerns: Manfred: : - there is no application in Fedora that uses GETNCNT or GETZCNT. : : - application that use only single-sop semop() are also safe, the : difference only affects complex apps. : : - portable application are also safe, the new behavior is standard : compliant. : : But that's it. The old behavior existed in Linux from 0.99.something : until now. Michael: : * These operations seem to be very little used. Grepping the public : source that is contained Fedora 20 source DVD, there appear to be no : uses. Of course, this says nothing about uses in private / : non-mainstream FOSS code, but it seems likely that the same pattern : is followed there. : : * The existing behavior is hard enough to understand that I suspect : that no one understood it well enough to rely on it anyway : (especially as that behavior contradicted both man page and POSIX). : : So, there's a chance of breakage, but I estimate that it's minute. Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-06 21:37:51 +00:00
if (sop->sem_num != semnum)
return 0;
ipc/sem.c: make semctl(,,{GETNCNT,GETZCNT}) standard compliant SUSv4 clearly defines how semncnt and semzcnt must be calculated: A task waits on exactly one semaphore: The semaphore from the first operation in the sop array that cannot proceed. The Linux implementation never followed the standard, it tried to count all semaphores that might be the reason why a task sleeps. This patch fixes that. Note: a) The implementation assumes that GETNCNT and GETZCNT are rare operations, therefore the code counts them only on demand. (If they wouldn't be rare, then the non-compliance would have been found earlier) b) compared to the initial version of the patch, the BUG_ONs were removed and it was clarified that the new behavior conforms to SUS. Back-compatibility concerns: Manfred: : - there is no application in Fedora that uses GETNCNT or GETZCNT. : : - application that use only single-sop semop() are also safe, the : difference only affects complex apps. : : - portable application are also safe, the new behavior is standard : compliant. : : But that's it. The old behavior existed in Linux from 0.99.something : until now. Michael: : * These operations seem to be very little used. Grepping the public : source that is contained Fedora 20 source DVD, there appear to be no : uses. Of course, this says nothing about uses in private / : non-mainstream FOSS code, but it seems likely that the same pattern : is followed there. : : * The existing behavior is hard enough to understand that I suspect : that no one understood it well enough to rely on it anyway : (especially as that behavior contradicted both man page and POSIX). : : So, there's a chance of breakage, but I estimate that it's minute. Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-06 21:37:51 +00:00
if (count_zero && sop->sem_op == 0)
return 1;
if (!count_zero && sop->sem_op < 0)
return 1;
return 0;
}
/* The following counts are associated to each semaphore:
* semncnt number of tasks waiting on semval being nonzero
* semzcnt number of tasks waiting on semval being zero
ipc/sem.c: make semctl(,,{GETNCNT,GETZCNT}) standard compliant SUSv4 clearly defines how semncnt and semzcnt must be calculated: A task waits on exactly one semaphore: The semaphore from the first operation in the sop array that cannot proceed. The Linux implementation never followed the standard, it tried to count all semaphores that might be the reason why a task sleeps. This patch fixes that. Note: a) The implementation assumes that GETNCNT and GETZCNT are rare operations, therefore the code counts them only on demand. (If they wouldn't be rare, then the non-compliance would have been found earlier) b) compared to the initial version of the patch, the BUG_ONs were removed and it was clarified that the new behavior conforms to SUS. Back-compatibility concerns: Manfred: : - there is no application in Fedora that uses GETNCNT or GETZCNT. : : - application that use only single-sop semop() are also safe, the : difference only affects complex apps. : : - portable application are also safe, the new behavior is standard : compliant. : : But that's it. The old behavior existed in Linux from 0.99.something : until now. Michael: : * These operations seem to be very little used. Grepping the public : source that is contained Fedora 20 source DVD, there appear to be no : uses. Of course, this says nothing about uses in private / : non-mainstream FOSS code, but it seems likely that the same pattern : is followed there. : : * The existing behavior is hard enough to understand that I suspect : that no one understood it well enough to rely on it anyway : (especially as that behavior contradicted both man page and POSIX). : : So, there's a chance of breakage, but I estimate that it's minute. Signed-off-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-06-06 21:37:51 +00:00
*
* Per definition, a task waits only on the semaphore of the first semop
* that cannot proceed, even if additional operation would block, too.
*/
static int count_semcnt(struct sem_array *sma, ushort semnum,
bool count_zero)
{
struct list_head *l;
struct sem_queue *q;
int semcnt;
semcnt = 0;
/* First: check the simple operations. They are easy to evaluate */
if (count_zero)
l = &sma->sems[semnum].pending_const;
else
l = &sma->sems[semnum].pending_alter;
list_for_each_entry(q, l, list) {
/* all task on a per-semaphore list sleep on exactly
* that semaphore
*/
semcnt++;
}
/* Then: check the complex operations. */
list_for_each_entry(q, &sma->pending_alter, list) {
semcnt += check_qop(sma, semnum, q, count_zero);
}
if (count_zero) {
list_for_each_entry(q, &sma->pending_const, list) {
semcnt += check_qop(sma, semnum, q, count_zero);
}
}
return semcnt;
}
/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
* as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
* remains locked on exit.
*/
static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
{
struct sem_undo *un, *tu;
struct sem_queue *q, *tq;
struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
int i;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
DEFINE_WAKE_Q(wake_q);
/* Free the existing undo structures for this semaphore set. */
ipc_assert_locked_object(&sma->sem_perm);
list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
list_del(&un->list_id);
spin_lock(&un->ulp->lock);
un->semid = -1;
list_del_rcu(&un->list_proc);
spin_unlock(&un->ulp->lock);
kfree_rcu(un, rcu);
}
/* Wake up all pending processes and let them fail with EIDRM. */
list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
unlink_queue(sma, q);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
}
list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
unlink_queue(sma, q);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
}
for (i = 0; i < sma->sem_nsems; i++) {
struct sem *sem = &sma->sems[i];
list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
unlink_queue(sma, q);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
}
list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
unlink_queue(sma, q);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
}
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&sem->sempid, NULL);
}
/* Remove the semaphore set from the IDR */
sem_rmid(ns, sma);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_q(&wake_q);
ns->used_sems -= sma->sem_nsems;
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
}
static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
{
switch (version) {
case IPC_64:
return copy_to_user(buf, in, sizeof(*in));
case IPC_OLD:
{
struct semid_ds out;
memset(&out, 0, sizeof(out));
ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
out.sem_otime = in->sem_otime;
out.sem_ctime = in->sem_ctime;
out.sem_nsems = in->sem_nsems;
return copy_to_user(buf, &out, sizeof(out));
}
default:
return -EINVAL;
}
}
static time64_t get_semotime(struct sem_array *sma)
{
int i;
time64_t res;
res = sma->sems[0].sem_otime;
for (i = 1; i < sma->sem_nsems; i++) {
time64_t to = sma->sems[i].sem_otime;
if (to > res)
res = to;
}
return res;
}
static int semctl_stat(struct ipc_namespace *ns, int semid,
int cmd, struct semid64_ds *semid64)
{
struct sem_array *sma;
int id = 0;
int err;
memset(semid64, 0, sizeof(*semid64));
rcu_read_lock();
ipc/sem: introduce semctl(SEM_STAT_ANY) There is a permission discrepancy when consulting shm ipc object metadata between /proc/sysvipc/sem (0444) and the SEM_STAT semctl command. The later does permission checks for the object vs S_IRUGO. As such there can be cases where EACCESS is returned via syscall but the info is displayed anyways in the procfs files. While this might have security implications via info leaking (albeit no writing to the sma metadata), this behavior goes way back and showing all the objects regardless of the permissions was most likely an overlook - so we are stuck with it. Furthermore, modifying either the syscall or the procfs file can cause userspace programs to break (ie ipcs). Some applications require getting the procfs info (without root privileges) and can be rather slow in comparison with a syscall -- up to 500x in some reported cases for shm. This patch introduces a new SEM_STAT_ANY command such that the sem ipc object permissions are ignored, and only audited instead. In addition, I've left the lsm security hook checks in place, as if some policy can block the call, then the user has no other choice than just parsing the procfs file. Link: http://lkml.kernel.org/r/20180215162458.10059-3-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reported-by: Robert Kettler <robert.kettler@outlook.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-10 23:35:26 +00:00
if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
sma = sem_obtain_object(ns, semid);
if (IS_ERR(sma)) {
err = PTR_ERR(sma);
goto out_unlock;
}
id = sma->sem_perm.id;
ipc/sem: introduce semctl(SEM_STAT_ANY) There is a permission discrepancy when consulting shm ipc object metadata between /proc/sysvipc/sem (0444) and the SEM_STAT semctl command. The later does permission checks for the object vs S_IRUGO. As such there can be cases where EACCESS is returned via syscall but the info is displayed anyways in the procfs files. While this might have security implications via info leaking (albeit no writing to the sma metadata), this behavior goes way back and showing all the objects regardless of the permissions was most likely an overlook - so we are stuck with it. Furthermore, modifying either the syscall or the procfs file can cause userspace programs to break (ie ipcs). Some applications require getting the procfs info (without root privileges) and can be rather slow in comparison with a syscall -- up to 500x in some reported cases for shm. This patch introduces a new SEM_STAT_ANY command such that the sem ipc object permissions are ignored, and only audited instead. In addition, I've left the lsm security hook checks in place, as if some policy can block the call, then the user has no other choice than just parsing the procfs file. Link: http://lkml.kernel.org/r/20180215162458.10059-3-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reported-by: Robert Kettler <robert.kettler@outlook.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-10 23:35:26 +00:00
} else { /* IPC_STAT */
sma = sem_obtain_object_check(ns, semid);
if (IS_ERR(sma)) {
err = PTR_ERR(sma);
goto out_unlock;
}
}
ipc/sem: introduce semctl(SEM_STAT_ANY) There is a permission discrepancy when consulting shm ipc object metadata between /proc/sysvipc/sem (0444) and the SEM_STAT semctl command. The later does permission checks for the object vs S_IRUGO. As such there can be cases where EACCESS is returned via syscall but the info is displayed anyways in the procfs files. While this might have security implications via info leaking (albeit no writing to the sma metadata), this behavior goes way back and showing all the objects regardless of the permissions was most likely an overlook - so we are stuck with it. Furthermore, modifying either the syscall or the procfs file can cause userspace programs to break (ie ipcs). Some applications require getting the procfs info (without root privileges) and can be rather slow in comparison with a syscall -- up to 500x in some reported cases for shm. This patch introduces a new SEM_STAT_ANY command such that the sem ipc object permissions are ignored, and only audited instead. In addition, I've left the lsm security hook checks in place, as if some policy can block the call, then the user has no other choice than just parsing the procfs file. Link: http://lkml.kernel.org/r/20180215162458.10059-3-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reported-by: Robert Kettler <robert.kettler@outlook.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-10 23:35:26 +00:00
/* see comment for SHM_STAT_ANY */
if (cmd == SEM_STAT_ANY)
audit_ipc_obj(&sma->sem_perm);
else {
err = -EACCES;
if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
goto out_unlock;
}
err = security_sem_semctl(&sma->sem_perm, cmd);
if (err)
goto out_unlock;
ipc_lock_object(&sma->sem_perm);
if (!ipc_valid_object(&sma->sem_perm)) {
ipc_unlock_object(&sma->sem_perm);
err = -EIDRM;
goto out_unlock;
}
kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
semid64->sem_otime = get_semotime(sma);
semid64->sem_ctime = sma->sem_ctime;
semid64->sem_nsems = sma->sem_nsems;
ipc_unlock_object(&sma->sem_perm);
rcu_read_unlock();
return id;
out_unlock:
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
rcu_read_unlock();
return err;
}
static int semctl_info(struct ipc_namespace *ns, int semid,
int cmd, void __user *p)
{
struct seminfo seminfo;
int max_id;
int err;
err = security_sem_semctl(NULL, cmd);
if (err)
return err;
memset(&seminfo, 0, sizeof(seminfo));
seminfo.semmni = ns->sc_semmni;
seminfo.semmns = ns->sc_semmns;
seminfo.semmsl = ns->sc_semmsl;
seminfo.semopm = ns->sc_semopm;
seminfo.semvmx = SEMVMX;
seminfo.semmnu = SEMMNU;
seminfo.semmap = SEMMAP;
seminfo.semume = SEMUME;
down_read(&sem_ids(ns).rwsem);
if (cmd == SEM_INFO) {
seminfo.semusz = sem_ids(ns).in_use;
seminfo.semaem = ns->used_sems;
} else {
seminfo.semusz = SEMUSZ;
seminfo.semaem = SEMAEM;
}
max_id = ipc_get_maxid(&sem_ids(ns));
up_read(&sem_ids(ns).rwsem);
if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
return -EFAULT;
return (max_id < 0) ? 0 : max_id;
}
static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
int val)
{
struct sem_undo *un;
struct sem_array *sma;
struct sem *curr;
int err;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
DEFINE_WAKE_Q(wake_q);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (val > SEMVMX || val < 0)
return -ERANGE;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
rcu_read_lock();
sma = sem_obtain_object_check(ns, semid);
if (IS_ERR(sma)) {
rcu_read_unlock();
return PTR_ERR(sma);
}
if (semnum < 0 || semnum >= sma->sem_nsems) {
rcu_read_unlock();
return -EINVAL;
}
if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
rcu_read_unlock();
return -EACCES;
}
err = security_sem_semctl(&sma->sem_perm, SETVAL);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (err) {
rcu_read_unlock();
return -EACCES;
}
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
if (!ipc_valid_object(&sma->sem_perm)) {
sem_unlock(sma, -1);
rcu_read_unlock();
return -EIDRM;
}
curr = &sma->sems[semnum];
ipc_assert_locked_object(&sma->sem_perm);
list_for_each_entry(un, &sma->list_id, list_id)
un->semadj[semnum] = 0;
curr->semval = val;
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&curr->sempid, task_tgid(current));
sma->sem_ctime = ktime_get_real_seconds();
/* maybe some queued-up processes were waiting for this */
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
do_smart_update(sma, NULL, 0, 0, &wake_q);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_q(&wake_q);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
return 0;
}
static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
int cmd, void __user *p)
{
struct sem_array *sma;
struct sem *curr;
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
int err, nsems;
ushort fast_sem_io[SEMMSL_FAST];
ushort *sem_io = fast_sem_io;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
DEFINE_WAKE_Q(wake_q);
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
rcu_read_lock();
sma = sem_obtain_object_check(ns, semid);
if (IS_ERR(sma)) {
rcu_read_unlock();
return PTR_ERR(sma);
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
}
nsems = sma->sem_nsems;
err = -EACCES;
if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
goto out_rcu_wakeup;
err = security_sem_semctl(&sma->sem_perm, cmd);
if (err)
goto out_rcu_wakeup;
err = -EACCES;
switch (cmd) {
case GETALL:
{
ushort __user *array = p;
int i;
sem_lock(sma, NULL, -1);
if (!ipc_valid_object(&sma->sem_perm)) {
err = -EIDRM;
goto out_unlock;
}
if (nsems > SEMMSL_FAST) {
if (!ipc_rcu_getref(&sma->sem_perm)) {
err = -EIDRM;
goto out_unlock;
}
sem_unlock(sma, -1);
rcu_read_unlock();
sem_io = kvmalloc_array(nsems, sizeof(ushort),
GFP_KERNEL);
if (sem_io == NULL) {
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
return -ENOMEM;
}
rcu_read_lock();
sem_lock_and_putref(sma);
if (!ipc_valid_object(&sma->sem_perm)) {
err = -EIDRM;
goto out_unlock;
}
}
for (i = 0; i < sma->sem_nsems; i++)
sem_io[i] = sma->sems[i].semval;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
err = 0;
if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
err = -EFAULT;
goto out_free;
}
case SETALL:
{
int i;
struct sem_undo *un;
if (!ipc_rcu_getref(&sma->sem_perm)) {
err = -EIDRM;
goto out_rcu_wakeup;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
rcu_read_unlock();
if (nsems > SEMMSL_FAST) {
sem_io = kvmalloc_array(nsems, sizeof(ushort),
GFP_KERNEL);
if (sem_io == NULL) {
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
return -ENOMEM;
}
}
if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
err = -EFAULT;
goto out_free;
}
for (i = 0; i < nsems; i++) {
if (sem_io[i] > SEMVMX) {
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
err = -ERANGE;
goto out_free;
}
}
rcu_read_lock();
sem_lock_and_putref(sma);
if (!ipc_valid_object(&sma->sem_perm)) {
err = -EIDRM;
goto out_unlock;
}
for (i = 0; i < nsems; i++) {
sma->sems[i].semval = sem_io[i];
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
}
ipc_assert_locked_object(&sma->sem_perm);
list_for_each_entry(un, &sma->list_id, list_id) {
for (i = 0; i < nsems; i++)
un->semadj[i] = 0;
}
sma->sem_ctime = ktime_get_real_seconds();
/* maybe some queued-up processes were waiting for this */
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
do_smart_update(sma, NULL, 0, 0, &wake_q);
err = 0;
goto out_unlock;
}
/* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
}
err = -EINVAL;
if (semnum < 0 || semnum >= nsems)
goto out_rcu_wakeup;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
if (!ipc_valid_object(&sma->sem_perm)) {
err = -EIDRM;
goto out_unlock;
}
curr = &sma->sems[semnum];
switch (cmd) {
case GETVAL:
err = curr->semval;
goto out_unlock;
case GETPID:
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
err = pid_vnr(curr->sempid);
goto out_unlock;
case GETNCNT:
err = count_semcnt(sma, semnum, 0);
goto out_unlock;
case GETZCNT:
err = count_semcnt(sma, semnum, 1);
goto out_unlock;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
out_unlock:
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
out_rcu_wakeup:
rcu_read_unlock();
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_q(&wake_q);
out_free:
if (sem_io != fast_sem_io)
kvfree(sem_io);
return err;
}
static inline unsigned long
copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
{
switch (version) {
case IPC_64:
if (copy_from_user(out, buf, sizeof(*out)))
return -EFAULT;
return 0;
case IPC_OLD:
{
struct semid_ds tbuf_old;
if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
return -EFAULT;
out->sem_perm.uid = tbuf_old.sem_perm.uid;
out->sem_perm.gid = tbuf_old.sem_perm.gid;
out->sem_perm.mode = tbuf_old.sem_perm.mode;
return 0;
}
default:
return -EINVAL;
}
}
/*
* This function handles some semctl commands which require the rwsem
* to be held in write mode.
* NOTE: no locks must be held, the rwsem is taken inside this function.
*/
static int semctl_down(struct ipc_namespace *ns, int semid,
int cmd, struct semid64_ds *semid64)
{
struct sem_array *sma;
int err;
struct kern_ipc_perm *ipcp;
down_write(&sem_ids(ns).rwsem);
rcu_read_lock();
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
&semid64->sem_perm, 0);
if (IS_ERR(ipcp)) {
err = PTR_ERR(ipcp);
goto out_unlock1;
}
[PATCH] Rework of IPC auditing 1) The audit_ipc_perms() function has been split into two different functions: - audit_ipc_obj() - audit_ipc_set_perm() There's a key shift here... The audit_ipc_obj() collects the uid, gid, mode, and SElinux context label of the current ipc object. This audit_ipc_obj() hook is now found in several places. Most notably, it is hooked in ipcperms(), which is called in various places around the ipc code permforming a MAC check. Additionally there are several places where *checkid() is used to validate that an operation is being performed on a valid object while not necessarily having a nearby ipcperms() call. In these locations, audit_ipc_obj() is called to ensure that the information is captured by the audit system. The audit_set_new_perm() function is called any time the permissions on the ipc object changes. In this case, the NEW permissions are recorded (and note that an audit_ipc_obj() call exists just a few lines before each instance). 2) Support for an AUDIT_IPC_SET_PERM audit message type. This allows for separate auxiliary audit records for normal operations on an IPC object and permissions changes. Note that the same struct audit_aux_data_ipcctl is used and populated, however there are separate audit_log_format statements based on the type of the message. Finally, the AUDIT_IPC block of code in audit_free_aux() was extended to handle aux messages of this new type. No more mem leaks I hope ;-) Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2006-04-02 21:07:33 +00:00
sma = container_of(ipcp, struct sem_array, sem_perm);
err = security_sem_semctl(&sma->sem_perm, cmd);
if (err)
goto out_unlock1;
switch (cmd) {
case IPC_RMID:
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
/* freeary unlocks the ipc object and rcu */
freeary(ns, ipcp);
goto out_up;
case IPC_SET:
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
err = ipc_update_perm(&semid64->sem_perm, ipcp);
if (err)
goto out_unlock0;
sma->sem_ctime = ktime_get_real_seconds();
break;
default:
err = -EINVAL;
goto out_unlock1;
}
out_unlock0:
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
out_unlock1:
rcu_read_unlock();
out_up:
up_write(&sem_ids(ns).rwsem);
return err;
}
long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg)
{
int version;
struct ipc_namespace *ns;
void __user *p = (void __user *)arg;
struct semid64_ds semid64;
int err;
if (semid < 0)
return -EINVAL;
version = ipc_parse_version(&cmd);
ns = current->nsproxy->ipc_ns;
switch (cmd) {
case IPC_INFO:
case SEM_INFO:
return semctl_info(ns, semid, cmd, p);
case IPC_STAT:
case SEM_STAT:
ipc/sem: introduce semctl(SEM_STAT_ANY) There is a permission discrepancy when consulting shm ipc object metadata between /proc/sysvipc/sem (0444) and the SEM_STAT semctl command. The later does permission checks for the object vs S_IRUGO. As such there can be cases where EACCESS is returned via syscall but the info is displayed anyways in the procfs files. While this might have security implications via info leaking (albeit no writing to the sma metadata), this behavior goes way back and showing all the objects regardless of the permissions was most likely an overlook - so we are stuck with it. Furthermore, modifying either the syscall or the procfs file can cause userspace programs to break (ie ipcs). Some applications require getting the procfs info (without root privileges) and can be rather slow in comparison with a syscall -- up to 500x in some reported cases for shm. This patch introduces a new SEM_STAT_ANY command such that the sem ipc object permissions are ignored, and only audited instead. In addition, I've left the lsm security hook checks in place, as if some policy can block the call, then the user has no other choice than just parsing the procfs file. Link: http://lkml.kernel.org/r/20180215162458.10059-3-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reported-by: Robert Kettler <robert.kettler@outlook.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-10 23:35:26 +00:00
case SEM_STAT_ANY:
err = semctl_stat(ns, semid, cmd, &semid64);
if (err < 0)
return err;
if (copy_semid_to_user(p, &semid64, version))
err = -EFAULT;
return err;
case GETALL:
case GETVAL:
case GETPID:
case GETNCNT:
case GETZCNT:
case SETALL:
return semctl_main(ns, semid, semnum, cmd, p);
case SETVAL: {
int val;
#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
/* big-endian 64bit */
val = arg >> 32;
#else
/* 32bit or little-endian 64bit */
val = arg;
#endif
return semctl_setval(ns, semid, semnum, val);
}
case IPC_SET:
if (copy_semid_from_user(&semid64, p, version))
return -EFAULT;
case IPC_RMID:
return semctl_down(ns, semid, cmd, &semid64);
default:
return -EINVAL;
}
}
SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
{
return ksys_semctl(semid, semnum, cmd, arg);
}
#ifdef CONFIG_COMPAT
struct compat_semid_ds {
struct compat_ipc_perm sem_perm;
compat_time_t sem_otime;
compat_time_t sem_ctime;
compat_uptr_t sem_base;
compat_uptr_t sem_pending;
compat_uptr_t sem_pending_last;
compat_uptr_t undo;
unsigned short sem_nsems;
};
static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
int version)
{
memset(out, 0, sizeof(*out));
if (version == IPC_64) {
struct compat_semid64_ds __user *p = buf;
return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
} else {
struct compat_semid_ds __user *p = buf;
return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
}
}
static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
int version)
{
if (version == IPC_64) {
struct compat_semid64_ds v;
memset(&v, 0, sizeof(v));
to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
v.sem_otime = in->sem_otime;
v.sem_ctime = in->sem_ctime;
v.sem_nsems = in->sem_nsems;
return copy_to_user(buf, &v, sizeof(v));
} else {
struct compat_semid_ds v;
memset(&v, 0, sizeof(v));
to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
v.sem_otime = in->sem_otime;
v.sem_ctime = in->sem_ctime;
v.sem_nsems = in->sem_nsems;
return copy_to_user(buf, &v, sizeof(v));
}
}
long compat_ksys_semctl(int semid, int semnum, int cmd, int arg)
{
void __user *p = compat_ptr(arg);
struct ipc_namespace *ns;
struct semid64_ds semid64;
int version = compat_ipc_parse_version(&cmd);
int err;
ns = current->nsproxy->ipc_ns;
if (semid < 0)
return -EINVAL;
switch (cmd & (~IPC_64)) {
case IPC_INFO:
case SEM_INFO:
return semctl_info(ns, semid, cmd, p);
case IPC_STAT:
case SEM_STAT:
ipc/sem: introduce semctl(SEM_STAT_ANY) There is a permission discrepancy when consulting shm ipc object metadata between /proc/sysvipc/sem (0444) and the SEM_STAT semctl command. The later does permission checks for the object vs S_IRUGO. As such there can be cases where EACCESS is returned via syscall but the info is displayed anyways in the procfs files. While this might have security implications via info leaking (albeit no writing to the sma metadata), this behavior goes way back and showing all the objects regardless of the permissions was most likely an overlook - so we are stuck with it. Furthermore, modifying either the syscall or the procfs file can cause userspace programs to break (ie ipcs). Some applications require getting the procfs info (without root privileges) and can be rather slow in comparison with a syscall -- up to 500x in some reported cases for shm. This patch introduces a new SEM_STAT_ANY command such that the sem ipc object permissions are ignored, and only audited instead. In addition, I've left the lsm security hook checks in place, as if some policy can block the call, then the user has no other choice than just parsing the procfs file. Link: http://lkml.kernel.org/r/20180215162458.10059-3-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Reported-by: Robert Kettler <robert.kettler@outlook.com> Cc: Eric W. Biederman <ebiederm@xmission.com> Cc: Kees Cook <keescook@chromium.org> Cc: Manfred Spraul <manfred@colorfullife.com> Cc: Michael Kerrisk <mtk.manpages@gmail.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-04-10 23:35:26 +00:00
case SEM_STAT_ANY:
err = semctl_stat(ns, semid, cmd, &semid64);
if (err < 0)
return err;
if (copy_compat_semid_to_user(p, &semid64, version))
err = -EFAULT;
return err;
case GETVAL:
case GETPID:
case GETNCNT:
case GETZCNT:
case GETALL:
case SETALL:
return semctl_main(ns, semid, semnum, cmd, p);
case SETVAL:
return semctl_setval(ns, semid, semnum, arg);
case IPC_SET:
if (copy_compat_semid_from_user(&semid64, p, version))
return -EFAULT;
/* fallthru */
case IPC_RMID:
return semctl_down(ns, semid, cmd, &semid64);
default:
return -EINVAL;
}
}
COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
{
return compat_ksys_semctl(semid, semnum, cmd, arg);
}
#endif
/* If the task doesn't already have a undo_list, then allocate one
* here. We guarantee there is only one thread using this undo list,
* and current is THE ONE
*
* If this allocation and assignment succeeds, but later
* portions of this code fail, there is no need to free the sem_undo_list.
* Just let it stay associated with the task, and it'll be freed later
* at exit time.
*
* This can block, so callers must hold no locks.
*/
static inline int get_undo_list(struct sem_undo_list **undo_listp)
{
struct sem_undo_list *undo_list;
undo_list = current->sysvsem.undo_list;
if (!undo_list) {
undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
if (undo_list == NULL)
return -ENOMEM;
spin_lock_init(&undo_list->lock);
refcount_set(&undo_list->refcnt, 1);
INIT_LIST_HEAD(&undo_list->list_proc);
current->sysvsem.undo_list = undo_list;
}
*undo_listp = undo_list;
return 0;
}
static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
{
struct sem_undo *un;
list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
if (un->semid == semid)
return un;
}
return NULL;
}
static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
{
struct sem_undo *un;
assert_spin_locked(&ulp->lock);
un = __lookup_undo(ulp, semid);
if (un) {
list_del_rcu(&un->list_proc);
list_add_rcu(&un->list_proc, &ulp->list_proc);
}
return un;
}
/**
* find_alloc_undo - lookup (and if not present create) undo array
* @ns: namespace
* @semid: semaphore array id
*
* The function looks up (and if not present creates) the undo structure.
* The size of the undo structure depends on the size of the semaphore
* array, thus the alloc path is not that straightforward.
* Lifetime-rules: sem_undo is rcu-protected, on success, the function
* performs a rcu_read_lock().
*/
static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
{
struct sem_array *sma;
struct sem_undo_list *ulp;
struct sem_undo *un, *new;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
int nsems, error;
error = get_undo_list(&ulp);
if (error)
return ERR_PTR(error);
rcu_read_lock();
spin_lock(&ulp->lock);
un = lookup_undo(ulp, semid);
spin_unlock(&ulp->lock);
if (likely(un != NULL))
goto out;
/* no undo structure around - allocate one. */
/* step 1: figure out the size of the semaphore array */
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
sma = sem_obtain_object_check(ns, semid);
if (IS_ERR(sma)) {
rcu_read_unlock();
return ERR_CAST(sma);
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
}
nsems = sma->sem_nsems;
if (!ipc_rcu_getref(&sma->sem_perm)) {
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
rcu_read_unlock();
un = ERR_PTR(-EIDRM);
goto out;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
rcu_read_unlock();
/* step 2: allocate new undo structure */
new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
if (!new) {
ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
return ERR_PTR(-ENOMEM);
}
/* step 3: Acquire the lock on semaphore array */
rcu_read_lock();
sem_lock_and_putref(sma);
if (!ipc_valid_object(&sma->sem_perm)) {
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
kfree(new);
un = ERR_PTR(-EIDRM);
goto out;
}
spin_lock(&ulp->lock);
/*
* step 4: check for races: did someone else allocate the undo struct?
*/
un = lookup_undo(ulp, semid);
if (un) {
kfree(new);
goto success;
}
/* step 5: initialize & link new undo structure */
new->semadj = (short *) &new[1];
new->ulp = ulp;
new->semid = semid;
assert_spin_locked(&ulp->lock);
list_add_rcu(&new->list_proc, &ulp->list_proc);
ipc_assert_locked_object(&sma->sem_perm);
list_add(&new->list_id, &sma->list_id);
un = new;
success:
spin_unlock(&ulp->lock);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
out:
return un;
}
static long do_semtimedop(int semid, struct sembuf __user *tsops,
unsigned nsops, const struct timespec64 *timeout)
{
int error = -EINVAL;
struct sem_array *sma;
struct sembuf fast_sops[SEMOPM_FAST];
struct sembuf *sops = fast_sops, *sop;
struct sem_undo *un;
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
int max, locknum;
bool undos = false, alter = false, dupsop = false;
struct sem_queue queue;
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
unsigned long dup = 0, jiffies_left = 0;
struct ipc_namespace *ns;
ns = current->nsproxy->ipc_ns;
if (nsops < 1 || semid < 0)
return -EINVAL;
if (nsops > ns->sc_semopm)
return -E2BIG;
if (nsops > SEMOPM_FAST) {
sops = kvmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
if (sops == NULL)
return -ENOMEM;
}
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
error = -EFAULT;
goto out_free;
}
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
if (timeout) {
if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
timeout->tv_nsec >= 1000000000L) {
error = -EINVAL;
goto out_free;
}
jiffies_left = timespec64_to_jiffies(timeout);
}
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
max = 0;
for (sop = sops; sop < sops + nsops; sop++) {
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
if (sop->sem_num >= max)
max = sop->sem_num;
if (sop->sem_flg & SEM_UNDO)
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
undos = true;
if (dup & mask) {
/*
* There was a previous alter access that appears
* to have accessed the same semaphore, thus use
* the dupsop logic. "appears", because the detection
* can only check % BITS_PER_LONG.
*/
dupsop = true;
}
if (sop->sem_op != 0) {
alter = true;
dup |= mask;
}
}
if (undos) {
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
/* On success, find_alloc_undo takes the rcu_read_lock */
un = find_alloc_undo(ns, semid);
if (IS_ERR(un)) {
error = PTR_ERR(un);
goto out_free;
}
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
} else {
un = NULL;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
rcu_read_lock();
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
sma = sem_obtain_object_check(ns, semid);
if (IS_ERR(sma)) {
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
rcu_read_unlock();
error = PTR_ERR(sma);
goto out_free;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
error = -EFBIG;
if (max >= sma->sem_nsems) {
rcu_read_unlock();
goto out_free;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
error = -EACCES;
if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
rcu_read_unlock();
goto out_free;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
if (error) {
rcu_read_unlock();
goto out_free;
}
ipc,sem: do not hold ipc lock more than necessary Instead of holding the ipc lock for permissions and security checks, among others, only acquire it when necessary. Some numbers.... 1) With Rik's semop-multi.c microbenchmark we can see the following results: Baseline (3.9-rc1): cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 151452270, ops/sec 5048409 + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock With this patchset: cpus 4, threads: 256, semaphores: 128, test duration: 30 secs total operations: 273156400, ops/sec 9105213 + 18.54% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.72% a.out [kernel.kallsyms] [k] sys_semtimedop + 7.70% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.58% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 6.54% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.71% a.out [kernel.kallsyms] [k] ipc_obtain_object_check 2) While on an Oracle swingbench DSS (data mining) workload the improvements are not as exciting as with Rik's benchmark, we can see some positive numbers. For an 8 socket machine the following are the percentages of %sys time incurred in the ipc lock: Baseline (3.9-rc1): 100 swingbench users: 8,74% 400 swingbench users: 21,86% 800 swingbench users: 84,35% With this patchset: 100 swingbench users: 8,11% 400 swingbench users: 19,93% 800 swingbench users: 77,69% [riel@redhat.com: fix two locking bugs] [sasha.levin@oracle.com: prevent releasing RCU read lock twice in semctl_main] [akpm@linux-foundation.org: coding-style fixes] Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Signed-off-by: Rik van Riel <riel@redhat.com> Reviewed-by: Chegu Vinod <chegu_vinod@hp.com> Acked-by: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Jason Low <jason.low2@hp.com> Cc: Emmanuel Benisty <benisty.e@gmail.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:29 +00:00
error = -EIDRM;
locknum = sem_lock(sma, sops, nsops);
/*
* We eventually might perform the following check in a lockless
* fashion, considering ipc_valid_object() locking constraints.
* If nsops == 1 and there is no contention for sem_perm.lock, then
* only a per-semaphore lock is held and it's OK to proceed with the
* check below. More details on the fine grained locking scheme
* entangled here and why it's RMID race safe on comments at sem_lock()
*/
if (!ipc_valid_object(&sma->sem_perm))
goto out_unlock_free;
/*
* semid identifiers are not unique - find_alloc_undo may have
* allocated an undo structure, it was invalidated by an RMID
* and now a new array with received the same id. Check and fail.
* This case can be detected checking un->semid. The existence of
* "un" itself is guaranteed by rcu.
*/
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (un && un->semid == -1)
goto out_unlock_free;
queue.sops = sops;
queue.nsops = nsops;
queue.undo = un;
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
queue.pid = task_tgid(current);
queue.alter = alter;
ipc/sem: optimize perform_atomic_semop() This is the main workhorse that deals with semop user calls such that the waitforzero or semval update operations, on the set, can complete on not as the sma currently stands. Currently, the set is iterated twice (setting semval, then backwards for the sempid value). Slowpaths, and particularly SEM_UNDO calls, must undo any altered sem when it is detected that the caller must block or has errored-out. With larger sets, there can occur situations where this involves a lot of cycles and can obviously be a suboptimal use of cached resources in shared memory. Ie, discarding CPU caches that are also calling semop and have the sembuf cached (and can complete), while the current lock holder doing the semop will block, error, or does a waitforzero operation. This patch proposes still iterating the set twice, but the first scan is read-only, and we perform the actual updates afterward, once we know that the call will succeed. In order to not suffer from the overhead of dealing with sops that act on the same sem_num, such (rare) cases use perform_atomic_semop_slow(), which is exactly what we have now. Duplicates are detected before grabbing sem_lock, and uses simple a 32/64-bit hash array variable to based on the sem_num we are working on. In addition add some comments to when we expect to the caller to block. [akpm@linux-foundation.org: coding-style fixes] [colin.king@canonical.com: ensure we left shift a ULL rather than a 32 bit integer] Link: http://lkml.kernel.org/r/20161028181129.7311-1-colin.king@canonical.com Link: http://lkml.kernel.org/r/20160921194603.GB21438@linux-80c1.suse Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Colin Ian King <colin.king@canonical.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:37 +00:00
queue.dupsop = dupsop;
error = perform_atomic_semop(sma, &queue);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
if (error == 0) { /* non-blocking succesfull path */
DEFINE_WAKE_Q(wake_q);
/*
* If the operation was successful, then do
* the required updates.
*/
if (alter)
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
do_smart_update(sma, sops, nsops, 1, &wake_q);
else
set_semotime(sma, sops);
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
sem_unlock(sma, locknum);
rcu_read_unlock();
wake_up_q(&wake_q);
goto out_free;
}
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
if (error < 0) /* non-blocking error path */
goto out_unlock_free;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
/*
* We need to sleep on this operation, so we put the current
* task into the pending queue and go to sleep.
*/
if (nsops == 1) {
struct sem *curr;
curr = &sma->sems[sops->sem_num];
if (alter) {
if (sma->complex_count) {
list_add_tail(&queue.list,
&sma->pending_alter);
} else {
list_add_tail(&queue.list,
&curr->pending_alter);
}
} else {
list_add_tail(&queue.list, &curr->pending_const);
}
} else {
if (!sma->complex_count)
merge_queues(sma);
if (alter)
list_add_tail(&queue.list, &sma->pending_alter);
else
list_add_tail(&queue.list, &sma->pending_const);
sma->complex_count++;
}
do {
queue.status = -EINTR;
queue.sleeper = current;
__set_current_state(TASK_INTERRUPTIBLE);
sem_unlock(sma, locknum);
rcu_read_unlock();
if (timeout)
jiffies_left = schedule_timeout(jiffies_left);
else
schedule();
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
/*
* fastpath: the semop has completed, either successfully or
* not, from the syscall pov, is quite irrelevant to us at this
* point; we're done.
*
* We _do_ care, nonetheless, about being awoken by a signal or
* spuriously. The queue.status is checked again in the
* slowpath (aka after taking sem_lock), such that we can detect
* scenarios where we were awakened externally, during the
* window between wake_q_add() and wake_up_q().
*/
error = READ_ONCE(queue.status);
if (error != -EINTR) {
/*
* User space could assume that semop() is a memory
* barrier: Without the mb(), the cpu could
* speculatively read in userspace stale data that was
* overwritten by the previous owner of the semaphore.
*/
smp_mb();
goto out_free;
}
rcu_read_lock();
locknum = sem_lock(sma, sops, nsops);
if (!ipc_valid_object(&sma->sem_perm))
goto out_unlock_free;
error = READ_ONCE(queue.status);
/*
* If queue.status != -EINTR we are woken up by another process.
* Leave without unlink_queue(), but with sem_unlock().
*/
if (error != -EINTR)
goto out_unlock_free;
/*
* If an interrupt occurred we have to clean up the queue.
*/
if (timeout && jiffies_left == 0)
error = -EAGAIN;
} while (error == -EINTR && !signal_pending(current)); /* spurious */
unlink_queue(sma, &queue);
out_unlock_free:
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, locknum);
rcu_read_unlock();
out_free:
if (sops != fast_sops)
kvfree(sops);
return error;
}
long ksys_semtimedop(int semid, struct sembuf __user *tsops,
unsigned int nsops, const struct timespec __user *timeout)
{
if (timeout) {
struct timespec64 ts;
if (get_timespec64(&ts, timeout))
return -EFAULT;
return do_semtimedop(semid, tsops, nsops, &ts);
}
return do_semtimedop(semid, tsops, nsops, NULL);
}
SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
unsigned int, nsops, const struct timespec __user *, timeout)
{
return ksys_semtimedop(semid, tsops, nsops, timeout);
}
#ifdef CONFIG_COMPAT
long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
unsigned int nsops,
const struct compat_timespec __user *timeout)
{
if (timeout) {
struct timespec64 ts;
if (compat_get_timespec64(&ts, timeout))
return -EFAULT;
return do_semtimedop(semid, tsems, nsops, &ts);
}
return do_semtimedop(semid, tsems, nsops, NULL);
}
COMPAT_SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsems,
unsigned int, nsops,
const struct compat_timespec __user *, timeout)
{
return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
}
#endif
SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
unsigned, nsops)
{
return do_semtimedop(semid, tsops, nsops, NULL);
}
/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
* parent and child tasks.
*/
int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
{
struct sem_undo_list *undo_list;
int error;
if (clone_flags & CLONE_SYSVSEM) {
error = get_undo_list(&undo_list);
if (error)
return error;
refcount_inc(&undo_list->refcnt);
tsk->sysvsem.undo_list = undo_list;
} else
tsk->sysvsem.undo_list = NULL;
return 0;
}
/*
* add semadj values to semaphores, free undo structures.
* undo structures are not freed when semaphore arrays are destroyed
* so some of them may be out of date.
* IMPLEMENTATION NOTE: There is some confusion over whether the
* set of adjustments that needs to be done should be done in an atomic
* manner or not. That is, if we are attempting to decrement the semval
* should we queue up and wait until we can do so legally?
* The original implementation attempted to do this (queue and wait).
* The current implementation does not do so. The POSIX standard
* and SVID should be consulted to determine what behavior is mandated.
*/
void exit_sem(struct task_struct *tsk)
{
struct sem_undo_list *ulp;
ulp = tsk->sysvsem.undo_list;
if (!ulp)
return;
tsk->sysvsem.undo_list = NULL;
if (!refcount_dec_and_test(&ulp->refcnt))
return;
for (;;) {
struct sem_array *sma;
struct sem_undo *un;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
int semid, i;
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
DEFINE_WAKE_Q(wake_q);
ipc/sem.c: add cond_resched in exit_sme In CONFIG_PREEMPT=n kernel a softlockup was observed while the for loop in exit_sem. Apparently it's possible for the loop to take quite a long time and it doesn't have a scheduling point in it. Since the codes is executing under an rcu read section this may also cause rcu stalls, which in turn block synchronize_rcu operations, which more or less de-stabilises the whole system. Fix this by introducing a cond_resched() at the beginning of the loop. So this patch fixes the following: NMI watchdog: BUG: soft lockup - CPU#10 stuck for 23s! [httpd:18119] CPU: 10 PID: 18119 Comm: httpd Tainted: G O 4.4.20-clouder2 #6 Hardware name: Supermicro X10DRi/X10DRi, BIOS 1.1 04/14/2015 task: ffff88348d695280 ti: ffff881c95550000 task.ti: ffff881c95550000 RIP: 0010:[<ffffffff81614bc7>] [<ffffffff81614bc7>] _raw_spin_lock+0x17/0x30 RSP: 0018:ffff881c95553e40 EFLAGS: 00000246 RAX: 0000000000000000 RBX: ffff883161b1eea8 RCX: 000000000000000d RDX: 0000000000000001 RSI: 000000000000000e RDI: ffff883161b1eea4 RBP: ffff881c95553ea0 R08: ffff881c95553e68 R09: ffff883fef376f88 R10: ffff881fffb58c20 R11: ffffea0072556600 R12: ffff883161b1eea0 R13: ffff88348d695280 R14: ffff883dec427000 R15: ffff8831621672a0 FS: 0000000000000000(0000) GS:ffff881fffb40000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 00007f3b3723e020 CR3: 0000000001c0a000 CR4: 00000000001406e0 Call Trace: ? exit_sem+0x7c/0x280 do_exit+0x338/0xb40 do_group_exit+0x43/0xd0 SyS_exit_group+0x14/0x20 entry_SYSCALL_64_fastpath+0x16/0x6e Link: http://lkml.kernel.org/r/1475154992-6363-1-git-send-email-kernel@kyup.com Signed-off-by: Nikolay Borisov <kernel@kyup.com> Cc: Herton R. Krzesinski <herton@redhat.com> Cc: Fabian Frederick <fabf@skynet.be> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:55:05 +00:00
cond_resched();
rcu_read_lock();
un = list_entry_rcu(ulp->list_proc.next,
struct sem_undo, list_proc);
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits The current semaphore code allows a potential use after free: in exit_sem we may free the task's sem_undo_list while there is still another task looping through the same semaphore set and cleaning the sem_undo list at freeary function (the task called IPC_RMID for the same semaphore set). For example, with a test program [1] running which keeps forking a lot of processes (which then do a semop call with SEM_UNDO flag), and with the parent right after removing the semaphore set with IPC_RMID, and a kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and CONFIG_DEBUG_SPINLOCK, you can easily see something like the following in the kernel log: Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64 000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b kkkkkkkk.kkkkkkk 010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff ....kkkk........ Prev obj: start=ffff88003b45c180, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff ...........7.... Next obj: start=ffff88003b45c200, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff ........h).<.... BUG: spinlock wrong CPU on CPU#2, test/18028 general protection fault: 0000 [#1] SMP Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: spin_dump+0x53/0xc0 Call Trace: spin_bug+0x30/0x40 do_raw_spin_unlock+0x71/0xa0 _raw_spin_unlock+0xe/0x10 freeary+0x82/0x2a0 ? _raw_spin_lock+0xe/0x10 semctl_down.clone.0+0xce/0x160 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 SyS_semctl+0x236/0x2c0 ? syscall_trace_leave+0xde/0x130 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89 RIP [<ffffffff810d6053>] spin_dump+0x53/0xc0 RSP <ffff88003750fd68> ---[ end trace 783ebb76612867a0 ]--- NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053] Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 3 PID: 18053 Comm: test Tainted: G D 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: native_read_tsc+0x0/0x20 Call Trace: ? delay_tsc+0x40/0x70 __delay+0xf/0x20 do_raw_spin_lock+0x96/0x140 _raw_spin_lock+0xe/0x10 sem_lock_and_putref+0x11/0x70 SYSC_semtimedop+0x7bf/0x960 ? handle_mm_fault+0xbf6/0x1880 ? dequeue_task_fair+0x79/0x4a0 ? __do_page_fault+0x19a/0x430 ? kfree_debugcheck+0x16/0x40 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 ? do_audit_syscall_entry+0x66/0x70 ? syscall_trace_enter_phase1+0x139/0x160 SyS_semtimedop+0xe/0x10 SyS_semop+0x10/0x20 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9 Kernel panic - not syncing: softlockup: hung tasks I wasn't able to trigger any badness on a recent kernel without the proper config debugs enabled, however I have softlockup reports on some kernel versions, in the semaphore code, which are similar as above (the scenario is seen on some servers running IBM DB2 which uses semaphore syscalls). The patch here fixes the race against freeary, by acquiring or waiting on the sem_undo_list lock as necessary (exit_sem can race with freeary, while freeary sets un->semid to -1 and removes the same sem_undo from list_proc or when it removes the last sem_undo). After the patch I'm unable to reproduce the problem using the test case [1]. [1] Test case used below: #include <stdio.h> #include <sys/types.h> #include <sys/ipc.h> #include <sys/sem.h> #include <sys/wait.h> #include <stdlib.h> #include <time.h> #include <unistd.h> #include <errno.h> #define NSEM 1 #define NSET 5 int sid[NSET]; void thread() { struct sembuf op; int s; uid_t pid = getuid(); s = rand() % NSET; op.sem_num = pid % NSEM; op.sem_op = 1; op.sem_flg = SEM_UNDO; semop(sid[s], &op, 1); exit(EXIT_SUCCESS); } void create_set() { int i, j; pid_t p; union { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; } un; /* Create and initialize semaphore set */ for (i = 0; i < NSET; i++) { sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT); if (sid[i] < 0) { perror("semget"); exit(EXIT_FAILURE); } } un.val = 0; for (i = 0; i < NSET; i++) { for (j = 0; j < NSEM; j++) { if (semctl(sid[i], j, SETVAL, un) < 0) perror("semctl"); } } /* Launch threads that operate on semaphore set */ for (i = 0; i < NSEM * NSET * NSET; i++) { p = fork(); if (p < 0) perror("fork"); if (p == 0) thread(); } /* Free semaphore set */ for (i = 0; i < NSET; i++) { if (semctl(sid[i], NSEM, IPC_RMID)) perror("IPC_RMID"); } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } int main(int argc, char **argv) { pid_t p; srand(time(NULL)); while (1) { p = fork(); if (p < 0) { perror("fork"); exit(EXIT_FAILURE); } if (p == 0) { create_set(); goto end; } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } end: return 0; } [akpm@linux-foundation.org: use normal comment layout] Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Rafael Aquini <aquini@redhat.com> CC: Aristeu Rozanski <aris@redhat.com> Cc: David Jeffery <djeffery@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-08-14 22:35:02 +00:00
if (&un->list_proc == &ulp->list_proc) {
/*
* We must wait for freeary() before freeing this ulp,
* in case we raced with last sem_undo. There is a small
* possibility where we exit while freeary() didn't
* finish unlocking sem_undo_list.
*/
spin_lock(&ulp->lock);
spin_unlock(&ulp->lock);
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits The current semaphore code allows a potential use after free: in exit_sem we may free the task's sem_undo_list while there is still another task looping through the same semaphore set and cleaning the sem_undo list at freeary function (the task called IPC_RMID for the same semaphore set). For example, with a test program [1] running which keeps forking a lot of processes (which then do a semop call with SEM_UNDO flag), and with the parent right after removing the semaphore set with IPC_RMID, and a kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and CONFIG_DEBUG_SPINLOCK, you can easily see something like the following in the kernel log: Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64 000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b kkkkkkkk.kkkkkkk 010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff ....kkkk........ Prev obj: start=ffff88003b45c180, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff ...........7.... Next obj: start=ffff88003b45c200, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff ........h).<.... BUG: spinlock wrong CPU on CPU#2, test/18028 general protection fault: 0000 [#1] SMP Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: spin_dump+0x53/0xc0 Call Trace: spin_bug+0x30/0x40 do_raw_spin_unlock+0x71/0xa0 _raw_spin_unlock+0xe/0x10 freeary+0x82/0x2a0 ? _raw_spin_lock+0xe/0x10 semctl_down.clone.0+0xce/0x160 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 SyS_semctl+0x236/0x2c0 ? syscall_trace_leave+0xde/0x130 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89 RIP [<ffffffff810d6053>] spin_dump+0x53/0xc0 RSP <ffff88003750fd68> ---[ end trace 783ebb76612867a0 ]--- NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053] Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 3 PID: 18053 Comm: test Tainted: G D 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: native_read_tsc+0x0/0x20 Call Trace: ? delay_tsc+0x40/0x70 __delay+0xf/0x20 do_raw_spin_lock+0x96/0x140 _raw_spin_lock+0xe/0x10 sem_lock_and_putref+0x11/0x70 SYSC_semtimedop+0x7bf/0x960 ? handle_mm_fault+0xbf6/0x1880 ? dequeue_task_fair+0x79/0x4a0 ? __do_page_fault+0x19a/0x430 ? kfree_debugcheck+0x16/0x40 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 ? do_audit_syscall_entry+0x66/0x70 ? syscall_trace_enter_phase1+0x139/0x160 SyS_semtimedop+0xe/0x10 SyS_semop+0x10/0x20 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9 Kernel panic - not syncing: softlockup: hung tasks I wasn't able to trigger any badness on a recent kernel without the proper config debugs enabled, however I have softlockup reports on some kernel versions, in the semaphore code, which are similar as above (the scenario is seen on some servers running IBM DB2 which uses semaphore syscalls). The patch here fixes the race against freeary, by acquiring or waiting on the sem_undo_list lock as necessary (exit_sem can race with freeary, while freeary sets un->semid to -1 and removes the same sem_undo from list_proc or when it removes the last sem_undo). After the patch I'm unable to reproduce the problem using the test case [1]. [1] Test case used below: #include <stdio.h> #include <sys/types.h> #include <sys/ipc.h> #include <sys/sem.h> #include <sys/wait.h> #include <stdlib.h> #include <time.h> #include <unistd.h> #include <errno.h> #define NSEM 1 #define NSET 5 int sid[NSET]; void thread() { struct sembuf op; int s; uid_t pid = getuid(); s = rand() % NSET; op.sem_num = pid % NSEM; op.sem_op = 1; op.sem_flg = SEM_UNDO; semop(sid[s], &op, 1); exit(EXIT_SUCCESS); } void create_set() { int i, j; pid_t p; union { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; } un; /* Create and initialize semaphore set */ for (i = 0; i < NSET; i++) { sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT); if (sid[i] < 0) { perror("semget"); exit(EXIT_FAILURE); } } un.val = 0; for (i = 0; i < NSET; i++) { for (j = 0; j < NSEM; j++) { if (semctl(sid[i], j, SETVAL, un) < 0) perror("semctl"); } } /* Launch threads that operate on semaphore set */ for (i = 0; i < NSEM * NSET * NSET; i++) { p = fork(); if (p < 0) perror("fork"); if (p == 0) thread(); } /* Free semaphore set */ for (i = 0; i < NSET; i++) { if (semctl(sid[i], NSEM, IPC_RMID)) perror("IPC_RMID"); } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } int main(int argc, char **argv) { pid_t p; srand(time(NULL)); while (1) { p = fork(); if (p < 0) { perror("fork"); exit(EXIT_FAILURE); } if (p == 0) { create_set(); goto end; } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } end: return 0; } [akpm@linux-foundation.org: use normal comment layout] Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Rafael Aquini <aquini@redhat.com> CC: Aristeu Rozanski <aris@redhat.com> Cc: David Jeffery <djeffery@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-08-14 22:35:02 +00:00
rcu_read_unlock();
break;
}
spin_lock(&ulp->lock);
semid = un->semid;
spin_unlock(&ulp->lock);
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits The current semaphore code allows a potential use after free: in exit_sem we may free the task's sem_undo_list while there is still another task looping through the same semaphore set and cleaning the sem_undo list at freeary function (the task called IPC_RMID for the same semaphore set). For example, with a test program [1] running which keeps forking a lot of processes (which then do a semop call with SEM_UNDO flag), and with the parent right after removing the semaphore set with IPC_RMID, and a kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and CONFIG_DEBUG_SPINLOCK, you can easily see something like the following in the kernel log: Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64 000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b kkkkkkkk.kkkkkkk 010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff ....kkkk........ Prev obj: start=ffff88003b45c180, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff ...........7.... Next obj: start=ffff88003b45c200, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff ........h).<.... BUG: spinlock wrong CPU on CPU#2, test/18028 general protection fault: 0000 [#1] SMP Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: spin_dump+0x53/0xc0 Call Trace: spin_bug+0x30/0x40 do_raw_spin_unlock+0x71/0xa0 _raw_spin_unlock+0xe/0x10 freeary+0x82/0x2a0 ? _raw_spin_lock+0xe/0x10 semctl_down.clone.0+0xce/0x160 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 SyS_semctl+0x236/0x2c0 ? syscall_trace_leave+0xde/0x130 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89 RIP [<ffffffff810d6053>] spin_dump+0x53/0xc0 RSP <ffff88003750fd68> ---[ end trace 783ebb76612867a0 ]--- NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053] Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 3 PID: 18053 Comm: test Tainted: G D 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: native_read_tsc+0x0/0x20 Call Trace: ? delay_tsc+0x40/0x70 __delay+0xf/0x20 do_raw_spin_lock+0x96/0x140 _raw_spin_lock+0xe/0x10 sem_lock_and_putref+0x11/0x70 SYSC_semtimedop+0x7bf/0x960 ? handle_mm_fault+0xbf6/0x1880 ? dequeue_task_fair+0x79/0x4a0 ? __do_page_fault+0x19a/0x430 ? kfree_debugcheck+0x16/0x40 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 ? do_audit_syscall_entry+0x66/0x70 ? syscall_trace_enter_phase1+0x139/0x160 SyS_semtimedop+0xe/0x10 SyS_semop+0x10/0x20 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9 Kernel panic - not syncing: softlockup: hung tasks I wasn't able to trigger any badness on a recent kernel without the proper config debugs enabled, however I have softlockup reports on some kernel versions, in the semaphore code, which are similar as above (the scenario is seen on some servers running IBM DB2 which uses semaphore syscalls). The patch here fixes the race against freeary, by acquiring or waiting on the sem_undo_list lock as necessary (exit_sem can race with freeary, while freeary sets un->semid to -1 and removes the same sem_undo from list_proc or when it removes the last sem_undo). After the patch I'm unable to reproduce the problem using the test case [1]. [1] Test case used below: #include <stdio.h> #include <sys/types.h> #include <sys/ipc.h> #include <sys/sem.h> #include <sys/wait.h> #include <stdlib.h> #include <time.h> #include <unistd.h> #include <errno.h> #define NSEM 1 #define NSET 5 int sid[NSET]; void thread() { struct sembuf op; int s; uid_t pid = getuid(); s = rand() % NSET; op.sem_num = pid % NSEM; op.sem_op = 1; op.sem_flg = SEM_UNDO; semop(sid[s], &op, 1); exit(EXIT_SUCCESS); } void create_set() { int i, j; pid_t p; union { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; } un; /* Create and initialize semaphore set */ for (i = 0; i < NSET; i++) { sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT); if (sid[i] < 0) { perror("semget"); exit(EXIT_FAILURE); } } un.val = 0; for (i = 0; i < NSET; i++) { for (j = 0; j < NSEM; j++) { if (semctl(sid[i], j, SETVAL, un) < 0) perror("semctl"); } } /* Launch threads that operate on semaphore set */ for (i = 0; i < NSEM * NSET * NSET; i++) { p = fork(); if (p < 0) perror("fork"); if (p == 0) thread(); } /* Free semaphore set */ for (i = 0; i < NSET; i++) { if (semctl(sid[i], NSEM, IPC_RMID)) perror("IPC_RMID"); } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } int main(int argc, char **argv) { pid_t p; srand(time(NULL)); while (1) { p = fork(); if (p < 0) { perror("fork"); exit(EXIT_FAILURE); } if (p == 0) { create_set(); goto end; } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } end: return 0; } [akpm@linux-foundation.org: use normal comment layout] Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Rafael Aquini <aquini@redhat.com> CC: Aristeu Rozanski <aris@redhat.com> Cc: David Jeffery <djeffery@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-08-14 22:35:02 +00:00
/* exit_sem raced with IPC_RMID, nothing to do */
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (semid == -1) {
rcu_read_unlock();
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits The current semaphore code allows a potential use after free: in exit_sem we may free the task's sem_undo_list while there is still another task looping through the same semaphore set and cleaning the sem_undo list at freeary function (the task called IPC_RMID for the same semaphore set). For example, with a test program [1] running which keeps forking a lot of processes (which then do a semop call with SEM_UNDO flag), and with the parent right after removing the semaphore set with IPC_RMID, and a kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and CONFIG_DEBUG_SPINLOCK, you can easily see something like the following in the kernel log: Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64 000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b kkkkkkkk.kkkkkkk 010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff ....kkkk........ Prev obj: start=ffff88003b45c180, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff ...........7.... Next obj: start=ffff88003b45c200, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff ........h).<.... BUG: spinlock wrong CPU on CPU#2, test/18028 general protection fault: 0000 [#1] SMP Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: spin_dump+0x53/0xc0 Call Trace: spin_bug+0x30/0x40 do_raw_spin_unlock+0x71/0xa0 _raw_spin_unlock+0xe/0x10 freeary+0x82/0x2a0 ? _raw_spin_lock+0xe/0x10 semctl_down.clone.0+0xce/0x160 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 SyS_semctl+0x236/0x2c0 ? syscall_trace_leave+0xde/0x130 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89 RIP [<ffffffff810d6053>] spin_dump+0x53/0xc0 RSP <ffff88003750fd68> ---[ end trace 783ebb76612867a0 ]--- NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053] Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 3 PID: 18053 Comm: test Tainted: G D 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: native_read_tsc+0x0/0x20 Call Trace: ? delay_tsc+0x40/0x70 __delay+0xf/0x20 do_raw_spin_lock+0x96/0x140 _raw_spin_lock+0xe/0x10 sem_lock_and_putref+0x11/0x70 SYSC_semtimedop+0x7bf/0x960 ? handle_mm_fault+0xbf6/0x1880 ? dequeue_task_fair+0x79/0x4a0 ? __do_page_fault+0x19a/0x430 ? kfree_debugcheck+0x16/0x40 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 ? do_audit_syscall_entry+0x66/0x70 ? syscall_trace_enter_phase1+0x139/0x160 SyS_semtimedop+0xe/0x10 SyS_semop+0x10/0x20 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9 Kernel panic - not syncing: softlockup: hung tasks I wasn't able to trigger any badness on a recent kernel without the proper config debugs enabled, however I have softlockup reports on some kernel versions, in the semaphore code, which are similar as above (the scenario is seen on some servers running IBM DB2 which uses semaphore syscalls). The patch here fixes the race against freeary, by acquiring or waiting on the sem_undo_list lock as necessary (exit_sem can race with freeary, while freeary sets un->semid to -1 and removes the same sem_undo from list_proc or when it removes the last sem_undo). After the patch I'm unable to reproduce the problem using the test case [1]. [1] Test case used below: #include <stdio.h> #include <sys/types.h> #include <sys/ipc.h> #include <sys/sem.h> #include <sys/wait.h> #include <stdlib.h> #include <time.h> #include <unistd.h> #include <errno.h> #define NSEM 1 #define NSET 5 int sid[NSET]; void thread() { struct sembuf op; int s; uid_t pid = getuid(); s = rand() % NSET; op.sem_num = pid % NSEM; op.sem_op = 1; op.sem_flg = SEM_UNDO; semop(sid[s], &op, 1); exit(EXIT_SUCCESS); } void create_set() { int i, j; pid_t p; union { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; } un; /* Create and initialize semaphore set */ for (i = 0; i < NSET; i++) { sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT); if (sid[i] < 0) { perror("semget"); exit(EXIT_FAILURE); } } un.val = 0; for (i = 0; i < NSET; i++) { for (j = 0; j < NSEM; j++) { if (semctl(sid[i], j, SETVAL, un) < 0) perror("semctl"); } } /* Launch threads that operate on semaphore set */ for (i = 0; i < NSEM * NSET * NSET; i++) { p = fork(); if (p < 0) perror("fork"); if (p == 0) thread(); } /* Free semaphore set */ for (i = 0; i < NSET; i++) { if (semctl(sid[i], NSEM, IPC_RMID)) perror("IPC_RMID"); } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } int main(int argc, char **argv) { pid_t p; srand(time(NULL)); while (1) { p = fork(); if (p < 0) { perror("fork"); exit(EXIT_FAILURE); } if (p == 0) { create_set(); goto end; } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } end: return 0; } [akpm@linux-foundation.org: use normal comment layout] Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Rafael Aquini <aquini@redhat.com> CC: Aristeu Rozanski <aris@redhat.com> Cc: David Jeffery <djeffery@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-08-14 22:35:02 +00:00
continue;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
ipc,sem: fix use after free on IPC_RMID after a task using same semaphore set exits The current semaphore code allows a potential use after free: in exit_sem we may free the task's sem_undo_list while there is still another task looping through the same semaphore set and cleaning the sem_undo list at freeary function (the task called IPC_RMID for the same semaphore set). For example, with a test program [1] running which keeps forking a lot of processes (which then do a semop call with SEM_UNDO flag), and with the parent right after removing the semaphore set with IPC_RMID, and a kernel built with CONFIG_SLAB, CONFIG_SLAB_DEBUG and CONFIG_DEBUG_SPINLOCK, you can easily see something like the following in the kernel log: Slab corruption (Not tainted): kmalloc-64 start=ffff88003b45c1c0, len=64 000: 6b 6b 6b 6b 6b 6b 6b 6b 00 6b 6b 6b 6b 6b 6b 6b kkkkkkkk.kkkkkkk 010: ff ff ff ff 6b 6b 6b 6b ff ff ff ff ff ff ff ff ....kkkk........ Prev obj: start=ffff88003b45c180, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff c0 fb 01 37 00 88 ff ff ...........7.... Next obj: start=ffff88003b45c200, len=64 000: 00 00 00 00 ad 4e ad de ff ff ff ff 5a 5a 5a 5a .....N......ZZZZ 010: ff ff ff ff ff ff ff ff 68 29 a7 3c 00 88 ff ff ........h).<.... BUG: spinlock wrong CPU on CPU#2, test/18028 general protection fault: 0000 [#1] SMP Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 2 PID: 18028 Comm: test Not tainted 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: spin_dump+0x53/0xc0 Call Trace: spin_bug+0x30/0x40 do_raw_spin_unlock+0x71/0xa0 _raw_spin_unlock+0xe/0x10 freeary+0x82/0x2a0 ? _raw_spin_lock+0xe/0x10 semctl_down.clone.0+0xce/0x160 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 SyS_semctl+0x236/0x2c0 ? syscall_trace_leave+0xde/0x130 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 8b 80 88 03 00 00 48 8d 88 60 05 00 00 48 c7 c7 a0 2c a4 81 31 c0 65 8b 15 eb 40 f3 7e e8 08 31 68 00 4d 85 e4 44 8b 4b 08 74 5e <45> 8b 84 24 88 03 00 00 49 8d 8c 24 60 05 00 00 8b 53 04 48 89 RIP [<ffffffff810d6053>] spin_dump+0x53/0xc0 RSP <ffff88003750fd68> ---[ end trace 783ebb76612867a0 ]--- NMI watchdog: BUG: soft lockup - CPU#3 stuck for 22s! [test:18053] Modules linked in: 8021q mrp garp stp llc nf_conntrack_ipv4 nf_defrag_ipv4 ip6t_REJECT nf_reject_ipv6 nf_conntrack_ipv6 nf_defrag_ipv6 xt_state nf_conntrack ip6table_filter ip6_tables binfmt_misc ppdev input_leds joydev parport_pc parport floppy serio_raw virtio_balloon virtio_rng virtio_console virtio_net iosf_mbi crct10dif_pclmul crc32_pclmul ghash_clmulni_intel pcspkr qxl ttm drm_kms_helper drm snd_hda_codec_generic i2c_piix4 snd_hda_intel snd_hda_codec snd_hda_core snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer snd soundcore crc32c_intel virtio_pci virtio_ring virtio pata_acpi ata_generic [last unloaded: speedstep_lib] CPU: 3 PID: 18053 Comm: test Tainted: G D 4.2.0-rc5+ #1 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014 RIP: native_read_tsc+0x0/0x20 Call Trace: ? delay_tsc+0x40/0x70 __delay+0xf/0x20 do_raw_spin_lock+0x96/0x140 _raw_spin_lock+0xe/0x10 sem_lock_and_putref+0x11/0x70 SYSC_semtimedop+0x7bf/0x960 ? handle_mm_fault+0xbf6/0x1880 ? dequeue_task_fair+0x79/0x4a0 ? __do_page_fault+0x19a/0x430 ? kfree_debugcheck+0x16/0x40 ? __do_page_fault+0x19a/0x430 ? __audit_syscall_entry+0xa8/0x100 ? do_audit_syscall_entry+0x66/0x70 ? syscall_trace_enter_phase1+0x139/0x160 SyS_semtimedop+0xe/0x10 SyS_semop+0x10/0x20 entry_SYSCALL_64_fastpath+0x12/0x71 Code: 47 10 83 e8 01 85 c0 89 47 10 75 08 65 48 89 3d 1f 74 ff 7e c9 c3 0f 1f 44 00 00 55 48 89 e5 e8 87 17 04 00 66 90 c9 c3 0f 1f 00 <55> 48 89 e5 0f 31 89 c1 48 89 d0 48 c1 e0 20 89 c9 48 09 c8 c9 Kernel panic - not syncing: softlockup: hung tasks I wasn't able to trigger any badness on a recent kernel without the proper config debugs enabled, however I have softlockup reports on some kernel versions, in the semaphore code, which are similar as above (the scenario is seen on some servers running IBM DB2 which uses semaphore syscalls). The patch here fixes the race against freeary, by acquiring or waiting on the sem_undo_list lock as necessary (exit_sem can race with freeary, while freeary sets un->semid to -1 and removes the same sem_undo from list_proc or when it removes the last sem_undo). After the patch I'm unable to reproduce the problem using the test case [1]. [1] Test case used below: #include <stdio.h> #include <sys/types.h> #include <sys/ipc.h> #include <sys/sem.h> #include <sys/wait.h> #include <stdlib.h> #include <time.h> #include <unistd.h> #include <errno.h> #define NSEM 1 #define NSET 5 int sid[NSET]; void thread() { struct sembuf op; int s; uid_t pid = getuid(); s = rand() % NSET; op.sem_num = pid % NSEM; op.sem_op = 1; op.sem_flg = SEM_UNDO; semop(sid[s], &op, 1); exit(EXIT_SUCCESS); } void create_set() { int i, j; pid_t p; union { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; } un; /* Create and initialize semaphore set */ for (i = 0; i < NSET; i++) { sid[i] = semget(IPC_PRIVATE , NSEM, 0644 | IPC_CREAT); if (sid[i] < 0) { perror("semget"); exit(EXIT_FAILURE); } } un.val = 0; for (i = 0; i < NSET; i++) { for (j = 0; j < NSEM; j++) { if (semctl(sid[i], j, SETVAL, un) < 0) perror("semctl"); } } /* Launch threads that operate on semaphore set */ for (i = 0; i < NSEM * NSET * NSET; i++) { p = fork(); if (p < 0) perror("fork"); if (p == 0) thread(); } /* Free semaphore set */ for (i = 0; i < NSET; i++) { if (semctl(sid[i], NSEM, IPC_RMID)) perror("IPC_RMID"); } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } int main(int argc, char **argv) { pid_t p; srand(time(NULL)); while (1) { p = fork(); if (p < 0) { perror("fork"); exit(EXIT_FAILURE); } if (p == 0) { create_set(); goto end; } /* Wait for forked processes to exit */ while (wait(NULL)) { if (errno == ECHILD) break; }; } end: return 0; } [akpm@linux-foundation.org: use normal comment layout] Signed-off-by: Herton R. Krzesinski <herton@redhat.com> Acked-by: Manfred Spraul <manfred@colorfullife.com> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Rafael Aquini <aquini@redhat.com> CC: Aristeu Rozanski <aris@redhat.com> Cc: David Jeffery <djeffery@redhat.com> Cc: <stable@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-08-14 22:35:02 +00:00
sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
/* exit_sem raced with IPC_RMID, nothing to do */
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
if (IS_ERR(sma)) {
rcu_read_unlock();
continue;
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
}
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_lock(sma, NULL, -1);
/* exit_sem raced with IPC_RMID, nothing to do */
if (!ipc_valid_object(&sma->sem_perm)) {
sem_unlock(sma, -1);
rcu_read_unlock();
continue;
}
un = __lookup_undo(ulp, semid);
if (un == NULL) {
/* exit_sem raced with IPC_RMID+semget() that created
* exactly the same semid. Nothing to do.
*/
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
continue;
}
/* remove un from the linked lists */
ipc_assert_locked_object(&sma->sem_perm);
list_del(&un->list_id);
/* we are the last process using this ulp, acquiring ulp->lock
* isn't required. Besides that, we are also protected against
* IPC_RMID as we hold sma->sem_perm lock now
*/
list_del_rcu(&un->list_proc);
/* perform adjustments registered in un */
for (i = 0; i < sma->sem_nsems; i++) {
struct sem *semaphore = &sma->sems[i];
if (un->semadj[i]) {
semaphore->semval += un->semadj[i];
/*
* Range checks of the new semaphore value,
* not defined by sus:
* - Some unices ignore the undo entirely
* (e.g. HP UX 11i 11.22, Tru64 V5.1)
* - some cap the value (e.g. FreeBSD caps
* at 0, but doesn't enforce SEMVMX)
*
* Linux caps the semaphore value, both at 0
* and at SEMVMX.
*
* Manfred <manfred@colorfullife.com>
*/
if (semaphore->semval < 0)
semaphore->semval = 0;
if (semaphore->semval > SEMVMX)
semaphore->semval = SEMVMX;
ipc/sem: Fix semctl(..., GETPID, ...) between pid namespaces Today the last process to update a semaphore is remembered and reported in the pid namespace of that process. If there are processes in any other pid namespace querying that process id with GETPID the result will be unusable nonsense as it does not make any sense in your own pid namespace. Due to ipc_update_pid I don't think you will be able to get System V ipc semaphores into a troublesome cache line ping-pong. Using struct pids from separate process are not a problem because they do not share a cache line. Using struct pid from different threads of the same process are unlikely to be a problem as the reference count update can be avoided. Further linux futexes are a much better tool for the job of mutual exclusion between processes than System V semaphores. So I expect programs that are performance limited by their interprocess mutual exclusion primitive will be using futexes. So while it is possible that enhancing the storage of the last rocess of a System V semaphore from an integer to a struct pid will cause a performance regression because of the effect of frequently updating the pid reference count. I don't expect that to happen in practice. This change updates semctl(..., GETPID, ...) to return the process id of the last process to update a semphore inthe pid namespace of the calling process. Fixes: b488893a390e ("pid namespaces: changes to show virtual ids to user") Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
2018-03-23 06:11:29 +00:00
ipc_update_pid(&semaphore->sempid, task_tgid(current));
}
}
/* maybe some queued-up processes were waiting for this */
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
do_smart_update(sma, NULL, 0, 1, &wake_q);
ipc,sem: fine grained locking for semtimedop Introduce finer grained locking for semtimedop, to handle the common case of a program wanting to manipulate one semaphore from an array with multiple semaphores. If the call is a semop manipulating just one semaphore in an array with multiple semaphores, only take the lock for that semaphore itself. If the call needs to manipulate multiple semaphores, or another caller is in a transaction that manipulates multiple semaphores, the sem_array lock is taken, as well as all the locks for the individual semaphores. On a 24 CPU system, performance numbers with the semop-multi test with N threads and N semaphores, look like this: vanilla Davidlohr's Davidlohr's + Davidlohr's + threads patches rwlock patches v3 patches 10 610652 726325 1783589 2142206 20 341570 365699 1520453 1977878 30 288102 307037 1498167 2037995 40 290714 305955 1612665 2256484 50 288620 312890 1733453 2650292 60 289987 306043 1649360 2388008 70 291298 306347 1723167 2717486 80 290948 305662 1729545 2763582 90 290996 306680 1736021 2757524 100 292243 306700 1773700 3059159 [davidlohr.bueso@hp.com: do not call sem_lock when bogus sma] [davidlohr.bueso@hp.com: make refcounter atomic] Signed-off-by: Rik van Riel <riel@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Acked-by: Davidlohr Bueso <davidlohr.bueso@hp.com> Cc: Chegu Vinod <chegu_vinod@hp.com> Cc: Jason Low <jason.low2@hp.com> Reviewed-by: Michel Lespinasse <walken@google.com> Cc: Peter Hurley <peter@hurleysoftware.com> Cc: Stanislav Kinsbursky <skinsbursky@parallels.com> Tested-by: Emmanuel Benisty <benisty.e@gmail.com> Tested-by: Sedat Dilek <sedat.dilek@gmail.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-05-01 02:15:44 +00:00
sem_unlock(sma, -1);
rcu_read_unlock();
ipc/sem: rework task wakeups Our sysv sems have been using the notion of lockless wakeups for a while, ever since commit 0a2b9d4c7967 ("ipc/sem.c: move wake_up_process out of the spinlock section"), in order to reduce the sem_lock hold times. This in-house pending queue can be replaced by wake_q (just like all the rest of ipc now), in that it provides the following advantages: o Simplifies and gets rid of unnecessary code. o We get rid of the IN_WAKEUP complexities. Given that wake_q_add() grabs reference to the task, if awoken due to an unrelated event, between the wake_q_add() and wake_up_q() window, we cannot race with sys_exit and the imminent call to wake_up_process(). o By not spinning IN_WAKEUP, we no longer need to disable preemption. In consequence, the wakeup paths (after schedule(), that is) must acknowledge an external signal/event, as well spurious wakeup occurring during the pending wakeup window. Obviously no changes in semantics that could be visible to the user. The fastpath is _only_ for when we know for sure that we were awoken due to a the waker's successful semop call (queue.status is not -EINTR). On a 48-core Haswell, running the ipcscale 'waitforzero' test, the following is seen with increasing thread counts: v4.8-rc5 v4.8-rc5 semopv2 Hmean sembench-sem-2 574733.00 ( 0.00%) 578322.00 ( 0.62%) Hmean sembench-sem-8 811708.00 ( 0.00%) 824689.00 ( 1.59%) Hmean sembench-sem-12 842448.00 ( 0.00%) 845409.00 ( 0.35%) Hmean sembench-sem-21 933003.00 ( 0.00%) 977748.00 ( 4.80%) Hmean sembench-sem-48 935910.00 ( 0.00%) 1004759.00 ( 7.36%) Hmean sembench-sem-79 937186.00 ( 0.00%) 983976.00 ( 4.99%) Hmean sembench-sem-234 974256.00 ( 0.00%) 1060294.00 ( 8.83%) Hmean sembench-sem-265 975468.00 ( 0.00%) 1016243.00 ( 4.18%) Hmean sembench-sem-296 991280.00 ( 0.00%) 1042659.00 ( 5.18%) Hmean sembench-sem-327 975415.00 ( 0.00%) 1029977.00 ( 5.59%) Hmean sembench-sem-358 1014286.00 ( 0.00%) 1049624.00 ( 3.48%) Hmean sembench-sem-389 972939.00 ( 0.00%) 1043127.00 ( 7.21%) Hmean sembench-sem-420 981909.00 ( 0.00%) 1056747.00 ( 7.62%) Hmean sembench-sem-451 990139.00 ( 0.00%) 1051609.00 ( 6.21%) Hmean sembench-sem-482 965735.00 ( 0.00%) 1040313.00 ( 7.72%) [akpm@linux-foundation.org: coding-style fixes] [sfr@canb.auug.org.au: merge fix for WAKE_Q to DEFINE_WAKE_Q rename] Link: http://lkml.kernel.org/r/20161122210410.5eca9fc2@canb.auug.org.au Link: http://lkml.kernel.org/r/1474225896-10066-3-git-send-email-dave@stgolabs.net Signed-off-by: Davidlohr Bueso <dbueso@suse.de> Acked-by: Manfred Spraul <manfred@colorfullife.com> Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-12-14 23:06:34 +00:00
wake_up_q(&wake_q);
kfree_rcu(un, rcu);
}
kfree(ulp);
}
#ifdef CONFIG_PROC_FS
static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
{
struct user_namespace *user_ns = seq_user_ns(s);
struct kern_ipc_perm *ipcp = it;
struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
time64_t sem_otime;
/*
* The proc interface isn't aware of sem_lock(), it calls
* ipc_lock_object() directly (in sysvipc_find_ipc).
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
* In order to stay compatible with sem_lock(), we must
* enter / leave complex_mode.
*/
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
complexmode_enter(sma);
sem_otime = get_semotime(sma);
seq_printf(s,
"%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
sma->sem_perm.key,
sma->sem_perm.id,
sma->sem_perm.mode,
sma->sem_nsems,
from_kuid_munged(user_ns, sma->sem_perm.uid),
from_kgid_munged(user_ns, sma->sem_perm.gid),
from_kuid_munged(user_ns, sma->sem_perm.cuid),
from_kgid_munged(user_ns, sma->sem_perm.cgid),
sem_otime,
sma->sem_ctime);
ipc/sem.c: fix complex_count vs. simple op race Commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") introduced a race: sem_lock has a fast path that allows parallel simple operations. There are two reasons why a simple operation cannot run in parallel: - a non-simple operations is ongoing (sma->sem_perm.lock held) - a complex operation is sleeping (sma->complex_count != 0) As both facts are stored independently, a thread can bypass the current checks by sleeping in the right positions. See below for more details (or kernel bugzilla 105651). The patch fixes that by creating one variable (complex_mode) that tracks both reasons why parallel operations are not possible. The patch also updates stale documentation regarding the locking. With regards to stable kernels: The patch is required for all kernels that include the commit 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") (3.10?) The alternative is to revert the patch that introduced the race. The patch is safe for backporting, i.e. it makes no assumptions about memory barriers in spin_unlock_wait(). Background: Here is the race of the current implementation: Thread A: (simple op) - does the first "sma->complex_count == 0" test Thread B: (complex op) - does sem_lock(): This includes an array scan. But the scan can't find Thread A, because Thread A does not own sem->lock yet. - the thread does the operation, increases complex_count, drops sem_lock, sleeps Thread A: - spin_lock(&sem->lock), spin_is_locked(sma->sem_perm.lock) - sleeps before the complex_count test Thread C: (complex op) - does sem_lock (no array scan, complex_count==1) - wakes up Thread B. - decrements complex_count Thread A: - does the complex_count test Bug: Now both thread A and thread C operate on the same array, without any synchronization. Fixes: 6d07b68ce16a ("ipc/sem.c: optimize sem_lock()") Link: http://lkml.kernel.org/r/1469123695-5661-1-git-send-email-manfred@colorfullife.com Reported-by: <felixh@informatik.uni-bremen.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Davidlohr Bueso <dave@stgolabs.net> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: <1vier1@web.de> Cc: <stable@vger.kernel.org> [3.10+] Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-11 20:54:50 +00:00
complexmode_tryleave(sma);
return 0;
}
#endif