Fundamental model of the current Linux kernel is to lazily init and
restore FPU instead of restoring the task state during context switch.
This changes that fundamental lazy model to the non-lazy model for
the processors supporting xsave feature.
Reasons driving this model change are:
i. Newer processors support optimized state save/restore using xsaveopt and
xrstor by tracking the INIT state and MODIFIED state during context-switch.
This is faster than modifying the cr0.TS bit which has serializing semantics.
ii. Newer glibc versions use SSE for some of the optimized copy/clear routines.
With certain workloads (like boot, kernel-compilation etc), application
completes its work with in the first 5 task switches, thus taking upto 5 #DNA
traps with the kernel not getting a chance to apply the above mentioned
pre-load heuristic.
iii. Some xstate features (like AMD's LWP feature) don't honor the cr0.TS bit
and thus will not work correctly in the presence of lazy restore. Non-lazy
state restore is needed for enabling such features.
Some data on a two socket SNB system:
* Saved 20K DNA exceptions during boot on a two socket SNB system.
* Saved 50K DNA exceptions during kernel-compilation workload.
* Improved throughput of the AVX based checksumming function inside the
kernel by ~15% as xsave/xrstor is faster than the serializing clts/stts
pair.
Also now kernel_fpu_begin/end() relies on the patched
alternative instructions. So move check_fpu() which uses the
kernel_fpu_begin/end() after alternative_instructions().
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1345842782-24175-7-git-send-email-suresh.b.siddha@intel.com
Merge 32-bit boot fix from,
Link: http://lkml.kernel.org/r/1347300665-6209-4-git-send-email-suresh.b.siddha@intel.com
Cc: Jim Kukunas <james.t.kukunas@linux.intel.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Avi Kivity <avi@redhat.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
use kernel_fpu_begin/end() instead of unconditionally accessing cr0 and
saving/restoring just the few used xmm/ymm registers.
This has some advantages like:
* If the task's FPU state is already active, then kernel_fpu_begin()
will just save the user-state and avoiding the read/write of cr0.
In general, cr0 accesses are much slower.
* Manual save/restore of xmm/ymm registers will affect the 'modified' and
the 'init' optimizations brought in the by xsaveopt/xrstor
infrastructure.
* Foward compatibility with future vector register extensions will be a
problem if the xmm/ymm registers are manually saved and restored
(corrupting the extended state of those vector registers).
With this patch, there was no significant difference in the xor throughput
using AVX, measured during boot.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1345842782-24175-5-git-send-email-suresh.b.siddha@intel.com
Cc: Jim Kukunas <james.t.kukunas@linux.intel.com>
Cc: NeilBrown <neilb@suse.de>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Currently for x86 and x86_32 binaries, fpstate in the user sigframe is copied
to/from the fpstate in the task struct.
And in the case of signal delivery for x86_64 binaries, if the fpstate is live
in the CPU registers, then the live state is copied directly to the user
sigframe. Otherwise fpstate in the task struct is copied to the user sigframe.
During restore, fpstate in the user sigframe is restored directly to the live
CPU registers.
Historically, different code paths led to different bugs. For example,
x86_64 code path was not preemption safe till recently. Also there is lot
of code duplication for support of new features like xsave etc.
Unify signal handling code paths for x86 and x86_64 kernels.
New strategy is as follows:
Signal delivery: Both for 32/64-bit frames, align the core math frame area to
64bytes as needed by xsave (this where the main fpu/extended state gets copied
to and excludes the legacy compatibility fsave header for the 32-bit [f]xsave
frames). If the state is live, copy the register state directly to the user
frame. If not live, copy the state in the thread struct to the user frame. And
for 32-bit [f]xsave frames, construct the fsave header separately before
the actual [f]xsave area.
Signal return: As the 32-bit frames with [f]xstate has an additional
'fsave' header, copy everything back from the user sigframe to the
fpstate in the task structure and reconstruct the fxstate from the 'fsave'
header (Also user passed pointers may not be correctly aligned for
any attempt to directly restore any partial state). At the next fpstate usage,
everything will be restored to the live CPU registers.
For all the 64-bit frames and the 32-bit fsave frame, restore the state from
the user sigframe directly to the live CPU registers. 64-bit signals always
restored the math frame directly, so we can expect the math frame pointer
to be correctly aligned. For 32-bit fsave frames, there are no alignment
requirements, so we can restore the state directly.
"lat_sig catch" microbenchmark numbers (for x86, x86_64, x86_32 binaries) are
with in the noise range with this change.
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1343171129-2747-4-git-send-email-suresh.b.siddha@intel.com
[ Merged in compilation fix ]
Link: http://lkml.kernel.org/r/1344544736.8326.17.camel@sbsiddha-desk.sc.intel.com
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
With this patch we provide the functionality to initialize the
Xen-SWIOTLB late in the bootup cycle - specifically for
Xen PCI-frontend. We still will work if the user had
supplied 'iommu=soft' on the Linux command line.
Note: We cannot depend on after_bootmem to automatically
determine whether this is early or not. This is because
when PCI IOMMUs are initialized it is after after_bootmem but
before a lot of "other" subsystems are initialized.
CC: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
[v1: Fix smatch warnings]
[v2: Added check for xen_swiotlb]
[v3: Rebased with new xen-swiotlb changes]
[v4: squashed xen/swiotlb: Depending on after_bootmem is not correct in]
Reviewed-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Make arch_uprobe_task->saved_trap_nr "unsigned int" and move it down
after ->saved_scratch_register, this changes sizeof() from 24 to 16.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
arch_uprobe_disable_step() correctly preserves X86_EFLAGS_TF and
returns to user-mode. But this means the application gets SIGTRAP
only after the next insn.
This means that UPROBE_CLEAR_TF logic is not really right. _enable
should only record the state of X86_EFLAGS_TF, and _disable should
check it separately from UPROBE_FIX_SETF.
Remove arch_uprobe_task->restore_flags, add ->saved_tf instead, and
change enable/disable accordingly. This assumes that the probed insn
was not trapped, see the next patch.
arch_uprobe_skip_sstep() logic has the same problem, change it to
check X86_EFLAGS_TF and send SIGTRAP as well. We will cleanup this
all after we fold enable/disable_step into pre/post_hol hooks.
Note: send_sig(SIGTRAP) is not actually right, we need send_sigtrap().
But this needs more changes, handle_swbp() does the same and this is
equally wrong.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
user_enable/disable_single_step() was designed for ptrace, it assumes
a single user and does unnecessary and wrong things for uprobes. For
example:
- arch_uprobe_enable_step() can't trust TIF_SINGLESTEP, an
application itself can set X86_EFLAGS_TF which must be
preserved after arch_uprobe_disable_step().
- we do not want to set TIF_SINGLESTEP/TIF_FORCED_TF in
arch_uprobe_enable_step(), this only makes sense for ptrace.
- otoh we leak TIF_SINGLESTEP if arch_uprobe_disable_step()
doesn't do user_disable_single_step(), the application will
be killed after the next syscall.
- arch_uprobe_enable_step() does access_process_vm() we do
not need/want.
Change arch_uprobe_enable/disable_step() to set/clear X86_EFLAGS_TF
directly, this is much simpler and more correct. However, we need to
clear TIF_BLOCKSTEP/DEBUGCTLMSR_BTF before executing the probed insn,
add set_task_blockstep(false).
Note: with or without this patch, there is another (hopefully minor)
problem. A probed "pushf" insn can see the wrong X86_EFLAGS_TF set by
uprobes. Perhaps we should change _disable to update the stack, or
teach arch_uprobe_skip_sstep() to emulate this insn.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
The arch specific implementation behaves like user_enable_single_step()
except that it does not disable single stepping if it was already
enabled by ptrace. This allows the debugger to single step over an
uprobe. The state of block stepping is not restored. It makes only sense
together with TF and if that was enabled then the debugger is notified.
Note: this is still not correct. For example, TIF_SINGLESTEP check
is not right, the application itself can set X86_EFLAGS_TF. And otoh
we leak TIF_SINGLESTEP (set by enable) if the probed insn is "popf".
See the next patches, we need the changes in arch/x86/kernel/step.c
first.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Following a relatively recent compiler change, make use of the
fact that for non-zero input BSF and TZCNT produce the same
result, and that CPUs not knowing of TZCNT will treat the
instruction as BSF (i.e. ignore what looks like a REP prefix to
them). The assumption here is that TZCNT would never have worse
performance than BSF.
For the moment, only do this when the respective generic-CPU
option is selected (as there are no specific-CPU options
covering the CPUs supporting TZCNT), and don't do that when size
optimization was requested.
Signed-off-by: Jan Beulich <jbeulich@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/504DEA1B020000780009A277@nat28.tlf.novell.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
On 64 bit x86 we save the current eflags in cpu_init for use in
ret_from_fork. Strictly speaking reserved bits in EFLAGS should
be read as written but in practise it is unlikely that EFLAGS
could ever be extended in this way and the kernel alread clears
any undefined flags early on.
The equivalent 32 bit code simply hard codes 0x0202 as the new
EFLAGS.
This change makes 64 bit use the same mechanism to setup the
initial EFLAGS on fork. Note that 64 bit resets EFLAGS before
calling schedule_tail() as opposed to 32 bit which calls
schedule_tail() first. Therefore the correct value for EFLAGS
has opposite IF bit.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Acked-by: Andi Kleen <ak@linux.intel.com>
Acked-by: "H. Peter Anvin" <hpa@zytor.com>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Andi Kleen <ak@linux.intel.com>
Link: http://lkml.kernel.org/r/20120824195847.GA31628@moon
Signed-off-by: Ingo Molnar <mingo@kernel.org>
If the caller passes a valid kmap_op to m2p_add_override, we use
kmap_op->dev_bus_addr to store the original mfn, but dev_bus_addr is
part of the interface with Xen and if we are batching the hypercalls it
might not have been written by the hypervisor yet. That means that later
on Xen will write to it and we'll think that the original mfn is
actually what Xen has written to it.
Rather than "stealing" struct members from kmap_op, keep using
page->index to store the original mfn and add another parameter to
m2p_remove_override to get the corresponding kmap_op instead.
It is now responsibility of the caller to keep track of which kmap_op
corresponds to a particular page in the m2p_override (gntdev, the only
user of this interface that passes a valid kmap_op, is already doing that).
CC: stable@kernel.org
Reported-and-Tested-By: Sander Eikelenboom <linux@eikelenboom.it>
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
* 'x86/platform' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (9690 commits)
x86: Document x86_init.paging.pagetable_init()
x86: xen: Cleanup and remove x86_init.paging.pagetable_setup_done()
x86: Move paging_init() call to x86_init.paging.pagetable_init()
x86: Rename pagetable_setup_start() to pagetable_init()
x86: Remove base argument from x86_init.paging.pagetable_setup_start
Linux 3.6-rc5
HID: tpkbd: work even if the new Lenovo Keyboard driver is not configured
Remove user-triggerable BUG from mpol_to_str
xen/pciback: Fix proper FLR steps.
uml: fix compile error in deliver_alarm()
dj: memory scribble in logi_dj
Fix order of arguments to compat_put_time[spec|val]
xen: Use correct masking in xen_swiotlb_alloc_coherent.
xen: fix logical error in tlb flushing
xen/p2m: Fix one-off error in checking the P2M tree directory.
powerpc: Don't use __put_user() in patch_instruction
powerpc: Make sure IPI handlers see data written by IPI senders
powerpc: Restore correct DSCR in context switch
powerpc: Fix DSCR inheritance in copy_thread()
powerpc: Keep thread.dscr and thread.dscr_inherit in sync
...
Some AMD systems may round the frequencies in ACPI tables to 100MHz
boundaries. We can obtain the real frequencies from MSRs, so add a quirk
to fix these frequencies up on AMD systems.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
The programming model for P-states on modern AMD CPUs is very similar to
that of Intel and VIA. It makes sense to consolidate this support into one
driver rather than duplicating functionality between two of them. This
patch adds support for AMDs with hardware P-state control to acpi-cpufreq.
Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Andre Przywara <andre.przywara@amd.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Optimize "rep ins" by allowing emulator to write back more than one
datum at a time. Introduce new operand type OP_MEM_STR which tells
writeback() that dst contains pointer to an array that should be written
back as opposite to just one data element.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Current code assumes that IO exit was due to instruction emulation
and handles execution back to emulator directly. This patch adds new
userspace IO exit completion callback that can be set by any other code
that caused IO exit to userspace.
Signed-off-by: Gleb Natapov <gleb@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
We never change emulate_ops[] at runtime so it should be r/o.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
Fix the following sparse warnings:
sys_ia32.c:293:38: warning: incorrect type in argument 2 (different address spaces)
sys_ia32.c:293:38: expected unsigned int [noderef] [usertype] <asn:1>*stat_addr
sys_ia32.c:293:38: got unsigned int *stat_addr
Ironically, sys_ia32.h was introduced to fix sparse warnings but
missed that one.
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Link: http://lkml.kernel.org/r/1346621506-30857-2-git-send-email-minipli@googlemail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
HPET_ID_VENDOR_8086 is defined but never used. It would be a redefine
of PCI_VENDOR_ID_INTEL if it was ever used.
Signed-off-by: Jon Mason <jdmason@kudzu.us>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Instead of populating the entire register file, read in registers
as they are accessed, and write back only the modified ones. This
saves a VMREAD and VMWRITE on Intel (for rsp, since it is not usually
used during emulation), and a two 128-byte copies for the registers.
Signed-off-by: Avi Kivity <avi@redhat.com>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Merging critical fixes from upstream required for development.
* upstream/master: (809 commits)
libata: Add a space to " 2GB ATA Flash Disk" DMA blacklist entry
Revert "powerpc: Update g5_defconfig"
powerpc/perf: Use pmc_overflow() to detect rolled back events
powerpc: Fix VMX in interrupt check in POWER7 copy loops
powerpc: POWER7 copy_to_user/copy_from_user patch applied twice
powerpc: Fix personality handling in ppc64_personality()
powerpc/dma-iommu: Fix IOMMU window check
powerpc: Remove unnecessary ifdefs
powerpc/kgdb: Restore current_thread_info properly
powerpc/kgdb: Bail out of KGDB when we've been triggered
powerpc/kgdb: Do not set kgdb_single_step on ppc
powerpc/mpic_msgr: Add missing includes
powerpc: Fix null pointer deref in perf hardware breakpoints
powerpc: Fixup whitespace in xmon
powerpc: Fix xmon dl command for new printk implementation
xfs: check for possible overflow in xfs_ioc_trim
xfs: unlock the AGI buffer when looping in xfs_dialloc
xfs: fix uninitialised variable in xfs_rtbuf_get()
powerpc/fsl: fix "Failed to mount /dev: No such device" errors
powerpc/fsl: update defconfigs
...
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
If the kernel is compiled with gcc 4.6.0 which supports -mfentry,
then use that instead of mcount.
With mcount, frame pointers are forced with the -pg option and we
get something like:
<can_vma_merge_before>:
55 push %rbp
48 89 e5 mov %rsp,%rbp
53 push %rbx
41 51 push %r9
e8 fe 6a 39 00 callq ffffffff81483d00 <mcount>
31 c0 xor %eax,%eax
48 89 fb mov %rdi,%rbx
48 89 d7 mov %rdx,%rdi
48 33 73 30 xor 0x30(%rbx),%rsi
48 f7 c6 ff ff ff f7 test $0xfffffffff7ffffff,%rsi
With -mfentry, frame pointers are no longer forced and the call looks
like this:
<can_vma_merge_before>:
e8 33 af 37 00 callq ffffffff81461b40 <__fentry__>
53 push %rbx
48 89 fb mov %rdi,%rbx
31 c0 xor %eax,%eax
48 89 d7 mov %rdx,%rdi
41 51 push %r9
48 33 73 30 xor 0x30(%rbx),%rsi
48 f7 c6 ff ff ff f7 test $0xfffffffff7ffffff,%rsi
This adds the ftrace hook at the beginning of the function before a
frame is set up, and allows the function callbacks to be able to access
parameters. As kprobes now can use function tracing (at least on x86)
this speeds up the kprobe hooks that are at the beginning of the
function.
Link: http://lkml.kernel.org/r/20120807194100.130477900@goodmis.org
Acked-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
All the original Xen headers have xen_pfn_t as mfn and pfn type, however
when they have been imported in Linux, xen_pfn_t has been replaced with
unsigned long. That might work for x86 and ia64 but it does not for arm.
Bring back xen_pfn_t and let each architecture define xen_pfn_t as they
see fit.
Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Acked-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
We still patch SMP instructions to UP variants if we boot with a
single CPU, but not at any other time. In particular, not if we
unplug CPUs to return to a single cpu.
Paul McKenney points out:
mean offline overhead is 6251/48=130.2 milliseconds.
If I remove the alternatives_smp_switch() from the offline
path [...] the mean offline overhead is 550/42=13.1 milliseconds
Basically, we're never going to get those 120ms back, and the
code is pretty messy.
We get rid of:
1) The "smp-alt-once" boot option. It's actually "smp-alt-boot", the
documentation is wrong. It's now the default.
2) The skip_smp_alternatives flag used by suspend.
3) arch_disable_nonboot_cpus_begin() and arch_disable_nonboot_cpus_end()
which were only used to set this one flag.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Paul McKenney <paul.mckenney@us.ibm.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/87vcgwwive.fsf@rustcorp.com.au
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The distinction between CONFIG_KVM_CLOCK and CONFIG_KVM_GUEST is
not so clear anymore, as demonstrated by recent bugs caused by poor
handling of on/off combinations of these options.
Merge CONFIG_KVM_CLOCK into CONFIG_KVM_GUEST.
Reported-By: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
In current code, if we map a readonly memory space from host to guest
and the page is not currently mapped in the host, we will get a fault
pfn and async is not allowed, then the vm will crash
We introduce readonly memory region to map ROM/ROMD to the guest, read access
is happy for readonly memslot, write access on readonly memslot will cause
KVM_EXIT_MMIO exit
Signed-off-by: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
This comment is no longer true. We support up to 2^16 CPUs
because __ticket_t is an u16 if NR_CPUS is larger than 256.
Signed-off-by: Richard Weinberger <richard@nod.at>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:
* Fix include order for bison/flex-generated C files, from Ben Hutchings
* Build fixes and documentation corrections from David Ahern
* Group parsing support, from Jiri Olsa
* UI/gtk refactorings and improvements from Namhyung Kim
* NULL deref fix for perf script, from Namhyung Kim
* Assorted cleanups from Robert Richter
* Let O= makes handle relative paths, from Steven Rostedt
* perf script python fixes, from Feng Tang.
* Improve 'perf lock' error message when the needed tracepoints
are not present, from David Ahern.
* Initial bash completion support, from Frederic Weisbecker
* Allow building without libelf, from Namhyung Kim.
* Support DWARF CFI based unwind to have callchains when %bp
based unwinding is not possible, from Jiri Olsa.
* Symbol resolution fixes, while fixing support PPC64 files with an .opt ELF
section was the end goal, several fixes for code that handles all
architectures and cleanups are included, from Cody Schafer.
* Add a description for the JIT interface, from Andi Kleen.
* Assorted fixes for Documentation and build in 32 bit, from Robert Richter
* Add support for non-tracepoint events in perf script python, from Feng Tang
* Cache the libtraceevent event_format associated to each evsel early, so that we
avoid relookups, i.e. calling pevent_find_event repeatedly when processing
tracepoint events.
[ This is to reduce the surface contact with libtraceevents and make clear what
is that the perf tools needs from that lib: so far parsing the common and per
event fields. ]
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
