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>
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>
Don't remove the __user annotation of the fpstate pointer, but
drop the superfluous void * cast instead.
This fixes the following sparse warnings:
xsave.c:135:15: warning: cast removes address space of expression
xsave.c:135:15: warning: incorrect type in argument 1 (different address spaces)
xsave.c:135:15: expected void const volatile [noderef] <asn:1>*<noident>
[...]
Signed-off-by: Mathias Krause <minipli@googlemail.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1346621506-30857-6-git-send-email-minipli@googlemail.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Use a more current logging style:
- Bare printks should have a KERN_<LEVEL> for consistency's sake
- Add pr_fmt where appropriate
- Neaten some macro definitions
- Convert some Ok output to OK
- Use "%s: ", __func__ in pr_fmt for summit
- Convert some printks to pr_<level>
Message output is not identical in all cases.
Signed-off-by: Joe Perches <joe@perches.com>
Cc: levinsasha928@gmail.com
Link: http://lkml.kernel.org/r/1337655007.24226.10.camel@joe2Laptop
[ merged two similar patches, tidied up the changelog ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Code paths like fork(), exit() and signal handling flush the fpu
state explicitly to the structures in memory.
BUG_ON() in __sanitize_i387_state() is checking that the fpu state
is not live any more. But for preempt kernels, task can be scheduled
out and in at any place and the preload_fpu logic during context switch
can make the fpu registers live again.
For example, consider a 64-bit Task which uses fpu frequently and as such
you will find its fpu_counter mostly non-zero. During its time slice, kernel
used fpu by doing kernel_fpu_begin/kernel_fpu_end(). After this, in the same
scheduling slice, task-A got a signal to handle. Then during the signal
setup path we got preempted when we are just before the sanitize_i387_state()
in arch/x86/kernel/xsave.c:save_i387_xstate(). And when we come back we
will have the fpu registers live that can hit the bug_on.
Similarly during core dump, other threads can context-switch in and out
(because of spurious wakeups while waiting for the coredump to finish in
kernel/exit.c:exit_mm()) and the main thread dumping core can run into this
bug when it finds some other thread with its fpu registers live on some other cpu.
So remove the paranoid check for now, even though it caught a bug in the
multi-threaded core dump case (fixed in the previous patch).
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Link: http://lkml.kernel.org/r/1336692811-30576-3-git-send-email-suresh.b.siddha@intel.com
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
While various modules include <asm/i387.h> to get access to things we
actually *intend* for them to use, most of that header file was really
pretty low-level internal stuff that we really don't want to expose to
others.
So split the header file into two: the small exported interfaces remain
in <asm/i387.h>, while the internal definitions that are only used by
core architecture code are now in <asm/fpu-internal.h>.
The guiding principle for this was to expose functions that we export to
modules, and leave them in <asm/i387.h>, while stuff that is used by
task switching or was marked GPL-only is in <asm/fpu-internal.h>.
The fpu-internal.h file could be further split up too, especially since
arch/x86/kvm/ uses some of the remaining stuff for its module. But that
kvm usage should probably be abstracted out a bit, and at least now the
internal FPU accessor functions are much more contained. Even if it
isn't perhaps as contained as it _could_ be.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/alpine.LFD.2.02.1202211340330.5354@i5.linux-foundation.org
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
This moves the bit that indicates whether a thread has ownership of the
FPU from the TS_USEDFPU bit in thread_info->status to a word of its own
(called 'has_fpu') in task_struct->thread.has_fpu.
This fixes two independent bugs at the same time:
- changing 'thread_info->status' from the scheduler causes nasty
problems for the other users of that variable, since it is defined to
be thread-synchronous (that's what the "TS_" part of the naming was
supposed to indicate).
So perfectly valid code could (and did) do
ti->status |= TS_RESTORE_SIGMASK;
and the compiler was free to do that as separate load, or and store
instructions. Which can cause problems with preemption, since a task
switch could happen in between, and change the TS_USEDFPU bit. The
change to TS_USEDFPU would be overwritten by the final store.
In practice, this seldom happened, though, because the 'status' field
was seldom used more than once, so gcc would generally tend to
generate code that used a read-modify-write instruction and thus
happened to avoid this problem - RMW instructions are naturally low
fat and preemption-safe.
- On x86-32, the current_thread_info() pointer would, during interrupts
and softirqs, point to a *copy* of the real thread_info, because
x86-32 uses %esp to calculate the thread_info address, and thus the
separate irq (and softirq) stacks would cause these kinds of odd
thread_info copy aliases.
This is normally not a problem, since interrupts aren't supposed to
look at thread information anyway (what thread is running at
interrupt time really isn't very well-defined), but it confused the
heck out of irq_fpu_usable() and the code that tried to squirrel
away the FPU state.
(It also caused untold confusion for us poor kernel developers).
It also turns out that using 'task_struct' is actually much more natural
for most of the call sites that care about the FPU state, since they
tend to work with the task struct for other reasons anyway (ie
scheduling). And the FPU data that we are going to save/restore is
found there too.
Thanks to Arjan Van De Ven <arjan@linux.intel.com> for pointing us to
the %esp issue.
Cc: Arjan van de Ven <arjan@linux.intel.com>
Reported-and-tested-by: Raphael Prevost <raphael@buro.asia>
Acked-and-tested-by: Suresh Siddha <suresh.b.siddha@intel.com>
Tested-by: Peter Anvin <hpa@zytor.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This creates three helper functions that do the TS_USEDFPU accesses, and
makes everybody that used to do it by hand use those helpers instead.
In addition, there's a couple of helper functions for the "change both
CR0.TS and TS_USEDFPU at the same time" case, and the places that do
that together have been changed to use those. That means that we have
fewer random places that open-code this situation.
The intent is partly to clarify the code without actually changing any
semantics yet (since we clearly still have some hard to reproduce bug in
this area), but also to make it much easier to use another approach
entirely to caching the CR0.TS bit for software accesses.
Right now we use a bit in the thread-info 'status' variable (this patch
does not change that), but we might want to make it a full field of its
own or even make it a per-cpu variable.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 5b1cbac377 ("i387: make irq_fpu_usable() tests more robust")
added a sanity check to the #NM handler to verify that we never cause
the "Device Not Available" exception in kernel mode.
However, that check actually pinpointed a (fundamental) race where we do
cause that exception as part of the signal stack FPU state save/restore
code.
Because we use the floating point instructions themselves to save and
restore state directly from user mode, we cannot do that atomically with
testing the TS_USEDFPU bit: the user mode access itself may cause a page
fault, which causes a task switch, which saves and restores the FP/MMX
state from the kernel buffers.
This kind of "recursive" FP state save is fine per se, but it means that
when the signal stack save/restore gets restarted, it will now take the
'#NM' exception we originally tried to avoid. With preemption this can
happen even without the page fault - but because of the user access, we
cannot just disable preemption around the save/restore instruction.
There are various ways to solve this, including using the
"enable/disable_page_fault()" helpers to not allow page faults at all
during the sequence, and fall back to copying things by hand without the
use of the native FP state save/restore instructions.
However, the simplest thing to do is to just allow the #NM from kernel
space, but fix the race in setting and clearing CR0.TS that this all
exposed: the TS bit changes and the TS_USEDFPU bit absolutely have to be
atomic wrt scheduling, so while the actual state save/restore can be
interrupted and restarted, the act of actually clearing/setting CR0.TS
and the TS_USEDFPU bit together must not.
Instead of just adding random "preempt_disable/enable()" calls to what
is already excessively ugly code, this introduces some helper functions
that mostly mirror the "kernel_fpu_begin/end()" functionality, just for
the user state instead.
Those helper functions should probably eventually replace the other
ad-hoc CR0.TS and TS_USEDFPU tests too, but I'll need to think about it
some more: the task switching functionality in particular needs to
expose the difference between the 'prev' and 'next' threads, while the
new helper functions intentionally were written to only work with
'current'.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Alignment of alloc_bootmem() depends on the value of
L1_CACHE_SHIFT. What we need here, however, is 64 byte alignment. Use
alloc_bootmem_align() and explicitly specify the alignment instead.
This fixes a kernel boot crash reported by Jody when the cpu in .config
is set to MPENTIUMII but the kernel is booted on a xsave-capable CPU.
Reported-by: Jody Bruchon <jody@nctritech.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
LKML-Reference: <20101116212442.059967454@sbsiddha-MOBL3.sc.intel.com>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Cc: <stable@kernel.org>
* 'x86-xsave-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, xsave: Make xstate_enable_boot_cpu() __init, protect on CPU 0
x86, xsave: Add __init attribute to setup_xstate_features()
x86, xsave: Make init_xstate_buf static
x86, xsave: Check cpuid level for XSTATE_CPUID (0x0d)
x86, xsave: Introduce xstate enable functions
x86, xsave: Separate fpu and xsave initialization
x86, xsave: Move boot cpu initialization to xsave_init()
x86, xsave: 32/64 bit boot cpu check unification in initialization
x86, xsave: Do not include asm/i387.h in asm/xsave.h
x86, xsave: Use xsaveopt in context-switch path when supported
x86, xsave: Sync xsave memory layout with its header for user handling
x86, xsave: Track the offset, size of state in the xsave layout
* 'x86-cleanups-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Clean up arch/x86/kernel/cpu/mtrr/cleanup.c: use ";" not "," to terminate statements
* 'x86-vmware-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, vmware: Preset lpj values when on VMware.
* 'x86-mtrr-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, mtrr: Use stop machine context to rendezvous all the cpu's
* 'x86-apic-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86/apic/es7000_32: Remove unused variable
* 'x86-fpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: Avoid unnecessary __clear_user() and xrstor in signal handling
* 'x86-vdso-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86, vdso: Unmap vdso pages
xstate_enable_boot_cpu() is, as the name implies, only used on the
boot CPU; furthermore, it invokes alloc_bootmem(), which is __init;
hence it needs to be tagged __init rather than __cpuinit.
Furthermore, it is *not* safe in the long run to rely on CPU 0 only
coming online during the early boot -- at some point we're going to
support offlining (and re-onlining) the boot CPU, and at that point we
must not call xstate_enable_boot_cpu() again.
The code is a fair bit more obscure than one would like, because the
__ref overrides aren't quite powerful enough.
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Acked-by: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: Robert Richter <robert.richter@amd.com>
LKML-Reference: <4C476236.1020302@zytor.com>
H. Peter Anvin
Suresh Siddha
Mathias Krause
Joe Perches
Linus Torvalds
Lucas De Marchi
H. Peter Anvin
Robert Richter