Enable the 64bit MCE_INTEL code (CMCI, thermal interrupts) for 32bit NEW_MCE.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The 64bit machine check code is in many ways much better than
the 32bit machine check code: it is more specification compliant,
is cleaner, only has a single code base versus one per CPU,
has better infrastructure for recovery, has a cleaner way to communicate
with user space etc. etc.
Use the 64bit code for 32bit too.
This is the second attempt to do this. There was one a couple of years
ago to unify this code for 32bit and 64bit. Back then this ran into some
trouble with K7s and was reverted.
I believe this time the K7 problems (and some others) are addressed.
I went over the old handlers and was very careful to retain
all quirks.
But of course this needs a lot of testing on old systems. On newer
64bit capable systems I don't expect much problems because they have been
already tested with the 64bit kernel.
I made this a CONFIG for now that still allows to select the old
machine check code. This is mostly to make testing easier,
if someone runs into a problem we can ask them to try
with the CONFIG switched.
The new code is default y for more coverage.
Once there is confidence the 64bit code works well on older hardware
too the CONFIG_X86_OLD_MCE and the associated code can be easily
removed.
This causes a behaviour change for 32bit installations. They now
have to install the mcelog package to be able to log
corrected machine checks.
The 64bit machine check code only handles CPUs which support the
standard Intel machine check architecture described in the IA32 SDM.
The 32bit code has special support for some older CPUs which
have non standard machine check architectures, in particular
WinChip C3 and Intel P5. I made those a separate CONFIG option
and kept them for now. The WinChip variant could be probably
removed without too much pain, it doesn't really do anything
interesting. P5 is also disabled by default (like it
was before) because many motherboards have it miswired, but
according to Alan Cox a few embedded setups use that one.
Forward ported/heavily changed version of old patch, original patch
included review/fixes from Thomas Gleixner, Bert Wesarg.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
First 32bit doesn't have oops_begin, so it's a barrier of using
this code on 32bit.
On closer examination it turns out oops_begin is not
a good idea in a machine check panic anyways. All oops_begin
does it so check for recursive/parallel oopses and implement the
"wait on oops" heuristic. But there's actually no good reason
to lock machine checks against oopses or prevent them
from recursion. Also "wait on oops" does not really make
sense for a machine check too.
Replace it with a manual bust_spinlocks/console_verbose.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
i386 has no idle notifiers, but the 64bit machine check
code uses them to wake up mcelog from a fatal machine check
exception.
For corrected machine checks found by the poller or
threshold interrupts going through an idle notifier is not needed
because the wake_up can is just done directly and doesn't
need the idle notifier. It is only needed for logging
exceptions.
To be honest I never liked the idle notifier even though I signed
off on it. On closer investigation the code actually turned out
to be nearly. Right now machine check exceptions on x86 are always
unrecoverable (lead to panic due to PCC), which means we never execute
the idle notifier path.
The only exception is the somewhat weird tolerant==3 case, which
ignores PCC. I'll fix this in a future patch in a much cleaner way.
So remove the "mcelog wakeup through idle notifier" code
from 64bit.
This allows to compile the 64bit machine check handler on 32bit
which doesn't have idle notifiers.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
It's the same function, so let's share it.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Give it the same name as on 32bit. This makes further merging easier.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Allows to call different machine check handlers from the low
level machine check entry vector.
This is needed for later when it will be used for 32bit too.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Various K7 have broken bank 0s. Don't enable it by default
Port from the 32bit code.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Quoting the comment:
* SDM documents that on family 6 bank 0 should not be written
* because it aliases to another special BIOS controlled
* register.
* But it's not aliased anymore on model 0x1a+
* Don't ignore bank 0 completely because there could be a valid
* event later, merely don't write CTL0.
This is mostly a port on the 32bit code, except that 32bit
always didn't write it and didn't have the 0x1a heuristic. I checked
with the CPU designers that the quirk is not required starting with
this model.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Replace unsigned long with u64s if they need to contain 64bit values.
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Decode more magic constants and turn them into symbols.
[ Sort definitions bitwise, introduce MCG_EXT_CNT - HS ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Decode magic constants and turn them into symbols.
[ Cleanup to use symbols already exists - HS ]
[ Impact: cleanup ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
The number of MCE banks supported by a CPU is a useful number to know,
so print it out during CPU initialization.
[ Impact: add printout ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
move mce_64.c => mce.c and glue it up in the Makefile.
Remove mce_32.c
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Prepare the 64-bit mce_64.c code side to be built on 32-bit.
[ includes ifdef relocation by Andi Kleen ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Cc: Andi Kleen <ak@firstfloor.org>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Prepare mce.h for unification, so that it will build on 32-bit x86
kernels too.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Prepare for unification, make two intel_init_thermal equal.
[ Impact: cleanup ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Make the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
This file has been modified many times along the years, by multiple
authors, so the general style and structure has diverged in a number
of areas making this file hard to read.
So fix the coding style match that of the rest of the x86 arch code.
[ Impact: cleanup ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
- Fix the comment formatting.
- The error path does not return 0, and printk lacks level and "\n".
- Move __setup("nomce") next to mcheck_disable().
- Improve readability etc.
[ Impact: cleanup ]
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Acked-by: Andi Kleen <ak@linux.intel.com>
LKML-Reference: <49CB3F38.7090703@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
* 'x86-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
x86: avoid back to back on_each_cpu in cpa_flush_array
x86, relocs: ignore R_386_NONE in kernel relocation entries
Cleanup cpa_flush_array() to avoid back to back on_each_cpu() calls.
[ Impact: optimizes fix 0af48f42df ]
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Slightly modified by trenn@suse.de -> only do this on fam 10h and fam 11h.
Currently powernow-k8 determines CPU frequency from ACPI PSS objects, but
according to AMD family 11h BKDG this frequency is just a rounded value:
"CoreFreq (MHz) = The CPU COF specified by MSRC001_00[6B:64][CpuFid]
rounded to the nearest 100 Mhz."
As a consequnce powernow-k8 reports wrong CPU frequency on some systems,
e.g. on Turion X2 Ultra:
powernow-k8: Found 1 AMD Turion(tm)X2 Ultra DualCore Mobile ZM-82
processors (2 cpu cores) (version 2.20.00)
powernow-k8: 0 : pstate 0 (2200 MHz)
powernow-k8: 1 : pstate 1 (1100 MHz)
powernow-k8: 2 : pstate 2 (600 MHz)
But this is wrong as frequency for Pstate2 is 550 MHz. x86info reports it
correctly:
#x86info -a |grep Pstate
...
Pstate-0: fid=e, did=0, vid=24 (2200MHz)
Pstate-1: fid=e, did=1, vid=30 (1100MHz)
Pstate-2: fid=e, did=2, vid=3c (550MHz) (current)
Solution is to determine the frequency directly from Pstate MSRs instead
of using rounded values from ACPI table.
Signed-off-by: Andreas Herrmann <andreas.herrmann3@amd.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Dave Jones <davej@redhat.com>
- Make the message shorter and easier to grep for
- Use printk_once instead of WARN_ONCE (functionality of these was mixed)
Signed-off-by: Thomas Renninger <trenn@suse.de>
Cc: Langsdorf, Mark <mark.langsdorf@amd.com>
Signed-off-by: Dave Jones <davej@redhat.com>
arch/x86/kernel/cpu/cpufreq/powernow-k7.c:172: warning: 'invalidate_entry' defined but not used
Reported-by: Toralf Förster <toralf.foerster@gmx.de>
Signed-off-by: Dave Jones <davej@redhat.com>
Some atom procs don't do freq scaling (such as the atom 330 on my own
littlefalls2 board). By adding the atom family here, we at least get
the benefit of passive cooling in a thermal emergency. Not sure how
to see that its actually helping any, but the driver does bind and
claim its functioning on my atom 330.
Signed-off-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Dave Jones <davej@redhat.com>
For relocatable 32bit kernels, boot/compressed/relocs.c processes
relocation entries in the kernel image and appends it to the kernel
image such that boot/compressed/head_32.S can relocate the kernel.
The kernel image is one statically linked object and only uses two
relocation types - R_386_PC32 and R_386_32, of the two only the latter
needs massaging during kernel relocation and thus handled by relocs.
R_386_PC32 is ignored and all other relocation types are considered
error.
When the target of a relocation resides in a discarded section,
binutils doesn't throw away the relocation record but nullifies it by
changing it to R_386_NONE, which unfortunately makes relocs fail.
The problem was triggered by yet out-of-tree x86 stack unwind patches
but given the binutils behavior, ignoring R_386_NONE is the right
thing to do.
The problem has been tracked down to binutils behavior by Jan Beulich.
[ Impact: fix build with certain binutils by ignoring R_386_NONE ]
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Jan Beulich <JBeulich@novell.com>
Cc: Ingo Molnar <mingo@elte.hu>
LKML-Reference: <4A1B8150.40702@kernel.org>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
* 'kvm-updates/2.6.30' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: Fix PDPTR reloading on CR4 writes
KVM: Make paravirt tlb flush also reload the PAE PDPTRs
The processor is documented to reload the PDPTRs while in PAE mode if any
of the CR4 bits PSE, PGE, or PAE change. Linux relies on this
behaviour when zapping the low mappings of PAE kernels during boot.
The code already handled changes to CR4.PAE; augment it to also notice changes
to PSE and PGE.
This triggered while booting an F11 PAE kernel; the futex initialization code
runs before any CR3 reloads and writes to a NULL pointer; the futex subsystem
ended up uninitialized, killing PI futexes and pulseaudio which uses them.
Cc: stable@kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>
The paravirt tlb flush may be used not only to flush TLBs, but also
to reload the four page-directory-pointer-table entries, as it is used
as a replacement for reloading CR3. Change the code to do the entire
CR3 reloading dance instead of simply flushing the TLB.
Cc: stable@kernel.org
Signed-off-by: Avi Kivity <avi@redhat.com>
Remap percpu allocator has subtle bug when combined with page
attribute changing. Remap percpu allocator aliases PMD pages for the
first chunk and as pageattr doesn't know about the alias it ends up
updating page attributes of the original mapping thus leaving the
alises in inconsistent state which might lead to subtle data
corruption. Please read the following threads for more information:
http://thread.gmane.org/gmane.linux.kernel/835783
The following is the proposed fix which teaches pageattr about percpu
aliases.
http://thread.gmane.org/gmane.linux.kernel/837157
However, the above changes are deemed too pervasive for upstream
inclusion for 2.6.30 release, so this patch essentially disables
the remap allocator for the time being.
Signed-off-by: Tejun Heo <tj@kernel.org>
LKML-Reference: <4A1A0A27.4050301@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
cpa_flush_array seems to prefer wbinvd() over clflush at 4M threshold.
clflush needs to be done on only one CPU as per instruction definition.
wbinvd() however, should be done on all CPUs.
[ Impact: fix missing flush which could cause data corruption ]
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
wbinvd is supported on all CPUs 486 or later. But,
pageattr.c is checking x86_model >= 4 before wbinvd(), which looks like
an oversight bug. It was first introduced at one place by changeset
d7c8f21a8c and got copied over to second
place in the same file later.
[ Impact: fix missing cache flush on early-model CPUs, potential data corruption ]
Signed-off-by: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Introduce "noxsave" boot parameter which will disable the cpu's xsave/xrstor
capabilities. Useful for debugging and working around xsave related issues.
[ Impact: make it possible to debug problems in the field ]
Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Remove ACPI 3 E820 extended memory attributes support. At least one
vendor actively set all the flags to zero, but left ECX on return at
24. This bug may be present in other BIOSes.
The breakage functionally means the ACPI 3 flags are probably
completely useless, and that no OS any time soon is going to rely on
their existence. Therefore, drop support completely. We may want to
revisit this question in the future, if we find ourselves actually
needing the flags.
This reverts all or part of the following checkins:
cd670599b7c549e71d07
However, retain the part from the latter commit that copies e820 into
a temporary buffer; that is an unrelated BIOS workaround. Put in a
comment to explain that part.
See https://bugzilla.redhat.com/show_bug.cgi?id=499396 for some
additional information.
[ Impact: detect all memory on affected machines ]
Reported-by: Thomas J. Baker <tjb@unh.edu>
Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Acked-by: Len Brown <len.brown@intel.com>
Cc: Chuck Ebbert <cebbert@redhat.com>
Cc: Kyle McMartin <kmcmartin@redhat.com>
Cc: Matt Domsch <matt_domsch@dell.com>
x86: DMI match for the Sony VGN-Z540N as it needs BIOS reboot,
see:
http://bugzilla.kernel.org/show_bug.cgi?id=12901
[ Impact: fix hung reboot on certain systems ]
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Cc: Len Brown <lenb@kernel.org>
LKML-Reference: <1242963350.32574.53.camel@rzhang-dt>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
tracing: Append prompt in /debug/tracing/README file
x86/function-graph: fix constraint for recording old return value
Xiaohui Xin and some other folks at Intel have been looking into what's
behind the performance hit of paravirt_ops when running native.
It appears that the hit is entirely due to the paravirtualized
spinlocks introduced by:
| commit 8efcbab674
| Date: Mon Jul 7 12:07:51 2008 -0700
|
| paravirt: introduce a "lock-byte" spinlock implementation
The extra call/return in the spinlock path is somehow
causing an increase in the cycles/instruction of somewhere around 2-7%
(seems to vary quite a lot from test to test). The working theory is
that the CPU's pipeline is getting upset about the
call->call->locked-op->return->return, and seems to be failing to
speculate (though I haven't seen anything definitive about the precise
reasons). This doesn't entirely make sense, because the performance
hit is also visible on unlock and other operations which don't involve
locked instructions. But spinlock operations clearly swamp all the
other pvops operations, even though I can't imagine that they're
nearly as common (there's only a .05% increase in instructions
executed).
If I disable just the pv-spinlock calls, my tests show that pvops is
identical to non-pvops performance on native (my measurements show that
it is actually about .1% faster, but Xiaohui shows a .05% slowdown).
Summary of results, averaging 10 runs of the "mmperf" test, using a
no-pvops build as baseline:
nopv Pv-nospin Pv-spin
CPU cycles 100.00% 99.89% 102.18%
instructions 100.00% 100.10% 100.15%
CPI 100.00% 99.79% 102.03%
cache ref 100.00% 100.84% 100.28%
cache miss 100.00% 90.47% 88.56%
cache miss rate 100.00% 89.72% 88.31%
branches 100.00% 99.93% 100.04%
branch miss 100.00% 103.66% 107.72%
branch miss rt 100.00% 103.73% 107.67%
wallclock 100.00% 99.90% 102.20%
The clear effect here is that the 2% increase in CPI is
directly reflected in the final wallclock time.
(The other interesting effect is that the more ops are
out of line calls via pvops, the lower the cache access
and miss rates. Not too surprising, but it suggests that
the non-pvops kernel is over-inlined. On the flipside,
the branch misses go up correspondingly...)
So, what's the fix?
Paravirt patching turns all the pvops calls into direct calls, so
_spin_lock etc do end up having direct calls. For example, the compiler
generated code for paravirtualized _spin_lock is:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq *0xffffffff805a5b30
<_spin_lock+22>: retq
The indirect call will get patched to:
<_spin_lock+0>: mov %gs:0xb4c8,%rax
<_spin_lock+9>: incl 0xffffffffffffe044(%rax)
<_spin_lock+15>: callq <__ticket_spin_lock>
<_spin_lock+20>: nop; nop /* or whatever 2-byte nop */
<_spin_lock+22>: retq
One possibility is to inline _spin_lock, etc, when building an
optimised kernel (ie, when there's no spinlock/preempt
instrumentation/debugging enabled). That will remove the outer
call/return pair, returning the instruction stream to a single
call/return, which will presumably execute the same as the non-pvops
case. The downsides arel 1) it will replicate the
preempt_disable/enable code at eack lock/unlock callsite; this code is
fairly small, but not nothing; and 2) the spinlock definitions are
already a very heavily tangled mass of #ifdefs and other preprocessor
magic, and making any changes will be non-trivial.
The other obvious answer is to disable pv-spinlocks. Making them a
separate config option is fairly easy, and it would be trivial to
enable them only when Xen is enabled (as the only non-default user).
But it doesn't really address the common case of a distro build which
is going to have Xen support enabled, and leaves the open question of
whether the native performance cost of pv-spinlocks is worth the
performance improvement on a loaded Xen system (10% saving of overall
system CPU when guests block rather than spin). Still it is a
reasonable short-term workaround.
[ Impact: fix pvops performance regression when running native ]
Analysed-by: "Xin Xiaohui" <xiaohui.xin@intel.com>
Analysed-by: "Li Xin" <xin.li@intel.com>
Analysed-by: "Nakajima Jun" <jun.nakajima@intel.com>
Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Xen-devel <xen-devel@lists.xensource.com>
LKML-Reference: <4A0B62F7.5030802@goop.org>
[ fixed the help text ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
