Re: [GIT PULL] One more vdso fix for 3.19

From: Andy Lutomirski
Date: Fri Jan 02 2015 - 13:19:46 EST


On Thu, Jan 1, 2015 at 1:23 PM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> * Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>
>> Hi Ingo et al,
>>
>> Please consider pulling for x86/urgent.
>>
>> Thanks,
>> Andy
>>
>>
>> The following changes since commit 394f56fe480140877304d342dec46d50dc823d46:
>>
>> x86_64, vdso: Fix the vdso address randomization algorithm
>> (2014-12-20 16:56:57 -0800)
>>
>> are available in the git repository at:
>>
>> git://git.kernel.org/pub/scm/linux/kernel/git/luto/linux.git
>> tags/pr-20141223-x86-vdso
>>
>> for you to fetch changes up to 1ddf0b1b11aa8a90cef6706e935fc31c75c406ba:
>>
>> x86, vdso: Use asm volatile in __getcpu (2014-12-23 13:05:30 -0800)
>>
>> ----------------------------------------------------------------
>> This is hopefully the last vdso fix for 3.19. It should be very
>> safe (it just adds a volatile).
>>
>> I don't think it fixes an actual bug (the __getcpu calls in the
>> pvclock code may not have been needed in the first place), but
>> discussion on that point is ongoing.
>>
>> It also fixes a big performance issue in 3.18 and earlier in which
>> the lsl instructions in vclock_gettime got hoisted so far up the
>> function that they happened even when the function they were in was
>> never called. n 3.19, the performance issue seems to be gone due to
>> the whims of my compiler and some interaction with a branch that's
>> now gone.
>>
>> I'll hopefully have a much bigger overhaul of the pvclock code
>> for 3.20, but it needs careful review.
>>
>> ----------------------------------------------------------------
>> Andy Lutomirski (1):
>> x86, vdso: Use asm volatile in __getcpu
>>
>> arch/x86/include/asm/vgtod.h | 6 ++++--
>> 1 file changed, 4 insertions(+), 2 deletions(-)
>
> Pulled into tip:x86/urgent, thanks Andy!
>
> I've attached the full commit below - I'm not sure it was sent to
> lkml.
>

It's here:

http://lkml.kernel.org/g/cover.1419295081.git.luto@xxxxxxxxxxxxxx

It was part of an "RFC" series, so it may have slipped under the
rader. This patch is Obviously Correct (tm) and was acked even in RFC
form, so I split it out. The rest needs much more careful review from
the paravirt people and might be ready in time for 3.20.

--Andy

> Thanks,
>
> Ingo
>
> ====================>
> From 1ddf0b1b11aa8a90cef6706e935fc31c75c406ba Mon Sep 17 00:00:00 2001
> From: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
> Date: Sun, 21 Dec 2014 08:57:46 -0800
> Subject: [PATCH] x86, vdso: Use asm volatile in __getcpu
>
> In Linux 3.18 and below, GCC hoists the lsl instructions in the
> pvclock code all the way to the beginning of __vdso_clock_gettime,
> slowing the non-paravirt case significantly. For unknown reasons,
> presumably related to the removal of a branch, the performance issue
> is gone as of
>
> e76b027e6408 x86,vdso: Use LSL unconditionally for vgetcpu
>
> but I don't trust GCC enough to expect the problem to stay fixed.
>
> There should be no correctness issue, because the __getcpu calls in
> __vdso_vlock_gettime were never necessary in the first place.
>
> Note to stable maintainers: In 3.18 and below, depending on
> configuration, gcc 4.9.2 generates code like this:
>
> 9c3: 44 0f 03 e8 lsl %ax,%r13d
> 9c7: 45 89 eb mov %r13d,%r11d
> 9ca: 0f 03 d8 lsl %ax,%ebx
>
> This patch won't apply as is to any released kernel, but I'll send a
> trivial backported version if needed.
>
> Fixes: 51c19b4f5927 x86: vdso: pvclock gettime support
> Cc: stable@xxxxxxxxxxxxxxx # 3.8+
> Cc: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
> Acked-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
> ---
> arch/x86/include/asm/vgtod.h | 6 ++++--
> 1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/include/asm/vgtod.h b/arch/x86/include/asm/vgtod.h
> index e7e9682a33e9..f556c4843aa1 100644
> --- a/arch/x86/include/asm/vgtod.h
> +++ b/arch/x86/include/asm/vgtod.h
> @@ -80,9 +80,11 @@ static inline unsigned int __getcpu(void)
>
> /*
> * Load per CPU data from GDT. LSL is faster than RDTSCP and
> - * works on all CPUs.
> + * works on all CPUs. This is volatile so that it orders
> + * correctly wrt barrier() and to keep gcc from cleverly
> + * hoisting it out of the calling function.
> */
> - asm("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
> + asm volatile ("lsl %1,%0" : "=r" (p) : "r" (__PER_CPU_SEG));
>
> return p;
> }



--
Andy Lutomirski
AMA Capital Management, LLC
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/