Re: [PATCH] perf/x86/intel: ignore CondChgd bit to avoid false NMI handling

[Peter Zijlstra <peterz@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1465 bytes --]

On Thu, Jun 12, 2014 at 04:00:11PM +0900, HATAYAMA Daisuke wrote:
> Also, I checked cpuid on the system with Neharlem processor where I
> have never seen CondChg bit is set.
> 
> [root@localhost ~]# ./cpuid -r
> CPU 0:
>    0x00000000 0x00: eax=0x0000000b ebx=0x756e6547 ecx=0x6c65746e edx=0x49656e69
>    0x00000001 0x00: eax=0x000206e6 ebx=0x40200800 ecx=0x00bce3bd edx=0xbfebfbff
> <snip>
>    0x0000000a 0x00: eax=0x07300403 ebx=0x00000044 ecx=0x00000000 edx=0x00000603
>                     ^^^^^^^^^^^^^^
> So, cpuid tells that CondChg bit is supported on this processor, too.

Yeah, I can't remember ever seeing that bit on nhm/wsm either. Weird
stuff that.

> > In any case, the proposed patch seems fine, just needs a better
> > changelog.
> > 
> 
> I see.
> 
> I'll write that the problem is that any NMI could be robbed by NMI
> watchdog explicitly. Now only patch title says this explicitly. This
> is your first comment.

Yeah, since that is the actual problem, its good to be clear on that.

> About CondChgd bit, I cannot write more than I see on actual
> system. If it's necessary to describe more about CondChgd bit, it
> would be appreciated if someone tell me more information about it.

I think we've found all 2 sentences the SDM has about that and unless
someone from Intel is going to come and explain why they wasted precious
silicon on this I suppose it will remain a mystery. No need to update on
that.

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]