Re: [regression 2.6.39-rc2][bisected] "perf, x86: P4 PMU - Read proper MSR register to catch" and NMIs

[Cyrill Gorcunov <gorcunov@openvz.org>]

On 04/07/2011 06:38 PM, Shaun Ruffell wrote:
> On Thu, Apr 07, 2011 at 07:18:50AM +0400, Cyrill Gorcunov wrote:
>> On Thursday, April 7, 2011, Don Zickus <dzickus@redhat.com> wrote:
>>> On Wed, Apr 06, 2011 at 05:30:36PM -0500, Shaun Ruffell wrote:
>>>>
>>>> I was able to bisect it down to commit 242214f9c1eeaae40, but I'm not
>>>> certain where to go from here.  Is this something that is already known
>>>> or is there more information I should try to collect?
>>>
>>> Nope, this is an ongoing issue.  What happened was the perf P4 nmi handler
>>> was swallowing all the NMIs.  My patch fixed that and exposed a double NMI
>>> problem.  We have been chasing it for a couple of months.  I think Cyril
>>> was finally able to duplicate it (as he wrote the P4 code).  I have
>>> confidence that he will find a fix for it soon. :-)
>>>
>>> Thanks for the report though!
>>
>> Hi, yeah, i got it too and i hope to fix this issue soon. Will ping as
>> only get working fix.
> 
> Don, Cyrill,
> 
> Thanks for the explanation and my apologies for not relating the
> previous discussions about this to what I was seeing. This issue would
> be a blocker for any 2.6.39 final right?

Well, could be, I didn't find the real reason for doubled nmi.
Still investigating. As a workaround you could simply disable
nmi-watchdog for a while in command line if it bothers you.

> 
> Cyrill, I would be more than happy to test any patches. It's relatively
> quick for me to reproduce.

OK, I'll prepare some patches for you to test. Still think on where it
could fails :( Stay tuned.

> 
> Thanks,
> Shaun

-- 
    Cyrill