Re: [PATCH -v3] perf, x86: try to handle unknown nmis with running perfctrs

From: Cyrill Gorcunov <gorcunov@gmail.com>
To: Don Zickus <dzickus@redhat.com>
Cc: Robert Richter <robert.richter@amd.com>,
	Ingo Molnar <mingo@elte.hu>,
	Peter Zijlstra <peterz@infradead.org>,
	Lin Ming <ming.m.lin@intel.com>,
	"fweisbec@gmail.com" <fweisbec@gmail.com>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"Huang, Ying" <ying.huang@intel.com>,
	Yinghai Lu <yinghai@kernel.org>, Andi Kleen <andi@firstfloor.org>
Subject: Re: [PATCH -v3] perf, x86: try to handle unknown nmis with running perfctrs
Date: Thu, 26 Aug 2010 19:34:04 +0400	[thread overview]
Message-ID: <20100826153404.GA6306@lenovo> (raw)
In-Reply-To: <20100826152246.GN4879@redhat.com>

On Thu, Aug 26, 2010 at 11:22:46AM -0400, Don Zickus wrote:
> On Thu, Aug 26, 2010 at 01:18:29PM +0400, Cyrill Gorcunov wrote:
> > On Thu, Aug 26, 2010 at 1:00 PM, Robert Richter <robert.richter@amd.com> wrote:
> > ...
> > >
> > > This could also be a race in the counter handling code, or we do not
> > > proper count the number of handled counters. Maybe 2 counters actually
> > > fired but we only noticed one counter and then accidentially cleared
> > > the 2nd without processing it.
> > >
> > > -Robert
> > >
> > 
> > Any chance to get it tested on P4 machine since it has a bit
> > different design?
> 
> Hmm, I take that back.  I guess I can reproduce this on my i5 that I had
> using Ingo's config.
> 
> Working on Robert's assumption, I added code to perf_event_intel.c that
> said if handled !=0 just add one to it (IOW always process handled as 0 or
>  something >1).  That seems to working good and catches the nmis that Ingo
> was seeing.
> 
> I'll keep looking for the race condition to better fix it.
> 
> Cheers,
> Don
> 

Sounds promising, mind to post new inter-diff? Ie what you have
changed from Robert's patch.

	-- Cyrill