Re: [PATCH] x86, UV: Fix NMI handler for UV platforms

[Don Zickus <dzickus@redhat.com>]

On Wed, Mar 23, 2011 at 11:45:20PM +0300, Cyrill Gorcunov wrote:
> On 03/23/2011 11:00 PM, Don Zickus wrote:
> > On Wed, Mar 23, 2011 at 01:53:20PM -0400, Don Zickus wrote:
> >> Let me know if the patch fixes that problem.  Then it will be one less
> >> thing to worry about. :-)
> > 
> > Ok, I was an idiot and made the patch against RHEL-6.  Here is the one
> > against 2.6.38.  Sorry about that.
> > 
> > Cheers,
> > Don
> > 
> > 
> > diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
> > index 87eab4a..62ec8e9 100644
> > --- a/arch/x86/kernel/cpu/perf_event.c
> > +++ b/arch/x86/kernel/cpu/perf_event.c
> > @@ -1375,7 +1375,7 @@ perf_event_nmi_handler(struct notifier_block *self,
> >  	if ((handled > 1) ||
> >  		/* the next nmi could be a back-to-back nmi */
> >  	    ((__this_cpu_read(pmu_nmi.marked) == this_nmi) &&
> > -	     (__this_cpu_read(pmu_nmi.handled) > 1))) {
> > +	     (__this_cpu_read(pmu_nmi.handled) > 0) && handled && this_nmi)) {
> 
> Don, why do you need to check for this_nmi here? it's zero for first nmi in a
> system (right?), so I fail to get the reason for such check. What I miss?

It was a stupid optimization, otherwise it _always_ traverses on the
first nmi.  I wasn't sure that is what I wanted.  Mainly I was trying to
wrap my head around the problem.  You can remove it to see if the problem
is still fixed.

I'm not a fan of this fix as it is getting a little ugly, but for now...

Cheers,
Don