Messages in this thread | | | Date | Wed, 4 Aug 2010 11:45:52 -0400 | From | Don Zickus <> | Subject | Re: A question of perf NMI handler |
| |
On Wed, Aug 04, 2010 at 05:02:41PM +0200, Peter Zijlstra wrote: > So then the first interrupt will see 3+ overflows, return 3+, and will > thus eat 2+ NMIs, only one of which will be the pending interrupt, > leaving 1+ NMIs from other sources to consume unhandled. > > In which case Yinghai will have to press his NMI button 2+ times before > it registers. > > That said, that might be a better situation than always consuming > unknown NMIs..
Well if the worse case scenario is only one extra NMI, then I can change the logic in my patch to eat a max of 1 possible NMI instead of 2 as in the example you gave above.
It still won't be 100% accurate but how often are people running perf where they need 4 counters, have to hit an external nmi button or run into broken firmware all at the same time? :-)
Cheers, Don
| |