Messages in this thread | | | Subject | Re: [PATCH -v3] perf, x86: try to handle unknown nmis with running perfctrs | From | Peter Zijlstra <> | Date | Fri, 20 Aug 2010 10:50:16 +0200 |
| |
On Thu, 2010-08-19 at 17:58 -0400, Don Zickus wrote: > On Thu, Aug 19, 2010 at 04:27:13PM +0200, Peter Zijlstra wrote: > > x86_pmu.drain_pebs(regs); > > + } > > > > for_each_set_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) { > > struct perf_event *event = cpuc->events[bit]; > > > > + handled++; > > + > > if (!test_bit(bit, cpuc->active_mask)) > > continue; > > Sorry I didn't notice this earlier, but I think you want that handled++ > after the if(..) continue pieces. Otherwise you will always have a number > >1. :-)
Only if there's any remaining set bits in the overflow status reg, right?
But if we want to be paranoid, we should also check if the event that generated the overflow actually had the INT flag enabled, I guess ;-)
| |