Messages in this thread | | | Date | Thu, 2 Sep 2010 15:11:12 +0200 | From | Robert Richter <> | Subject | Re: [PATCH 4/4] [x86] perf: fix accidentally ack'ing a second event on intel perf counter |
| |
On 02.09.10 04:13:19, Stephane Eranian wrote: > Robert, > > Do you have the test program you used to test this? > I believe the NHM hack does not solve the problem, it > just makes it harder to appear.
For testing back-to-back nmis I have used:
perf record -e cycles -e instructions -e cache-references -e cache-misses -e branch-misses -a -- sleep 10
with load on all cpus. But I couldn't reproduce this particular problem as I do not have such a system available. I think it might trigger also with only one counter running. What the observed from the status bits, only one counter was involved.
> > I suspect the real issue is that the GLOBAL_STATUS > bitmask cannot be trusted. I'd like to verify this.
So yes, it looks like it is a cpu bug with a race then clearing the status. I didn't check the errata list, maybe it is already known.
> > Has the problem appear only on Nehalem or also on > Westmere?
I don't know.
-Robert
-- Advanced Micro Devices, Inc. Operating System Research Center
| |