]> git.baikalelectronics.ru Git - kernel.git/commit
powerpc: perf: Fix frequency calculation for overflowing counters
authorAnton Blanchard <anton@samba.org>
Mon, 17 Jan 2011 05:17:42 +0000 (16:17 +1100)
committerIngo Molnar <mingo@elte.hu>
Mon, 17 Jan 2011 10:43:02 +0000 (11:43 +0100)
commit65b76604fcf013198800442067d28514f78639e9
tree3a55c96dfb709415ee2c4e54e242c4f1a7cd01e2
parent804a21ced561638ea2a6d158ebc8ec0fd5b99ffa
powerpc: perf: Fix frequency calculation for overflowing counters

When profiling a benchmark that is almost 100% userspace, I noticed some wildly
inaccurate profiles that showed almost all time spent in the kernel.

Closer examination shows we were programming a tiny number of cycles into the
PMU after each overflow (about ~200 away from the next overflow). This gets us
stuck in a loop which we eventually break out of by throttling the PMU (there
are regular throttle/unthrottle events in the log).

It looks like we aren't setting event->hw.last_period to something same and the
frequency to period calculations in perf are going haywire.

With the following patch we find the correct period after a few interrupts and
stay there. I also see no more throttle events.

Signed-off-by: Anton Blanchard <anton@samba.org>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@ghostprotocols.net>
LKML-Reference: <20110117161742.5feb3761@kryten>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/powerpc/kernel/perf_event.c