2012-02-16 04:48:44

by Anton Blanchard

[permalink] [raw]
Subject: [PATCH] powerpc: perf: power_pmu_start restores incorrect values, breaking frequency events


perf on POWER stopped working after commit e050e3f0a71b (perf: Fix
broken interrupt rate throttling). That patch exposed a bug in
the POWER perf_events code.

Since the PMCs count upwards and take an exception when the top bit
is set, we want to write 0x80000000 - left in power_pmu_start. We were
instead programming in left which effectively disables the counter
until we eventually hit 0x80000000. This could take seconds or longer.

With the patch applied I get the expected number of samples:

# taskset -c 0 yes > /dev/null &
# perf record -C 0 -a sleep 10
# perf report -D | grep SAMPLE | tail -1
SAMPLE events: 9948

Signed-off-by: Anton Blanchard <[email protected]>
---
Cc: <[email protected]>

Index: linux-build/arch/powerpc/kernel/perf_event.c
===================================================================
--- linux-build.orig/arch/powerpc/kernel/perf_event.c 2012-02-16 15:07:57.465384699 +1100
+++ linux-build/arch/powerpc/kernel/perf_event.c 2012-02-16 15:11:48.449579581 +1100
@@ -865,6 +865,7 @@ static void power_pmu_start(struct perf_
{
unsigned long flags;
s64 left;
+ unsigned long val;

if (!event->hw.idx || !event->hw.sample_period)
return;
@@ -880,7 +881,12 @@ static void power_pmu_start(struct perf_

event->hw.state = 0;
left = local64_read(&event->hw.period_left);
- write_pmc(event->hw.idx, left);
+
+ val = 0;
+ if (left < 0x80000000L)
+ val = 0x80000000L - left;
+
+ write_pmc(event->hw.idx, val);

perf_event_update_userpage(event);
perf_pmu_enable(event->pmu);


2012-02-16 04:58:27

by Paul Mackerras

[permalink] [raw]
Subject: Re: [PATCH] powerpc: perf: power_pmu_start restores incorrect values, breaking frequency events

On Thu, Feb 16, 2012 at 03:48:22PM +1100, Anton Blanchard wrote:
>
> perf on POWER stopped working after commit e050e3f0a71b (perf: Fix
> broken interrupt rate throttling). That patch exposed a bug in
> the POWER perf_events code.
>
> Since the PMCs count upwards and take an exception when the top bit
> is set, we want to write 0x80000000 - left in power_pmu_start. We were
> instead programming in left which effectively disables the counter
> until we eventually hit 0x80000000. This could take seconds or longer.
>
> With the patch applied I get the expected number of samples:
>
> # taskset -c 0 yes > /dev/null &
> # perf record -C 0 -a sleep 10
> # perf report -D | grep SAMPLE | tail -1
> SAMPLE events: 9948
>
> Signed-off-by: Anton Blanchard <[email protected]>

Acked-by: Paul Mackerras <[email protected]>

2012-02-19 11:13:21

by Stephane Eranian

[permalink] [raw]
Subject: Re: [PATCH] powerpc: perf: power_pmu_start restores incorrect values, breaking frequency events

Glad to see you fixed the PPC problem.

On Thu, Feb 16, 2012 at 5:57 AM, Paul Mackerras <[email protected]> wrote:
> On Thu, Feb 16, 2012 at 03:48:22PM +1100, Anton Blanchard wrote:
>>
>> perf on POWER stopped working after commit e050e3f0a71b (perf: Fix
>> broken interrupt rate throttling). That patch exposed a bug in
>> the POWER perf_events code.
>>
>> Since the PMCs count upwards and take an exception when the top bit
>> is set, we want to write 0x80000000 - left in power_pmu_start. We were
>> instead programming in left which effectively disables the counter
>> until we eventually hit 0x80000000. This could take seconds or longer.
>>
>> With the patch applied I get the expected number of samples:
>>
>> # taskset -c 0 yes > /dev/null &
>> # perf record -C 0 -a sleep 10
>> # perf report -D | grep SAMPLE | tail -1
>>           SAMPLE events:       9948
>>
>> Signed-off-by: Anton Blanchard <[email protected]>
>
> Acked-by: Paul Mackerras <[email protected]>