perf: Per PMU context reschedule and misc

[PATCH 0/5] perf: Per PMU context reschedule and misc

Posted by Peter Zijlstra 1 year, 4 months ago

Hi,

This is 'fallout' from Namhyung posting his per-pmu ctx_resched() patches. It
started with me trying to clean up and get rid of corner cases, and then got
involved when Kan noted the time keeping issue.

Anyway, please review / test.

Re: [PATCH 0/5] perf: Per PMU context reschedule and misc

Posted by Liang, Kan 1 year, 4 months ago


On 2024-08-07 7:29 a.m., Peter Zijlstra wrote:
> Hi,
> 
> This is 'fallout' from Namhyung posting his per-pmu ctx_resched() patches. It
> started with me trying to clean up and get rid of corner cases, and then got
> involved when Kan noted the time keeping issue.
> 
> Anyway, please review / test.
> 

Except for the tiny nit, the patch series looks good to me.

Reviewed-by: Kan Liang <kan.liang@linux.intel.com>

Thanks,
Kan

Re: [PATCH 0/5] perf: Per PMU context reschedule and misc

Posted by Namhyung Kim 1 year, 4 months ago

Hi Peter,

On Wed, Aug 7, 2024 at 4:56 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> Hi,
>
> This is 'fallout' from Namhyung posting his per-pmu ctx_resched() patches. It
> started with me trying to clean up and get rid of corner cases, and then got
> involved when Kan noted the time keeping issue.
>
> Anyway, please review / test.

It works blazingly fast!

  # ./stress-pmu
  delta: 0.000307 sec (3 usec/op)

I found a problem with my patch that it called __pmu_ctx_sched_out() for
nothing (I guess is_active only has EVENT_TIME).  I thought ctx_sched_out()
would stop if it doesn't change EVENT_ALL but it iterated all PMUs anyway.

But with this change we don't need ctx_sched_out(EVENT_TIME) anymore.

Reviewed-by: Namhyung Kim <namhyung@kernel.org>

Thanks,
Namhyung