2021-10-15 13:20:19

by Athira Rajeev

[permalink] [raw]
Subject: [V2] perf/core: Avoid calling perf_mux_hrtimer_restart multiple times when scheduling event groups

Perf uses multiplexing if there are more events to be scheduled than the
available counters. With multiplexing, event groups will be rotated at
specified interval of time which is set using "hrtimer". During event
scheduling, if any of the event group fails to schedule, multiplexing
gets enabled by setting "rotate_necessary" for that event context and
starting the hrtimer by calling "perf_mux_hrtimer_restart".

Patch adds an optimisation to avoid calling "perf_mux_hrtimer_restart"
multiple times if already rotate_necessary is set for that context.
Even though "perf_mux_hrtimer_restart" will just return if hrtimer is
already active, this could avoid the overhead of calling this function
multiple times if there are many event groups. Patch adds the check to
avoid calling perf_mux_hrtimer_restart() for each event group on
every schedule.

Signed-off-by: Athira Rajeev <[email protected]>
---
Changelog:
From v1 -> v2:
- Rebased on top of latest kernel source.

kernel/events/core.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index f23ca260307f..9e1a06646ad0 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3753,8 +3753,10 @@ static int merge_sched_in(struct perf_event *event, void *data)
perf_cgroup_event_disable(event, ctx);
perf_event_set_state(event, PERF_EVENT_STATE_ERROR);
} else {
- ctx->rotate_necessary = 1;
- perf_mux_hrtimer_restart(cpuctx);
+ if (!ctx->rotate_necessary) {
+ ctx->rotate_necessary = 1;
+ perf_mux_hrtimer_restart(cpuctx);
+ }
group_update_userpage(event);
}
}
--
2.27.0


2021-10-18 03:26:08

by Song Liu

[permalink] [raw]
Subject: Re: [V2] perf/core: Avoid calling perf_mux_hrtimer_restart multiple times when scheduling event groups



> On Oct 15, 2021, at 12:04 AM, Athira Rajeev <[email protected]> wrote:
>
> Perf uses multiplexing if there are more events to be scheduled than the
> available counters. With multiplexing, event groups will be rotated at
> specified interval of time which is set using "hrtimer". During event
> scheduling, if any of the event group fails to schedule, multiplexing
> gets enabled by setting "rotate_necessary" for that event context and
> starting the hrtimer by calling "perf_mux_hrtimer_restart".
>
> Patch adds an optimisation to avoid calling "perf_mux_hrtimer_restart"
> multiple times if already rotate_necessary is set for that context.
> Even though "perf_mux_hrtimer_restart" will just return if hrtimer is
> already active, this could avoid the overhead of calling this function
> multiple times if there are many event groups. Patch adds the check to
> avoid calling perf_mux_hrtimer_restart() for each event group on
> every schedule.
>
> Signed-off-by: Athira Rajeev <[email protected]>

Acked-by: Song Liu <[email protected]>

2021-10-18 11:09:41

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [V2] perf/core: Avoid calling perf_mux_hrtimer_restart multiple times when scheduling event groups

On Sat, Oct 16, 2021 at 01:08:04AM +0000, Song Liu wrote:
>
>
> > On Oct 15, 2021, at 12:04 AM, Athira Rajeev <[email protected]> wrote:
> >
> > Perf uses multiplexing if there are more events to be scheduled than the
> > available counters. With multiplexing, event groups will be rotated at
> > specified interval of time which is set using "hrtimer". During event
> > scheduling, if any of the event group fails to schedule, multiplexing
> > gets enabled by setting "rotate_necessary" for that event context and
> > starting the hrtimer by calling "perf_mux_hrtimer_restart".
> >
> > Patch adds an optimisation to avoid calling "perf_mux_hrtimer_restart"
> > multiple times if already rotate_necessary is set for that context.
> > Even though "perf_mux_hrtimer_restart" will just return if hrtimer is
> > already active, this could avoid the overhead of calling this function
> > multiple times if there are many event groups. Patch adds the check to
> > avoid calling perf_mux_hrtimer_restart() for each event group on
> > every schedule.
> >
> > Signed-off-by: Athira Rajeev <[email protected]>
>
> Acked-by: Song Liu <[email protected]>

Thanks!

2021-10-21 12:24:36

by Athira Rajeev

[permalink] [raw]
Subject: Re: [V2] perf/core: Avoid calling perf_mux_hrtimer_restart multiple times when scheduling event groups



> On 18-Oct-2021, at 4:37 PM, Peter Zijlstra <[email protected]> wrote:
>
> On Sat, Oct 16, 2021 at 01:08:04AM +0000, Song Liu wrote:
>>
>>
>>> On Oct 15, 2021, at 12:04 AM, Athira Rajeev <[email protected]> wrote:
>>>
>>> Perf uses multiplexing if there are more events to be scheduled than the
>>> available counters. With multiplexing, event groups will be rotated at
>>> specified interval of time which is set using "hrtimer". During event
>>> scheduling, if any of the event group fails to schedule, multiplexing
>>> gets enabled by setting "rotate_necessary" for that event context and
>>> starting the hrtimer by calling "perf_mux_hrtimer_restart".
>>>
>>> Patch adds an optimisation to avoid calling "perf_mux_hrtimer_restart"
>>> multiple times if already rotate_necessary is set for that context.
>>> Even though "perf_mux_hrtimer_restart" will just return if hrtimer is
>>> already active, this could avoid the overhead of calling this function
>>> multiple times if there are many event groups. Patch adds the check to
>>> avoid calling perf_mux_hrtimer_restart() for each event group on
>>> every schedule.
>>>
>>> Signed-off-by: Athira Rajeev <[email protected]>
>>
>> Acked-by: Song Liu <[email protected]>
>
> Thanks!

Thanks for checking the patch.

Hi Peterz,
Should I resend this patch with Acked-by or can you consider pulling it, if the patch looks fine to you.

Thanks
Athira.