* [PATCH v2] perf/core: Avoid context switch overheads
@ 2017-08-08 10:00 石祤
2017-08-08 10:37 ` Peter Zijlstra
0 siblings, 1 reply; 4+ messages in thread
From: 石祤 @ 2017-08-08 10:00 UTC (permalink / raw)
To: yang_oliver, peterz, mingo, acme, alexander.shishkin,
linux-kernel, tglx, eranian, torvalds, jolsa, linxiulei
Cc: leilei.lin
From: "leilei.lin" <leilei.lin@alibaba-inc.com>
A performance issue caused by less strickly check in task
sched when these tasks were once attached by per-task perf_event.
A task will alloc task->perf_event_ctxp[ctxn] when it was called
by perf_event_open, and task->perf_event_ctxp[ctxn] would not
ever be freed to NULL.
__perf_event_task_sched_in()
if (task->perf_event_ctxp[ctxn]) // here is always true
perf_event_context_sched_in() // operate pmu
50% at most performance overhead was observed under some extreme
test case. Therefor, add a more strick check as to ctx->nr_events,
when ctx->nr_events == 0, it's no need to continue.
Signed-off-by: leilei.lin <leilei.lin@alibaba-inc.com>
---
kernel/events/core.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 426c2ff..3d86695 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3180,6 +3180,13 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
return;
perf_ctx_lock(cpuctx, ctx);
+ /*
+ * We must check ctx->nr_events while holding ctx->lock, such
+ * that we serialize against perf_install_in_context().
+ */
+ if (!cpuctx->task_ctx && !ctx->nr_events)
+ goto unlock;
+
perf_pmu_disable(ctx->pmu);
/*
* We want to keep the following priority order:
@@ -3193,6 +3200,8 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
perf_event_sched_in(cpuctx, ctx, task);
perf_pmu_enable(ctx->pmu);
+
+unlock:
perf_ctx_unlock(cpuctx, ctx);
}
--
2.8.4.31.g9ed660f
^ permalink raw reply related [flat|nested] 4+ messages in thread
* Re: [PATCH v2] perf/core: Avoid context switch overheads
2017-08-08 10:00 [PATCH v2] perf/core: Avoid context switch overheads 石祤
@ 2017-08-08 10:37 ` Peter Zijlstra
2017-08-09 0:28 ` 林守磊
0 siblings, 1 reply; 4+ messages in thread
From: Peter Zijlstra @ 2017-08-08 10:37 UTC (permalink / raw)
To: 石祤
Cc: yang_oliver, mingo, acme, alexander.shishkin, linux-kernel, tglx,
eranian, torvalds, jolsa, leilei.lin
On Tue, Aug 08, 2017 at 06:00:45PM +0800, 石祤 wrote:
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 426c2ff..3d86695 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -3180,6 +3180,13 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
> return;
>
> perf_ctx_lock(cpuctx, ctx);
> + /*
> + * We must check ctx->nr_events while holding ctx->lock, such
> + * that we serialize against perf_install_in_context().
> + */
> + if (!cpuctx->task_ctx && !ctx->nr_events)
> + goto unlock;
Do we really need the cpuctx->task_ctx test? I think that task_ctx is
'tight' these days. We never have it set unless there are events
scheduled for that context.
I even think the cpuctx->task_ctx == ctx test right above here is
superfluous these days. That could only happen when the
perf_install_in_context() IPI came before perf_event_task_sched_in(),
but we removed the arch option to do context switches with IRQs enabled.
> +
> perf_pmu_disable(ctx->pmu);
> /*
> * We want to keep the following priority order:
> @@ -3193,6 +3200,8 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
> cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
> perf_event_sched_in(cpuctx, ctx, task);
> perf_pmu_enable(ctx->pmu);
> +
> +unlock:
> perf_ctx_unlock(cpuctx, ctx);
> }
>
> --
> 2.8.4.31.g9ed660f
>
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [PATCH v2] perf/core: Avoid context switch overheads
2017-08-08 10:37 ` Peter Zijlstra
@ 2017-08-09 0:28 ` 林守磊
0 siblings, 0 replies; 4+ messages in thread
From: 林守磊 @ 2017-08-09 0:28 UTC (permalink / raw)
To: Peter Zijlstra
Cc: yang_oliver, mingo, acme, alexander.shishkin, linux-kernel, tglx,
Stephane Eranian, torvalds, jolsa, leilei.lin
2017-08-08 18:37 GMT+08:00 Peter Zijlstra <peterz@infradead.org>:
> On Tue, Aug 08, 2017 at 06:00:45PM +0800, 石祤 wrote:
>
>> diff --git a/kernel/events/core.c b/kernel/events/core.c
>> index 426c2ff..3d86695 100644
>> --- a/kernel/events/core.c
>> +++ b/kernel/events/core.c
>> @@ -3180,6 +3180,13 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
>> return;
>>
>> perf_ctx_lock(cpuctx, ctx);
>> + /*
>> + * We must check ctx->nr_events while holding ctx->lock, such
>> + * that we serialize against perf_install_in_context().
>> + */
>> + if (!cpuctx->task_ctx && !ctx->nr_events)
>> + goto unlock;
>
> Do we really need the cpuctx->task_ctx test? I think that task_ctx is
> 'tight' these days. We never have it set unless there are events
> scheduled for that context.
>
> I even think the cpuctx->task_ctx == ctx test right above here is
> superfluous these days. That could only happen when the
> perf_install_in_context() IPI came before perf_event_task_sched_in(),
> but we removed the arch option to do context switches with IRQs enabled.
>
It looks that cpuctx->task_ctx exists somewhere else, so I thought it was
conservative making this patch.
For a centain, during my process of debugging I didn't figure out any value
of cpuctx->task_ctx. I shall make a v3.
Thanks
>> +
>> perf_pmu_disable(ctx->pmu);
>> /*
>> * We want to keep the following priority order:
>> @@ -3193,6 +3200,8 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
>> cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
>> perf_event_sched_in(cpuctx, ctx, task);
>> perf_pmu_enable(ctx->pmu);
>> +
>> +unlock:
>> perf_ctx_unlock(cpuctx, ctx);
>> }
>>
>> --
>> 2.8.4.31.g9ed660f
>>
^ permalink raw reply [flat|nested] 4+ messages in thread
* [PATCH v2] perf/core: Avoid context switch overheads
@ 2017-08-08 4:43 石祤
0 siblings, 0 replies; 4+ messages in thread
From: 石祤 @ 2017-08-08 4:43 UTC (permalink / raw)
To: yang_oliver, peterz, mingo, acme, alexander.shishkin,
linux-kernel, tglx, eranian, torvalds, jolsa, linxiulei
Cc: leilei.lin
From: "leilei.lin" <leilei.lin@alibaba-inc.com>
A performance issue caused by less strickly check in task
sched when these tasks were once attached by per-task perf_event.
A task will alloc task->perf_event_ctxp[ctxn] when it was called
by perf_event_open, and task->perf_event_ctxp[ctxn] would not
ever be freed to NULL.
__perf_event_task_sched_in()
if (task->perf_event_ctxp[ctxn]) // here is always true
perf_event_context_sched_in() // operate pmu
50% at most performance overhead was observed under some extreme
test case. Therefor, add a more strick check as to ctx->nr_events,
when ctx->nr_events == 0, it's no need to continue.
Signed-off-by: leilei.lin <leilei.lin@alibaba-inc.com>
---
kernel/events/core.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 426c2ff..3d86695 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3180,6 +3180,13 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
return;
perf_ctx_lock(cpuctx, ctx);
+ /*
+ * We must check ctx->nr_events while holding ctx->lock, such
+ * that we serialize against perf_install_in_context().
+ */
+ if (!cpuctx->task_ctx && !ctx->nr_events)
+ goto unlock;
+
perf_pmu_disable(ctx->pmu);
/*
* We want to keep the following priority order:
@@ -3193,6 +3200,8 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
perf_event_sched_in(cpuctx, ctx, task);
perf_pmu_enable(ctx->pmu);
+
+unlock:
perf_ctx_unlock(cpuctx, ctx);
}
--
2.8.4.31.g9ed660f
^ permalink raw reply related [flat|nested] 4+ messages in thread
end of thread, other threads:[~2017-08-09 0:28 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-08-08 10:00 [PATCH v2] perf/core: Avoid context switch overheads 石祤
2017-08-08 10:37 ` Peter Zijlstra
2017-08-09 0:28 ` 林守磊
-- strict thread matches above, loose matches on Subject: below --
2017-08-08 4:43 石祤
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.