Re: [PATCH v2] perf/core: Optimize event reschedule for a PMU

["Liang, Kan" <kan.liang@linux.intel.com>]

On 2024-08-02 3:11 p.m., Peter Zijlstra wrote:
> On Fri, Aug 02, 2024 at 08:50:23PM +0200, Peter Zijlstra wrote:
>> On Fri, Aug 02, 2024 at 08:43:50PM +0200, Peter Zijlstra wrote:
>>> On Fri, Aug 02, 2024 at 08:38:41PM +0200, Peter Zijlstra wrote:
>>>> On Fri, Aug 02, 2024 at 02:30:19PM -0400, Liang, Kan wrote:
>>>>>> @@ -2792,7 +2833,14 @@ static int  __perf_install_in_context(void *info)
>>>>>>  	if (reprogram) {
>>>>>>  		ctx_sched_out(ctx, EVENT_TIME);
>>>
>>> Clearly I should read better...
>>>
>>>>>>  		add_event_to_ctx(event, ctx);
>>>>>> -		ctx_resched(cpuctx, task_ctx, get_event_type(event));
>>>>>> +		if (ctx->nr_events == 1) {
>>>>>> +			/* The first event needs to set ctx->is_active. */
>>>>>> +			ctx_resched(cpuctx, task_ctx, NULL, get_event_type(event));
>>>>>> +		} else {
>>>>>> +			ctx_resched(cpuctx, task_ctx, event->pmu_ctx->pmu,
>>>>>> +				    get_event_type(event));
>>>>>> +			ctx_sched_in(ctx, EVENT_TIME);
>>>>>
>>>>> The changelog doesn't mention the time difference much. As my
>>>>> understanding, the time is shared among PMUs in the same ctx.
>>>>> When perf does ctx_resched(), the time is deducted.
>>>>> There is no problem to stop and restart the global time when perf
>>>>> re-schedule all PMUs.
>>>>> But if only one PMU is re-scheduled while others are still running, it
>>>>> may be a problem to stop and restart the global time. Other PMUs will be
>>>>> impacted.
>>>
>>> So yeah, this stops ctx time but not all PMUs.
>>
>> But isn't this already the case? We don't have perf_ctx_disable() here
>> currently. 
>>
>> Bah, this heat is melting my brain.
> 
> I think all it wants is to update time and ensure the added event and
> the resched all use the same time, which could be done differently.
>

Yes. I think that's what the current code tries to do.
But it seems the current code doesn't do it clearly either.

ctx_sched_out(ctx, EVENT_TIME); <-- disable the time
ctx_resched()
    perf_ctx_disable()          <-- disable all PMUs

    perf_event_sched_in()
        ctx_sched_in()          <-- enable the time
    perf_ctx_enable()           <-- enable all PMUs

I think the ctx_sched_out(ctx, EVENT_TIME) should be moved after the
perf_ctx_disable();.
Hope it can be fixed by the different way.

> But I'll have to continue staring at this later.

Sure.

Thanks,
Kan