Re: [PATCH v2] sched/freq: move call to cpufreq_update_util

[Dietmar Eggemann]

On 12.11.19 16:05, Vincent Guittot wrote:
> Le Tuesday 12 Nov 2019 Ã 15:48:13 (+0100), Vincent Guittot a Ãcrit :
>> update_cfs_rq_load_avg() calls cfs_rq_util_change() everytime pelt decays,
>> which might be inefficient when cpufreq driver has rate limitation.
>>
>> When a task is attached on a CPU, we have call path:
>>
>> update_blocked_averages()
>>   update_cfs_rq_load_avg()
>>     cfs_rq_util_change -- > trig frequency update
>>   attach_entity_load_avg()
>>     cfs_rq_util_change -- > trig frequency update

This looks like attach_entity_load_avg() is called from
update_blocked_averages(). Do you refer to the attach_entity_load_avg()
call from attach_entity_cfs_rq() or update_load_avg() here? I assume the
former.

>> The 1st frequency update will not take into account the utilization of the
>> newly attached task and the 2nd one might be discard because of rate
>> limitation of the cpufreq driver.
>>
>> update_cfs_rq_load_avg() is only called by update_blocked_averages()
>> and update_load_avg() so we can move the call to
>> {cfs_rq,cpufreq}_util_change() into these 2 functions. It's also

s/cpufreq_util_change()/cpufreq_update_util() ?

[...]

>> I have just rebased the patch on latest tip/sched/core and made it a proper
>> patchset after Doug reported that the problem has diseappeared according to
>> his 1st results but tests results are not all based on the same v5.4-rcX
>> and with menu instead of teo governor.

I had some minor tweaks to do putting this on a0e813f26ebc ("sched/core:
Further clarify sched_class::set_next_task()") ? I saw the '[tip:
sched/urgent] sched/pelt: Fix update of blocked PELT ordering' tip-bot
msg this morning though.

[...]

>> @@ -7493,9 +7495,9 @@ static void update_blocked_averages(int cpu)
>>  	 * that RT, DL and IRQ signals have been updated before updating CFS.
>>  	 */
>>  	curr_class = rq->curr->sched_class;
>> -	update_rt_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &rt_sched_class);
>> -	update_dl_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &dl_sched_class);
>> -	update_irq_load_avg(rq, 0);
>> +	decayed |= update_rt_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &rt_sched_class);
>> +	decayed |= update_dl_rq_load_avg(rq_clock_pelt(rq), rq, curr_class == &dl_sched_class);
>> +	decayed |= update_irq_load_avg(rq, 0);

Why not 'decayed  = update_cfs_rq_load_avg()' like in the
!CONFIG_FAIR_GROUP_SCHED case?

[...]