Re: [PATCH v3 2/7] sched/fair: Decay task PELT values during migration

[Dietmar Eggemann]

On 08/03/2022 19:19, Vincent Donnefort wrote:

Patch header:

s/during migration/during wakeup migration ?

To make sure it's not related to lb migration (TASK_ON_RQ_MIGRATING).

[...]

> Neither clock_task, clock_pelt nor clock can be accessed without the
> runqueue lock. The new cfs_rq last_update_lag is therefore created and
> encode those three values when the last_update_time value for that very

s/encode/encodes ... It's not really encoding?

[...]

> Now we have an estimation for A + B, we can create an estimator for the
> PELT value at the time of the migration. We need for this purpose to
> inject last_update_time which is a combination of both clock_pelt and
> lost_idle_time. The latter is a time value which is completely lost form a

s/form/from

> PELT point of view and must be ignored. And finally, we can write:
> 
>   rq_clock_pelt_estimator() = last_update_time + A + B
>                             = last_update_time +
>                                    sched_clock_cpu() - last_update_lag

rq_clock_pelt_estimator() did exist in v2 but does not in v3 anymore.
Might be misleading when people search for it.

[...]

> @@ -3688,6 +3704,7 @@ update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq)
>  			   cfs_rq->last_update_time_copy,
>  			   sa->last_update_time);
>  #endif
> +	update_cfs_rq_lag(cfs_rq);
>  
>  	return decayed;
>  }
> @@ -6852,6 +6869,44 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
>  
>  static void detach_entity_cfs_rq(struct sched_entity *se);
>  
> +#ifdef CONFIG_NO_HZ_COMMON

Couldn't you put update_cfs_rq_lag() and migrate_se_pelt_lag() under one
CONFIG_NO_HZ_COMMON?

> +static inline void migrate_se_pelt_lag(struct sched_entity *se)

[...]

> +	/*
> +	 * The lag estimation comes with a cost we don't want to pay all the
> +	 * time. Hence, limiting to the case where the source CPU is idle and
> +	 * we know we are at the greatest risk to have an outdated clock.
> +	 */
> +	if (!is_idle)
> +		return;
> +
> +	last_update_lag = u64_u32_load(cfs_rq->last_update_lag);

So each taskgroup has its own last_update_lag. I guess it works since we
sync se in migrate_task_rq_fair() -> remove_entity_load_avg() ->
sync_entity_load_avg() with its cfs_rq.

[...]

> @@ -6859,6 +6914,9 @@ static void detach_entity_cfs_rq(struct sched_entity *se);
>   */
>  static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
>  {
> +	struct sched_entity *se = &p->se;
> +	struct cfs_rq *cfs_rq = cfs_rq_of(se);

This line can stay in the if condition below since cfs_rq is only used
there. The brackets are also still there (A).

[...]

> @@ -6866,8 +6924,6 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
>  	 * the task on the new runqueue.
>  	 */
>  	if (READ_ONCE(p->__state) == TASK_WAKING) {
> -		struct sched_entity *se = &p->se;         <--- (A)
> -		struct cfs_rq *cfs_rq = cfs_rq_of(se);

[...]