Re: [PATCH] sched: sync with the cfs_rq when changing sched class

["T. Zhou" <t.s.zhou@hotmail.com>]

Hi,

On Thu, Aug 13, 2015 at 02:55:55PM +0900, byungchul.park@lge.com wrote:
> +static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
> +{
> +	se->avg.last_update_time = cfs_rq->avg.last_update_time;
> +	cfs_rq->avg.load_avg += se->avg.load_avg;
> +	cfs_rq->avg.load_sum += se->avg.load_sum;
> +	cfs_rq->avg.util_avg += se->avg.util_avg;
> +	cfs_rq->avg.util_sum += se->avg.util_sum;
> +}

I see this function is used in enqueue_entity_load_avg() also.
In tip tree code, enqueue_entity_load_avg() uses cfs_rq_clock_task()
as se->last_update_time in migration condition.
Here, you use cfs_rq->avg.last_update_time as se->last_update_time.
If se->last_update_time is different, next decay may be different too.
Just from inspecting code, maybe some reasonable there?

Thanks
> +static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
> +{
> +	__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
> +			&se->avg, se->on_rq * scale_load_down(se->load.weight),
> +			cfs_rq->curr == se, NULL);
> +
> +	cfs_rq->avg.load_avg =
> +		max_t(long, cfs_rq->avg.load_avg - se->avg.load_avg, 0);
> +	cfs_rq->avg.load_sum =
> +		max_t(s64, cfs_rq->avg.load_sum - se->avg.load_sum, 0);
> +	cfs_rq->avg.util_avg =
> +		max_t(long, cfs_rq->avg.util_avg - se->avg.util_avg, 0);
> +	cfs_rq->avg.util_sum =
> +		max_t(s32, cfs_rq->avg.util_sum - se->avg.util_sum, 0);
> +}
> +
>  /* Add the load generated by se into cfs_rq's load average */
>  static inline void
>  enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
> @@ -2717,27 +2742,20 @@ enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
>  	u64 now = cfs_rq_clock_task(cfs_rq);
>  	int migrated = 0, decayed;
>  
> -	if (sa->last_update_time == 0) {
> -		sa->last_update_time = now;
> +	if (sa->last_update_time == 0)
>  		migrated = 1;
> -	}
> -	else {
> +	else
>  		__update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
> -			se->on_rq * scale_load_down(se->load.weight),
> -			cfs_rq->curr == se, NULL);
> -	}
> +				se->on_rq * scale_load_down(se->load.weight),
> +				cfs_rq->curr == se, NULL);
>  
>  	decayed = update_cfs_rq_load_avg(now, cfs_rq);
>  
>  	cfs_rq->runnable_load_avg += sa->load_avg;
>  	cfs_rq->runnable_load_sum += sa->load_sum;
>  
> -	if (migrated) {
> -		cfs_rq->avg.load_avg += sa->load_avg;
> -		cfs_rq->avg.load_sum += sa->load_sum;
> -		cfs_rq->avg.util_avg += sa->util_avg;
> -		cfs_rq->avg.util_sum += sa->util_sum;
> -	}
> +	if (migrated)
> +		attach_entity_load_avg(cfs_rq, se);
>  
>  	if (decayed || migrated)
>  		update_tg_load_avg(cfs_rq, 0);
> @@ -7911,17 +7929,7 @@ static void switched_from_fair(struct rq *rq, struct task_struct *p)
>  
>  #ifdef CONFIG_SMP
>  	/* Catch up with the cfs_rq and remove our load when we leave */
> -	__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq), &se->avg,
> -		se->on_rq * scale_load_down(se->load.weight), cfs_rq->curr == se, NULL);
> -
> -	cfs_rq->avg.load_avg =
> -		max_t(long, cfs_rq->avg.load_avg - se->avg.load_avg, 0);
> -	cfs_rq->avg.load_sum =
> -		max_t(s64, cfs_rq->avg.load_sum - se->avg.load_sum, 0);
> -	cfs_rq->avg.util_avg =
> -		max_t(long, cfs_rq->avg.util_avg - se->avg.util_avg, 0);
> -	cfs_rq->avg.util_sum =
> -		max_t(s32, cfs_rq->avg.util_sum - se->avg.util_sum, 0);
> +	detach_entity_load_avg(cfs_rq, se);
>  #endif
>  }
>  
> @@ -7938,6 +7946,11 @@ static void switched_to_fair(struct rq *rq, struct task_struct *p)
>  	 */
>  	se->depth = se->parent ? se->parent->depth + 1 : 0;
>  #endif
> +
> +#ifdef CONFIG_SMP
> +	/* synchronize task with its cfs_rq */
> +	attach_entity_load_avg(cfs_rq_of(&p->se), &p->se);
> +#endif
>  	if (!task_on_rq_queued(p))
>  		return;
>  
> @@ -8023,16 +8036,7 @@ static void task_move_group_fair(struct task_struct *p, int queued)
>  
>  #ifdef CONFIG_SMP
>  	/* synchronize task with its prev cfs_rq */
> -	if (!queued)
> -		__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
> -				&se->avg, se->on_rq * scale_load_down(se->load.weight),
> -				cfs_rq->curr == se, NULL);
> -
> -	/* remove our load when we leave */
> -	cfs_rq->avg.load_avg = max_t(long, cfs_rq->avg.load_avg - se->avg.load_avg, 0);
> -	cfs_rq->avg.load_sum = max_t(s64, cfs_rq->avg.load_sum - se->avg.load_sum, 0);
> -	cfs_rq->avg.util_avg = max_t(long, cfs_rq->avg.util_avg - se->avg.util_avg, 0);
> -	cfs_rq->avg.util_sum = max_t(s32, cfs_rq->avg.util_sum - se->avg.util_sum, 0);
> +	detach_entity_load_avg(cfs_rq, se);
>  #endif
>  	set_task_rq(p, task_cpu(p));
>  	se->depth = se->parent ? se->parent->depth + 1 : 0;
> @@ -8042,11 +8046,7 @@ static void task_move_group_fair(struct task_struct *p, int queued)
>  
>  #ifdef CONFIG_SMP
>  	/* Virtually synchronize task with its new cfs_rq */
> -	p->se.avg.last_update_time = cfs_rq->avg.last_update_time;
> -	cfs_rq->avg.load_avg += p->se.avg.load_avg;
> -	cfs_rq->avg.load_sum += p->se.avg.load_sum;
> -	cfs_rq->avg.util_avg += p->se.avg.util_avg;
> -	cfs_rq->avg.util_sum += p->se.avg.util_sum;
> +	attach_entity_load_avg(cfs_rq, se);
>  #endif
>  }

-- 
Tao
--