Re: [patch 11/16] sched: replace update_shares weight distribution with per-entity computation

["\"Jan H. Schönherr\"" <schnhrr@cs.tu-berlin.de>]

Am 23.08.2012 16:14, schrieb pjt@google.com:
> From: Paul Turner <pjt@google.com>
> 
> Now that the machinery in place is in place to compute contributed load in a
> bottom up fashion; replace the shares distribution code within update_shares()
> accordingly.

[snip]

>  static int update_shares_cpu(struct task_group *tg, int cpu)
>  {
> +	struct sched_entity *se;
>  	struct cfs_rq *cfs_rq;
>  	unsigned long flags;
>  	struct rq *rq;
>  
> -	if (!tg->se[cpu])
> -		return 0;
> -
>  	rq = cpu_rq(cpu);
> +	se = tg->se[cpu];
>  	cfs_rq = tg->cfs_rq[cpu];
>  
>  	raw_spin_lock_irqsave(&rq->lock, flags);
>  
>  	update_rq_clock(rq);
> -	update_cfs_load(cfs_rq, 1);
>  	update_cfs_rq_blocked_load(cfs_rq, 1);
>  
> -	/*
> -	 * We need to update shares after updating tg->load_weight in
> -	 * order to adjust the weight of groups with long running tasks.
> -	 */
> -	update_cfs_shares(cfs_rq);
> +	if (se) {
> +		update_entity_load_avg(se, 1);
> +		/*
> +		 * We can pivot on the runnable average decaying to zero for
> +		 * list removal since the parent average will always be >=
> +		 * child.
> +		 */
> +		if (se->avg.runnable_avg_sum)
> +			update_cfs_shares(cfs_rq);
> +		else
> +			list_del_leaf_cfs_rq(cfs_rq);

The blocked load, which we decay from this function, is not part of
se->avg.runnable_avg_sum. Is list removal a good idea while there might be
blocked load? We only get here, because we are on that list... don't we end up
with a wrong task group load then?

Regards
Jan

> +	} else {
> +		update_rq_runnable_avg(rq, rq->nr_running);
> +	}
>  
>  	raw_spin_unlock_irqrestore(&rq->lock, flags);
>