Re: [PATCH v3] sched: modify how to compute a slice and check a preemptability

[Mike Galbraith <umgwanakikbuti@gmail.com>]

On Sun, 2015-07-19 at 18:11 +0900, byungchul.park@lge.com wrote:

> @@ -3226,6 +3226,12 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
>  	struct sched_entity *se;
>  	s64 delta;
>  
> +	/*
> +	 * Ensure that a task executes at least for sysctl_sched_min_granularity
> +	 */
> +	if (delta_exec < sysctl_sched_min_granularity)
> +		return;
> +

Think about what this does to a low weight task, or any task in a low
weight group.  The scheduler equalizes runtimes for a living, there is
no free lunch.  Any runtime larger than fair share that you graciously
grant to random task foo doesn't magically appear out of the vacuum, it
comes out of task foo's wallet. If you drag that hard coded minimum down
into the depths of group scheduling, yeah, every task will get a nice
juicy slice of CPU.. eventually, though you may not live to see it.

(yeah, overrun can and will happen at all depths due to tick
granularity, but you guaranteed it, so I inflated severity a bit;)

>  	ideal_runtime = sched_slice(cfs_rq, curr);
>  	delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
>  	if (delta_exec > ideal_runtime) {
> @@ -3243,9 +3249,6 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
>  	 * narrow margin doesn't have to wait for a full slice.
>  	 * This also mitigates buddy induced latencies under load.
>  	 */
> -	if (delta_exec < sysctl_sched_min_granularity)
> -		return;
> -

That was about something entirely different.  Feel free to remove it
after verifying that it has outlived it's original purpose, but please
don't just move it about at random.

	-Mike