Re: [RFC PATCH 07/16] sched/fair: Fix forced idle sibling starvation corner case(Internet mail)

[benbjiang(蒋彪)]

Hi,

> On Jul 1, 2020, at 5:32 AM, Vineeth Remanan Pillai <vpillai@xxxxxxxxxxxxxxxx> wrote:
> 
> From: vpillai <vpillai@xxxxxxxxxxxxxxxx>
> 
> If there is only one long running local task and the sibling is
> forced idle, it  might not get a chance to run until a schedule
> event happens on any cpu in the core.
> 
> So we check for this condition during a tick to see if a sibling
> is starved and then give it a chance to schedule.
> 
> Signed-off-by: Vineeth Remanan Pillai <vpillai@xxxxxxxxxxxxxxxx>
> Signed-off-by: Julien Desfossez <jdesfossez@xxxxxxxxxxxxxxxx>
> ---
> kernel/sched/fair.c | 39 +++++++++++++++++++++++++++++++++++++++
> 1 file changed, 39 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index ae17507533a0..49fb93296e35 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -10613,6 +10613,40 @@ static void rq_offline_fair(struct rq *rq)
> 
> #endif /* CONFIG_SMP */
> 
> +#ifdef CONFIG_SCHED_CORE
> +static inline bool
> +__entity_slice_used(struct sched_entity *se)
> +{
> +	return (se->sum_exec_runtime - se->prev_sum_exec_runtime) >
> +		sched_slice(cfs_rq_of(se), se);
> +}
> +
> +/*
> + * If runqueue has only one task which used up its slice and if the sibling
> + * is forced idle, then trigger schedule to give forced idle task a chance.
> + */
> +static void resched_forceidle_sibling(struct rq *rq, struct sched_entity *se)
> +{
> +	int cpu = cpu_of(rq), sibling_cpu;
> +
> +	if (rq->cfs.nr_running > 1 || !__entity_slice_used(se))
> +		return;
> +
> +	for_each_cpu(sibling_cpu, cpu_smt_mask(cpu)) {
> +		struct rq *sibling_rq;
> +		if (sibling_cpu == cpu)
> +			continue;
> +		if (cpu_is_offline(sibling_cpu))
> +			continue;
> +
> +		sibling_rq = cpu_rq(sibling_cpu);
> +		if (sibling_rq->core_forceidle) {
> +			resched_curr(sibling_rq);
> +		}
> +	}
> +}
> +#endif
> +
> /*
>  * scheduler tick hitting a task of our scheduling class.
>  *
> @@ -10636,6 +10670,11 @@ static void task_tick_fair(struct rq *rq, struct task_struct *curr, int queued)
> 
> 	update_misfit_status(curr, rq);
> 	update_overutilized_status(task_rq(curr));
> +
> +#ifdef CONFIG_SCHED_CORE
> +	if (sched_core_enabled(rq))
> +		resched_forceidle_sibling(rq, &curr->se);
> +#endif
Hi,

resched_forceidle_sibling depends on tick, but there could be no tick in 1s(scheduler_tick_max_derferment) after
entering nohz_full mode. 
And when enable nohz_full, cpu will enter nohz_full mode frequently when *there is only one long running local task*.
That means the siblings rescheduling would be delayed much more than sched_slice(), could be unfair and result in
big latency.

Should we restrict cpu with forced-idle sibling entering nohz_full mode by adding specific flag and checking it before
stop tick?

Or we can do rescheduling on siblings in task_tick_idle by checking starvation time? :)

Thx
Regard,
Jiang 

> }
> 
> /*
> -- 
> 2.17.1
> 
>