Re: [PATCH v4] sched/deadline: remove useless param from setup_new_dl_entity

[Xunlei Pang <xpang@redhat.com>]

On 2016/07/21 at 22:46, Juri Lelli wrote:
> On 21/07/16 15:36, Juri Lelli wrote:
>> On 21/07/16 15:21, Juri Lelli wrote:
>>> Hi,
>>>
>>> On 18/07/16 21:37, Xunlei Pang wrote:
>>>> On 2016/07/18 at 21:04, Juri Lelli wrote:
>>>>> On 15/07/16 18:39, Xunlei Pang wrote:
>>>>>> On 2016/07/13 at 18:58, Juri Lelli wrote:
>>>>> [...]
>>>>>
>>>>>> Since this is only called for queued cases now, there is no need to
>>>>>> check boosted stuff here. As enqueue_task(ENQUEUE_REPLENISH)
>>>>>> is called before check_class_changed() in rt_mutex_setprio().
>>>>>>
>>>>> But we don't do the same in setscheduler, right?
>>>> If p is deadline PI-boosted, setscheduler() won't call change its sched_class.
>>>> If p isn't deadline PI-boosted, then pi_task is NULL.
>>>>
>>>> So, I think the added code won't hit. Did I miss something?
>>>>
>>> No, I think you are right.
>>>
>> Oh, and we need to filter the call after rt_mutex_setprio has
>> already issued a replenishment.
>>
> Does something like this on top of v4 make sense?
>
> ---
>  kernel/sched/deadline.c | 22 +++++++---------------
>  1 file changed, 7 insertions(+), 15 deletions(-)
>
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index dc56f5be0112..6f05ac78711c 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -350,9 +350,8 @@ static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
>  {
>  	struct dl_rq *dl_rq = dl_rq_of_se(dl_se);
>  	struct rq *rq = rq_of_dl_rq(dl_rq);
> -	struct task_struct *pi_task;
> -	struct sched_dl_entity *pi_se = dl_se;
>  
> +	WARN_ON(dl_se->dl_boosted);
>  	WARN_ON(dl_time_before(rq_clock(rq), dl_se->deadline));
>  
>  	/*
> @@ -364,21 +363,12 @@ static inline void setup_new_dl_entity(struct sched_dl_entity *dl_se)
>  		return;
>  
>  	/*
> -	 * Use the scheduling parameters of the top pi-waiter task,
> -	 * if we have one from which we can inherit a deadline.
> -	 */
> -	if (dl_se->dl_boosted &&
> -	    (pi_task = rt_mutex_get_top_task(dl_task_of(dl_se))) &&
> -	    dl_prio(pi_task->normal_prio))
> -		pi_se = &pi_task->dl;
> -
> -	/*
>  	 * We use the regular wall clock time to set deadlines in the
>  	 * future; in fact, we must consider execution overheads (time
>  	 * spent on hardirq context, etc.).
>  	 */
> -	dl_se->deadline = rq_clock(rq) + pi_se->dl_deadline;
> -	dl_se->runtime = pi_se->dl_runtime;
> +	dl_se->deadline = rq_clock(rq) + dl_se->dl_deadline;
> +	dl_se->runtime = dl_se->dl_runtime;
>  }
>  
>  /*
> @@ -1734,9 +1724,11 @@ static void switched_to_dl(struct rq *rq, struct task_struct *p)
>  	if (task_on_rq_queued(p)) {
>  		/*
>  		 * If p is not queued we will update its parameters at next
> -		 * wakeup.
> +		 * wakeup. If p is dl_boosted we already updated its params in
> +		 * rt_mutex_setprio()->enqueue_task(..., ENQUEUE_REPLENISH).
>  		 */
> -		if (dl_time_before(p->dl.deadline, rq_clock(rq)))
> +		if (dl_time_before(p->dl.deadline, rq_clock(rq)) &&
> +				!p->dl.dl_boosted)

Hi Juri,

It looks good to me, only one question:
For on_rq boosted to deadline, p->dl.deadline has been updated after rq_lock(rq) by
rt_mutex_setprio()->enqueue_task(..., ENQUEUE_REPLENISH) and no rq clock update
afterwards, so dl_time_before() will be false, seems p->dl.dl_boosted check is needless.

Regards,
Xunlei

>  			setup_new_dl_entity(&p->dl);
>  
>  		if (rq->curr != p) {