Re: [PATCH 2/2] sched/dl: Cleanup prio_changed_dl()

[Kirill Tkhai <tkhai@yandex.ru>]

29.09.2014, 19:03, "Juri Lelli" <juri.lelli@arm.com>:
> Hi Kirill,
>
> On 27/09/14 09:41, Kirill Tkhai wrote:
>>  From: Kirill Tkhai <ktkhai@parallels.com>
>>
>>  Now rq->curr task can't be dequeued in this function
>>  (it's possible only in __schedule()).
>
> Could you please detail about "before", with a reference to the commit
> that made this further change necessary?

I wrote about unlocked context switching, but I've looked one more time
and found that it used to be impossible before too.

We set "rq->curr = next" before prepare_lock_switch(),
so dequeued task wasn't able to be curr.

I'll update both descriptions in v2.

Thanks,
Kirill

>
> Thanks a lot,
>
> - Juri
>>  Also, delete "else" branch which is dead code.
>>
>>  Signed-off-by: Kirill Tkhai <ktkhai@parallels.com>
>>  ---
>>   kernel/sched/deadline.c |   50 +++++++++++++++++++++++------------------------
>>   1 file changed, 24 insertions(+), 26 deletions(-)
>>
>>  diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
>>  index 63f8b4a..38b4f19 100644
>>  --- a/kernel/sched/deadline.c
>>  +++ b/kernel/sched/deadline.c
>>  @@ -1638,35 +1638,33 @@ static void switched_to_dl(struct rq *rq, struct task_struct *p)
>>   static void prio_changed_dl(struct rq *rq, struct task_struct *p,
>>                               int oldprio)
>>   {
>>  - if (task_on_rq_queued(p) || rq->curr == p) {
>>  + if (!task_on_rq_queued(p))
>>  + return;
>>   #ifdef CONFIG_SMP
>>  - /*
>>  - * This might be too much, but unfortunately
>>  - * we don't have the old deadline value, and
>>  - * we can't argue if the task is increasing
>>  - * or lowering its prio, so...
>>  - */
>>  - if (!rq->dl.overloaded)
>>  - pull_dl_task(rq);
>>  -
>>  - /*
>>  - * If we now have a earlier deadline task than p,
>>  - * then reschedule, provided p is still on this
>>  - * runqueue.
>>  - */
>>  - if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline) &&
>>  -    rq->curr == p)
>>  - resched_curr(rq);
>>  -#else
>>  - /*
>>  - * Again, we don't know if p has a earlier
>>  - * or later deadline, so let's blindly set a
>>  - * (maybe not needed) rescheduling point.
>>  - */
>>  + /*
>>  + * This might be too much, but unfortunately
>>  + * we don't have the old deadline value, and
>>  + * we can't argue if the task is increasing
>>  + * or lowering its prio, so...
>>  + */
>>  + if (!rq->dl.overloaded)
>>  + pull_dl_task(rq);
>>  + /*
>>  + * If we now have a earlier deadline task than p,
>>  + * then reschedule, provided p is still on this
>>  + * runqueue.
>>  + */
>>  + if (dl_time_before(rq->dl.earliest_dl.curr, p->dl.deadline) &&
>>  +    rq->curr == p)
>>                   resched_curr(rq);
>>  +#else
>>  + /*
>>  + * Again, we don't know if p has a earlier
>>  + * or later deadline, so let's blindly set a
>>  + * (maybe not needed) rescheduling point.
>>  + */
>>  + resched_curr(rq);
>>   #endif /* CONFIG_SMP */
>>  - } else
>>  - switched_to_dl(rq, p);
>>   }
>>
>>   const struct sched_class dl_sched_class = {