Re: [PATCH] sched/deadline: Remove unnecessary condition in push_dl_task()

[Juri Lelli]

On 15/02/17 09:25, Steven Rostedt wrote:
> On Wed, 15 Feb 2017 10:47:49 +0000
> Juri Lelli <juri.lelli@xxxxxxx> wrote:
> 
> > [+Steve, Luca]
> > 
> > Hi,
> > 
> > On 15/02/17 14:11, Byungchul Park wrote:
> > > Once pick_next_pushable_dl_task(rq) return a task, it guarantees that
> > > the task's cpu is rq->cpu, so task_cpu(next_task) is always rq->cpu if
> > > task == next_task. Remove a redundant condition and make code simpler.
> > > 
> > > Signed-off-by: Byungchul Park <byungchul.park@xxxxxxx>
> > > ---
> > >  kernel/sched/deadline.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> > > index 27737f3..ad8d577 100644
> > > --- a/kernel/sched/deadline.c
> > > +++ b/kernel/sched/deadline.c
> > > @@ -1483,7 +1483,7 @@ static int push_dl_task(struct rq *rq)
> > >  		 * then possible that next_task has migrated.
> > >  		 */
> > >  		task = pick_next_pushable_dl_task(rq);
> > > -		if (task_cpu(next_task) == rq->cpu && task == next_task) {
> > > +		if (task == next_task) {  
> > 
> > Seems a sensible optimization to me. Actually, we are doing the same for
> > rt.c; Steve, Peter, do you think we should optimize that as well?
> > 
> 
> Are we doing the same for push_rt_task()? I don't see it, and I don't
> see it in tip/sched/core either. What I have is:
> 
>   if (task_cpu(next_task) == rq->cpu && task == next_task) {

Sorry, bad wording on my side. I meant the are currently checking the
same conditions both for DL and for RT, and we should probably optimize
RT as well if we are going to take this patch.

> 
> But that said, I believe this patch is correct, and we should change
> rt.c as well.
> 

That's what I meant. :)

> 
>   task = pick_next_pushable_dl_task(rq);
> 
> Which has:
> 
>   BUG_ON(rq->cpu != task_cpu(task))
> 
> when it returns a task other than NULL. Which means that task_cpu(task)
> must be rq->cpu. Then if task == next_task, then task_cpu(next_task)
> must be rq->cpu as well.

Right.

> 
> Reviewed-by: Steven Rostedt (VMware) <rostedt@xxxxxxxxxxx>
> 

You can also add mine

Reviewed-by: Juri Lelli <juri.lelli@xxxxxxx>

> Mind fixing rt.c if it hasn't been fixed already.
> 
> -- Steve