Re: [RFC][PATCH 3/4] sched: Change sched_class::set_cpus_allowed calling context

[Peter Zijlstra <peterz@infradead.org>]

On Mon, May 18, 2015 at 10:32:16AM +0200, Peter Zijlstra wrote:
> On Mon, May 18, 2015 at 03:37:43PM +0800, pang.xunlei@zte.com.cn wrote:
> > Hi Peter,
> > 
> > With this modification, I think the pushing action in my previous patch 
> > "Check to push the task away after its affinity was changed" will not
> > be able to be implemented inside sched_class::set_cpus_allowed().
> 
> Ah, right, I knew there was a patch I needed to look at.

So basically you want to do:

+check_push:
+       if (weight > 1 &&
+           !task_running(rq, p) &&
+           !test_tsk_need_resched(rq->curr) &&
+           !cpumask_subset(new_mask, &p->cpus_allowed)) {
+               /* Update new affinity and try to push. */
+               cpumask_copy(&p->cpus_allowed, new_mask);
+               p->nr_cpus_allowed = weight;
+               push_rt_tasks(rq);
+               return true;
+       }

in set_cpus_allowed_rt(), which would not work because of us calling
put_prev_task(), which does enqueue_pushable_task() and would allow
pick_next_pushable_task() to select the current task, which would then
BUG_ON().

Note however that you already test for !task_running(), which precludes
that entire argument, because if @p is not running, we did not call
put_prev_task() etc..

So I think the above would still work; albeit it needs a comment on why
etc..