Re: [RFC] Repeated rto_push_irq_work_func() invocation.

[Steven Rostedt]

On Wed, 2 Oct 2024 17:05:43 +0200
Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote:

> On 2024-10-02 10:37:49 [-0400], Steven Rostedt wrote:
> > On Wed, 2 Oct 2024 13:21:05 +0200
> > Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx> wrote:
> >   
> > > Would it make sense to avoid scheduling rto_push_work if rq->curr has
> > > NEED_RESCHED set and make the scheduler do push_rt_task()?  
> > 
> > Can we safely check rq->curr without taking any locks on that CPU?
> > 
> > I guess tasks do not get freed while a CPU has preemption disabled, so it
> > may be safe to do:
> > 
> > 	task = READ_ONCE(rq->curr);
> > 	if (test_task_need_resched(task))
> > 		/* skip */
> > 
> > ??  
> 
> +rcu_read_lock() but yes. This would be one part. You need to check if

Yeah, preempt_disable() is pretty much equivalent today to rcu_read_lock(),
as synchronize_rcu() and synchronize_sched() have been merged into one.

> the task on pull-list has lower priority than the current task on rq.
> This would be need to be moved to somewhere in schedule() probably after
> pick_next_task().

Does it matter? The target CPU has NEED_RESCHED set, which should handle
the pushing logic anyway, right?

Hmm, I probably should look deeper to make sure that anytime a schedule
happens where there's overloaded RT tasks, it tries to push.

-- Steve