Re: [PATCH 04/10] nohz: Remove useless argument on tick_nohz_task_switch()

[Frederic Weisbecker]

On Mon, Aug 03, 2015 at 02:39:59PM +0200, Peter Zijlstra wrote:
> On Thu, Jul 23, 2015 at 06:42:09PM +0200, Frederic Weisbecker wrote:
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 78b4bad10..4d34035 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -2489,7 +2489,7 @@ static struct rq *finish_task_switch(struct task_struct *prev)
> >  		put_task_struct(prev);
> >  	}
> >  
> > -	tick_nohz_task_switch(current);
> > +	tick_nohz_task_switch();
> >  	return rq;
> >  }
> 
> OK, so I just noticed this one. Please explain? WTF does it make sense
> to have per task tick state?
> 
> If we have >1 tasks, we need the tick. If we have 1 task, per-task ==
> per-cpu.
> 
> So what gives?

tick_nohz_task_switch() is removed in the end of the patchset. There is no
more per task dependency after this specific set.

Now we just had a talk with Chris about Posix cpu timers whose tick dependency
turned from per-task/per-process to global dependency. When a posix timer is
queued somewhere, we keep the tick everywhere (still talking about this patchset,
not linus tree).

Since some nohz full users may still want to run posix timers on housekeepers,
I'll turn it to a per task dependency again. But this context switch call will
still be removed and the scheduler tick dependency stays a per-cpu dependency
and not a task one.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/