Re: [PATCH v2 1/2] sched: schedule_tail() should disable preemption

[Oleg Nesterov <oleg@redhat.com>]

On 10/09, Peter Zijlstra wrote:
>
> On Thu, Oct 09, 2014 at 07:28:34PM +0200, Peter Zijlstra wrote:
> > On Thu, Oct 09, 2014 at 06:57:13PM +0200, Oleg Nesterov wrote:
> > > @@ -2333,10 +2336,12 @@ context_switch(struct rq *rq, struct task_struct *prev,
> > >  #endif
> > >
> > >  	context_tracking_task_switch(prev, next);
> > > +
> > > +	pc = preempt_count();
> >
> > The only problem here is that you can loose PREEMPT_NEED_RESCHED, I
> > haven't thought about whether that is a problem here or not.

No, it must not be set. It was cleared by clear_preempt_need_resched()
in __schedule(), and nobody can set it.

(schedule_tail() also relies on fact it runs with irqs disabled).

But,

> Also, if you make that preempt_count_save(), to mirror the restore,

Yes, agreed. That is why I said we probably want preempt_count_raw()
(or _save). This way we avoid the unnecessary "&= ~PREEMPT_NEED_RESCHED"
on x86.

> can both preserve PREEMPT_NEED_RESCHED and avoid emitting code on !x86.

This too, although gcc should optimize this code out anyway. At least
it seems to do on x86 if I make preempt_count_restore() empty.

Oleg.