Re: [RFC][PATCH 16/17] sched: Move the second half of ttwu() to the remote cpu

[Oleg Nesterov <oleg@redhat.com>]

On 01/04, Peter Zijlstra wrote:
>
> On Tue, 2011-01-04 at 15:28 +0100, Oleg Nesterov wrote:
> > On 12/24, Peter Zijlstra wrote:
> > >
> > > +static void
> > > +ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags)
> > > +{
> > > +#ifdef CONFIG_SMP
> > > +	if (task_cpu(p) != cpu_of(rq))
> > > +		set_task_cpu(p, cpu_of(rq));
> > > +#endif
> >
> > This looks a bit suspicious.
> >
> > If this is called by sched_ttwu_pending() we are holding rq->lock,
> > not task_rq_lock(). It seems, we can race with, say, migration
> > thread running on task_cpu().
>
> I don't think so, nobody should be migrating a TASK_WAKING task.

I am not sure...

Suppose that p was TASK_INTERRUPTIBLE and p->on_rq == 1 before, when
set_cpus_allowed_ptr() was called. To simplify, suppose that
the caller is preempted right after it drops p->pi_lock and before
it does stop_one_cpu(migration_cpu_stop).

After that p can complete chedule() and deactivate itself.

Now, try_to_wake_up() can set TASK_WAKING, choose another CPU,
and do ttwu_queue_remote().

Finally, the caller of set_cpus_allowed_ptr() resumes and
schedules migration_cpu_stop.



It is very possible I missed something, but what is the new
locking rules for set_task_cpu() anyway? I mean, which rq->lock
it needs?

Oleg.