Re: [PATCH] sched: fix tsk->pi_lock isn't held when do_set_cpus_allowed()

[Peter Zijlstra]

On Tue, Aug 25, 2015 at 12:10:32PM +0200, Peter Zijlstra wrote:
> On Tue, Aug 25, 2015 at 12:05:27PM +0200, Peter Zijlstra wrote:
> > On Tue, Aug 25, 2015 at 03:59:54PM +0800, Wanpeng Li wrote:
> > > +++ b/kernel/cpuset.c
> > > @@ -2376,8 +2376,12 @@ void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask)
> > >  
> > >  void cpuset_cpus_allowed_fallback(struct task_struct *tsk)
> > >  {
> > > +	unsigned long flags;
> > > +
> > >  	rcu_read_lock();
> > > +	raw_spin_lock_irqsave(&tsk->pi_lock, flags);
> > >  	do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus);
> > > +	raw_spin_lock_irqsave(&tsk->pi_lock, flags);
> > >  	rcu_read_unlock();
> > 
> > Aside from the double lock thing that was already pointed out, I think
> > this is wrong, because the select_task_rq() call can already have
> > pi_lock held.
> > 
> > Taking it again would result in a deadlock.
> > 
> > Consider for instance:
> > 
> > try_to_wake_up()
> >   raw_spin_lock_irqsave(->pi_lock)
> >   select_task_rq()
> >     select_ballback_rq()
> >       cpuset_cpus_allowed_fallback()
> >         raw_spin_lock_irqsave(->pi_lock)
> > 
> > 
> > The problem is with the migration path and should be fixed there.
> 
> Another problem, migration_call() will have rq->lock held, so you're
> proposing to acquire pi_lock while holding rq->lock, this is an
> inversion from the regular nesting order.
> 

So Possibly, Maybe (I'm still to wrecked to say for sure), something
like this would work:

	WARN_ON(debug_locks && (lockdep_is_held(&p->pi_lock) ||
				(p->on_rq && lockdep_is_held(&rq->lock))));

Instead of those two separate lockdep asserts.

Please consider carefully.
--
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/