Re: [REPOST-v2] sched: Prevent wakeup to enter critical section needlessly

[Oleg Nesterov <oleg@redhat.com>]

On 11/21, Ivo Sieben wrote:
> Hi
>
> 2012/11/19 Oleg Nesterov <oleg@redhat.com>:
> >
> > Because on a second thought I suspect this change is wrong.
> >
> > Just for example, please look at kauditd_thread(). It does
> >
> >         set_current_state(TASK_INTERRUPTIBLE);
> >
> >         add_wait_queue(&kauditd_wait, &wait);
> >
> >         if (!CONDITION)         // <-- LOAD
> >                 schedule();
> >
> > And the last LOAD can leak into the critical section protected by
> > wait_queue_head_t->lock, and it can be reordered with list_add()
> > inside this critical section. In this case we can race with wake_up()
> > unless it takes the same lock.
> >
> > Oleg.
> >
>
> I agree that I should solve my problem using the waitqueue_active()
> function locally. I'll abandon this patch and fix it in the
> tty_ldisc.c.
>
> But we try to understand your fault scenario: How can the LOAD leak
> into the critical section? As far as we understand the spin_unlock()
> function also contains a memory barrier
                           ^^^^^^^^^^^^^^

Not really, in general unlock is a one-way barrier.

> to prevent such a reordering
> from happening.

Please look at the comment above prepare_to_wait(), for example. Or
look at wmb() in try_to_wake_up().

I guess this is not possible on x86, but in general

	X;
	LOCK();
	UNLOCK();
	Y;

can be reordered as

	LOCK();
	Y;
	X;
	UNLOCK();

UNLOCK + LOCK is the full memory barrier.

Oleg.