Re: [RFC v4] wait: prevent waiter starvation in __wait_on_bit_lock

From: Dmitry Adamushko
Date: Thu Jan 22 2009 - 19:26:40 EST

2009/1/22 Oleg Nesterov <oleg@xxxxxxxxxx>:
> On 01/21, Johannes Weiner wrote:
>> @@ -187,6 +187,31 @@ __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
>> }
>> } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
>> finish_wait(wq, &q->wait);
>> + if (unlikely(ret)) {
>> + /*
>> + * Contenders are woken exclusively. If we were woken
>> + * by an unlock we have to take the lock ourselves and
>> + * wake the next contender on unlock. But the waiting
>> + * function failed, we do not take the lock and won't
>> + * unlock in the future. Make sure the next contender
>> + * does not wait forever on an unlocked bit.
>> + *
>> + * We can also get here without being woken through
>> + * the waitqueue, so there is a small chance of doing a
>> + * bogus wake up between an unlock clearing the bit and
>> + * the next contender being woken up and setting it again.
>> + *
>> + * It does no harm, though, the scheduler will ignore it
>> + * as the process in question is already running.
>> + *
>> + * The unlock path clears the bit and then wakes up the
>> + * next contender. If the next contender is us, the
>> + * barrier makes sure we also see the bit cleared.
>> + */
>> + smp_rmb();
>> + if (!test_bit(q->key.bit_nr, q->key.flags)))
>> + __wake_up_bit(wq, q->key.flags, q->key.bit_nr);
> I think this is correct, and (unfortunately ;) you are right:
> we need rmb() even after finish_wait().

Hum, I think it's actually not necessary in this particular case when
(1) "the next contender is us" and (2) we are in the "ret != 0" path
so that the only thing we really care about -- if we were exclusivly
woken up, then wake up somebody else [*].

"the next contender is us" implies that we were still on the 'wq'
queue when __wake_up_bit() -> __wake_up() has been called, meaning
that wq->lock has also been taken (in __wake_up()).

Now, on our side, we are definitely on the 'wq' queue before calling
finish_wait(), meaning that we also take the wq->lock.

In short, wq->lock is a sync. mechanism in this case. The scheme is as follows:

our side:

[ finish_wait() ]

delete us from the 'wq'

test_bit() [ read a bit ]

waker's side:

smp_mb__after_clear_bit() --- is a must to ensure that we fetch the
'wq' (and do a waitqueue_active(wq) check) in __wake_up_bit() _only_
after clearing the bit.

[ __wake_up_bit(); -> __wake_up() ] --> we are on the 'wq' (see conditions [*])
wake 'us' up here

Now the point is, without smp_rmb() on the side of wait_on_bit(),
test_bit() [ which is a LOAD op ]can get _into_ the wq->lock section,
smth like this:

[ finish_wait() ]

test_bit() [ read a bit ]
delete us from the 'wq'

If (1) is true (we were woken up indeed), it means that __wake_up()
(from __wake_up_bit()) has been executed before we were able to enter

By the moment __wake_up_bit() was executed (we were woken up), the bit
was already cleared -- that's guaranteed by a full MB on the
wake_up_bit side (in our case [*] wq->lock would do it even without
the MB) -> meaning that we don't miss !test_bit() int this particular
case [*].

p.s. if the explanation is vague or heh even wrong, it's definitely
due to the lack of sleep ;-))

> For example, don't we have the similar problems with
> wait_event_interruptible_exclusive() ?

yes, I think so.

> Oleg.

Best regards,
Dmitry Adamushko
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at