On Mon, 27 Nov 2006, Oleg Nesterov wrote:
> I still can't relax, another attempt to "prove" this should not be
> possible on CPUs supported by Linux :)
>
> Let's suppose it is possible, then it should also be possible if CPU_1
> does spin_lock() instead of mb() (spin_lock can't be "stronger"), yes?
>
> Now,
>
> int COND;
> wait_queue_head_t wq;
>
> my_wait()
> {
> add_wait_queue(&wq);
> for (;;) {
> set_current_state(TASK_UNINTERRUPTIBLE);
>
> if (COND)
> break;
>
> schedule();
> }
> remove_wait_queue(&wq);
> }
>
> my_wake()
> {
> COND = 1;
> wake_up(&wq);
> }
>
> this should be correct, but it is not!
>
> my_wait:
>
> task->state = TASK_UNINTERRUPTIBLE; // STORE
>
> mb();
>
> if (COND) break; // LOAD
>
>
> my_wake:
>
> COND = 1; // STORE
>
> spin_lock(WQ.lock);
> spin_lock(runqueue.lock);
>
> // try_to_wake_up()
> if (!(task->state & TASK_UNINTERRUPTIBLE)) // LOAD
> goto out;
>
>
> So, my_wait() gets COND == 0, and goes to schedule in 'D' state.
> try_to_wake_up() reads ->state == TASK_RUNNING, and does nothing.
This is a very good point. I don't know what the resolution is; Paul will
have to explain the situation.
Alan
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
