Re: [PATCH] Document that wake_up(), complete() and co. imply a full memory barrier

[Oleg Nesterov <oleg@redhat.com>]

On 04/22, Ingo Molnar wrote:
>
> * David Howells <dhowells@redhat.com> wrote:
>
> The dangerous pattern is lockless code doing wakeups. But lockless
> code always has to use proper barriers or atomics anyway, and has to
> be aware of the fact that kernel primitives they call are not
> necessarily full memory barriers.
>
> In fact i'd encourage to _not_ document try_to_lock() as a write
> barrier either - but rather have explicit barriers where they are
> needed. Then we could remove that barrier from try_to_wake_up() too
> ;-)

Well. I think in that case almost every try_to_wake_up/wake_up_process
needs mb().

For example:

	do_nanosleep:

		for (;;) {
			set_current_state(TASK_INTERRUPTIBLE);
			if (likely(t->task))
				schedule();
			break;
		}

	hrtimer_wakeup:

		task = t->task
		t->task = NULL;
		wake_up_process(task);

If try_to_wake_up() has no the barrier semantics, we can miss the event.
"t->task = NULL" and the reading of task->state must not be reordered.

Oleg.