Re: [PATCH 1/3] wait-simple: Introduce the simple waitqueue implementation

[Peter Zijlstra <peterz@infradead.org>]

On Thu, Dec 12, 2013 at 12:10:15PM -0500, Steven Rostedt wrote:
> So if we break up your code above, we have:
> 
> 	raw_spin_lock_irqsave(&head->lock, flags);
> 	w->task = current;
> 	if (list_empty(&w->node)) {
> 		list_add(&w->node, &head->list);
> 		smp_mb();
> 	}
> 	__set_current_state(state);
> 	raw_spin_unlock_irqrestore(&head->lock, flags);
> 
> 	if (!cond)
> 		schedule();
> 

the unlock is semi-permeable and would allow the cond test to cross over
and even be satisfied before the state write.

> 
> vs
> 
> 	cond = true;
> 
> 	raw_spin_lock_irqsave(&head->lock, flags);
> 	woken = __swait_wake_locked(head, state, num);
> 	raw_spin_unlock_irqrestore(&head->lock, flags);

Same here, the lock is semi-permeable and would allow the cond store to
leak down.

In the first case we really need the implied mb of set_current_state(),
the the second case the actual wakeup would still provide the required
barrier.