Re: [PATCH 08/14] hrtimer: Allow hrtimer::function() to free the timer

[Peter Zijlstra]

On Mon, Jun 08, 2015 at 04:27:49PM +0200, Oleg Nesterov wrote:
> On 06/08, Peter Zijlstra wrote:
> >
> > On Mon, Jun 08, 2015 at 11:14:17AM +0200, Peter Zijlstra wrote:
> > > > Finally. Suppose that timer->function() returns HRTIMER_RESTART
> > > > and hrtimer_active() is called right after __run_hrtimer() sets
> > > > cpu_base->running = NULL. I can't understand why hrtimer_active()
> > > > can't miss ENQUEUED in this case. We have wmb() in between, yes,
> > > > but then hrtimer_active() should do something like
> > > >
> > > > 	active = cpu_base->running == timer;
> > > > 	if (!active) {
> > > > 		rmb();
> > > > 		active = state != HRTIMER_STATE_INACTIVE;
> > > > 	}
> > > >
> > > > No?
> > >
> > > Hmm, good point. Let me think about that. It would be nice to be able to
> > > avoid more memory barriers.
> 
> Yes, but otoh, can't we avoid seqcount_t altogether?
> 
> To remind, we assume that
> 
> 	- "false positive" is fine. If we observe ENQUEUED or ->running
> 	  we can safely return true. It doesn't matter if the timer becomes
> 	  "inactive" right after return.
> 
> 	- we need to fix migrate_hrtimer_list() and __hrtimer_start_range_ns()
> 	  to preserve ENQUEUED. This fixes the races with hrtimer_is_queued()
> 	  and hrtimer_active() we currently have.
> 
> Now, can't we simply do
> 
> 	__run_hrtimer()
> 	{
> 
> 		cpu_base->running = timer;
> 
> 		wmb();				// 1
> 
> 		__remove_hrtimer(INACTIVE);	// clears ENQUEUED
> 
> 		fn();				// autorearm can set ENQUEUED again
> 
> 		wmb();				// 2
> 
> 		cpu_base->running = NULL;	// XXX
> 	}
> 
> 	hrtimer_active(timer)
> 	{
> 		if (timer->state & ENQUEUED)
> 			return true;
> 
> 		rmb();				// pairs with 1
> 
> 
> 		// We do not care if we race with __hrtimer_start_range_ns().
> 		// The running timer can't change its base.
> 		// If it was ENQUEUED, we rely on the previous check.
> 
> 		base = timer->base->cpu_base;
> 		read_barrier_depends();
> 		if (base->running == timer)
> 			return true;
> 
> 		rmb();				// pairs with 2
> 
> 		// Avoid the race with auto-rearming timer. If we see the
> 		// result of XXX above we should also see ENQUEUED if it
> 		// was set by __run_hrtimer() or timer->function().
> 		//
> 		// We do not care if another thread does hrtimer_start()
> 		// and we miss ENQUEUED. In this case we can the "inactive"
> 		// window anyway, we can pretend that hrtimer_start() was
> 		// called after XXX above. So we can equally pretend that
> 		// hrtimer_active() was called in this window.
> 		//
> 		if (timer->state & ENQUEUED)
> 			return true;
> 
> 		return false;
> 	}
> 
> Most probably I missed something... I'll try to think more, but perhaps
> you see a hole immediately?

This is something I proposed earlier; Kirill said:

  lkml.kernel.org/r/2134411433408823@xxxxxxxxxxxxxxx

Which I read like the below, imagine our timer expires periodically and
rearms itself:

 acquire
 cpu_base->running = timer;
 wmb
 timer->state = INACTIVE;
 release
				[R] timer->state (== INACTIVE)
 fn()
 acquire
 timer->state = ACTIVE
 wmb
 cpu_base->running = NULL
 release

				[R] cpu_base->running (== NULL)

 acquire
 cpu_base->running = timer;
 wmb
 timer->state = INACTIVE;
 release

				[R] timer->state (== INACTIVE)
 fn()
 acquire
 timer->state = ACTIVE
 wmb
 cpu_base->running = NULL
 release


And we have a false negative.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/