Re: [PATCH 3/4] scheduler: replace migration_thread with cpu_stop

[Peter Zijlstra <peterz@infradead.org>]

On Tue, 2010-05-04 at 09:17 +0200, Tejun Heo wrote:
> Hello,
> 
> On 05/03/2010 03:26 PM, Peter Zijlstra wrote:
> > On Thu, 2010-04-22 at 18:09 +0200, Tejun Heo wrote:
> >> @@ -2909,7 +2912,9 @@ redo:
> >>                         }
> >>                         raw_spin_unlock_irqrestore(&busiest->lock, flags);
> >>                         if (active_balance)
> >> -                               wake_up_process(busiest->migration_thread);
> >> +                               stop_one_cpu_nowait(cpu_of(busiest),
> >> +                                       active_load_balance_cpu_stop, busiest,
> >> +                                       &busiest->active_balance_work); 
> > 
> > So who guarantees busiest->active_balance_work isn't already enqueued by
> > some other cpu's load-balancer run?
> > 
> 
> Hmmm... maybe I'm mistaken but isn't that guaranteed by
> busiest->active_balance which is protected by the rq lock?
> active_load_balance_cpu_stop is scheduled iff busiest->active_balance
> was changed from zero and only active_load_balance_cpu_stop() can
> clear it at the end of its execution at which point the
> active_balance_work is safe to reuse.

Ah, indeed. It wasn't obvious from looking at the patch, but when
looking at the full code it fairly easy to see.