Re: [RFC][PATCH 5/5] sched: Reduce ttwu rq->lock contention

[Peter Zijlstra <peterz@infradead.org>]

On Thu, 2010-12-16 at 20:03 +0100, Peter Zijlstra wrote:
> On Thu, 2010-12-16 at 19:58 +0100, Peter Zijlstra wrote:
> > > > +    * Since we've already set TASK_WAKING this task's CPU cannot
> > > > +    * change from under us.
> > > 
> > > I think it can. Yes, we've set TASK_WAKING. But, at least the task
> > > itself can change its state back to TASK_RUNNING without calling
> > > schedule. Say, __wait_event()-like code.
> > 
> > Oh crud, you're right, that's going to make all this cmpxchg stuff lots
> > more interesting :/ 
> 
> Hrmph, we can add a task_is_waking() test to the rq->lock in schedule(),
> like we have for __task_rq_lock():
> 
>   local_irq_save(flags);
> again:
>   while (task_is_waking(current))
> 	cpu_relax();
>   raw_spin_lock(rq->lock);
>   if (task_is_waking(current)) {
> 	raw_spin_unlock(rq->lock);
> 	goto again;
>   }
> 
> 
> But that's not particularly pretty...

OK, I'm just not thinking straight here, if we're not passing through
schedule() the above clearly won't help any...

Damn, that's a nasty case to solve...

  current->state = TASK_UNINTERRUPTIBLE;
  while (!true)
    schedule();
  current->state = TASK_RUNNING;

... /me goes make a strong pot of tea