On Wed, 2012-07-11 at 17:11 -0700, Paul Turner wrote:
> >>       for_each_leaf_cfs_rq(rq, cfs_rq) {
> >> +             __update_blocked_averages_cpu(cfs_rq->tg, rq->cpu);
> >>
> >> +             /*
> >> +              * Periodically release the lock so that a cfs_rq with many
> >> +              * children cannot hold it for an arbitrary period of time.
> >> +              */
> >> +             if (num_updates++ % 20 == 0) {
> >> +                     raw_spin_unlock_irqrestore(&rq->lock, flags);
> >> +                     cpu_relax();
> >> +                     raw_spin_lock_irqsave(&rq->lock, flags);
> >
> > Gack.. that's not real pretty is it.. Esp. since we're still holding RCU
> > lock and are thus (mostly) still not preemptable.
> >
> > How much of a problem was this?, the changelog is silent on this.
> 
> So the holding of RCU about these operations is nothing new (and
> indeed they should be much faster than before).
> 
> As above, the bound is only for the crazy-large-numbers of cgroups
> case where we don't want to sit on with interrupts disabled forever.
> I suspect it wants to be larger, but picked a fairly conservative
> number to start with since I also think it's not a big performance
> factor either way.

How about you leave this ugly out for now (its unrelated to the rest of
the changes anyway).. and we can revisit this later?