Re: RFC [patch] sched,cgroup_sched: convince RT_GROUP_SCHED throttle to work

[Mike Galbraith <efault@gmx.de>]

On Sun, 2012-04-15 at 05:37 +0200, Mike Galbraith wrote: 
> On Sat, 2012-04-14 at 13:10 +0200, Peter Zijlstra wrote: 
> > On Tue, 2012-04-10 at 11:08 +0200, Mike Galbraith wrote:
> >  
> > > --- a/kernel/sched/rt.c
> > > +++ b/kernel/sched/rt.c
> > > @@ -782,6 +782,19 @@ static int do_sched_rt_period_timer(stru
> > >  	const struct cpumask *span;
> > >  
> > >  	span = sched_rt_period_mask();
> > > +#ifdef CONFIG_RT_GROUP_SCHED
> > > +	/*
> > > +	 * FIXME: isolated CPUs should really leave the root task group,
> > 
> > No no, that's the wrong fix, the right fix is to remove isolcpus :-)
> 
> Yeah, isolcpus needs to die, but...
> 
> > I guess the alternative 'fix' is to not account the rt_runtime on
> > isolated cpus.. does something like the below actually work?
> 
> I haven't tried it, because the exact same thing happens when you
> isolate via cpusets directly below root.  One timer, two (or more)
> rd->span, so _somebody_ is screwed.

You _could_ bail on !rq->sd I suppose, but the way I hacked around it,
the user can keep the throttle for testing/troubleshooting their
isolated setup, and turn it off in production.  OTOH, auto throttle
disable for all isolated sets could work just as well.

-Mike