Re: [PATCH v2 5/7] sched: Implement shared runqueue in CFS

[David Vernet]

On Tue, Jul 11, 2023 at 12:18:32PM +0200, Peter Zijlstra wrote:
> On Mon, Jul 10, 2023 at 03:03:40PM -0500, David Vernet wrote:
> 
> > diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> > index 1451f5aa82ac..3ad437d4ea3d 100644
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> 
> > @@ -9842,6 +9843,7 @@ void __init sched_init_smp(void)
> >  
> >  	init_sched_rt_class();
> >  	init_sched_dl_class();
> > +	init_sched_fair_class_late();
> >  
> >  	sched_smp_initialized = true;
> >  }
> 
> > @@ -12854,3 +12999,34 @@ __init void init_sched_fair_class(void)
> >  #endif /* SMP */
> >  
> >  }
> > +
> > +__init void init_sched_fair_class_late(void)
> > +{
> > +#ifdef CONFIG_SMP
> > +	int i;
> > +	struct shared_runq *shared_runq;
> > +	struct rq *rq;
> > +	struct rq *llc_rq;
> > +
> > +	for_each_possible_cpu(i) {
> > +		if (per_cpu(sd_llc_id, i) == i) {
> > +			llc_rq = cpu_rq(i);
> > +
> > +			shared_runq = kzalloc_node(sizeof(struct shared_runq),
> > +					       GFP_KERNEL, cpu_to_node(i));
> > +			INIT_LIST_HEAD(&shared_runq->list);
> > +			spin_lock_init(&shared_runq->lock);
> > +			llc_rq->cfs.shared_runq = shared_runq;
> > +		}
> > +	}
> > +
> > +	for_each_possible_cpu(i) {
> > +		rq = cpu_rq(i);
> > +		llc_rq = cpu_rq(per_cpu(sd_llc_id, i));
> > +
> > +		if (rq == llc_rq)
> > +			continue;
> > +		rq->cfs.shared_runq = llc_rq->cfs.shared_runq;
> > +	}
> > +#endif /* SMP */
> > +}
> 
> I don't think this is right; the llc_id thing depends on the online
> mask, not on possible mask. So if you boot with a number of CPUs offline
> and late bring them online, you're screwy (IIRC this is actually a very
> common thing in virt land).
> 
> Additionally, llc_id depends on the sched_domain tree, if someone were
> to go create partitions, they can split an LLC and llc_id would split
> right along with it.
> 
> I think you need to move this into sched/topology.c and handle hotplug /
> domain (re) creation.

Yeah, you're right. This falls apart if we hotplug when we do domain
recreation. I'll address this in v3.

> And yes, that's going to be a pain, because you might need to re-hash
> existing lists.

Eh, it needs to be done. I played around with this for a bit before
sending the v1 RFC code but ended up keeping the series simple because
it was RFC, and fixing this is pretty involved. I should have taken care
of it regardless before dropping the RFC tag, so glad you pointed it
out.