Re: [Patch v4 18/22] sched/cache: Enable cache aware scheduling for multi LLCs NUMA node

[Tim Chen]

On Thu, 2026-04-09 at 15:37 +0200, Peter Zijlstra wrote:
> On Wed, Apr 01, 2026 at 02:52:30PM -0700, Tim Chen wrote:
> 
> > diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> > index 995a42cb4697..0b1fc1b0709d 100644
> > --- a/kernel/sched/topology.c
> > +++ b/kernel/sched/topology.c
> > @@ -809,6 +809,7 @@ enum s_alloc {
> >  };
> >  
> >  #ifdef CONFIG_SCHED_CACHE
> > +DEFINE_STATIC_KEY_FALSE(sched_cache_present);
> >  static bool alloc_sd_llc(const struct cpumask *cpu_map,
> >  			 struct s_data *d)
> >  {
> > @@ -2674,6 +2675,7 @@ static int
> >  build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *attr)
> >  {
> >  	enum s_alloc alloc_state = sa_none;
> > +	bool has_multi_llcs = false;
> >  	struct sched_domain *sd;
> >  	struct s_data d;
> >  	struct rq *rq = NULL;
> > @@ -2784,10 +2786,12 @@ build_sched_domains(const struct cpumask *cpu_map, struct sched_domain_attr *att
> >  				 * between LLCs and memory channels.
> >  				 */
> >  				nr_llcs = sd->span_weight / child->span_weight;
> > -				if (nr_llcs == 1)
> > +				if (nr_llcs == 1) {
> >  					imb = sd->span_weight >> 3;
> > -				else
> > +				} else {
> >  					imb = nr_llcs;
> > +					has_multi_llcs = true;
> > +				}
> >  				imb = max(1U, imb);
> >  				sd->imb_numa_nr = imb;
> >  
> 
> Well that all don't apply :/
> 
> Also, this seems sub-optimal, we can have multi-llc without NUMA.
> 
> Doesn't the LLC domain having a parent more or less imply there being
> multiple and we should be doing this, irrespective of having NUMA?

If we have NUMA and single LLC, original thought was that NUMA balancing
would be taking care of moving the tasks to preferred node. And that's
why we thought it is not necessary to enable cache aware scheduling for
that case. Yes, if there's no NUMA and we have a parent group for LLCs,
agree that we should enable it.

Tim