Re: [PATCH 1/2] sched/fair: consider hk_mask early in triggering ilb

[Shrikanth Hegde]

On 3/20/26 9:07 AM, K Prateek Nayak wrote:
> Hello Shrikanth,
>
> On 3/19/2026 12:23 PM, Shrikanth Hegde wrote:
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index b19aeaa51ebc..02cca2c7a98d 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -7392,6 +7392,7 @@ static inline unsigned int cfs_h_nr_delayed(struct rq *rq)
> >   static DEFINE_PER_CPU(cpumask_var_t, load_balance_mask);
> >   static DEFINE_PER_CPU(cpumask_var_t, select_rq_mask);
> >   static DEFINE_PER_CPU(cpumask_var_t, should_we_balance_tmpmask);
> > +static DEFINE_PER_CPU(cpumask_var_t, kick_ilb_tmpmask);
>
> nit. We can rename and reuse select_rq_mask. Wakeups happen with IRQs
> disabled and kick happens from the hrtimer handler so it should be safe
> to reuse that and save some space.
>
> Thoughts?

May be. but it could be a confusing name. sched_tmpmask?

We could similar stuff already to load_balance_mask, select_rq_mask.
So, i would prefer to keep it separate.

> [..snip..]
>
> > @@ -12715,27 +12716,41 @@ static void nohz_balancer_kick(struct rq *rq)
> >   	 */
> >   	nohz_balance_exit_idle(rq);
> >   
> > +	/* ILB considers only HK_TYPE_KERNEL_NOISE housekeeping CPUs */
> > +
> >   	if (READ_ONCE(nohz.has_blocked_load) &&
> > -	    time_after(now, READ_ONCE(nohz.next_blocked)))
> > +	    time_after(now, READ_ONCE(nohz.next_blocked))) {
> >   		flags = NOHZ_STATS_KICK;
> > +		cpumask_and(ilb_cpus, nohz.idle_cpus_mask,
> > +			    housekeeping_cpumask(HK_TYPE_KERNEL_NOISE));
> > +	}
> >   
> >   	/*
> > -	 * Most of the time system is not 100% busy. i.e nohz.nr_cpus > 0
> > -	 * Skip the read if time is not due.
> > +	 * Most of the time system is not 100% busy. i.e there are idle
> > +	 * housekeeping CPUs.
> > +	 *
> > +	 * So, Skip the reading idle_cpus_mask if time is not due.
> >   	 *
> >   	 * If none are in tickless mode, there maybe a narrow window
> >   	 * (28 jiffies, HZ=1000) where flags maybe set and kick_ilb called.
> >   	 * But idle load balancing is not done as find_new_ilb fails.
> > -	 * That's very rare. So read nohz.nr_cpus only if time is due.
> > +	 * That's very rare. So check (idle_cpus_mask & HK_TYPE_KERNEL_NOISE)
> > +	 * only if time is due.
> > +	 *
> >   	 */
> >   	if (time_before(now, nohz.next_balance))
> >   		goto out;
> >   
> > +	/* Avoid the double computation */
> > +	if (flags != NOHZ_STATS_KICK)
> > +		cpumask_and(ilb_cpus, nohz.idle_cpus_mask,
> > +			    housekeeping_cpumask(HK_TYPE_KERNEL_NOISE));
> > +
> >   	/*
> >   	 * None are in tickless mode and hence no need for NOHZ idle load
> >   	 * balancing
> >   	 */
> > -	if (unlikely(cpumask_empty(nohz.idle_cpus_mask)))
> > +	if (unlikely(cpumask_empty(ilb_cpus)))
> >   		return;
>
> We can just use the return value from the previous cpumask_and() for
> this and save on another cpumask iteration.

Makes sense. Will do.

> >   
> >   	if (rq->nr_running >= 2) {