Re: [PATCH v3 3/4] sched/fair: Allow load balancing between CPUs of identical capacity

[Ricardo Neri]

On Fri, May 15, 2026 at 11:16:00PM +0800, Chen, Yu C wrote:
> On 5/15/2026 2:34 AM, Ricardo Neri wrote:
> > sched_balance_find_src_rq() avoids selecting a runqueue with a single
> > running task as busiest if doing so results in migrating the task to a
> > CPU with less than ~5% of extra capacity. It also unintentionally
> > prevents migrations between CPUs of identical capacity.
> > 
> > When CONFIG_SCHED_CLUSTER is enabled, load should be balanced across
> > clusters of CPUs with the same capacity. Allowing migration between CPUs
> > of identical capacity is necessary to meet this goal.
> > 
> > Use arch_scale_cpu_capacity() to reflect architectural capacity, excluding
> > runtime reductions due to side activity or thermal pressure. Guard this
> > check with the sched_cluster_active static key so that systems without
> > cluster topology are unaffected.
> > 
> > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
> > ---
> > Changes in v3:
> >   * Reverted the inverted capacity check; the inverted form incorrectly
> >     allows migrations to CPUs of slightly less capacity.
> >   * Guarded the check for architectural capacity with the
> >     sched_cluster_active static key.
> > 
> > Changes in v2:
> >   * Used arch_scale_cpu_capacity() instead of capacity_of() to ignore
> >     runtime variability.
> >   * Inverted the check for runtime capacity. (Christian)
> >   * Reworded patch description for clarity.
> > ---
> >   kernel/sched/fair.c | 6 ++++++
> >   1 file changed, 6 insertions(+)
> > 
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index dcc02ceb44b5..d2a4c529f67f 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -11846,8 +11846,14 @@ static struct rq *sched_balance_find_src_rq(struct lb_env *env,
> >   		 * eventually lead to active_balancing high->low capacity.
> >   		 * Higher per-CPU capacity is considered better than balancing
> >   		 * average load.
> > +		 *
> > +		 * CONFIG_SCHED_CLUSTER requires balancing load across clusters
> > +		 * of identical capacity. Use architectural capacity to ignore
> > +		 * runtime variability.
> >   		 */
> >   		if (env->sd->flags & SD_ASYM_CPUCAPACITY &&
> > +		    (!static_branch_unlikely(&sched_cluster_active) ||
> > +		     arch_scale_cpu_capacity(env->dst_cpu) != arch_scale_cpu_capacity(i)) &&
> >   		    !capacity_greater(capacity_of(env->dst_cpu), capacity) &&
> 
> As stated in the commit log, the existing logic blocks task migrations
> between CPUs
> with identical capacity, which is based on capacity_of() comparison rather
> than
> arch_scale_cpu_capacity.

Thanks for your review!

> Could I kindly ask why replacing
> !capacity_greater(capacity_of(env->dst_cpu), capacity)

This check only allows balance if dst_cpu has more than 1.05x runtime
capacity than the CPU under consideration (i in the loop in this function).

> with
> capacity_greater(capacity, capacity_of(env->dst_cpu))
> does not achieve the expected effect?

The problem with inverting the check is that it would allow balance if
dst_cpu has more than 0.95x runtime capacity than the CPU under
consideration, effectively allowing migrations to lower-capacity CPUs.

> This would theoretically enable migration among equal-capacity CPUs, and
> in most cases e-cores in different clusters should return 0 thus
> load balance is allowed.

Yes, it would meet the condition you describe, but also has the side
effect described above.

For this reason I decided to target the specific case of identical
architectural capacity.


> 
> thanks,
> Chenyu
>