Re: [PATCH v3 08/20] sched/fair: load balance only among preferred CPUs

[Shrikanth Hegde <sshegde@linux.ibm.com>]

Hi Prateek.
Thanks for reviewing the patchset and tags.

On 6/4/26 10:50 AM, K Prateek Nayak wrote:
> Hello Shrikanth,
> 
> On 5/14/2026 8:51 PM, Shrikanth Hegde wrote:
>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
>> index a704285ac55a..0a851d4b0d7e 100644
>> --- a/kernel/sched/fair.c
>> +++ b/kernel/sched/fair.c
>> @@ -12087,6 +12087,9 @@ static int sched_balance_rq(int this_cpu, struct rq *this_rq,
>>   
>>   	cpumask_and(cpus, sched_domain_span(sd), cpu_active_mask);
>>   
>> +	/* Spread load among preferred CPUs */
>> +	cpumask_and(cpus, cpus, cpu_preferred_mask);
> 
> nit.
> 
> Instead of the onine_mask, can we keep the cpu_preferred_mask closer in sync
> with the active_mask?

I tried to make it active_mask. When there is high steal time,
a CPU marked as not-active isn't available for workload which pins
them. That would break user affinities.

> 
> That way, if we can ensure that a preferred CPU is always a subset of the
> active_mask, we can save on a couple of cpumask_and() operations by just
> looking at preferred_mask instead of both the active_mask and the
> preferred_mask.
> 

active mask used heavily today. I didn't want to merge the two.

> Are there any complications in that approach to track it via
> sched_cpu_activate() / sched_cpu_deactivate()?

Sorry i didn;t get this part. You mean track preferred mask with it?

We end up calling sched domain rebuild too if we go via sched_cpu_activate
         build_sched_domains+8
         partition_sched_domains_locked+1316
         partition_sched_domains+80
         rebuild_sched_domains_locked+332
         cpuset_handle_hotplug+1456
         sched_cpu_deactivate+608
         cpuhp_invoke_callback+504
         cpuhp_thread_fun+324
         smpboot_thread_fn+556


> 

What is active_mask guaranteed to provide today? Are there any ABI for it
or we are free to change it?