Re: [PATCH v2 05/23] sched/cache: Assign preferred LLC ID to processes

[Chen, Yu C]

On 1/7/2026 12:49 PM, Jianyong Wu wrote:
> Hi Tim,
>
> On 12/4/2025 7:07 AM, Tim Chen wrote:
> > +
> > +    if (mm->mm_sched_cpu != -1) {
> > +        mm_sched_llc = llc_id(mm->mm_sched_cpu);
> > +
> > +#ifdef CONFIG_NUMA_BALANCING
> > +        /*
> > +         * Don't assign preferred LLC if it
> > +         * conflicts with NUMA balancing.
> > +         */
> > +        if (p->numa_preferred_nid >= 0 &&
> > +            cpu_to_node(mm->mm_sched_cpu) != p->numa_preferred_nid)
> > +            mm_sched_llc = -1;
>
> Scenario: NUMA balancing is initially enabled and numa_preferred_nid is 
> set. After disabling NUMA balancing, numa_preferred_nid is not cleared, 
> leading to unintended impact of NUMA balancing on cache-aware 
> scheduling. Should we check if NUMA balancing is disabled before 
> checking numa_preferred_nid?

Good point, sched_numa_balancing should be checked here.

thanks,
Chenyu

> Thanks
> Jianyong Wu