Re: [PATCH v10 3/5] sched: Enforce user requested affinity

[Waiman Long]

On 10/7/22 10:57, Waiman Long wrote:
> On 10/7/22 06:01, Peter Zijlstra wrote:
> > On Thu, Sep 22, 2022 at 02:00:39PM -0400, Waiman Long wrote:
> > > @@ -9647,6 +9656,9 @@ void __init sched_init(void)
> > >               cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> > >           per_cpu(select_rq_mask, i) = (cpumask_var_t)kzalloc_node(
> > >               cpumask_size(), GFP_KERNEL, cpu_to_node(i));
> > > +        per_cpu(runqueues.scratch_mask, i) =
> > > +            (cpumask_var_t)kzalloc_node(cpumask_size(),
> > > +                            GFP_KERNEL, cpu_to_node(i));
> > >       }
> > >   #endif /* CONFIG_CPUMASK_OFFSTACK */
> > That doesn't actually apply; I've made it:
> >
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -9748,6 +9748,7 @@ void __init sched_init(void)
> >             rq->core_cookie = 0UL;
> >   #endif
> > + zalloc_cpumask_var_node(&per_cpu(runqueues.scratch_mask, i), 
> > GFP_KERNEL, cpu_to_node(i));
> >       }
> >         set_load_weight(&init_task, false);
> Sorry, I should have worked on the latest tip tree instead.

To be consistent with the surround context, it may be better to change 
it to

+        zalloc_cpumask_var_node(rq->scratch_mask, GFP_KERNEL, 
cpu_to_node(i));

Cheers,
Longman