Re: [PATCH v6 3/5] sched: Enforce user requested affinity

[Waiman Long]

On 8/31/22 05:21, Peter Zijlstra wrote:
> On Wed, Aug 31, 2022 at 11:18:22AM +0200, Peter Zijlstra wrote:
> > On Thu, Aug 25, 2022 at 09:01:17PM -0400, Waiman Long wrote:
> >
> > > @@ -9352,6 +9381,11 @@ int sched_cpu_activate(unsigned int cpu)
> > >   		sched_update_numa(cpu, true);
> > >   		sched_domains_numa_masks_set(cpu);
> > >   		cpuset_cpu_active();
> > > +		/*
> > > +		 * Preallocated scratch cpumask
> > > +		 */
> > > +		if (!rq->scratch_mask)
> > > +			rq->scratch_mask = kmalloc(cpumask_size(), GFP_KERNEL);
> > >   	}
> > this is too late; I think you'll have to add a sched_cpu_prepare() and
> > simply fail the cpu-up when the allocation fails.
> Alternatively, waste some more memory and add yet another per-cpu
> cpumask.

A percpu cpumask is probably a better idea. However, I don't need that 
as early as the other cpumasks like __cpu_possible_mask. Maybe I can do 
a percpu memory allocation early in the pre-smp boot.

Cheers,
Longman