Re: [RFC PATCH] sched/fair: Refactor can_migrate_task() to elimate looping

[I Hsin Cheng]

On Thu, Jan 09, 2025 at 10:46:27AM +0100, Peter Zijlstra wrote:
> On Thu, Jan 09, 2025 at 01:29:47AM +0800, I Hsin Cheng wrote:
> >  kernel/sched/fair.c | 16 ++++++++++------
> >  1 file changed, 10 insertions(+), 6 deletions(-)
> > 
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index 2d16c8545..ce46f61da 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -9404,12 +9404,16 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
> >  			return 0;
> >  
> >  		/* Prevent to re-select dst_cpu via env's CPUs: */
> > -		for_each_cpu_and(cpu, env->dst_grpmask, env->cpus) {
> > -			if (cpumask_test_cpu(cpu, p->cpus_ptr)) {
> > -				env->flags |= LBF_DST_PINNED;
> > -				env->new_dst_cpu = cpu;
> > -				break;
> > -			}
> > +		struct cpumask dst_mask;
> 
> Except you cannot put cpumask on-stack...
> 
> > +
> > +		cpumask_and(&dst_mask, env->dst_grpmask, env->cpus);
> > +		cpumask_and(&dst_mask, &dst_mask, p->cpus_ptr);
> > +
> > +		cpu = cpumask_first(&dst_mask);
> > +
> > +		if (cpu < nr_cpu_ids) {
> > +			env->flags |= LBF_DST_PINNED;
> > +			env->new_dst_cpu = cpu;
> >  		}
> >  
> >  		return 0;
> > -- 
> > 2.43.0
> >

> > Except you cannot put cpumask on-stack...

Oh I'm sorry, may I ask the reason? is it because cpumask tends to be
very large?
I assume we're not supposed to change the value of "env->dst_grpmask" or
"env->cpus" here, so in order to achieve something like this we need
another cpumask, explicitly allocate a cpumask on heap for this section
would be an overkill I think ?

Or maybe we can have kernel maintain a global cpumask so anyone wishing
to perform some operations will get to do it without maintain allocate a
cpumask themselves?