Re: [PATCH 1/2] sched: Make select_task_rq() require cpu_active() for user tasks

[Paul Burton]

Hi Peter,

On Mon, May 28, 2018 at 04:49:24PM +0200, Peter Zijlstra wrote:
> On Sat, May 26, 2018 at 08:46:47AM -0700, Paul Burton wrote:
> 
> > --- a/kernel/sched/core.c
> > +++ b/kernel/sched/core.c
> > @@ -1565,7 +1565,8 @@ int select_task_rq(struct task_struct *p, int cpu, int sd_flags, int wake_flags)
> >  	 *   not worry about this generic constraint ]
> >  	 */
> >  	if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
> > -		     !cpu_online(cpu)))
> > +		     !cpu_online(cpu) ||
> > +		     (!cpu_active(cpu) && !(p->flags & PF_KTHREAD))))
> >  		cpu = select_fallback_rq(task_cpu(p), p);
> 
> That is not quite right.. and I find that the wrong patch:
> 
>   955dbdf4ce87 ("sched: Allow migrating kthreads into online but inactive CPUs")
> 
> got merged over my suggested alternative :-(
> 
>   http://lkml.kernel.org/r/20170725165821.cejhb7v2s3kecems@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
> 
> So, lets first fix that

Thanks for the link - just knowing that the intention is that only
per-CPU kthreads are allowed on !active CPUs is useful.

> , and then your patch becomes something like the below I think.
> 
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -1562,7 +1562,7 @@ int select_task_rq(struct task_struct *p
>  	 *   not worry about this generic constraint ]
>  	 */
>  	if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) ||
> -		     !cpu_online(cpu)))
> +		     (is_per_cpu_kthread(p) ? !cpu_online(cpu) : !cpu_active(cpu)))
>  		cpu = select_fallback_rq(task_cpu(p), p);
>  
>  	return cpu;

Yes this looks good to me.

Are you planning to submit your change to introduce
is_per_cpu_kthread(), or shall I?

Thanks,
    Paul