Re: [PATCH] sched/fair: Don't migrate with src_cpu == dst_cpu

[Scott Wood <swood@redhat.com>]

On Thu, 2020-12-03 at 09:47 +0100, Peter Zijlstra wrote:
> On Thu, Dec 03, 2020 at 12:04:49AM -0600, Scott Wood wrote:
> > Besides being a waste of time to try to move tasks to where they already
> > are, this avoids triggering the WARN_ON_ONCE(is_migration_disabled(p))
> > in set_task_cpu().
> > 
> > Signed-off-by: Scott Wood <swood@redhat.com>
> > ---
> > Patch is against tip/master.  Assertion was seen by running rteval on
> > the
> > RT tree.
> > 
> >  kernel/sched/fair.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index e7e21ac479a2..f443626164d4 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -7574,7 +7574,8 @@ int can_migrate_task(struct task_struct *p, struct
> > lb_env *env)
> >  
> >  		/* 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)) {
> > +			if (cpu != env->src_cpu &&
> > +			    cpumask_test_cpu(cpu, p->cpus_ptr)) {
> >  				env->flags |= LBF_DST_PINNED;
> >  				env->new_dst_cpu = cpu;
> >  				break;
> 
> Do we have _any_ clue as to how we ended up in that situation? The above
> sounds like it should be a WARN and we should avoid getting here in the
> first place.

My initial impression was that there simply wasn't anything stopping it from
happening, but digging deeper it looks like it's specific to NUMA domains
with overlapping CPUs.

-Scott