Re: [PATCH 1/1] hotplug cpu: migrate a task within its cpuset

[Nathan Lynch <ntl@pobox.com>]

Hi-

Ingo Molnar wrote:
> 
> * Cliff Wickman <cpw@sgi.com> wrote:
> 
> > With this patch, migrate the task to:
> >  1) to any cpu on the same node as the disabled cpu, which is both online
> >     and among that task's cpus_allowed
> >  2) to any online cpu within the task's cpuset
> >  3) to any cpu which is both online and among that task's cpus_allowed
> > 
> > Diffed against 2.6.21-rc3 (Andrew's current top of tree)
> 
> looks good to me.
> 
> Acked-by: Ingo Molnar <mingo@elte.hu>
> 
> > +	/* try to stay on the same cpuset */
> > +	if (dest_cpu == NR_CPUS) {
> > +		p->cpus_allowed = cpuset_cpus_allowed(p);
> > +		dest_cpu = any_online_cpu(p->cpus_allowed);
> > +	}
> 
> what's the practical effect of this - when moving the last CPU offline 
> from a node you got jobs migrated to really alien nodes? Thus i think we 
> should queue this up for v2.6.21 too, correct? It's a NOP on systems 
> that do not set up cpusets, so it's low-risk.

See my earlier reply to this patch.  Calling cpuset_cpus_allowed
(which takes a mutex) here is a bug, since move_task_off_dead_cpu must
be called with interrupts disabled.


> btw., unrelated to your patch, there's this bit right after the code 
> above:
> 
>         /* No more Mr. Nice Guy. */
>         if (dest_cpu == NR_CPUS) {
>                 rq = task_rq_lock(p, &flags);
>                 cpus_setall(p->cpus_allowed);
>                 dest_cpu = any_online_cpu(p->cpus_allowed);
> 
> out of consistency, shouldnt the cpus_setall() rather be:
> 
> 		p->cpus_allowed = cpu_possible_map;
> 
> ? It shouldnt make any real difference but it looks more consistent.

The default value of cpus_allowed is CPU_MASK_ALL, I thought -- at
least that's what we set init's to early on.  Though, as you say, it
shouldn't make any difference.