Re: Subject: [PATCH] sched: fixed erroneous all_pinned logic.

[Peter Zijlstra]

On Fri, 2011-04-08 at 14:57 +0400, Vladimir Davydov wrote:
> 
> On Apr 8, 2011, at 4:24 AM, Ken Chen wrote:
> 
> > @@ -2302,6 +2292,9 @@ static int move_tasks(
> > #endif
> >       } while (load_moved && max_load_move > total_load_moved);
> > 
> > +     if (total_load_moved)
> > +             *all_pinned = 0;
> > +
> >       return total_load_moved > 0;
> > }
> > 
> > @@ -3300,7 +3293,7 @@ static int load_balance(
> >                       struct sched_domain *sd, enum cpu_idle_type idle,
> >                       int *balance)
> > {
> > -     int ld_moved, all_pinned = 0, active_balance = 0;
> > +     int ld_moved, all_pinned = 1, active_balance = 0;
> >       struct sched_group *group;
> >       unsigned long imbalance;
> >       struct rq *busiest;
> 
> As far as I understand, this patch sets the all_pinned flag if and
> only if we fail to move any tasks during the load balance. However,
> the migration can fail because e.g. all tasks are cache hot on their
> cpus (can_migrate_task() returns 0 in this case), and in this case we
> shouldn't treat all tasks as cpu bound, should we?-- 

Hmm, you've got a good point there... (that'll teach me to read email in
date order).

Ken would it work to only push the all_pinned = 1 higher and not also
the all_pinned = 0?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/