Re: bug in sched.c:activate_task()

[Con Kolivas]

Nick Piggin writes:

> Con Kolivas wrote:
>
> > Chen, Kenneth W writes:
> >
> > > Update p->timestamp to "now" in activate_task() doesn't look right
> > > to me at all.  p->timestamp records last time it was running on a
> > > cpu.  activate_task shouldn't update that variable when it queues
> > > a task on the runqueue.
> > >
> > > This bug (and combined with others) triggers improper load balancing.
> >
> >
> > The updated timestamp was placed there by Ingo to detect on-runqueue 
> > time. If it is being used for load balancing then it is being used in 
> > error.
>
> Load balancing wants to know if a task is considered cache hot.

Yes I know. It used to be performed based on jiffies which was adequate 
resolution for cache warmth at the time. The timestamp was being used for on 
runqueue length measurement before the load balancing was modified to use 
that value.

Con

-
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/