On 02/03/21 18:42, Valentin Schneider wrote:
> >> @@ -9805,9 +9810,6 @@ static int load_balance(int this_cpu, struct rq
> >> *this_rq,
> >> active_load_balance_cpu_stop, busiest,
> >> &busiest->active_balance_work);
> >> }
> >> -
> >> - /* We've kicked active balancing, force task migration.
> >> */
> >> - sd->nr_balance_failed = sd->cache_nice_tries+1;
> >
> > This has an impact on future calls to need_active_balance() too, no? We
> > enter
> > this path because need_active_balance() returned true; one of the
> > conditions it
> > checks for is
> >
> > return unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2);
> >
> > So since we used to reset nr_balanced_failed to cache_nice_tries+1, the
> > above
> > condition would be false in the next call or two IIUC. But since we remove
> > that, we could end up here again soon.
> >
> > Was this intentional?
> >
>
> Partially, I'd say :-)
>
> If you look at active_load_balance_cpu_stop(), it does
>
> sd->nr_balance_failed = 0;
>
> when it successfully pulls a task. So we get a reset of the failed counter
> on pull, which I've preserved. As for interactions with later
> need_active_balance(), the commit that introduced the current counter write
> (which is over 15 years old!):
>
> 3950745131e2 ("[PATCH] sched: fix SMT scheduling problems")
>
> only states the task_hot() issue; thus I'm doubtful whether said
> interaction was intentional.
The '+1' was added in that comment. 'Original' code was just resetting the
nr_balance_failed cache_nice_tries, so that we don't do another one too soon
I think.
With this change, no active balance is allowed until later. Which makes sense.
I can't see why we would have allowed another kick sooner tbh. But as you say,
this is ancient piece of logic.
I agree I can't see a reason to worry about this (potential) change of
behavior.
Thanks
--
Qais Yousef
