On 02/16/2018 09:49 AM, Peter Zijlstra wrote:
> On Fri, Feb 16, 2018 at 10:14:02AM +0100, Juri Lelli wrote:
>>> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
>>> index 7d324b706e67..4215438667e5 100644
>>> --- a/kernel/sched/sched.h
>>> +++ b/kernel/sched/sched.h
>>> @@ -650,8 +650,12 @@ struct root_domain {
>>> cpumask_var_t span;
>>> cpumask_var_t online;
>>>
>>> - /* Indicate more than one runnable task for any CPU */
>>> - bool overload;
>>> + /*
>>> + * Indicate whether the idle balance can be used to solve
>>> + * imbalance within the root domain.
>>> + * e.g. There is more than one runnable task for any CPU
>>> + */
>>> + bool should_idle_balance;
>>
>> Current name is however consistent with RT/DL's naming convention
I saw that it was already used elsewhere in fair but didn't know about
RT/DL, thanks for pointing that out.
>
> Yeah, not a fan either. We've consistently used the term to mean
> nr_running>1. The thing to fix there is the stupid bool, not the name.
>
So yeah the other thing that doesn't help here is that we're cramming
several meanings into rq->rd->overload:
- is there an overloaded group
- is there a group with misfit task(s)
So it didn't make sense to keep it named "overload". Perhaps a better way of
handling this would be to keep exposing which is which instead of merging it
all in a bool. Something along those lines:
@update_sg_lb_stats():
[...]
nr_running = rq->nr_running;
if (nr_running > 1)
- *overload = true;
+ sds->balance_status |= LB_STATUS_OVERLOAD
[...]
if (env->sd->flags & SD_ASYM_CPUCAPACITY &&
!sgs->group_misfit_task_load && rq->misfit_task_load) {
sgs->group_misfit_task_load = rq->misfit_task_load;
- *should_idle_balance = true;
+ sds->balance_status |= LB_STATUS_MISFIT
}
@update_sd_lb_stats():
[...]
if (!env->sd->parent) {
/* update overload indicator if we are at root domain */
- if (env->dst_rq->rd->overload != overload)
- env->dst_rq->rd->overload = overload;
+ if (env->dst_rq->rd->balance_status != sds->balance_status)
+ env->dst_rq->rd->balance_status = sds->balance_status
}
