Re: [PATCH 1/6] sched/fair: remove update of blocked load from newidle_balance

[Valentin Schneider]

On 05/02/21 12:48, Vincent Guittot wrote:
> @@ -10517,16 +10499,11 @@ static void nohz_newidle_balance(struct rq *this_rq)
>           time_before(jiffies, READ_ONCE(nohz.next_blocked)))
>               return;
>

I was wondering whether all the conditions above were still relevant. I
think they are, but this one:

        /* Will wake up very soon. No time for doing anything else*/
        if (this_rq->avg_idle < sysctl_sched_migration_cost)
                return;

should have its comment updated to something like:

        /*
         * Will wake up very soon. Blocked load will be updated
         * periodically, no need to wake an idle CPU.
         */

given kick_ilb() isn't the costliest of things.

> -	raw_spin_unlock(&this_rq->lock);
>       /*
> -	 * This CPU is going to be idle and blocked load of idle CPUs
> -	 * need to be updated. Run the ilb locally as it is a good
> -	 * candidate for ilb instead of waking up another idle CPU.
> -	 * Kick an normal ilb if we failed to do the update.
> +	 * Blocked load of idle CPUs need to be updated.
> +	 * Kick an ILB to update statistics.
>        */
> -	if (!_nohz_idle_balance(this_rq, NOHZ_STATS_KICK, CPU_NEWLY_IDLE))
> -		kick_ilb(NOHZ_STATS_KICK);
> -	raw_spin_lock(&this_rq->lock);

With this change, the return value of _nohz_idle_balance() is no longer
used. This means we could get rid of the tracking of whether it iterated
over all nohz CPUs or not.

> +	kick_ilb(NOHZ_STATS_KICK);
>  }
>
>  #else /* !CONFIG_NO_HZ_COMMON */