Re: [PATCH] sched/fair: Fix the logic about active_balance in load_balance()

[Dietmar Eggemann]

On 02/08/2020 06:51, Qi Zheng wrote:
> I think the unbalance scenario here should be that we need to
> do active balance but it is not actually done. So fix it.
> 
> Signed-off-by: Qi Zheng <arch0.zheng@xxxxxxxxx>
> ---
>  kernel/sched/fair.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 2ba8f230feb9..6d8c53718b67 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -9710,7 +9710,7 @@ static int load_balance(int this_cpu, struct rq *this_rq,
>  	} else
>  		sd->nr_balance_failed = 0;
>  
> -	if (likely(!active_balance) || voluntary_active_balance(&env)) {
> +	if (likely(!active_balance) && voluntary_active_balance(&env)) {
>  		/* We were unbalanced, so reset the balancing interval */
>  		sd->balance_interval = sd->min_interval;
>  	} else {
> 

Active balance is potentially already been done when we reach this code.

See 'if (need_active_balance(&env))' and 'if (!busiest->active_balance)'
further up.

Here we only reset sd->balance_interval in case:
(A) the last load balance wasn't an active one
(B) the reason for the active load balance was:
    (1) asym packing
    (2) capacity of src_cpu is reduced compared to the one of dst_cpu
    (3) misfit handling

(B) is done to not unnecessarily increase of balance interval, see
commit 46a745d90585 ("sched/fair: Fix unnecessary increase of balance
interval").