Re: [PATCH 3/5] sched/fair: Wait before decaying max_newidle_lb_cost

From: Dietmar Eggemann
Date: Fri Oct 29 2021 - 06:01:28 EST


On 19/10/2021 14:35, Vincent Guittot wrote:
> Decay max_newidle_lb_cost only when it has not been updated for a while
> and ensure to not decay a recently changed value.
>
> Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> ---
> include/linux/sched/topology.h | 2 +-
> kernel/sched/fair.c | 36 +++++++++++++++++++++++++---------
> kernel/sched/topology.c | 2 +-
> 3 files changed, 29 insertions(+), 11 deletions(-)
>
> diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
> index 2f9166f6dec8..c07bfa2d80f2 100644
> --- a/include/linux/sched/topology.h
> +++ b/include/linux/sched/topology.h
> @@ -105,7 +105,7 @@ struct sched_domain {
>
> /* idle_balance() stats */
> u64 max_newidle_lb_cost;
> - unsigned long next_decay_max_lb_cost;
> + unsigned long last_decay_max_lb_cost;
>
> u64 avg_scan_cost; /* select_idle_sibling */
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index c4c36865321b..e50fd751e1df 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -10239,6 +10239,30 @@ void update_max_interval(void)
> max_load_balance_interval = HZ*num_online_cpus()/10;
> }
>
> +static inline bool update_newidle_cost(struct sched_domain *sd, u64 cost)
> +{
> + if (cost > sd->max_newidle_lb_cost) {
> + /*
> + * Track max cost of a domain to make sure to not delay the
> + * next wakeup on the CPU.
> + */
> + sd->max_newidle_lb_cost = cost;
> + sd->last_decay_max_lb_cost = jiffies;

That's the actual change of the patch: sd->last_decay_max_lb_cost being
moved forward also when newidle cost is updated from newidle_balance() ?

> + } else if (time_after(jiffies, sd->last_decay_max_lb_cost + HZ)) {
> + /*
> + * Decay the newidle max times by ~1% per second to ensure that
> + * it is not outdated and the current max cost is actually
> + * shorter.

I assume that `max cost` refers here to a local variable of the only
caller of update_newidle_cost(..., 0) - rebalance_domains()?

"the current max cost" has to be shorter so that
rq->max_idle_balance_cost also decays in this case. Is this what this
comment tries to say here?

[...]