Re: [PATCH v5 1/2] sched/deadline: optimize sched_dl_global_validate()

From: Peter Zijlstra
Date: Wed Oct 07 2020 - 12:55:53 EST


On Wed, Oct 07, 2020 at 11:12:29PM +0800, Peng Liu wrote:
> +/* Used for dl_bw check and update. */
> +static u32 dl_generation;

> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
> index 28709f6b0975..53477e8b26b0 100644
> --- a/kernel/sched/sched.h
> +++ b/kernel/sched/sched.h
> @@ -798,6 +798,13 @@ struct root_domain {
> */
> cpumask_var_t dlo_mask;
> atomic_t dlo_count;
> +
> + /*
> + * Indicate whether a root_domain's dl_bw has been checked or
> + * updated. It's monotonously increasing, then wrap around.
> + */
> + u32 visit_gen;
> +
> struct dl_bw dl_bw;
> struct cpudl cpudl;
>
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index dd7770226086..90f3e5558fa2 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -516,6 +516,7 @@ static int init_rootdomain(struct root_domain *rd)
> init_irq_work(&rd->rto_push_work, rto_push_irq_work_func);
> #endif
>
> + rd->visit_gen = 0;
> init_dl_bw(&rd->dl_bw);
> if (cpudl_init(&rd->cpudl) != 0)
> goto free_rto_mask;

I'm fairly sure I made the generation a u64, the above is susceptible to
a false positive due to wrap-around.

Increase the generation to -1, create a new root domain, then the next
generation is 0 and we'll skip the new domain, even though it should be
updated.