Re: [PATCH 15/16] sched/topology: optimize topology_span_sane()

[Peter Zijlstra]

On Mon, Jul 18, 2022 at 12:28:43PM -0700, Yury Norov wrote:

>  kernel/sched/topology.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index 05b6c2ad90b9..ad32d0a43424 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -2211,6 +2211,8 @@ static struct sched_domain *build_sched_domain(struct sched_domain_topology_leve
>  static bool topology_span_sane(struct sched_domain_topology_level *tl,
>  			      const struct cpumask *cpu_map, int cpu)
>  {
> +	const struct cpumask *mc = tl->mask(cpu);
> +	const struct cpumask *mi;
>  	int i;
>  
>  	/* NUMA levels are allowed to overlap */
> @@ -2226,14 +2228,18 @@ static bool topology_span_sane(struct sched_domain_topology_level *tl,
>  	for_each_cpu(i, cpu_map) {
>  		if (i == cpu)
>  			continue;
> +
> +		mi = tl->mask(i);
> +		if (mi == mc)
> +			continue;
> +
>  		/*
>  		 * We should 'and' all those masks with 'cpu_map' to exactly
>  		 * match the topology we're about to build, but that can only
>  		 * remove CPUs, which only lessens our ability to detect
>  		 * overlaps
>  		 */
> -		if (!cpumask_equal(tl->mask(cpu), tl->mask(i)) &&
> -		    cpumask_intersects(tl->mask(cpu), tl->mask(i)))
> +		if (!cpumask_equal(mc, mi) && cpumask_intersects(mc, mi))
>  			return false;
>  	}

This is once again a super slow path; but I don't suppose you're making
the code worse in this case.