Re: [PATCH 6/7] sched/fair: Remove on_null_domain() and redundant checks

[Peter Zijlstra]

On Wed, Apr 03, 2024 at 05:05:38PM +0200, Pierre Gondois wrote:
> CPUs with a NULL sched domain are removed from the HKR_TYPE_SCHED
> isolation mask. The two following checks are equialent:
> - !housekeeping_runtime_test_cpu(cpu, HKR_TYPE_SCHED)
> - on_null_domain(rq)
> 
> Remove on_null_domain() and the redundant checks.
> 
> Signed-off-by: Pierre Gondois <pierre.gondois@xxxxxxx>
> ---
>  kernel/sched/fair.c | 18 ++++++------------
>  1 file changed, 6 insertions(+), 12 deletions(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 3e0f2a0f153f..9657c8f2176b 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -11830,11 +11830,6 @@ static void sched_balance_domains(struct rq *rq, enum cpu_idle_type idle)
>  
>  }
>  
> -static inline int on_null_domain(struct rq *rq)
> -{
> -	return unlikely(!rcu_dereference_sched(rq->sd));
> -}
> -
>  #ifdef CONFIG_NO_HZ_COMMON
>  /*
>   * NOHZ idle load balancing (ILB) details:
> @@ -12040,7 +12035,7 @@ void nohz_balance_exit_idle(struct rq *rq)
>  	SCHED_WARN_ON(rq != this_rq());
>  
>  	/* If we're a completely isolated CPU, we don't play: */
> -	if (on_null_domain(rq))
> +	if (!housekeeping_runtime_test_cpu(cpu_of(rq), HKR_TYPE_SCHED))
>  		return;
>  
>  	if (likely(!rq->nohz_tick_stopped))

This seems broken, the whole null domain can happen with cpusets, but
this housekeeping nonsense is predicated on CPU_ISOLATION and none of
that is mandatory for CPUSETS.