Re: [PATCH v3] sched/fair: bring back select_idle_smt, but differently

From: Peter Zijlstra <peterz@infradead.org>
To: Vincent Guittot <vincent.guittot@linaro.org>
Cc: Rik van Riel <riel@surriel.com>,
	Mel Gorman <mgorman@techsingularity.net>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Kernel Team <kernel-team@fb.com>, Ingo Molnar <mingo@kernel.org>,
	Valentin Schneider <valentin.schneider@arm.com>
Subject: Re: [PATCH v3] sched/fair: bring back select_idle_smt, but differently
Date: Wed, 7 Apr 2021 11:54:37 +0200	[thread overview]
Message-ID: <YG2BXRm60IhpumD8@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <20210407094217.GA2926@vingu-book>

On Wed, Apr 07, 2021 at 11:42:17AM +0200, Vincent Guittot wrote:
> I would really prefer to keep that out of select_idle_cpu which aims to merge in one
> single loop the walk through sd_llc. In the case of select_idle_smt, this is done outside
> the loop:

Fair enough.

> @@ -6317,11 +6339,21 @@ static int select_idle_sibling(struct task_struct *p, int prev, int target)
>  		}
>  	}
>  
> +	if (static_branch_likely(&sched_smt_present)) {
> +		smt = test_idle_cores(target, false);
> +		if (!smt && cpus_share_cache(prev, target)) {
> +			/* No idle core. Check if prev has an idle sibling. */
> +			i = select_idle_smt(p, sd, prev);
> +			if ((unsigned int)i < nr_cpumask_bits)
> +				return i;
> +		}
> +	}
> +
>  	sd = rcu_dereference(per_cpu(sd_llc, target));
>  	if (!sd)
>  		return target;

It needs to be here, otherwise you're using @sd uninitialized.

> -	i = select_idle_cpu(p, sd, target);
> +	i = select_idle_cpu(p, sd, smt, target);
>  	if ((unsigned)i < nr_cpumask_bits)
>  		return i;

Let me have another poke at it.