Date: Tue, 6 Jun 2017 18:33:53 +0200
From: Peter Zijlstra <peterz@infradead.org>
To: Jeffrey Hugo <jhugo@codeaurora.org>
Cc: Ingo Molnar <mingo@redhat.com>, linux-kernel@vger.kernel.org,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Austin Christ <austinwc@codeaurora.org>,
        Tyler Baicar <tbaicar@codeaurora.org>,
        Timur Tabi <timur@codeaurora.org>
Subject: Re: [PATCH V4 1/2] sched/fair: Fix load_balance() affinity redo path
Message-ID: <20170606163353.twc64yj7n7uazj66@hirez.programming.kicks-ass.net>
References: <1496442432-330-1-git-send-email-jhugo@codeaurora.org>
 <1496442432-330-2-git-send-email-jhugo@codeaurora.org>
 <20170606123011.aehmelhnakvzjybb@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20170606123011.aehmelhnakvzjybb@hirez.programming.kicks-ass.net>
User-Agent: NeoMutt/20170113 (1.7.2)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1372
Lines: 41

On Tue, Jun 06, 2017 at 02:30:11PM +0200, Peter Zijlstra wrote:

> So with that the problem is that active_load_balance_cpu_stop() calls
> into can_migrate_task() with ->idle = CPU_IDLE and !dst_grpmask, which
> then goes *bang*. Now active_load_balance_cpu_stop() doesn't need to
> re-evaluate anything, so ideally it would just skip this entirely,
> right?
> 
> So why not do #3:
> 
> ---
>  kernel/sched/fair.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 47a0c552c77b..fd639d32fa4c 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8523,6 +8523,13 @@ static int active_load_balance_cpu_stop(void *data)
>  			.src_cpu	= busiest_rq->cpu,
>  			.src_rq		= busiest_rq,
>  			.idle		= CPU_IDLE,

So I realized that setting CPU_NEW_IDLE here does the 'same' thing.
Except that would be a 'visible' change in the schedstat output (not
sure anybody really cares about that).

> +			/*
> +			 * can_migrate_task() doesn't need to compute new_dst_cpu
> +			 * for active balancing. Since we have CPU_IDLE, but no
> +			 * @dst_grpmask we need to make that test go away with lying
> +			 * about DST_PINNED.
> +			 */
> +			.flags		= LBF_DST_PINNED,

But given this is trivial and doesn't have any visible side effects,
this seems 'better'.

>  		};
>  
>  		schedstat_inc(sd->alb_count);