Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 15 Oct 2018 11:20:32 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Peng Hao <peng.hao2@zte.com.cn>
Cc:     mingo@redhat.com, rostedt@goodmis.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3] sched/rt : return accurate release rq lock info
Message-ID: <20181015092032.GO9867@hirez.programming.kicks-ass.net>
References: <1538778131-44406-1-git-send-email-peng.hao2@zte.com.cn>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1538778131-44406-1-git-send-email-peng.hao2@zte.com.cn>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Oct 06, 2018 at 06:22:11AM +0800, Peng Hao wrote:
> find_lock_lowest_rq may or not releease rq lock when return
> lowest_rq=NULL, but it is fuzzy.
> If not releasing rq lock, it is unnecessary to re-call
> pick_next_pushable_task.
> When CONFIG_PREEMPT=n, not releasing rq lock and return
> lowest_rq=null frequently happens in a simple test case:
> Four different rt priority tasks run on limited two cpus.
> Thanks for Steven Rostedt's advice.

Can we please write a more coherent Changelog, the above is very hard to
read.

Maybe something along the lines of:

Subject: sched/rt: Reduce push_rt_task() retries

Improve push_rt_task() by propagating the double_lock_balance() usage
from find_lock_lowest_rq(), thereby reducing the number of cases where
we have to assume rq->lock was dropped.


> Signed-off-by: Peng Hao <peng.hao2@zte.com.cn>
> ---
>  kernel/sched/rt.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
> index 2e2955a..be0fc43 100644
> --- a/kernel/sched/rt.c
> +++ b/kernel/sched/rt.c
> @@ -1754,7 +1754,7 @@ static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
>  				     !task_on_rq_queued(task))) {
>  
>  				double_unlock_balance(rq, lowest_rq);
> -				lowest_rq = NULL;
> +				lowest_rq = RETRY_TASK;
>  				break;
>  			}
>  		}

I'm confused.. should not:

		/* try again */
		double_unlock_balance(rq, lowest_rq);
		lowest_rq = NULL;

also return RETRY_TASK? That also is in the double_lock_balance() path
and will this have had rq->lock() released.