Re: [PATCH -tip] kernel/sched,wake_q: Branch predict wake_q_add() cmpxchg

From: Waiman Long <longman@redhat.com>
To: Davidlohr Bueso <dave@stgolabs.net>,
	Peter Zijlstra <peterz@infradead.org>
Cc: Yongji Xie <elohimes@gmail.com>,
	mingo@redhat.com, will.deacon@arm.com,
	linux-kernel@vger.kernel.org, xieyongji@baidu.com,
	zhangyu31@baidu.com, liuqi16@baidu.com, yuanlinsi01@baidu.com,
	nixun@baidu.com, lilin24@baidu.com,
	andrea.parri@amarulasolutions.com
Subject: Re: [PATCH -tip] kernel/sched,wake_q: Branch predict wake_q_add() cmpxchg
Date: Mon, 3 Dec 2018 11:10:58 -0500	[thread overview]
Message-ID: <c1cd0892-9ef4-ffad-ce93-39977389071c@redhat.com> (raw)
In-Reply-To: <20181203053130.gwkw6kg72azt2npb@linux-r8p5>

On 12/03/2018 12:31 AM, Davidlohr Bueso wrote:
> The cmpxchg will fail when the task is already in the process
> of waking up, and as such is an extremely rare occurrence.
> Micro-optimize the call and put an unlikely() around it.
>
> To no surprise, when using CONFIG_PROFILE_ANNOTATED_BRANCHES
> under a number of workloads the incorrect rate was a mere 1-2%.
>
> Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
> ---
> kernel/sched/core.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 091e089063be..f7747cf6e427 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -408,7 +408,7 @@ void wake_q_add(struct wake_q_head *head, struct
> task_struct *task)
>      * This cmpxchg() executes a full barrier, which pairs with the full
>      * barrier executed by the wakeup in wake_up_q().
>      */
> -    if (cmpxchg(&node->next, NULL, WAKE_Q_TAIL))
> +    if (unlikely(cmpxchg(&node->next, NULL, WAKE_Q_TAIL)))
>         return;
>
>     get_task_struct(task);

Acked-by: Waiman Long <longman@redhat.com>