Date:   Fri, 17 Dec 2021 20:48:47 +0100
From:   Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To:     Waiman Long <longman@redhat.com>
Cc:     Johannes Weiner <hannes@cmpxchg.org>,
        Michal Hocko <mhocko@kernel.org>,
        Vladimir Davydov <vdavydov.dev@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel@vger.kernel.org, cgroups@vger.kernel.org,
        linux-mm@kvack.org, Thomas Gleixner <tglx@linutronix.de>
Subject: Re: [PATCH-next v3] mm/memcg: Properly handle memcg_stock access for
 PREEMPT_RT
Message-ID: <Ybzpn59+ecDCnULt@linutronix.de>
References: <20211214144412.447035-1-longman@redhat.com>
 <Ybx3ubNFfGpCqhn0@linutronix.de>
 <b82efbad-1eb2-9441-ab0b-cbb3d2b5eac6@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <b82efbad-1eb2-9441-ab0b-cbb3d2b5eac6@redhat.com>
Precedence: bulk

On 2021-12-17 13:46:53 [-0500], Waiman Long wrote:
> > annotation. Looking at the history, I'm also impressed by that fact that
> > disabling/ enabling interrupts is *so* expensive that all this is
> > actually worth it.
> 
> For !RT with voluntary or no preemption, preempt_disable() is just a
> compiler barrier. So it is definitely cheaper than disabling interrupt. The
> performance benefit is less with preemptible but !RT kernel. Microbenchmark
> testing shows a performance improvement of a few percents depending on the
> exact benchmark.

Thanks for confirming. I got the feeling that this optimisation is for
!CONFIG_PREEMPTION. So I instead of depending on CONFIG_PREEMPT_RT I'm
leaning towards CONFIG_PREEMPT instead.

> Cheers,
> Longman

Sebastian