Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 11 Oct 2021 07:32:39 +0000
From:   Hyeonggon Yoo <42.hyeyoo@gmail.com>
To:     Christoph Lameter <cl@gentwo.de>
Cc:     linux-mm@kvack.org, linux-kernel@vger.kernel.org,
        Pekka Enberg <penberg@kernel.org>,
        David Rientjes <rientjes@google.com>,
        Joonsoo Kim <iamjoonsoo.kim@lge.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Vlastimil Babka <vbabka@suse.cz>
Subject: Re: [PATCH] mm, slub: Use prefetchw instead of prefetch
Message-ID: <20211011073239.GA63896@kvm.asia-northeast3-a.c.our-ratio-313919.internal>
References: <20211008133602.4963-1-42.hyeyoo@gmail.com>
 <alpine.DEB.2.22.394.2110110919050.130815@gentwo.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <alpine.DEB.2.22.394.2110110919050.130815@gentwo.de>
Precedence: bulk

On Mon, Oct 11, 2021 at 09:20:36AM +0200, Christoph Lameter wrote:
> On Fri, 8 Oct 2021, Hyeonggon Yoo wrote:
> 
> > It's certain that an object will be not only read, but also
> > written after allocation.
> 
> get_freepointer is used in multiple code path not only in allocation. It
> is for example used when scanning through a freelist.
> 
> With this change all objects get needlessly dirtied and the cross cpu
> cache contention is increased.

I didn't touch get_freepointer and there's
only one caller of prefetch_freepointer.

My change was not adding additional prefetch on get_freepointer,
but changing existing prefetch into prefetchw.

The prefetch was introcued by commit 0ad9500e16fe ("slub: prefetch next
freelist pointer in slab_alloc()") that you ACKed in 2011.

Do you think removeing existing prefetch is better than changing it
from prefetch to prefetchw?