From mboxrd@z Thu Jan  1 00:00:00 1970
From: Olivier Matz <olivier.matz@6wind.com>
Subject: advice to pmd maintainters: do not touch mbuf
	initialized fields
Date: Thu, 8 Jun 2017 11:06:07 +0200
Message-ID: <20170608110607.30887c26@platinum>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Cc: Ferruh Yigit <ferruh.yigit@intel.com>
To: dev@dpdk.org
Return-path: <dev-bounces@dpdk.org>
Received: from mail-wm0-f42.google.com (mail-wm0-f42.google.com [74.125.82.42])
 by dpdk.org (Postfix) with ESMTP id 5C59F2BB9
 for <dev@dpdk.org>; Thu,  8 Jun 2017 11:06:10 +0200 (CEST)
Received: by mail-wm0-f42.google.com with SMTP id x70so73025706wme.0
 for <dev@dpdk.org>; Thu, 08 Jun 2017 02:06:10 -0700 (PDT)
List-Id: DPDK patches and discussions <dev.dpdk.org>
List-Unsubscribe: <http://dpdk.org/ml/options/dev>,
 <mailto:dev-request@dpdk.org?subject=unsubscribe>
List-Archive: <http://dpdk.org/ml/archives/dev/>
List-Post: <mailto:dev@dpdk.org>
List-Help: <mailto:dev-request@dpdk.org?subject=help>
List-Subscribe: <http://dpdk.org/ml/listinfo/dev>,
 <mailto:dev-request@dpdk.org?subject=subscribe>
Errors-To: dev-bounces@dpdk.org
Sender: "dev" <dev-bounces@dpdk.org>

Hi,

In the structure reorganization patchset [1], some work was done
to ensure that the values of the following fields are always initialized
for mbuf that are inside the mbuf pool (not yet allocated).

In other words, calling rte_mbuf_raw_alloc() or rte_mempool_get() return
mbufs that are initialized as following:

  m->next = NULL
  m->nb_segs = 1
  m->refcnt = 1

This property can be used in the Rx path of PMDs where the mbuf
structure is filled from the hardware descriptors: the initialization of
these fields can be skipped to gain few cycles per packet, and to avoid
an access to the second cache line (m->next) before it is prefetched.

It is advised to do this optimization in the PMDs for this reason, and
also because we could want to move some of these fields to the second
mbuf cache line in the future: accessing them in the Rx path would cost.

This work has already been done at several places:

  http://dpdk.org/browse/dpdk/commit/?id=ebb7bcabb
  http://dpdk.org/browse/next/dpdk-next-net/commit/?id=e211a7569
  http://dpdk.org/browse/next/dpdk-next-net/commit/?id=a37cf9005
  http://dpdk.org/browse/next/dpdk-next-net/commit/?id=fb83a23ee
  ...

Feel free to ask if you have any question.

Thanks,
Olivier

[1] http://dpdk.org/ml/archives/dev/2017-April/063081.html