From mboxrd@z Thu Jan 1 00:00:00 1970 From: Eric Dumazet Subject: Re: [PATCH] tcp: gso: do not generate out of order packets Date: Wed, 15 May 2013 21:46:48 -0700 Message-ID: <1368679608.4519.105.camel@edumazet-glaptop> References: <1368668281.4519.75.camel@edumazet-glaptop> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , netdev , Tom Herbert , Neal Cardwell , Yuchung Cheng To: Maciej =?UTF-8?Q?=C5=BBenczykowski?= Return-path: Received: from mail-da0-f41.google.com ([209.85.210.41]:51109 "EHLO mail-da0-f41.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752275Ab3EPEqu (ORCPT ); Thu, 16 May 2013 00:46:50 -0400 Received: by mail-da0-f41.google.com with SMTP id y19so1417023dan.14 for ; Wed, 15 May 2013 21:46:50 -0700 (PDT) In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On Wed, 2013-05-15 at 20:58 -0700, Maciej =C5=BBenczykowski wrote: > Could we instead move the queue mapping into the skb somehow instead? Quite frankly all the ndo_select_queue() / ARFS / XPS stuff are already a mess. You need a socket pointer to be able to change the queue numbers we cache in it. We lack this information at GSO segmentation. Let's do GSO properly and provide skbs exactly like if they were built by TCP stack. (Well, there _is_ some difference, as the truesize will have less overhead)