From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Hutchings Subject: Re: [PATCH 05/34]bnx2x: Setting the GSO_TYPE with LRO Date: Wed, 14 Jan 2009 20:27:22 +0000 Message-ID: <1231964842.3010.46.camel@achroite> References: <20090114201826.GA23648@gondor.apana.org.au> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Cc: Eilon Greenstein , davem@davemloft.net, netdev@vger.kernel.org To: Herbert Xu Return-path: Received: from smarthost03.mail.zen.net.uk ([212.23.3.142]:36065 "EHLO smarthost03.mail.zen.net.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752260AbZANU12 (ORCPT ); Wed, 14 Jan 2009 15:27:28 -0500 In-Reply-To: <20090114201826.GA23648@gondor.apana.org.au> Sender: netdev-owner@vger.kernel.org List-ID: On Thu, 2009-01-15 at 07:18 +1100, Herbert Xu wrote: > Eilon Greenstein wrote: > >> > > This is far from bullet proof solution, but IMHO it is better set then > > not this way, you stand a chance (depending on the output device > > capabilities) > > The question is simple, does your hardware guarantee that on > output GSO will turn the packet into exactly the same sequence > of packets that was seen on input, with no changes (apart from > what the stack would have done to them anyway, e.g., TTL update) > whatsoever? > > If not then you mustn't set this and also you must disable it > if forwarding/bridging is used. The driver just has to cooperate with dev_disable_lro() in order for the forwarding/bridging code to disable it. Then skb_warn_if_lro() should catch the case where the user mistakenly turns it back on. However, setting gso_type on input subverts this check. Ben. -- Ben Hutchings, Senior Software Engineer, Solarflare Communications Not speaking for my employer; that's the marketing department's job. They asked us to note that Solarflare product names are trademarked.