From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: [PATCH net-next] gro: relax ID check in inet_gro_receive() Date: Thu, 21 Mar 2013 14:56:27 -0400 (EDT) Message-ID: <20130321.145627.565133121162775865.davem@davemloft.net> References: <20130321.114616.279859400813363663.davem@davemloft.net> <1363882091.4431.20.camel@edumazet-glaptop> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: eric.dumazet@gmail.com, netdev@vger.kernel.org, dmitry@broadcom.com, eilong@broadcom.com, pshelar@nicira.com, hkchu@google.com, maze@google.com To: dkravkov@gmail.com Return-path: Received: from shards.monkeyblade.net ([149.20.54.216]:45892 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751643Ab3CUS4a (ORCPT ); Thu, 21 Mar 2013 14:56:30 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: From: Dmitry Kravkov Date: Thu, 21 Mar 2013 20:11:37 +0200 >> Well, I suspect this hardware is not capable of doing the proper ID >> manipulation twice. (inner and outer header) > > This is correct: ID only for one of the headers can be handled with > current FW/HW, for other DF is set. DF does not matter. Regardless of DF, we must set the ID field correctly. It is abundantly clear that the current GRE tunnel segmentation is not generating packets according to our well documented rules, in that we must be able to precisely create exactly the original packet stream from the segmented frame. Someone needs to send me patches to revert the bnx2x GRE segmentation support, and any software implementation in our tree that has the same bug. If someone doesn't do it, I will revert all of this code myself. You simply will have to cope with not having this optimization until your hardware can do it properly and according to our well established rules for segmentation offloads. Thanks.