From mboxrd@z Thu Jan 1 00:00:00 1970 From: David Miller Subject: Re: Get rxhash fixes and RFS support in tun Date: Thu, 21 Nov 2013 13:03:39 -0500 (EST) Message-ID: <20131121.130339.204009506339126583.davem@davemloft.net> References: <20131120.215025.590189385589840672.davem@davemloft.net> Mime-Version: 1.0 Content-Type: Text/Plain; charset=us-ascii Content-Transfer-Encoding: 7bit Cc: netdev@vger.kernel.org, edumazet@google.com, hkchu@google.com To: therbert@google.com Return-path: Received: from shards.monkeyblade.net ([149.20.54.216]:43372 "EHLO shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754224Ab3KUSDl (ORCPT ); Thu, 21 Nov 2013 13:03:41 -0500 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: From: Tom Herbert Date: Thu, 21 Nov 2013 08:28:18 -0800 > Since we're doing deep inspection now in flow_dissector, we should > already have the discovered the hash on the inner header if it's a > standard encap. For AH and normal IP tunnels, this might be fine. But this is not possible for ESP since the inner headers are encrypted when the flow_dissector takes a look. And again, there is the issue of which hardware devices are doing this properly. And for those that do perform inner-tunnel deep inspection for rxhash generation, how many layers of tunnels are they able to look beneath? My impression is that they support one level, at best, which means the rxhash is basically reliable only under a specific set of conditions. I need you to elaborate with your knowledge of what hardware actually supports and does in this area a bit more before I can have a real opinion on what we should be doing. Thanks.