From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Borkmann Date: Tue, 16 Aug 2016 22:59:17 +0000 Subject: Re: [PATCH net] sctp: linearize early if it's not GSO Message-Id: <57B39AC5.7000002@iogearbox.net> List-Id: References: In-Reply-To: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Marcelo Ricardo Leitner , netdev@vger.kernel.org Cc: linux-sctp@vger.kernel.org, Neil Horman , Vlad Yasevich On 08/17/2016 12:35 AM, Marcelo Ricardo Leitner wrote: > Because otherwise when crc computation is still needed it's way more > expensive than on a linear buffer to the point that it affects > performance. > > It's so expensive that netperf test gives a perf output as below: > > Overhead Shared Object Symbol > 69,44% [kernel] [k] gf2_matrix_square > 2,84% [kernel] [k] crc32_generic_combine.part.0 > 2,78% [kernel] [k] _raw_spin_lock_bh What kernel is this, seems not net kernel? $ git grep -n gf2_matrix_square $ git grep -n crc32_generic_combine $ Maybe RHEL? Did you consider backporting 6d514b4e7737 et al? > And performance goes from 2Gbit/s to 0.5Gbit/s on this test. Doing the > linearization before checksumming is enough to restore it. > > Fixes: 3acb50c18d8d ("sctp: delay as much as possible skb_linearize") > Signed-off-by: Marcelo Ricardo Leitner From mboxrd@z Thu Jan 1 00:00:00 1970 From: Daniel Borkmann Subject: Re: [PATCH net] sctp: linearize early if it's not GSO Date: Wed, 17 Aug 2016 00:59:17 +0200 Message-ID: <57B39AC5.7000002@iogearbox.net> References: Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Cc: linux-sctp@vger.kernel.org, Neil Horman , Vlad Yasevich To: Marcelo Ricardo Leitner , netdev@vger.kernel.org Return-path: Received: from www62.your-server.de ([213.133.104.62]:51909 "EHLO www62.your-server.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752053AbcHPW7T (ORCPT ); Tue, 16 Aug 2016 18:59:19 -0400 In-Reply-To: Sender: netdev-owner@vger.kernel.org List-ID: On 08/17/2016 12:35 AM, Marcelo Ricardo Leitner wrote: > Because otherwise when crc computation is still needed it's way more > expensive than on a linear buffer to the point that it affects > performance. > > It's so expensive that netperf test gives a perf output as below: > > Overhead Shared Object Symbol > 69,44% [kernel] [k] gf2_matrix_square > 2,84% [kernel] [k] crc32_generic_combine.part.0 > 2,78% [kernel] [k] _raw_spin_lock_bh What kernel is this, seems not net kernel? $ git grep -n gf2_matrix_square $ git grep -n crc32_generic_combine $ Maybe RHEL? Did you consider backporting 6d514b4e7737 et al? > And performance goes from 2Gbit/s to 0.5Gbit/s on this test. Doing the > linearization before checksumming is enough to restore it. > > Fixes: 3acb50c18d8d ("sctp: delay as much as possible skb_linearize") > Signed-off-by: Marcelo Ricardo Leitner