From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933533Ab2DKWZe (ORCPT ); Wed, 11 Apr 2012 18:25:34 -0400 Received: from violet.fr.zoreil.com ([92.243.8.30]:54050 "EHLO violet" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S933500Ab2DKWZd (ORCPT ); Wed, 11 Apr 2012 18:25:33 -0400 Date: Thu, 12 Apr 2012 00:21:24 +0200 From: Francois Romieu To: Svenning =?utf-8?B?U8O4cmVuc2Vu?= Cc: Bjarke Istrup Pedersen , linux-kernel@vger.kernel.org, netdev@vger.kernel.org, Andreas Mohr Subject: Re: via-rhine: Problem with lost link after a while Message-ID: <20120411222124.GA20605@electric-eye.fr.zoreil.com> References: <20120410204249.GA26627@electric-eye.fr.zoreil.com> <20120410225534.GA28480@electric-eye.fr.zoreil.com> <4F8553BF.6030303@secomea.com> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: <4F8553BF.6030303@secomea.com> User-Agent: Mutt/1.4.2.2i X-Organisation: Land of Sunshine Inc. Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Svenning Sørensen : [...] > Regarding the "Tx descriptor busy" test: no, I didn't see it > trigger, I just put it there just in case because I suspected there > could be a race due to the lock-free tx path. > But I'm glad if you're confident that it can't happen :) Almost :o) Without the patch it may happen in mainline. Once the napi Tx completion function checks if queueing is stopped before enabling it, it should not happen though. The patch should fix the race where queuing is not enabled as well. That being said, I would welcome a pony^W^W some testing on a multi-core system with lots of Tx and enough irq from a different (non via-rhine) source to starve the softirq processing. -- Ueimor