From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756641AbYBPPxP (ORCPT ); Sat, 16 Feb 2008 10:53:15 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751443AbYBPPxA (ORCPT ); Sat, 16 Feb 2008 10:53:00 -0500 Received: from einhorn.in-berlin.de ([192.109.42.8]:36853 "EHLO einhorn.in-berlin.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751185AbYBPPw7 (ORCPT ); Sat, 16 Feb 2008 10:52:59 -0500 X-Envelope-From: stefanr@s5r6.in-berlin.de Date: Sat, 16 Feb 2008 16:51:55 +0100 (CET) From: Stefan Richter Subject: Re: [PATCH 9/9] firewire: fw-sbp2: fix I/O errors during reconnect To: Jarod Wilson , Kristian Hoegsberg cc: linux1394-devel@lists.sourceforge.net, linux-kernel@vger.kernel.org In-Reply-To: Message-ID: References: <47B08F6B.2020502@redhat.com> <47B0AE54.8090602@s5r6.in-berlin.de> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=us-ascii Content-Disposition: INLINE Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org I wrote: > The only remedy seems to be to block the SCSI device until reconnect. In the longer term, we should look into keeping commands enqueued in fw-sbp2 during the reconnect phase. Plus we need full parallelism of sbp2_reconnect and sbp2_login between all attached targets. The current single-threaded scheme does not work too well for more than one target attached to the same bus. The old stack handles that case in a single thread too but is better tuned to sbp2's needs. The old stack blocks the Scsi_Hosts earlier, guaranteedly performs all reconnects and re-logins before any new login, and has a small delay between self-ID-complete and reconnect. I looked into changing fw-sbp2's own workqueue delays and into adding a counter to fw-sbp2 to track outstanding reconnects in order to defer new logins, but it is too messy. -- Stefan Richter -=====-==--- --=- =---- http://arcgraph.de/sr/