From: Takashi Iwai Subject: Re: crash/reboot with rawmidi on ice1712 dual opteron Date: Tue, 24 Apr 2007 16:32:36 +0200 Message-ID: References: <461FD727.8080400@bome.com> <4626361D.3080801@bome.com> <462E12F8.3090005@64studio.com> Mime-Version: 1.0 (generated by SEMI 1.14.6 - "Maruoka") Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Received: from mx1.suse.de (ns.suse.de [195.135.220.2]) by alsa0.perex.cz (Postfix) with ESMTP id DFAB22458F for ; Tue, 24 Apr 2007 16:32:36 +0200 (CEST) In-Reply-To: <462E12F8.3090005@64studio.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: alsa-devel-bounces@alsa-project.org Errors-To: alsa-devel-bounces@alsa-project.org To: Daniel James Cc: alsa-devel@alsa-project.org, Florian List-Id: alsa-devel@alsa-project.org At Tue, 24 Apr 2007 15:23:52 +0100, Daniel James wrote: > > Hi Takashi, hi Florian, > > I found a similar MIDI crashing bug on a dual Opteron machine in 2005, > even with only one processor installed, which I reported at the time on > alsa-devel. The card was an M-Audio Audiophile 24/96, normally reliable. > > I couldn't replicate this problem on an Asus single processor Opteron > board, so I concluded it was a quirk of my dual socket Tyan S2875 > motherboard: > > http://article.gmane.org/gmane.linux.alsa.devel/24682/ > http://article.gmane.org/gmane.linux.alsa.devel/25323/ > > Last time I tested it, earlier this year I think, the bug was still > there. I even flashed the BIOS of the Tyan board in case it was a BIOS > bug, but it made no difference. I installed a 32-bit distro and that > made no difference either. > > Maybe there is something more generally wrong here, which only affects > dual-processor AMD64 hardware. I can make the Tyan machine available > over SSH if that helps, it has a fixed IP address. The most important thing is to find out what triggers which result. As far as I understand from Florian's analysis, the io-port access results in a machine reboot, not a kernel panic or so. It's scary because the controls is completely out of kernel. Perhaps an easiest but foolishest way to trace this is to put printk at each io-port access and any other important points, and give some sleep at each point, then watch the kernel message. You can get rid of spin_lock_*() around that, just for testing. Takashi