From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S261997AbVEQW0s (ORCPT ); Tue, 17 May 2005 18:26:48 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S262030AbVEQWXw (ORCPT ); Tue, 17 May 2005 18:23:52 -0400 Received: from fire.osdl.org ([65.172.181.4]:59066 "EHLO smtp.osdl.org") by vger.kernel.org with ESMTP id S262017AbVEQWQt (ORCPT ); Tue, 17 May 2005 18:16:49 -0400 Date: Tue, 17 May 2005 15:16:48 -0700 From: Andrew Morton To: "Martin J. Bligh" Cc: dev@sw.ru, torvalds@osdl.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] NMI watchdog config option (was: Re: [PATCH] NMI lockup and AltSysRq-P dumping calltraces on _all_ cpus via NMI IPI) Message-Id: <20050517151648.2abff61e.akpm@osdl.org> In-Reply-To: <293160000.1116338500@[10.10.2.4]> References: <42822B5F.8040901@sw.ru> <768860000.1116282855@flay> <42899797.2090702@sw.ru> <20050517001542.40e6c6b7.akpm@osdl.org> <293160000.1116338500@[10.10.2.4]> X-Mailer: Sylpheed version 1.0.0 (GTK+ 1.2.10; i386-vine-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org "Martin J. Bligh" wrote: > > > So much has changed in there that we might have fixed it by accident, and I > > do recall a couple of fundamental and subtle NMI bugs being fixed. So > > yeah, it might be worth enabling it by default again. Care to send a patch > > which does that? > > There are some unfixable machine issues - for instance, the IBM > Netfinity 8500R corrupts one of the registers (ebx?) every time we get > an NMI for us, and panics. Probably other boxes you mention above have > similar issues? But it's not our code that's at fault ... That sounds like an instant crash. The problems which were reported a few years back were different - mysterious lockups after hours or days of operation. > In light of this, I don't think it's a good idea to enable NMI by default, > at least not without a blacklist function of some sort? OK, thanks - I'll leave things as they stand.