From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756247Ab0KJTOy (ORCPT ); Wed, 10 Nov 2010 14:14:54 -0500 Received: from mx2.mail.elte.hu ([157.181.151.9]:56680 "EHLO mx2.mail.elte.hu" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755862Ab0KJTOx (ORCPT ); Wed, 10 Nov 2010 14:14:53 -0500 Date: Wed, 10 Nov 2010 20:14:32 +0100 From: Ingo Molnar To: Don Zickus Cc: fweisbec@gmail.com, Peter Zijlstra , LKML , akpm@linux-foundation.org, sergey.senozhatsky@gmail.com Subject: Re: [PATCH v3] watchdog: touch_nmi_watchdog should only touch local cpu not every one Message-ID: <20101110191432.GB30227@elte.hu> References: <1289240913-16677-1-git-send-email-dzickus@redhat.com> <20101110074941.GD29493@elte.hu> <20101110160516.GZ4823@redhat.com> <20101110180711.GB22410@elte.hu> <20101110190352.GA31794@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20101110190352.GA31794@redhat.com> User-Agent: Mutt/1.5.20 (2009-08-17) X-ELTE-SpamScore: -2.0 X-ELTE-SpamLevel: X-ELTE-SpamCheck: no X-ELTE-SpamVersion: ELTE 2.0 X-ELTE-SpamCheck-Details: score=-2.0 required=5.9 tests=BAYES_00 autolearn=no SpamAssassin version=3.2.5 -2.0 BAYES_00 BODY: Bayesian spam probability is 0 to 1% [score: 0.0000] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Don Zickus wrote: > On Wed, Nov 10, 2010 at 07:07:11PM +0100, Ingo Molnar wrote: > > > > * Don Zickus wrote: > > > > > > Hm, the flip side is that if a CPU is stuck spewing backtraces, we will now make > > > > all the other CPUs a lot more noisy - which might only 'lock up' because this > > > > CPU is stuck spewing oopses, right? > > > > > > When you say the other CPUs will be a lot more noisy, is that because they are > > > busy processing backtraces for the first cpu to spew? I guess I don't understand > > > how the other CPUs could have their interrupts off the whole time while the first > > > cpu is spewing a backtrace (just trying to educate myself). > > > > Say the kernel crashes on a CPU and keeps spewing new oopses, while write-holding > > tasklist_lock. > > > > Any other CPU that delivers a signal from IRQ context, trying to take the > > tasklist_lock, will loop indefinitely until that crashing CPU releases the lock. > > > > In that case the 'secondary' NMI warnings from all other CPUs (eventually every CPU > > gets stuck in such a scenario) will start spewing NMI lockup messages. > > > > Dunno. Maybe we should do your change - but also have an option to 'shut up' the > > kernel after the first hard oops [not warning]. That would silence the secondary NMI > > watchdog messages as well. > > You mean inside the panic() routine? like a ratelimit? A ratelimit, but some really serious one - like only one crash displayed per 10 minutes, or so - to give the user time to make a picture of the first crash, if it's still visible on the screen or so. Thanks, Ingo