From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: KVM host freezing Date: Mon, 06 Jun 2011 11:34:32 +0300 Message-ID: <4DEC9118.3000404@redhat.com> References: <20110602082512.GF14747@torres.zugschlus.de> <4DE7930A.3060704@redhat.com> <20110602203606.GD19565@torres.zugschlus.de> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org To: Marc Haber Return-path: Received: from mx1.redhat.com ([209.132.183.28]:20612 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752908Ab1FFIeg (ORCPT ); Mon, 6 Jun 2011 04:34:36 -0400 In-Reply-To: <20110602203606.GD19565@torres.zugschlus.de> Sender: kvm-owner@vger.kernel.org List-ID: On 06/02/2011 11:36 PM, Marc Haber wrote: > On Thu, Jun 02, 2011 at 04:41:30PM +0300, Avi Kivity wrote: > > On 06/02/2011 11:25 AM, Marc Haber wrote: > > >Is there any possibility that the freezes have to do with the > > >"unhandles (rd|wr)msr" messages? > > > > Very unlikely. > > What does that mean anyway? The guest read or wrote a model specific register which kvm does not implement. > It's looked with sufficiently high > priority to get spewed onto the console. It's done so if the problem is the cause of a guest malfunction, we'll have a clue. > > >When else could be the cause? > > > > > >In the mean time, I have taken the box offline and am running memtest. > > >Up to now, everything seems to be fine. > > > > > >Any hints will be appreciated. > > > > You might try setting up netconsole to get reliable logging. > > logging of what? Of things written to syslog before the freeze occurs > so that they don't reach the disk reliably? I could set up logging > (which facility/priority?) to the serial console. dmesg. Syslog is unlikely to get anything during a hard freeze. > > Do you have NMIs? 'grep NMI /proc/interrupts'. > > about one per minute. > > > Does running 'perf top -F 10000' make the hang come sooner? > > If that's important, I'll make some effort to compile a statically > linked perf for 2.6.39. Is it? > Yes. But why statically linked? The default should work fine. -- error compiling committee.c: too many arguments to function