From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([208.118.235.92]:59619) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1UIGse-0004jm-Sw for qemu-devel@nongnu.org; Wed, 20 Mar 2013 07:08:34 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1UIGsd-0002rc-Iz for qemu-devel@nongnu.org; Wed, 20 Mar 2013 07:08:28 -0400 Received: from mail-ve0-f176.google.com ([209.85.128.176]:36938) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1UIGsd-0002rQ-FA for qemu-devel@nongnu.org; Wed, 20 Mar 2013 07:08:27 -0400 Received: by mail-ve0-f176.google.com with SMTP id cz10so1253404veb.35 for ; Wed, 20 Mar 2013 04:08:27 -0700 (PDT) Sender: Paolo Bonzini Message-ID: <514998A3.9090903@redhat.com> Date: Wed, 20 Mar 2013 12:08:19 +0100 From: Paolo Bonzini MIME-Version: 1.0 References: <1363609123-20748-1-git-send-email-alex@alex.org.uk> <51471767.8030604@redhat.com> <7AC8953FE45335FB794B6DFE@Ximines.local> <51471F14.7030209@redhat.com> <6D0F4ACDA3B7FCF1A50F8B52@Ximines.local> <5147298C.8080900@redhat.com> <51473E82.1020806@redhat.com> <861AFE1A9C44444FD8BAEE16@Ximines.local> <5147512E.5050501@eu.citrix.com> <5147562E.1090203@redhat.com> <5148414C.7060303@redhat.com> <51484328.4030301@eu.citrix.com> <5148489C.70808@redhat.com> <51484A2B.10202@eu.citrix.com> <0E09404F0675BA19C550B743@nimrod.local> <51499157.2060202@eu.citrix.com> In-Reply-To: <51499157.2060202@eu.citrix.com> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [Xen-devel] [PATCHv3] QEMU(upstream): Disable xen's use of O_DIRECT by default as it results in crashes. List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: George Dunlap Cc: Ian Campbell , Stefano Stabellini , Ian Jackson , "qemu-devel@nongnu.org" , xen-devel , Alex Bligh , Anthony Liguori Il 20/03/2013 11:37, George Dunlap ha scritto: >> That is true. My guess is that nobody really migrates HVM guests without >> PV drivers installed (it's not even possible on XenServer but xl let you >> do that if you want to). When the PV drivers initialize at boot time, >> the IDE disk is closed. Therefore we wouldn't have this problem. >> >> Maybe we should prevent HVM guest migration without PV drivers with xl >> too. Ian, what do you think? > > We should be able to make it safe if we just make sure that qemu does a > metadata sync / FS sync when we ask qemu to write the save file. To make it safe, just use cache=none. There are downsides (for example tmpfs does not support it), but perhaps you can use a global option or environment variable to toggle the behavior. Is xl still using the driver:subdriver:/path,readonly syntax? That's really inflexible compared to libvirt's XML... Paolo