From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: vm performance degradation after kvm live migration or save-restore with EPT enabled Date: Thu, 08 Aug 2013 14:29:04 +0200 Message-ID: <52038F10.5070900@redhat.com> References: Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Xiao Guangrong , Xiejunyong , "Huangweidong (C)" , KVM , Gleb Natapov , "Michael S. Tsirkin" , Luonengjun , Xiahai , Marcelo Tosatti , "paolo.bonzini@gmail.com" , qemu-devel , Bruce Rogers , Zanghongyong , Xin Rong Fu , Yi Li , "xiaoguangrong@linux.vnet.ibm.com" , Hanweidong , =?UTF-8?B?QW5kcmVhcyBGw6RyYmVy?= To: "Zhanghaoyu (A)" Return-path: Received: from mail-ee0-f46.google.com ([74.125.83.46]:43977 "EHLO mail-ee0-f46.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1757776Ab3HHM3M (ORCPT ); Thu, 8 Aug 2013 08:29:12 -0400 Received: by mail-ee0-f46.google.com with SMTP id c13so1459115eek.19 for ; Thu, 08 Aug 2013 05:29:11 -0700 (PDT) In-Reply-To: Sender: kvm-owner@vger.kernel.org List-ID: On 08/08/2013 01:31 PM, Zhanghaoyu (A) wrote: > And, rebuild the kvm-kmod, then re-insmod kvm-intel.ko with ept=3D0= , > > I virsh-save the VM, and then virsh-restore the VM, the performance > degradation disappeared, and no GFN printed. > > But, I also made a test on the first bad > commit(612819c3c6e67bac8fceaa7cc402f13b1b63f7e4), and applied above > patch too, > > With EPT disabled, as soon as the restoring completed, the GFNs=E2=80= =99 > flooding is starting, take some examples to show as below, > > but, after a period of time, only gfn =3D 240 printed constantly. An= d > some processes restore failed, so the performance cannot be measured. This is 0xF0000 which is the BIOS. It makes some sense, though I'm not= =20 sure why a running OS would poke there. It can be a useful hint, but please do as Gleb said. Try your kernel=20 (+kvm-kmod) with current QEMU, current kernel with your QEMU, current=20 kernel with current QEMU. Make a table of which kernel/QEMU combos wor= k=20 and which do not. Then we can try to understand if the bug still exist= s=20 and perhaps get hints about how your vendor could backport the fix. Paolo From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:34333) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1V7PLF-0004Ie-JJ for qemu-devel@nongnu.org; Thu, 08 Aug 2013 08:29:30 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1V7PL6-0001Yv-Av for qemu-devel@nongnu.org; Thu, 08 Aug 2013 08:29:21 -0400 Received: from mail-ea0-x22e.google.com ([2a00:1450:4013:c01::22e]:56671) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1V7PL6-0001Xc-4v for qemu-devel@nongnu.org; Thu, 08 Aug 2013 08:29:12 -0400 Received: by mail-ea0-f174.google.com with SMTP id z15so1383117ead.19 for ; Thu, 08 Aug 2013 05:29:11 -0700 (PDT) Sender: Paolo Bonzini Message-ID: <52038F10.5070900@redhat.com> Date: Thu, 08 Aug 2013 14:29:04 +0200 From: Paolo Bonzini MIME-Version: 1.0 References: In-Reply-To: Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 8bit Subject: Re: [Qemu-devel] vm performance degradation after kvm live migration or save-restore with EPT enabled List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: "Zhanghaoyu (A)" Cc: "xiaoguangrong@linux.vnet.ibm.com" , Marcelo Tosatti , "Huangweidong (C)" , Xiao Guangrong , Gleb Natapov , "Michael S. Tsirkin" , "paolo.bonzini@gmail.com" , Xiejunyong , Luonengjun , qemu-devel , Xiahai , Zanghongyong , Xin Rong Fu , Yi Li , KVM , Bruce Rogers , Hanweidong , =?UTF-8?B?QW5kcmVhcyBGw6RyYmVy?= On 08/08/2013 01:31 PM, Zhanghaoyu (A) wrote: > And, rebuild the kvm-kmod, then re-insmod kvm-intel.ko with ept=0, > > I virsh-save the VM, and then virsh-restore the VM, the performance > degradation disappeared, and no GFN printed. > > But, I also made a test on the first bad > commit(612819c3c6e67bac8fceaa7cc402f13b1b63f7e4), and applied above > patch too, > > With EPT disabled, as soon as the restoring completed, the GFNs’ > flooding is starting, take some examples to show as below, > > but, after a period of time, only gfn = 240 printed constantly. And > some processes restore failed, so the performance cannot be measured. This is 0xF0000 which is the BIOS. It makes some sense, though I'm not sure why a running OS would poke there. It can be a useful hint, but please do as Gleb said. Try your kernel (+kvm-kmod) with current QEMU, current kernel with your QEMU, current kernel with current QEMU. Make a table of which kernel/QEMU combos work and which do not. Then we can try to understand if the bug still exists and perhaps get hints about how your vendor could backport the fix. Paolo