From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: high load with win7 & usb tablet Date: Fri, 10 Sep 2010 15:50:46 +0300 Message-ID: <4C8A29A6.60108@redhat.com> References: <4C80E839.50604@msgid.tls.msk.ru> <4C8A2254.7090505@redhat.com> <4C8A282A.5010506@msgid.tls.msk.ru> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: KVM list To: Michael Tokarev Return-path: Received: from mx1.redhat.com ([209.132.183.28]:51668 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753642Ab0IJMuy (ORCPT ); Fri, 10 Sep 2010 08:50:54 -0400 In-Reply-To: <4C8A282A.5010506@msgid.tls.msk.ru> Sender: kvm-owner@vger.kernel.org List-ID: On 09/10/2010 03:44 PM, Michael Tokarev wrote: > 10.09.2010 16:19, Avi Kivity wrote: >> On 09/03/2010 03:21 PM, Michael Tokarev wrote: >>> Hello. >>> >>> I noticed that window7, unlike windowsXP as I played with >>> previously, poses quite high load on host when idle. On >>> my machine, host cpu usage is about 25% when one single-cpu >>> win7 guest is running with "-usbdevice tablet" option. >>> Here's a typical kvm_stats when such a guest is idling: > [] >> That's pretty good for an idle guest. > Hm. Well. Sometimes it goes "a bit" further than that: > > efer_reload 0 0 > exits 6307547 12445 This still isn't worth 25% cpu. >> What does the guest think it is doing? The Windows performance monitor? > Nothing. It shows perfmon.exe using 3% CPU, and that's about > it. I'm attaching a .png screenshot of the KVM window with > w7 perfmon running. Unfortunately I don't have English > version handy, it's all in Russian, and I weren't able to > change system language to English. But it is still readable. > > During this shot, host CPU was about 30..35% in use - higher > than usual 25%, but I guess it's due to perfmon. Please build qemu with --disable-strip and use 'perf' to profile it. > Also, I haven't found a way to see, for example, individual > interrupt activity in windows (like /proc/interrupts in > linux). irq_injections is very low anyway. -- I have a truly marvellous patch that fixes the bug which this signature is too narrow to contain.