From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sheng Yang Subject: Re: [PATCH 0/3] Emulate MSI-X mask bits for assigned devices Date: Mon, 11 Oct 2010 17:34:24 +0800 Message-ID: <201010111734.24896.sheng@linux.intel.com> References: <1285667052-24907-1-git-send-email-sheng@linux.intel.com> <20100930164451.GC29881@amt.cnet> Mime-Version: 1.0 Content-Type: Text/Plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Cc: Avi Kivity , kvm@vger.kernel.org To: Marcelo Tosatti Return-path: Received: from mga01.intel.com ([192.55.52.88]:53249 "EHLO mga01.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753483Ab0JKJeU (ORCPT ); Mon, 11 Oct 2010 05:34:20 -0400 In-Reply-To: <20100930164451.GC29881@amt.cnet> Sender: kvm-owner@vger.kernel.org List-ID: On Friday 01 October 2010 00:44:51 Marcelo Tosatti wrote: > On Tue, Sep 28, 2010 at 05:44:09PM +0800, Sheng Yang wrote: > > Hi Avi & Marcelo > > > > This patchset would add emulation of MSI-X mask bit for assigned devices > > in QEmu. > > > > BTW: We are also purposed an acceleration of MSI-X mask bit for KVM - to > > get it done in kernel. That's because sometime MSI-X mask bit was > > accessed very frequently by guest and emulation in QEmu cost a lot(tens > > to hundreds percent CPU utilization sometime), e.g. guest using Linux > > 2.6.18 or under some heavy workload. The method has been proved > > efficient in Xen, but it would require VMM to handle MSI-X table. Is > > that OK with you? > > It would be good to understand where that cost is, before. Eg. do you > commit kvm irq route frequently when guest unmasks? I don't think it would cause performance issue, but maybe need another bulk of codes in kernel. I would post the in-kernel patch later, which is easier for discuss. :) -- regards Yang, Sheng