From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sheng Yang Subject: Re: [PATCH 0/4] MSI-X Enabling Date: Wed, 24 Dec 2008 11:00:19 +0800 Message-ID: <200812241100.20552.sheng@linux.intel.com> References: <1230019973-16833-1-git-send-email-sheng@linux.intel.com> <20081223201932.GA6783@amt.cnet> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Cc: Avi Kivity , kvm@vger.kernel.org To: Marcelo Tosatti Return-path: Received: from mga09.intel.com ([134.134.136.24]:1156 "EHLO mga09.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751233AbYLXDAg (ORCPT ); Tue, 23 Dec 2008 22:00:36 -0500 In-Reply-To: <20081223201932.GA6783@amt.cnet> Content-Disposition: inline Sender: kvm-owner@vger.kernel.org List-ID: On Wednesday 24 December 2008 04:19:32 Marcelo Tosatti wrote: > On Tue, Dec 23, 2008 at 04:12:49PM +0800, Sheng Yang wrote: > > Hi Avi > > > > This patchset would enable MSI-X support. > > > > The main change is a new kind of fake device - intercepted MMIO device is > > added for supporting to intercept one page contained MSI-X entries table. > > > > I also consider of doing this intercept in userspace. But I think it's > > would be more complex to push it in userspace, for we also have to > > prevent KVM from creating page table for the page. > > Unmapping the page in userspace (with munmap) from the PCI MMIO map will > disallow KVM from creating pagetable for the page. > > > And in the future, we would add hook > > function in the read/write operation to enabling mask bit for each MSI-X > > entry, may be with pending bits. > > Can you explain it in more detail? Can't do it from userspace? > > Are there performance considerations to access in this page? > > > So I decided to put it in the kernel. > > > > Another thing is I am not very confident about modify the position of > > handling guest memory and MMIO in emulate read/write function. Is it > > would cause any side affect? > > Should not be necessary if done in userspace. Oops... After reconsider, I think I have made a design mistake here. MMIO should can be emulated in userspace... I would update the patch. -- regards Yang, Sheng