From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([208.118.235.92]:59185) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1SAu79-00047K-Dn for qemu-devel@nongnu.org; Thu, 22 Mar 2012 22:20:31 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1SAu6z-0000D1-8N for qemu-devel@nongnu.org; Thu, 22 Mar 2012 22:20:26 -0400 Received: from ozlabs.org ([203.10.76.45]:36220) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1SAu6y-0000By-SJ for qemu-devel@nongnu.org; Thu, 22 Mar 2012 22:20:17 -0400 Date: Fri, 23 Mar 2012 13:02:17 +1100 From: David Gibson Message-ID: <20120323020217.GE9087@truffala.fritz.box> References: <1332382488-12869-1-git-send-email-david@gibson.dropbear.id.au> <4F6B2E21.8000001@amd.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <4F6B2E21.8000001@amd.com> Subject: Re: [Qemu-devel] [0/13] RFC: Guest visible IOMMU List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Wei Wang Cc: Joerg Roedel , mst@redhat.com, qemu-devel@nongnu.org, eduard.munteanu@linux360.ro On Thu, Mar 22, 2012 at 02:50:25PM +0100, Wei Wang wrote: > On 03/22/2012 03:14 AM, David Gibson wrote: > >Here's another version of my series implementing the nevessary > >framework for emulation of guest-visible IOMMUs. I've incorporated > >the feedback from the last posting. I'm fairly confident that > >preliminary patches 1-3 are ready to go in, the rest might still need > >more consideration. > > > >In particular, if anyone has thoughts on how to handle the case where > >an IOMMU mapping is invalidated by the guest where a qemu-side device > >currently has that piece of IOVA-space dma_memory_map()ed, I'd be > >interested to hear it. > > > > > > > > Hi David, > I am Xen developer. Currently I am working on a hardware based iommu > emulation for amd iommu v2 systems (we need this to enable iommuv2 > drivers in guest OS and therefore to passthroug gpgpu to hvm guest). > In this work, only a virtual iommu device has to be registered on > qemu bus and dma translations are still performed by iommu hardware. > So I wonder if it is possible that I could integrate my work with > yours. I had a proof-of-concept patch for qemu: > > http://lists.xen.org/archives/html/xen-devel/2011-12/msg01269.html > > my last post was: > http://lists.xen.org/archives/html/xen-devel/2012-03/msg00591.html I'd be very glad to see AMD IOMMU support integrated with my infrastructure. Eduard - Gabriel's original patch series included an AMD IOMMU emulation, but since I'm not really familiar with that IOMMU, and don't have the hardware to test it, I didn't feel confident to attempt porting it to my new infrastructure. I notice that you seem to be aiming mostly at pass-through. Pass-through has some complications of its own - this patch series was aimed firstly at IOMMU emulation, although I'm also concerned with pass-through (we're currently struggling with how best to handle pass-through for pSeries machines). -- David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson