From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932416Ab2DJVDo (ORCPT ); Tue, 10 Apr 2012 17:03:44 -0400 Received: from mx1.redhat.com ([209.132.183.28]:60217 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932381Ab2DJVDn (ORCPT ); Tue, 10 Apr 2012 17:03:43 -0400 Message-ID: <1334091801.3799.387.camel@bling.home> Subject: Re: [RFC PATCH 0/3] IOMMU groups From: Alex Williamson To: david@gibson.dropbear.id.au, dwmw2@infradead.org Cc: iommu@lists.linux-foundation.org, qemu-devel@nongnu.org, joerg.roedel@amd.com, kvm@vger.kernel.org, linux-kernel@vger.kernel.org Date: Tue, 10 Apr 2012 15:03:21 -0600 In-Reply-To: <20120402203721.28977.95285.stgit@bling.home> References: <20120402203721.28977.95285.stgit@bling.home> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Ping. Does this approach look like it could satisfy your desire for a more integrated group layer? I'd really like to move VFIO forward, we've been stalled on this long enough. David Woodhouse, I think this provides the quirking you're looking for for device like the Ricoh, do you have any other requirements for a group layer? Thanks, Alex On Mon, 2012-04-02 at 15:14 -0600, Alex Williamson wrote: > This series attempts to make IOMMU device grouping a slightly more > integral part of the device model. iommu_device_groups were originally > introduced to support the VFIO user space driver interface which needs > to understand the granularity of device isolation in order to ensure > security of devices when assigned for user access. This information > was provided via a simple group identifier from the IOMMU driver allowing > VFIO to walk devices and assemble groups itself. > > The feedback received from this was that groups should be the effective > unit of work for the IOMMU API. The existing model of allowing domains > to be created and individual devices attached ignores many of the > restrictions of the IOMMU, whether by design, by topology or by defective > devices. Additionally we should be able to use the grouping information > at the dma ops layer for managing domains and quirking devices. > > This series is a sketch at implementing only those aspects and leaving > everything else about the multifaceted hairball of Isolation groups for > another API. Please comment and let me know if this seems like the > direction we should be headed. Thanks, > > Alex > > > --- > > Alex Williamson (3): > iommu: Create attach/detach group interface > iommu: Create basic group infrastructure and update AMD-Vi & Intel VT-d > iommu: Introduce iommu_group > > > drivers/iommu/amd_iommu.c | 50 ++++++---- > drivers/iommu/intel-iommu.c | 76 ++++++++-------- > drivers/iommu/iommu.c | 210 ++++++++++++++++++++++++++++++++++++++----- > include/linux/device.h | 2 > include/linux/iommu.h | 43 +++++++++ > 5 files changed, 301 insertions(+), 80 deletions(-)