From: ebiederm@xmission.com (Eric W. Biederman) Subject: Re: [PATCH v2 2/2] Intel-IOMMU, intr-remap: source-id checking Date: Wed, 20 May 2009 05:02:23 -0700 Message-ID: References: <1242757912-6041-1-git-send-email-weidong.han@intel.com> <1242757912-6041-3-git-send-email-weidong.han@intel.com> <20090519115055.GB14305@elte.hu> <20090520082114.GC6736@elte.hu> <20090520094310.GF9835@amd.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Ingo Molnar , Yinghai Lu , Weidong Han , dwmw2@infradead.org, suresh.b.siddha@intel.com, linux-kernel@vger.kernel.org, iommu@lists.linux-foundation.org, kvm@vger.kernel.org To: Joerg Roedel Return-path: Received: from out01.mta.xmission.com ([166.70.13.231]:56445 "EHLO out01.mta.xmission.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754771AbZETMC3 (ORCPT ); Wed, 20 May 2009 08:02:29 -0400 In-Reply-To: <20090520094310.GF9835@amd.com> (Joerg Roedel's message of "Wed\, 20 May 2009 11\:43\:10 +0200") Sender: kvm-owner@vger.kernel.org List-ID: Joerg Roedel writes: > On Wed, May 20, 2009 at 10:21:14AM +0200, Ingo Molnar wrote: >> >> * Eric W. Biederman wrote: >> > Being a major nitpick, I have to point out that the code is not >> > structured to support other iommus, and I think AMD has one that >> > can do this as well. >> >> (Joerg Cc:-ed) > > The AMD IOMMU does also have an interrupt remapping feature. But it is > currently unsupported in the Linux driver. When this is going to be > implemented I think we can go the same way as for device passthrough in > KVM. This was an Intel-only feature too and was later adapted to support > AMD IOMMU too. The Intel IOMMU is also interesting because it allows traffic from buggy ioapcis to be remapped safely in process context, adding a large margin of safety and flexibility to the existing irq migration and cpu hotunplug scenarios. Can we use the AMD IOMMU to achieve the same objective? I would really like to say from point X on AMD and Intel systems don't have to worry about irq migrations used caused by old buggy ioapic state machines, which I have previously confirmed in both AMD and Intel ioapics. Eric