From mboxrd@z Thu Jan 1 00:00:00 1970 From: Joerg Roedel Subject: Re: Any side effect if turn off amd_iommu Date: Wed, 13 May 2015 15:44:39 +0200 Message-ID: <20150513134439.GJ20611@8bytes.org> References: <20150326120547.GU4441@8bytes.org> <20150330160910.GZ4441@8bytes.org> <20150504110802.GE15736@8bytes.org> <20150505151319.GO15736@8bytes.org> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline In-Reply-To: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: iommu-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org Errors-To: iommu-bounces-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org To: AceLan Kao Cc: iommu-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org List-Id: iommu@lists.linux-foundation.org Hi AceLan Kao, On Wed, May 13, 2015 at 03:37:57PM +0800, AceLan Kao wrote: > The method I described is not making any guess, we need a table to > write down the correct ioapic value for those machines, we can add the > BIOS version check(DMI_BIOS_VERSION) if you think the behavior may > change with different version of BIOSes. You don't know which IO-APIC is which one without a correct IVRS table. AMD desktop systems have usually have up to two IO-APICs, one in the SB and one in the NB. But the ids of these IO-APICs are not fixed and can depend on the BIOS version as well as on the processor that is plugged into the board. So since we don't know which IO-APIC is which one we can't make any good guesses that work for everybody. When we implement workarounds we may fix it for one user, but risk breakage for others. > If you still feel uncomfortable, I'm okay with any solutions, but I > don't get it how to do it in your way. > We can found those machines from bug reports(I have some machines on > my hand, too), so we need a table to record those machines? And how to > disable interrupt remapping, will it lead to any side effect? Disabling interrupt remapping has no big side effect. You only have to load vfio with a special parameter so allow unsafe interrupts to do device assignment with KVM. Joerg