Path: csiph.com!eternal-september.org!feeder.eternal-september.org!aioe.org!news.servidellagleba.it!bofh.it!news.nic.it!robomod From: Joerg Roedel Newsgroups: linux.kernel Subject: Re: [PATCH v2] PCI: Add ATS-disable quirk for AMD Stoney GPUs Date: Thu, 20 Apr 2017 14:20:02 +0200 Message-ID: References: X-Original-To: Lukas Wunner MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline User-Agent: Mutt/1.5.24 (2015-08-30) Sender: robomod@news.nic.it List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Approved: robomod@news.nic.it Lines: 27 Organization: linux.* mail to news gateway X-Original-Cc: Joerg Roedel , Bjorn Helgaas , linux-pci@vger.kernel.org, linux-kernel@vger.kernel.org, Daniel Drake , Alexander Deucher , Samuel Sieb , David Woodhouse , Paul Menzel X-Original-Date: Thu, 20 Apr 2017 14:11:42 +0200 X-Original-Message-ID: <20170420121142.GI5077@suse.de> X-Original-References: <1491575538-22694-1-git-send-email-joro@8bytes.org> <20170408074107.GA3585@wunner.de> X-Original-Sender: linux-kernel-owner@vger.kernel.org Xref: csiph.com linux.kernel:1627408 On Sat, Apr 08, 2017 at 09:41:07AM +0200, Lukas Wunner wrote: > On Fri, Apr 07, 2017 at 04:32:18PM +0200, Joerg Roedel wrote: > > From: Joerg Roedel > > > > ATS is broken on this hardware and causes IOMMU stalls and > > system failure. Disable ATS on these devices to make them > > usable again with IOMMU enabled. > > AMD Stoney Ridge is an x86 CPU + GPU combo and this quirk pertains > to the GPU, right? > > In that case the quirk should go to arch/x86. Paul Menzel (+cc) > has just complained on linux-pci@ that final fixups are taking half > a second, and I think that could be reduced if more efforts were > spent to move arch-specific quirks out of the catch-all in > drivers/pci/quirks.c. The affected hardware here might be x86-only, but ATS is not. If a broken ATS-capable plug-in card appears, we need this in generic code anyway. Also has anyone profiled why the fixups take so long (and on what hardware)? Maybe the fixup-device matching can be improved instead of cluttering arch-code with pci-fixups. Joerg