From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH 5/9] PCI: host: brcmstb: add dma-ranges for inbound traffic Date: Wed, 18 Oct 2017 08:53:16 +0200 Message-ID: <20171018065316.GA11183@lst.de> References: <1507761269-7017-1-git-send-email-jim2101024@gmail.com> <1507761269-7017-6-git-send-email-jim2101024@gmail.com> <589c04cb-061b-a453-3188-79324a02388e@arm.com> <20171017081422.GA19475@lst.de> 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: Jim Quinlan Cc: Mark Rutland , linux-mips-6z/3iImG2C8G8FEW9MqTrA@public.gmane.org, Florian Fainelli , devicetree-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, linux-pci , Kevin Cernekee , Will Deacon , linux-kernel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org, Ralf Baechle , iommu-cunTk1MwBs9QetFLy7KEm3xJsTq8ys+cHZ5vskTnxNA@public.gmane.org, Rob Herring , bcm-kernel-feedback-list , Gregory Fong , Catalin Marinas , Bjorn Helgaas , Brian Norris , Christoph Hellwig , linux-arm-kernel-IAPFreCvJWM7uuMidbF8XUB+6BGkLq7r@public.gmane.org List-Id: iommu@lists.linux-foundation.org On Tue, Oct 17, 2017 at 12:11:55PM -0400, Jim Quinlan wrote: > My understanding is that dma_pfn_offset is that it is a single > constant offset from RAM, in our case, to map to PCIe space. Yes. > But in > my commit message I detail how our PCIe controller presents memory > with multiple regions with multiple different offsets. If an EP device > maps to a region on the host memory, yes we can set the dma_pfn_offset > for that device for that location within that range,. But if the > device then unmaps and allocates from another region with a different > offset, it won't work. If I set dma_pfn_offset I have to assume that > the device is using only one region of memory only, not more than one, > and that it is not unmapping that region and mapping another (with a > different offset). Can I make those assumptions? No, we can't make that assumption unfortunately. But how is your code going to work if the mapping spans multiple of your translation regions? Also I really don't think the stacking of dma_ops as in this patch is a good idea. For MIPS you should do the variant suggested in the patch description and hook into dma_to_phys/phys_to_dma helpers, and for ARM/ARM64 you should talk to the maintainers on how they want the translation integrated. From mboxrd@z Thu Jan 1 00:00:00 1970 Received: with ECARTIS (v1.0.0; list linux-mips); Wed, 18 Oct 2017 08:53:27 +0200 (CEST) Received: from verein.lst.de ([213.95.11.211]:53757 "EHLO newverein.lst.de" rhost-flags-OK-OK-OK-OK) by eddie.linux-mips.org with ESMTP id S23990498AbdJRGxUjWUdI (ORCPT ); Wed, 18 Oct 2017 08:53:20 +0200 Received: by newverein.lst.de (Postfix, from userid 2407) id ECE9A68C40; Wed, 18 Oct 2017 08:53:16 +0200 (CEST) Date: Wed, 18 Oct 2017 08:53:16 +0200 From: Christoph Hellwig To: Jim Quinlan Cc: Christoph Hellwig , Robin Murphy , linux-kernel@vger.kernel.org, Mark Rutland , linux-mips@linux-mips.org, Florian Fainelli , devicetree@vger.kernel.org, linux-pci , Kevin Cernekee , Will Deacon , Ralf Baechle , Rob Herring , bcm-kernel-feedback-list , Gregory Fong , Catalin Marinas , Bjorn Helgaas , Brian Norris , linux-arm-kernel@lists.infradead.org, Marek Szyprowski , iommu@lists.linux-foundation.org Subject: Re: [PATCH 5/9] PCI: host: brcmstb: add dma-ranges for inbound traffic Message-ID: <20171018065316.GA11183@lst.de> References: <1507761269-7017-1-git-send-email-jim2101024@gmail.com> <1507761269-7017-6-git-send-email-jim2101024@gmail.com> <589c04cb-061b-a453-3188-79324a02388e@arm.com> <20171017081422.GA19475@lst.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.17 (2007-11-01) Return-Path: X-Envelope-To: <"|/home/ecartis/ecartis -s linux-mips"> (uid 0) X-Orcpt: rfc822;linux-mips@linux-mips.org Original-Recipient: rfc822;linux-mips@linux-mips.org X-archive-position: 60441 X-ecartis-version: Ecartis v1.0.0 Sender: linux-mips-bounce@linux-mips.org Errors-to: linux-mips-bounce@linux-mips.org X-original-sender: hch@lst.de Precedence: bulk List-help: List-unsubscribe: List-software: Ecartis version 1.0.0 List-Id: linux-mips X-List-ID: linux-mips List-subscribe: List-owner: List-post: List-archive: X-list: linux-mips On Tue, Oct 17, 2017 at 12:11:55PM -0400, Jim Quinlan wrote: > My understanding is that dma_pfn_offset is that it is a single > constant offset from RAM, in our case, to map to PCIe space. Yes. > But in > my commit message I detail how our PCIe controller presents memory > with multiple regions with multiple different offsets. If an EP device > maps to a region on the host memory, yes we can set the dma_pfn_offset > for that device for that location within that range,. But if the > device then unmaps and allocates from another region with a different > offset, it won't work. If I set dma_pfn_offset I have to assume that > the device is using only one region of memory only, not more than one, > and that it is not unmapping that region and mapping another (with a > different offset). Can I make those assumptions? No, we can't make that assumption unfortunately. But how is your code going to work if the mapping spans multiple of your translation regions? Also I really don't think the stacking of dma_ops as in this patch is a good idea. For MIPS you should do the variant suggested in the patch description and hook into dma_to_phys/phys_to_dma helpers, and for ARM/ARM64 you should talk to the maintainers on how they want the translation integrated. From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Return-Path: Date: Wed, 18 Oct 2017 08:53:16 +0200 From: Christoph Hellwig To: Jim Quinlan Subject: Re: [PATCH 5/9] PCI: host: brcmstb: add dma-ranges for inbound traffic Message-ID: <20171018065316.GA11183@lst.de> References: <1507761269-7017-1-git-send-email-jim2101024@gmail.com> <1507761269-7017-6-git-send-email-jim2101024@gmail.com> <589c04cb-061b-a453-3188-79324a02388e@arm.com> <20171017081422.GA19475@lst.de> MIME-Version: 1.0 In-Reply-To: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Mark Rutland , linux-mips@linux-mips.org, Florian Fainelli , devicetree@vger.kernel.org, linux-pci , Kevin Cernekee , Will Deacon , linux-kernel@vger.kernel.org, Ralf Baechle , iommu@lists.linux-foundation.org, Rob Herring , bcm-kernel-feedback-list , Gregory Fong , Catalin Marinas , Bjorn Helgaas , Brian Norris , Robin Murphy , Christoph Hellwig , linux-arm-kernel@lists.infradead.org, Marek Szyprowski Content-Type: text/plain; charset="us-ascii" Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+bjorn=helgaas.com@lists.infradead.org List-ID: On Tue, Oct 17, 2017 at 12:11:55PM -0400, Jim Quinlan wrote: > My understanding is that dma_pfn_offset is that it is a single > constant offset from RAM, in our case, to map to PCIe space. Yes. > But in > my commit message I detail how our PCIe controller presents memory > with multiple regions with multiple different offsets. If an EP device > maps to a region on the host memory, yes we can set the dma_pfn_offset > for that device for that location within that range,. But if the > device then unmaps and allocates from another region with a different > offset, it won't work. If I set dma_pfn_offset I have to assume that > the device is using only one region of memory only, not more than one, > and that it is not unmapping that region and mapping another (with a > different offset). Can I make those assumptions? No, we can't make that assumption unfortunately. But how is your code going to work if the mapping spans multiple of your translation regions? Also I really don't think the stacking of dma_ops as in this patch is a good idea. For MIPS you should do the variant suggested in the patch description and hook into dma_to_phys/phys_to_dma helpers, and for ARM/ARM64 you should talk to the maintainers on how they want the translation integrated. _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel From mboxrd@z Thu Jan 1 00:00:00 1970 From: hch@lst.de (Christoph Hellwig) Date: Wed, 18 Oct 2017 08:53:16 +0200 Subject: [PATCH 5/9] PCI: host: brcmstb: add dma-ranges for inbound traffic In-Reply-To: References: <1507761269-7017-1-git-send-email-jim2101024@gmail.com> <1507761269-7017-6-git-send-email-jim2101024@gmail.com> <589c04cb-061b-a453-3188-79324a02388e@arm.com> <20171017081422.GA19475@lst.de> Message-ID: <20171018065316.GA11183@lst.de> To: linux-arm-kernel@lists.infradead.org List-Id: linux-arm-kernel.lists.infradead.org On Tue, Oct 17, 2017 at 12:11:55PM -0400, Jim Quinlan wrote: > My understanding is that dma_pfn_offset is that it is a single > constant offset from RAM, in our case, to map to PCIe space. Yes. > But in > my commit message I detail how our PCIe controller presents memory > with multiple regions with multiple different offsets. If an EP device > maps to a region on the host memory, yes we can set the dma_pfn_offset > for that device for that location within that range,. But if the > device then unmaps and allocates from another region with a different > offset, it won't work. If I set dma_pfn_offset I have to assume that > the device is using only one region of memory only, not more than one, > and that it is not unmapping that region and mapping another (with a > different offset). Can I make those assumptions? No, we can't make that assumption unfortunately. But how is your code going to work if the mapping spans multiple of your translation regions? Also I really don't think the stacking of dma_ops as in this patch is a good idea. For MIPS you should do the variant suggested in the patch description and hook into dma_to_phys/phys_to_dma helpers, and for ARM/ARM64 you should talk to the maintainers on how they want the translation integrated.