arm: msm: Add System MMU support.
in [Kernel]
|
Prev: [PATCH 3/4] Checkpatch: prefer usleep_range over usleep
Next: [PATCH 0/4 -tip] delay documentation and checkpatch additions
From: Stepan Moskovchenko on 3 Aug 2010 14:50 On 8/3/2010 2:23 AM, Roedel, Joerg wrote: > On Mon, Aug 02, 2010 at 04:29:38PM -0400, Zach Pfeffer wrote: >> On Mon, Aug 02, 2010 at 09:58:02AM +0200, Roedel, Joerg wrote: >>> In the means of the IOMMU-API the domain is the abstraction of an >>> address space (in other words a page table). The IOMMU(s) which this domain >>> is later assigned to are determined by the iommu_attach_device calls. >>> I think the right way to go here is to create the concept of a >>> device-context in the IOMMU-API and add functions like >>> >>> iommu_attach_context(struct iommu_domain *domain, >>> struct iommu_context *ctxt); >>> iommu_detach_context(struct iommu_context *ctxt); >>> >>> This would work if you can determine in your iommu-driver which iommu >>> you need to program for which device. What do you think? >>> >> Joerg, I'd like to make sure I understand this. A domain is an address >> space separate from the actual page-tables that may be part of an >> iommu_context, correct? After I iommu_attach_context the ctxt will >> reflect the address space of the domain, correct? > A domain is defined by a single page-table which can be modified using > the iommu_map/iommu_unmap function calls. I am not completly sure what > you mean by an iommu_context. Can you describe what it means in your > context? > > Joerg > Joerg, I think with some rework, all my use cases can be handled by your existing iommu API. If the domain is treated basically a page table, there will be some changes, but I think it can be done. I will push a new version of my driver in a few days. One thing that may be helpful for the future, however, is maybe something like adding iommu_tlb_flush to the ops. I suppose this would either have to take a device, or the domain would need to keep a list of devices it had been attached to (so that their TLBs can be invalidated). But I suppose on the other hand, iommu_map/unmap may be able to just implicitly invalidate the TLB also, since TLB invalidation often follows map/unmap. What are your thoughts? Thanks Steve -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Roedel, Joerg on 4 Aug 2010 05:50
On Tue, Aug 03, 2010 at 02:43:20PM -0400, Stepan Moskovchenko wrote: > On 8/3/2010 2:23 AM, Roedel, Joerg wrote: > > A domain is defined by a single page-table which can be modified using > > the iommu_map/iommu_unmap function calls. I am not completly sure what > > you mean by an iommu_context. Can you describe what it means in your > > context? > > Joerg, > I think with some rework, all my use cases can be handled by your > existing iommu API. If the domain is treated basically a page table, > there will be some changes, but I think it can be done. I will push a > new version of my driver in a few days. > > One thing that may be helpful for the future, however, is maybe > something like adding iommu_tlb_flush to the ops. I suppose this would > either have to take a device, or the domain would need to keep a list of > devices it had been attached to (so that their TLBs can be invalidated). > But I suppose on the other hand, iommu_map/unmap may be able to just > implicitly invalidate the TLB also, since TLB invalidation often follows > map/unmap. What are your thoughts? Sounds good. I am curious for your patches :-) For the TLB-flush question, I think it would make sense to add iommu tlb flushing functions to the IOMMU-API. We currently flush the TLB implicitly in the map/unmap calls but thats very inefficient. It would be better to have a seperate function for it in the API. The right parameter for such a function is a domain. The IOMMU driver knows which devices are attached to a domain an could easily flush all TLBs. One alternative I can think of: An iommu_domain_commit() function which syncs software changes of a domain to the hardware. The map/unmap calls have to save which parts of the tlb need to be flushed and commit does flush those parts then (or flush everything). Joerg -- AMD Operating System Research Center Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach General Managers: Alberto Bozzo, Andrew Bowd Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |