RFC: Unified DMA allocation algorithms
Hi all,
I've been looking at a better way to do custom dma allocation algorithms in a similar style to Ion heaps. Most drivers/clients have come up with a series of semi-standard ways to get memory (CMA, memblock_reserve, discontiguous pages etc.) . As these allocation schemes get more and more complex, there needs to be a since place where all clients (Ion based driver vs. DRM driver vs. ???) can independently take advantage of any optimizations and call a single API for the backing allocations.
The dma_map_ops take care of almost everything needed for abstraction but the question is where should new allocation algorithms be located? Most of the work has been added to either arm/mm/dma-mapping.c or dma-contiguous.c . My current thought:
1) split out the dma_map_ops currently in dma-mapping.c into separate files (dma-mapping-common.c, dma-mapping-iommu.c)
2) Extend dma-contiguous.c to support memblock_reserve memory
3) Place additional algorithms in either arch/arm/mm or drivers/base/dma-alloc/ as appropriate to the code. This is the part where I'm most unsure about the direction.
I don't have anything written yet but I plan to draft some patches assuming the proposed approach sounds reasonable and no one else has started on something similar already.
Thoughts? Opinions?
Thanks,
Laura
Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
Quoted 22 lines · Laura Abbott
Hi all,
I've been looking at a better way to do custom dma allocation algorithms in a similar style to Ion heaps. Most
drivers/clientshave come up with a series of semi-standard ways to get memory (CMA,memblock_reserve, discontiguous pages etc.) . As these allocation schemes get more and more complex, there needs to be a since place where all clients (Ion based driver vs. DRM driver vs. ???) can independently take advantage of any optimizations and call a single API for the backing allocations.The
dma_map_opstake care of almost everything needed for abstraction but the question is where should new allocation algorithms be located? Most of the work has been added to eitherarm/mm/dma-mapping.cordma-contiguous.c. My current thought:1) split out the
dma_map_opscurrently indma-mapping.cinto separate files (dma-mapping-common.c,dma-mapping-iommu.c)
2) Extenddma-contiguous.cto supportmemblock_reservememory
3) Place additional algorithms in eitherarch/arm/mmordrivers/base/dma-alloc/as appropriate to the code. This is the part where I'm most unsure about the direction.
Quoted 7 lines
I don't have anything written yet but I plan to draft some patches assuming the proposed approach sounds reasonable and no one else has started on something similar already.
Thoughts? Opinions?
From my (oblivious to all the arm madness) pov the big thing is
getting dma allocations working for more than one struct device. This way we could get rid of to "where do I need to allocate buffers" duplication between the kernel and userspace (which needs to know that to pick the right ion heap), which is my main gripe with ion ;-)
Rob Clark sent out a quick rfc for that a while back:
http://lists.linaro.org/pipermail/linaro-mm-sig/2012-July/002250.html
But that's by far not good enough for arm, especially now that cma gets tightly bound to individual devices with the dt bindings. Also, no one really followed up on Rob's patches, and personally I don't really care that much since x86 is a bit saner ... But it should be good enough for contiguous allocations, which leaves only really crazy stuff unsolved.
So I think when you want to rework the various algorithms for allocating dma mem and consolidate them it should also solve this little multi-dev issue.
Adding tons more people/lists who might be interested.
Cheers, Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch
Quoted 1 line · Laura Abbott
Hi all,
Hi Laura,
Quoted 7 lines
I've been looking at a better way to do custom dma allocation algorithms in a similar style to Ion heaps. Most
drivers/clientshave come up with a series of semi-standard ways to get memory (CMA,memblock_reserve, discontiguous pages etc.) . As these allocation schemes get more and more complex, there needs to be a since place where all clients (Ion based driver vs. DRM driver vs. ???) can independently take advantage of any optimizations and call a single API for the backing allocations.
Makes sense.
Quoted 12 lines
The
dma_map_opstake care of almost everything needed for abstraction but the question is where should new allocation algorithms be located? Most of the work has been added to eitherarm/mm/dma-mapping.cordma-contiguous.c. My current thought:1) split out the
dma_map_opscurrently indma-mapping.cinto separate files (dma-mapping-common.c,dma-mapping-iommu.c)
2) Extenddma-contiguous.cto supportmemblock_reservememory
3) Place additional algorithms in eitherarch/arm/mmordrivers/base/dma-alloc/as appropriate to the code. This is the part where I'm most unsure about the direction.
Quoted 5 lines
I don't have anything written yet but I plan to draft some patches assuming the proposed approach sounds reasonable and no one else has started on something similar already.
Thoughts? Opinions?
The only thing I'd add (which unfortunately doesn't help to answer your questions!) is that it would be good to see the various allocation strategies parameterised by IOMMU constraints. For example:
- Input address sizes / windows
- Mapping granules (i.e. not just PAGE_SIZE)
- Streaming hints for physically contiguous allocation
dma-mapping does some of this, but it's still not ideal or generalised (for example, it allocates a bitmap for the entire virtual space for each domain). I'd welcome any generalisation in this area.
Will