CMA on AArch64

CMA on AArch64

[Laura Abbott]

Hi,

I noticed there is currently no CMA support for AArch64. Is this already 
on someone's TODO list or is this still open?

Thanks,
Laura

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation

CMA on AArch64

[Catalin Marinas]

Hi Laura,

On Wed, Nov 20, 2013 at 09:41:18PM +0000, Laura Abbott wrote:
> I noticed there is currently no CMA support for AArch64. Is this already 
> on someone's TODO list or is this still open?

It's more like a wish list but there were other things that took
priority. The arm64 DMA mapping code currently only supports swiotlb and
our next plan is for IOMMU mapping. But I would like parts of the DMA
IOMMU code under arch/arm to be made into a library (similar to the
lib/swiotlb.c) so that other architectures can easily make use of it.

Yet another feature is DMA cache coherency management (primarily
streaming DMA) and I have a patch but I want this driven via DT rather
than drivers or SoC code calling set_dma_ops explicitly (the bindings
definition is work in progress).

Anyway, I think the CMA support wouldn't interfere much with the above
and could be merged. If you have time, please feel free to send some
patches ;).

-- 
Catalin

CMA on AArch64

[Russell King - ARM Linux]

On Thu, Nov 21, 2013 at 12:06:00PM +0000, Catalin Marinas wrote:
> Hi Laura,
> 
> On Wed, Nov 20, 2013 at 09:41:18PM +0000, Laura Abbott wrote:
> > I noticed there is currently no CMA support for AArch64. Is this already 
> > on someone's TODO list or is this still open?
> 
> It's more like a wish list but there were other things that took
> priority. The arm64 DMA mapping code currently only supports swiotlb and
> our next plan is for IOMMU mapping. But I would like parts of the DMA
> IOMMU code under arch/arm to be made into a library (similar to the
> lib/swiotlb.c) so that other architectures can easily make use of it.

If you're referring to dmabounce.c, that's a 100% no-way-ever-even-if-
hell-freezes-over NACK.

CMA on AArch64

[Catalin Marinas]

On Thu, Nov 21, 2013 at 12:11:35PM +0000, Russell King - ARM Linux wrote:
> On Thu, Nov 21, 2013 at 12:06:00PM +0000, Catalin Marinas wrote:
> > On Wed, Nov 20, 2013 at 09:41:18PM +0000, Laura Abbott wrote:
> > > I noticed there is currently no CMA support for AArch64. Is this already 
> > > on someone's TODO list or is this still open?
> > 
> > It's more like a wish list but there were other things that took
> > priority. The arm64 DMA mapping code currently only supports swiotlb and
> > our next plan is for IOMMU mapping. But I would like parts of the DMA
> > IOMMU code under arch/arm to be made into a library (similar to the
> > lib/swiotlb.c) so that other architectures can easily make use of it.
> 
> If you're referring to dmabounce.c, that's a 100% no-way-ever-even-if-
> hell-freezes-over NACK.

Does dmabounce.c implement IOMMU support? I'm talking about functions
like __alloc_iova, __iommu_create_mapping etc.

The generic swiotlb code has enough bouncing capabilities for my needs.

-- 
Catalin

[Linaro-mm-sig] CMA on AArch64

[Hwang Chih-Chyuan]

> Yet another feature is DMA cache coherency management (primarily
> streaming DMA) and I have a patch but I want this driven via DT rather
> than drivers or SoC code calling set_dma_ops explicitly (the bindings
> definition is work in progress).
>

Hi Catalin,

Is there the plan or the progress about this part (streaming DMA)?  :) 		 	   		  

[Linaro-mm-sig] CMA on AArch64

[Catalin Marinas]

On Wed, Feb 26, 2014 at 09:50:35AM +0000, Hwang Chih-Chyuan wrote:
> > Yet another feature is DMA cache coherency management (primarily
> > streaming DMA) and I have a patch but I want this driven via DT rather
> > than drivers or SoC code calling set_dma_ops explicitly (the bindings
> > definition is work in progress).
> 
> Is there the plan or the progress about this part (streaming DMA)?  :)

You can grab a patch here:

https://git.kernel.org/cgit/linux/kernel/git/cmarinas/linux-aarch64.git/commit/?h=devel&id=96f083d416c0d01687ed5b37074831f461838455

However, I'd like to see some progress on defining the system topology
via DT rather than driver manually setting the dma ops. Dave Martin is
looking at this but I don't have an ETA yet.

-- 
Catalin