[RFC PATCH 3/5] iommu: implement common IOMMU ops for DMA mapping

[Robin Murphy]

Hi Joerg,

On 27/01/15 00:21, Joerg Roedel wrote:
> Hi Robin,
>
> thanks for the patch, I think it is good start to move forward. See my
> comments below.
>
> On Mon, Jan 12, 2015 at 08:48:55PM +0000, Robin Murphy wrote:
>> Taking inspiration from the existing arch/arm code, break out some
>> generic functions to interface the DMA-API to the IOMMU-API. This will
>> do the bulk of the heavy lifting for IOMMU-backed dma-mapping.
>>
>> Whilst the target is arm64, rather than introduce yet another private
>> implementation, place this in common code as the first step towards
>> consolidating the numerous versions spread around between architecture
>> code and IOMMU drivers.
>
> Is there any reason you implemented that only for AArch64? I think it
> makes a lot of sense to have a common implementation of that across
> architectures (AArch64, ARM and X86).

Laz^WPragmatism - I'm expecting quite a lot of changes to get this 
looking good, so keeping the series as lean as possible to aid 
reviewing/rebasing/etc. seemed sensible. In the same vein, since the 
other architectures already have code that works, my priority is getting 
something in place to fill the gap in arm64 (my current remit is "get 
the SMMUs on Juno working"); it seemed logical to minimise disruption 
and dependencies by aiming to get this merged with the one user, then 
start porting the others (and making the inevitable necessary tweaks) 
once it's in.

I'll adjust the commit message to make that clearer - on re-reading it, 
it does come across as rather vague about that intent.

>>
>> Signed-off-by: Robin Murphy <robin.murphy at arm.com>
>> ---
>>   include/linux/dma-iommu.h |  78 ++++++++
>>   lib/Kconfig               |   8 +
>>   lib/Makefile              |   1 +
>>   lib/dma-iommu.c           | 455 ++++++++++++++++++++++++++++++++++++++++++++++
>
> I'd like this to live in drivers/iommu, as most other dma-api
> implementations for iommu-drivers also live there.
>

That's reasonable - I was trying to model this on SWIOTLB, so it ended 
up in the same place. Mind you, I suppose there's a fair argument for 
moving SWIOTLB over to drivers/iommu too.

>> +/*
>> + * This little guy is filling in until iommu_map_sg lands and we can hook that
>> + * up instead (which is going to be rather involved thanks to page alignment)
>> + */
>
> iommu_map_sg should be there now. You can convert this function to it.
>

Indeed, that comment is pretty ancient and the 'proper' one was already 
half-done when I posted this; It'll be in v2.

Thanks,
Robin.

>> +static int __iommu_dma_map_sg_simple(struct device *dev, struct scatterlist *sg,
>> +		int nents, enum dma_data_direction dir, struct dma_attrs *attrs,
>> +		bool coherent)
>> +{
>> +	struct scatterlist *s;
>> +	int i;
>> +
>> +	for_each_sg(sg, s, nents, i) {
>> +		sg_dma_address(s) = __iommu_dma_map_page(dev, sg_page(s), s->offset,
>> +						s->length, dir, coherent);
>> +		sg_dma_len(s) = s->length;
>> +	}
>> +	return nents;
>> +}
>
> So long,
>
>          Joerg
>
>