Hi Ohad,

On Friday 26 October 2012 07:50:56 Ohad Ben-Cohen wrote:

On Thu, Oct 25, 2012 at 11:39 PM, Tony Lindgren [off-list ref] wrote:

quoted

quoted

quoted

Joerg and Ohad, do you have any opinions on this?

I agree that there's some merit in having a separate header file for
IOVMM, since it's a different layer from the IOMMU API.

But in reality it's tightly coupled with OMAP's IOMMU, and ideally it
really should go away and be replaced with the DMA API.

For this reason, and for the fact that anyway there's only a single
user for it (omap3isp) and there will never be any more, maybe we
shouldn't pollute include/linux anymore.

Anyone volunteering to remove IOVMM and adapt omap3isp to the DMA API
instead ? ;)

That's on my to-do list, as well as porting the OMAP3 ISP driver to videobuf2, 
adding DT support, moving to the common clock framework (when that will be 
available for the OMAP3), supporting missing V4L2 features, ... All this in my 
spare time of course, otherwise it wouldn't be fun, would it ? ;-)

I would also like to move the tidspbridge to the DMA API, but I think we'll 
need to move step by step there, and using the OMAP IOMMU and IOVMM APIs as an 
intermediate step would allow splitting patches in reviewable chunks. I know 
it's a step backwards in term of OMAP IOMMU usage, but that's in my opinion a 
temporary nuisance to make the leap easier.

-- 
Regards,

Laurent Pinchart