[RFC 2/6] omap: iovmm: generic iommu api migration

Laurent Pinchart laurent.pinchart at ideasonboard.com
Tue Jun 7 07:26:52 EDT 2011

Hi Ohad,

On Tuesday 07 June 2011 12:28:53 Ohad Ben-Cohen wrote:
> On Tue, Jun 7, 2011 at 12:05 PM, Laurent Pinchart wrote:
> > pgsz isn't used anymore, you can remove it.
> Ok.
> >> +             order = get_order(bytes);
> > 
> > Does iommu_map() handle offsets correctly, or does it expect pa to be
> > aligned to an order (or other) boundary ?
> Right now we have a BUG_ON if pa is unaligned, but that can be changed
> if needed (do we want it to handle offsets ?).

At least for the OMAP3 ISP we need to, as video buffers don't necessarily 
start on page boundaries.

> > As Russell pointed out, we should use sg->length instead of
> > sg_dma_length(sg). sg_dma_length(sg) is only valid after the scatter
> > list has been DMA-mapped, which doesn't happen in the iovmm driver. This
> > applies to all sg_dma_len(sg) calls.
> I'll make sure I don't introduce such calls, but it sounds like a
> separate patch should take care of the existing ones; pls tell me if
> you want me to send one.

A separate patch is indeed needed, yes. As you're already working on iommu it 
might be simpler if you add it to your tree. Otherwise I can send it.


Laurent Pinchart

More information about the linux-arm-kernel mailing list