Re: [PATCH v4 1/3] dt-bindings: media: qcom,qcm2290-venus: document shikra Iris compatible

[Dmitry Baryshkov]

On Fri, Jul 31, 2026 at 06:14:15PM +0530, Vikash Garodia wrote:
> 
> On 7/31/2026 5:27 PM, Dmitry Baryshkov wrote:
> > On Tue, Jul 21, 2026 at 11:02:25PM +0530, Vikash Garodia wrote:
> > > Document the iris video accelerator used on shikra platforms by adding
> > > the qcom,shikra-iris compatible.
> > > 
> > > Although QCM2290 and shikra share the same video hardware and overall
> > > integration, their SMMU programming differs. QCM2290 exposes separate
> > > stream IDs for the video hardware and the Xtensa path, requiring two
> > > explicit IOMMU entries, whereas shikra uses a masked SMR to collapse
> > > equivalent stream IDs into a single mapping. Due to QCM2290’s SID layout
> > > and Xtensa isolation requirements, such SMR masking is not applicable on
> > > QCM2290 platforms.
> > 
> > This means that there are still two different streams, one for pixel and
> > one for non-pixel data. Stop using the mask and describe them
> > separately. One will go to the pixel node, another will go to the
> > non-pixel node. As you wrote, the hardware is the same as QCM2290, which
> > has two separate streams.
> 
> AFAIU, SMR mask is an optimization to reduce the stream id configuration to
> SMMU hardware registers, separating out is _not_ recommended which again
> brings in additional register for programming the additional stream.

We will need to separate it later anyway, to handle two different IOVA
regions. So why not perform it right from the beginning?

We should not be introducing a mechanism (memory-region) which is
limited only to this platform, if all other plaforms will use something
different (subnodes).


-- 
With best wishes
Dmitry