On Wed, Oct 15, 2025 at 05:14:02PM +0800, Liu Ying wrote:
> On 10/14/2025, Marek Vasut wrote:
> > On 10/14/25 5:11 PM, Frank Li wrote:
> >> On Tue, Oct 14, 2025 at 04:03:37PM +0200, Marek Vasut wrote:
> >>> On 10/13/25 6:56 PM, Frank Li wrote:
> >>>> On Sat, Oct 11, 2025 at 06:51:20PM +0200, Marek Vasut wrote:
> >>>>> Rework dc_subdev_get_id() to drop ARRAY_SIZE() use and use empty 
> >>>>> trailing
> >>>>> entry in each ID look up array instead. This allows passing of those 
> >>>>> arrays
> >>>>> around as OF match data, which will be useful when using this pipeline 
> >>>>> on
> >>>>> i.MX95, which has different address-to-ID mapping.
> >>>>>
> >>>>> Signed-off-by: Marek Vasut <[email protected]>
> >>>>
> >>>> This change is okay. but my questions is why need map register to id.
> >>>
> >>> This seems to be a recurring pattern in the driver, where some components
> >>> need to find other components to link with them. The mapping is fixed, and
> >>> since the DT does not encode link IDs, the resolution of the mapping has 
> >>> to
> >>> happen by mapping the component base addresses to the IDs first.
> >>
> >> In graphic link, port@<n>, n should be id? why not use it?
> > I suspect you could model the relationships between the DC blocks using OF
> > graph, yes. I also suspect that description would be very complex in
> > DT, considering the amount of blocks and links this device contains. I
> > suspect this is why there is no such DT description using OF graph.
>
> Yes.  The design decision was made to avoid using OF graph to describe
> links between DC blocks due to the complexity.

Any previous discuission? Using registers base address to determiate ID
is not solution.

Frank

>
> >
> > I think it might also be good to talk to Liu directly about the original 
> > design
> > decision and why this id mapping was done the way it was done,
> > they should know better than me.
>
>
> --
> Regards,
> Liu Ying

Reply via email to