Re: [PATCH] dt-bindings: display: panel: Make DSI panels port optional

[Dmitry Baryshkov]

On Wed, Nov 11, 2026 at 02:54:54PM +0100, Krzysztof Kozlowski wrote:
> On 11/11/2026 14:52, Maxime Ripard wrote:
> > On Wed, Nov 11, 2026 at 01:44:23PM +0100, Krzysztof Kozlowski wrote:
> >> On 11/11/2026 13:37, Dmitry Baryshkov wrote:
> >>> On Wed, Nov 11, 2026 at 01:31:39PM +0100, Krzysztof Kozlowski wrote:
> >>>> On 11/11/2026 13:30, Krzysztof Kozlowski wrote:
> >>>>> On 11/11/2026 12:13, Maxime Ripard wrote:
> >>>>>> On Wed, Nov 11, 2026 at 12:03:39PM +0100, Krzysztof Kozlowski wrote:
> >>>>>>> DSI panels can be placed under DSI bus directly and many Linux DRM
> >>>>>>> drivers will work fine with it: mipi_dsi_host_register() adds device for
> >>>>>>> such panel and DRM driver does not use OF graph to configure lanes.
> >>>>>>
> >>>>>> That's not true. We do use the endpoints to configure the lanes through
> >>>>>> the data-lanes property. See drm_of_get_data_lanes*
> >>>>>
> >>>>> Some drivers use endpoints, some not. As discussed on other thread,
> >>>>> there are more than which do not.
> >>>>
> >>>> "there are more than one which do not."
> >>>
> >>> Which means that those should be fixed. We have legacy cases where the
> >>> drivers assumed that, but we should not target legacy.
> >>
> >> These legacy drivers work fine for 10 or 15 years, like Exynos.
> >>
> >> You can improve them, for sure, but if no one does that then requiring a
> >> graph is a mistake there, from the panel point of view.
> > 
> > At the very least the commit log needs some work then, because the
> > situation you describe in it is simply wrong, and quite different to the
> > situation you then described.
> 
> 
> Commit msg said:
> "many Linux DRM drivers will work fine with it"
> 
> So many->some? Is this the only problem? What else is not described in
> commit msg?

The commit message points out that a minority of drivers use and require
the port, while majority of the drivers don't. I think up to now only
the Exynos and sunXi were mentioned. From my point of view, it would be
better to change corresponding DSI host bindings to allow port bindings
and then define the OF graph connecting DSI host to DSI panel (which
might be done gradually, etc.) even if the OF graph is not used nowadays
by those drivers instead of providing a false signal that it's okay to
have DSI links without OF graph connection. It's not future-proof and
not okay, especially for the new designs.

-- 
With best wishes
Dmitry