Re: [RFC PATCH 1/8] drm: writeback: Refactor drm_writeback_connector structure

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

On Mon, Aug 11, 2025 at 01:22:30PM +0300, Dmitry Baryshkov wrote:
> On Mon, Aug 11, 2025 at 12:44:29PM +0300, Laurent Pinchart wrote:
> > On Mon, Aug 11, 2025 at 02:57:00PM +0530, Suraj Kandpal wrote:
> > > Some drivers cannot work with the current design where the connector
> > > is embedded within the drm_writeback_connector such as intel and
> > > some drivers that can get it working end up adding a lot of checks
> > > all around the code to check if it's a writeback conenctor or not.
> > > To solve this we move the drm_writeback_connector within the
> > > drm_connector and remove the drm_connector base which was in
> > > drm_writeback_connector. We do all other required
> > > modifications that come with these changes along with addition
> > > of new function which returns the drm_connector when
> > > drm_writeback_connector is present.
> > > All drivers will be expected to allocate the drm_connector.
> > > 
> > > Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
> > > ---
> > >  drivers/gpu/drm/drm_writeback.c | 33 ++++++++++------
> > >  include/drm/drm_connector.h     | 60 +++++++++++++++++++++++++++++
> > >  include/drm/drm_writeback.h     | 68 ++++-----------------------------
> > >  3 files changed, 89 insertions(+), 72 deletions(-)
> > > 
> > > @@ -2305,6 +2360,11 @@ struct drm_connector {
> > >  	 * @cec: CEC-related data.
> > >  	 */
> > >  	struct drm_connector_cec cec;
> > > +
> > > +	/**
> > > +	 * @writeback: Writeback related valriables.
> > > +	 */
> > > +	struct drm_writeback_connector writeback;
> > 
> > No, sorry, that's a bad idea. Most connectors have nothing to do with
> > writeback, you shouldn't introduce writeback-specific fields here.
> > drm_writeback_connector happens to be a drm_connector because of
> > historical reasons (it was decided to reuse the connector API exposed to
> > userspace instead of exposing a completely separate API in order to
> > simplify the implementation), but that does not mean that every
> > connector is related to writeback.
> > 
> > I don't know what issues the Intel driver(s) have with
> > drm_writeback_connector, but you shouldn't make things worse for
> > everybody due to a driver problem.
> 
> Suraj is trying to solve a problem that in Intel code every drm_connector
> must be an intel_connector too. His previous attempt resulted in a loose
> abstraction where drm_writeback_connector.base wasn't initialized in
> some cases (which is a bad idea IMO).
> 
> I know the historical reasons for drm_writeback_connector, but I think
> we can do better now.
> 
> So, I think, a proper approach would be:
> 
> struct drm_connector {
>     // other fields
> 
>     union {
>         struct drm_connector_hdmi hdmi; // we already have it
>         struct drm_connector_wb wb;  // this is new
>     };
> 
>     // rest of the fields.
> };

I still don't like that. This really doesn't belong here. If anything,
the drm_connector for writeback belongs to drm_crtc.

If the issue is that some drivers need a custom drm_connector subclass,
then I'd rather turn the connector field of drm_writeback_connector into
a pointer.

> I plan to add drm_connector_dp in a similar way, covering DP needs
> (currently WIP).

-- 
Regards,

Laurent Pinchart