Hi Daniel, On Wednesday 04 Jan 2017 17:13:23 Laurent Pinchart wrote: > On Wednesday 04 Jan 2017 15:58:25 Daniel Vetter wrote: > > On Wed, Jan 04, 2017 at 04:33:57PM +0200, Laurent Pinchart wrote: > >> On Wednesday 04 Jan 2017 14:51:48 Daniel Vetter wrote: > >>> Hm, something like drm_bridge_panel_bridge_init(dev, panel) should be > >>> enough, or not? My idea is to use this for the case where the only > >>> thing in dt is the panel, with no real bridge chip. And I think we > >>> don't need anything beyond that one _init function, plus maybe some > >>> additional paramaters ... > >> > >> There should be no bridge then. If you want the DRM core to manage > >> panels automatically, then we should create specific helpers for that, > >> not abuse the bridge infrastructure. Bridges should be instantiated from > >> a hardware device and bound to drivers as usual. > > > > I guess that's the part where I disagree: Just because there's physically > > no bridge doesn't mean we shouldn't just treat it as one in the software > > abstraction. If it looks and acts like a bridge (even an empty one), then > > imo it can be a bridge. > > > > If you insist on panels being panels, then I guess we need some other kind > > of glue to bind them into arbitrary bridge chains. But given that the > > callbacks match very closely, I don't see the point. > > > > In an idea world a panel would probably derive from a drm_bridge, but > > we're not in that universe unfortunately ;-) > > Or both would derive from another object, but I agree that's how it should > work. That's what I want to achieve, one step at a time. Creating dummy > bridges isn't a step in that direction in my opinion, so I'd rather not do > that, but work towards the right abstraction. Do you object getting this patch merged as-is as a first step in the right direction ? :-) -- Regards, Laurent Pinchart
