[RFC V2 0/3] drm/bridge: panel and chaining

[Daniel Vetter]

On Thu, May 15, 2014 at 12:32:58PM +0200, Thierry Reding wrote:
> On Fri, May 09, 2014 at 09:59:34AM -0400, Rob Clark wrote:
> > On Fri, May 9, 2014 at 5:08 AM, Andrzej Hajda <a.hajda at samsung.com> wrote:
> [...]
> > > 4. And the last thing, it is more about the concept/design. drm_bridge,
> > > drm_hw_block suggests that those interfaces describes the whole device:
> > > bridge, panel, whatever.
> > 
> > hmm, I don't think this is the case.  I can easily see things like:
> > 
> >   struct foo_panel {
> >     struct drm_panel base;
> >     struct drm_bridge bridge;
> >     ...
> >   }
> > 
> > where a panel implementation implements both panel and bridge.  In
> > fact that is kinda what I was encouraging.
> 
> For lack of a better entry point into the discussion, let me pick this
> example. It seems like in general we all agree that the basic structure
> would have to be something like this:
> 
> 	CRTC -> encoder -> bridge [ -> bridge ... ] -> panel
> 
> Where panel could optionally be a bridge/panel hybrid as Rob pointed
> out. I'm not sure if there's panels like this, but I suspect the use
> case would be where a panel had built-in controls to modify the image in
> some fashion (brightness, saturation, ...). I think those things exist
> in DCS for example.
> 
> What's missing here, and if I understand correctly what Sean said about
> the connector type, what we need to solve is how to wire up the
> connector so that it reflects the correct type. So the above would have
> to become something like this:
> 
> 	CRTC -> encoder -> bridge [ -> bridge ... ] -> panel
> 		connector -----------------------------^

This is kinda always how I've thought it should play out: The last item in
the chain actually implements the drm connector, everyone else just
ignores it.

> One of the problems with that approach is that panel is more than just a
> video sink. It also controls the panel (and optionally backlight) power.
> The standard DRM connector helpers don't work well in that case, because
> drm_helper_connector_dpms() forwards changes to the CRTC and encoder,
> though that could possibly be solved by wrapping it in a small custom
> implementation that also controls the panel. Anyway, that's really just
> an implementation detail.

I don't see an issue with the dpms really. At worst the overall kms driver
(which provides the crtc and encoder implementation) needs to pass a
suitable dpms implementation for the drm_bridge to use in the connector.
Or we could just move this vfunc into the core kms funtion table since
everyone uses the same (well except i915, but that's a different story).

dpms changes would then still be driven through the crtc -> encoder ->
bridge ... chain.

> So once a complete chain from encoder to panel has been created, we need
> a way to find the appropriate connector type. Perhaps to achieve that it
> would be useful to instantiate the connector by the bridge that's
> connected to the panel. So essentially something like this:
> 
> 	struct drm_bridge {
> 		/* to allow chaining */
> 		struct drm_bridge *next;
> 		/* for bridges directly connected to a panel or monitor */
> 		struct drm_connector *connector;
> 		/* for bridges directly connected to a panel */
> 		struct drm_panel *panel;

Imo these two shouldn't be here, but instead should be embedded into the
overall structure the drm_bridge driver is using.
> 
> 		...
> 	};
> 
> To make this work in a generic way, I think we're missing one bridge
> instance. Currently the bridge in an encoder is completely optional, so
> if there is no bridge in the system this needs to be special cased. One
> way to unify this would be to make drm_encoder implement drm_bridge, or
> an alternative would be to make drm_panel implement a bridge. Both can
> possibly be dummies stubbed out with simple helpers.

Yeah, imo if the panel isn't just a dumb thing but needs to actively take
part in the modeset dance, it simply needs to implement itself as a
drm_bridge+panel combo and do the magic in the various hooks provided.

> I wonder if this would have any consequences on Dave's DP MST work,
> since presumably an MST hub could be considered a bridge. In that case I
> guess there would need to be support for multiple "next" bridges,
> perhaps a simple doubly-linked list instead of a next pointer. The first
> bridge would then be used to model the hub's input and child bridges
> would be used to model each
> of the outputs.

DP is standardize. No need to wrestle the complexity through a drm_bridge,
since code reuse anywhere else (non-DP) will be know to be zero. And for
all the guys implementing a DP output can simply use the helpers. So I
don't see an upside of this idea.

Imo we should restrict drm_bridge to all the crazy transcoder chips/blocks
commonly found on SoCs which all do non-standard stuff and where we
actually have an established or anticipated need for sharing.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch