New LCD connector type?

[Daniel Vetter]

On Mon, May 09, 2016 at 11:07:53PM +0300, Laurent Pinchart wrote:
> Hi Ville,
> 
> On Monday 09 May 2016 15:07:21 Ville Syrjälä wrote:
> > On Mon, May 09, 2016 at 02:43:02PM +0300, Laurent Pinchart wrote:
> > > On Saturday 07 May 2016 21:11:23 Ville Syrjälä wrote:
> > >> On Fri, May 06, 2016 at 10:01:29PM +0300, Sergei Shtylyov wrote:
> > >>> (Resending as I couldn't find this message from January in the list
> > >>> archives.)
> > >>> 
> > >>> Hello.
> > >>> 
> > >>> We have a board which has Newhaven Display LCD [1] connected via a
> > >>> what seems to be a custom 40-signal LCD connector. That connector
> > >>> includes (6-bit per component) RGB signals, DISP, H/VSYNC, and the
> > >>> pixel clock signals; in addition, there are I2C and IRQ signals.
> > >>> 
> > >>> Now we need to add support for this kind of connector to the R-Car DU
> > >>> driver in order to get the LCD panel resolution from the device tree
> > >>> (see drivers/gpu/drm/rcar-du/rcar_du_*con.c for the examples) and so
> > >>> we'd probably need a new DRM_MODE_CONNECTOR_* constant in
> > >>> include/uapi/drm/drm_mode.h in order to do that... Laurent advised me
> > >>> to ask for the opinions on the dri-devel mailing list first, which I'm
> > >>> doing. :-)
> > >> 
> > >> I'd just make it LVDS.
> > > 
> > > Even if the connector doesn't carry LVDS signals ?
> > 
> > No one should really have to care about such details for any purely
> > internal connector. IMO the fact that the thing we have is called
> > LVDS could be considered a historical accident.
> > 
> > We do have eDP and DSI types too though, which I might not have even
> > bothered to add myself. These days eDP does make some sense as we expose
> > the AUX ch via /dev so eDP vs. other internal connectors does have some
> > user visible difference at least. And I guess you could justify DSI by
> > the fact that it's a widely used standard.
> >
> > If people feel anal about this, I guess you could always add some kind
> > of generic internal connector type to account for everything that isn't
> > LVDS/eDP/DSI. But a new type means that you'll have to modify userspace
> > to deal with the new connector type, at least if you want to name the
> > connector based on its type (which is the norm).
> 
> I don't have a too strong opinion on the topic, but given that we've added a 
> few connector types semi-randomly (or at least without giving it too much 
> thoughts, as you described as "historical accident") I think we should at 
> least decide what we want connector types to represent.
> 
> For instance, on the topic of DSI, what could have been more interesting than 
> the physical protocol is whether the panel supports command mode and partial 
> updates, but the DSI connector type doesn't tell anything about that except 
> hinting that command mode could be supported.

I think for command mode we'd want some prop on either the connector or
the crtc. There's lots more stuff that does manual upload or similar
things, where vrefresh becomes less well defined ... In any case userspace
that renders to the frontbuffer _must_ call the DIRTYFB ioctl to flush
stuff out, to make sure it all works on command mode DSI, PSR eDP and any
of the other cases.

> If the connector type is solely used to describe the shape of the user-
> accessible connector then I agree than an "internal connector" type could have 
> covered LVDS, DPI and DSI (as well as DBI and possibly eDP). One of the 
> question is whether it could be useful to convey more information than that to 
> userspace.

That was the original intention at least - it was meant to help users find
the right connector on their machine. There's also encoder types for
completeness, but those are already genuinely useless.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch