Re: [PATCH v10 08/12] drm/bridge: lvds-codec: Implement basic bus format negotiation

[Boris Brezillon <boris.brezillon@collabora.com>]

On Tue, 25 Feb 2020 07:15:43 +0100
Sam Ravnborg <sam@ravnborg.org> wrote:

> Hi Boris/Laurent.
> 
> > > +
> > > +	err = of_property_read_u32(np, "bus-width", &input_bus_width);
> > > +	of_node_put(np);
> > > +
> > > +	if (err) {
> > > +		lvds_codec->input_fmt = MEDIA_BUS_FMT_FIXED;
> > > +	} else if (input_bus_width == 18) {
> > > +		lvds_codec->input_fmt = MEDIA_BUS_FMT_RGB666_1X18;
> > > +	} else if (input_bus_width == 24) {
> > > +		lvds_codec->input_fmt = MEDIA_BUS_FMT_RGB888_1X24;
> > > +	} else {
> > > +		dev_dbg(dev, "unsupported bus-width value %u on port 0\n",
> > > +			input_bus_width);
> > > +		return -ENOTSUPP;  
> > 
> > ENOTSUPP is "Operation not supported", I'd go for -EINVAL.
> >   
> > > +	}  
> > 
> > Doesn't this apply to LVDS encoders only ? For LVDS decoders I don't
> > think we want to report an RGB format on the input.  
> 
> In panel-lvds we use the property "data-mapping" for the same purpose.
> To specify the MEDIA_BUS format.

I started with data-mapping, and was told (by Laurent IIRC) that
bus-width would be more appropriate for a DPI (AKA RGB) bus. I think it
has to do with the fact that fully-parallel buses always have one color
bit per-signal, while serial or partially-parallel buses can have
several color-bits per-signal, the assignment being described by this
'data-mapping' property. This being said, I can see a case where
data-mapping would be needed for DPI buses => RGB component ordering. A
24bit bus does not distinguish between RGB888 and BGR888.

> 
> It would be good to standardize on the same property, and maybe have the
> same binding descriptions for all.

As for the standardization, I'm all for it, but let's do that in a
second step, please.