On 12/05/14 12:34, Javier Martinez Canillas wrote:

Maybe we can remove this hackery by relying on the fact that a
compatible string can have a set of values that goes from more
specific to more general. So you can have something like:

compatible = "sony,panel-foobar", "omapdss,panel-foobar"

So right now only "omapdss,panel-foobar" will be matched and later
when we have common panel drivers then that driver could handle the
"sony,panel-foobar" compatible string.

Other platforms could do something similar and have

compatible = "sony,panel-foobar", "baz,panel-foobar"

That way you won't break DT backward compatible but still not require
hacks on arch/arm/mach-omap2/display.c.

We do the same for OMAP boards, we now have the following compatible string:

compatible = "isee,omap3-igep0020", "ti,omap36xx", "ti,omap3";

There isn't a struct machine_desc that matches "isee,omap3-igep0020"
but later if we find that we need some board specific initialization
we could add one without breaking existing DTS. In fact it used to be
a single machine_desc that matched "ti,omap3" for both omap36xx and
omap34xx but later when some DT clocks changes were introduced we
needed to split both SoC families.

I think that's a different thing. "isee,omap3-igep0020" is a proper
compatible string, if the board is "isee,...".  No matter what software
you run, that's correct and fine.

The omapdss case is different, there the "omapdss" points to a sw thing,
it does not describe the hardware. It's only needed as we don't have
proper sw drivers for the devices.

That said, I think what you describe would work. But I would rather keep
the .dts files clean and correct, and keep that hacks hidden inside the
kernel.

 Tomi