On Sun, Sep 26, 2021 at 4:36 PM Sven Peter [off-list ref] wrote:

On Sun, Sep 26, 2021, at 15:10, Linus Walleij wrote:

quoted

If these four blocks in the SoC are different, they should have
different compatible strings, because they are not, by
definition, compatible.

I'd prefer to have a single compatible and get the npins from some
property and I don't think that's necessarily over-generalizing.
AFAICT Apple has been using the exact same MMIO interface for years
and I'd expect them to continue using it in the future. The only thing
that seems to change is the number of pins available and their assignment.
If we just have a single compatible we can likely support the M1X/2 or
however Apple calls the next SoCs with just a simple DTB change without
touching any driver code.

Admittedly the word "compatible" in DT context is a bit fuzzy around the
edges. Sometimes it is more like "parameterizable" which is what you
are saying here.

I don't have a very strong opinion on it, both approaches are possible.
Being able to define new SoCs are not always helpful for developers
as the data in DT can become opaque and hard to understand if it
is too general.

Yours,
Linus Walleij