Re: [RFC v7 1/6] dt-bindings: gpio: fix microchip,mpfs-gpio interrupt descriptions

[Conor Dooley <conor@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 2034 bytes --]

On Wed, Jul 24, 2024 at 03:25:38PM +0200, Krzysztof Kozlowski wrote:
> On 23/07/2024 13:27, Conor Dooley wrote:
> > The microchip,mpfs-gpio binding suffered greatly due to being written
> > with a narrow minded view of the controller, and the interrupt bits
> > ended up incorrect. It was mistakenly assumed that the interrupt
> > configuration was set by platform firmware, based on the FPGA
> > configuration, and that the GPIO DT nodes were the only way to really
> > communicate interrupt configuration to software.
> > 
> > Instead, the mux should be a device in its own right, and the GPIO
> > controllers should be connected to it, rather than to the PLIC.
> > Now that a binding exists for that mux, try to fix the misconceptions
> > in the GPIO controller binding.
> > 
> > Firstly, it's not possible for this controller to have fewer than 14
> > GPIOs, and thus 14 interrupts also. There are three controllers, with
> > 14, 24 & 32 GPIOs each.
> > 
> > The example is wacky too - it follows from the incorrect understanding
> > that the GPIO controllers are connected to the PLIC directly. They are
> > not however, with a mux sitting in between. Update the example to use
> > the mux as a parent, and the interrupt numbers at the mux for GPIO2 as
> > the example - rather than the strange looking, repeated <53>.
> > 
> 
> You make ngpios required, which could be an ABI break except that there
> is no Linux user for this, so there is no ABI break, right? If so, would
> be nice to mention it. Rest looks good:

No upstream user at least, and I don't believe that there are any
non-linux projects using the GPIO controllers via DT. I could, I
suppose, not make it required and use 32 as a default - but that could
cause problems with existing devicetrees where all 3 controllers omitted
the property, despite having differing numbers of GPIOs.

Now that I look again, the driver actually doesn't enforce the presence
of the property and I think it should fail to probe if not present.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]