Re: [PATCH 02/12] dt-bindings: interrupt-controller: stm32-exti: Add irq nexus child node

[Antonio Borneo <antonio.borneo@foss.st.com>]

On Thu, 2024-02-22 at 16:43 -0700, Rob Herring wrote:
> On Fri, Feb 16, 2024 at 10:47:47AM +0100, Antonio Borneo wrote:
> > The mapping of EXTI interrupts to its parent interrupt controller
> > is both SoC and instance dependent.
> > The current implementation requires adding a new table to the
> > driver's code and a new compatible for each new EXTI instance.
> > 
> > Add to the binding an interrupt nexus child node that will be
> > used on the new EXTI instances and can be optionally used on the
> > existing instances.
> > The property 'interrupt-map' in the nexus node maps each EXTI
> > interrupt to the parent interrupt.
> > Align #address-cells and #interrupt-cells between the EXTI node
> > and its nexus node.
> 
> Looks like an abuse of interrupt-map. You avoid adding yourself to the 
> abuser list by putting it in a child node. Clever. (See list in 
> drivers/of/irq.c if you don't know what I'm talking about)

Hi Rob,
thanks for the review.

Yes, I know already about the abuser list but, from the commit
message and the associated comment, I interpret it as an incorrect
use of the property interrupt-map with custom syntax thus relying 
on custom parsing code.
The child nexus node in this series allows using the default parser
in kernel.

From your reply, looks like my interpretation is incorrect and I
missed the real concern about the abuser list.
Could you please explain why this use of interrupt-map is incorrect
and/or which are the correct use cases?

> I assume the EXTI has 0..N interrupts. Just define 'interrupts' with N 
> entries with each entry mapping EXTI interrupt N to 'interrupts' entry 
> N.

Yes, EXTI has 0..N interrupts that can be mapped to multiple
parent interrupt controllers and the mapping table has holes.
While the DT in this series only use one interrupt parent, a second
parent will follow.
So 'interrupts-extended' property would be a better matching than
'interrupts' to handle the multiple parents.

But how to code the missing entries in an 'interrupts-extended' list?
As in the example in Documentation/devicetree/bindings/dma/apple,admac.yaml ?

The 'interrupt-map' contains the matching EXTI index, thus allowing
a 'sparse' map where holes are simply ignored.

Best Regards,
Antonio