On Fri, Nov 1, 2019 at 3:30 PM Stefano Stabellini stefano.stabellini@xilinx.com wrote:

On Fri, 1 Nov 2019, Rob Herring wrote:

...

...

...

Action items:

• All: if you can think of simple solution to interrupt-map info duplication, let Stefano know & he will try to add to the example.
• Stefano: Remove ranges from amba_rpu
• Stefano: In next couple weeks, look at GPIO controller extensions & write in detail & send to list so we can discuss.

I am trying to look for the interrupt-related GPIO extensions, and the only one I could find is: socionext,interrupt-ranges from Documentation/devicetree/bindings/gpio/gpio-uniphier.txt:

• socionext,interrupt-ranges: Specifies an interrupt number mapping between this GPIO controller and its interrupt parent, in the form of arbitrary number of <child-interrupt-base parent-interrupt-base length> triplets.

No, there's numerous examples that *should* have used interrupt-map, but invented their own thing.

...

which is very similar to what we need here, but it looks like a socionext specific property while I seem to recall that Rob said that it was made "standard".

Rob, is this the property you had in mind? Or is there another one?

No. It's '*-map-pass-thru' and in the DT spec. Here's the commit adding it[1]. The motivation for this was GPIO, but the spec was written generically.

Rob

[1] https://github.com/devicetree-org/devicetree-specification/commit/32c96cd64d...

Thank you, that was very helpful! So if I read it right, it would turn it into this at the interrupt-multiplexer node:

    #interrupt-cells = <3>;

    /* copy all attributes from children to parents */
    interrupt-map-pass-thru = <0xffffffff 0xffffffff 0xffffffff>;

    /* mask all childres bits to always match the first 0x0 entries */
    interrupt-map-mask = <0x0 0x0 0x0>;

    /* child address cells, child interrupt cells, parent, parent interrupt cells */
    interrupt-map = <0x0 0x0 0x0 &gic_a72 0x0 0x0 0x0>,
                    <0x0 0x0 0x0 &gic_r5 0x0 0x0 0x0>;

The zero interrupt-map-mask would make sure that any children interrupts would match the two zero interrupt-map entries. The fully set interrupt-map-pass-thru would make sure that any children interrupts numbers and flags would be copied over to the parents.

I believe that's right. Of course, there's no kernel support for interrupt-map-pass-thru (there is for GPIO), and it in no way solves the problem of which entry to select. A simple way to solve this is to look for 'interrupt-parent' in the /cpus node when not found in the root node. The OS would just have to know which /cpus node to look in. That wouldn't work for existing OSs, but it's not major changes to dts files or OSs. For your case, interrupt-map is not even needed. For the complex case, you'd split interrupt-map for each GIC into each /cpus node.

Rob