Nested indirect buses

[mbolivar-nordic]

Is nesting indirect buses allowed? It seems to introduce a lot of
complication if so, and my guess is that it should just be explicitly
forbidden in the specification, but I'd like to know.

For example:

/ {
	parent {
		compatible = "indirect-bus";
		child {
			compatible = "indirect-bus";
			peripheral@deadbeef { /* ... */ };
		};
	};
};

If this is allowed, what are the semantics for mapping in
peripheral@deadbeef above in the address-map property of a
"cpus,cluster" node?

• If I map in parent, do I automatically get peripheral@deadbeef
  too even if I don't map child?

  • If so, via what address map translation through child?
  • If not, does that mean I can simply ignore any "indirect-bus"
    node, along with all of its children, that aren't mapped
    explicitly (assuming peripheral@deadbeef is not present in the
    address-map property)?

• If I map in child but not parent, how can I generate an
  equivalent devicetree? Do I need to lift child to be under the
  root node or something like that?

[sstabellini]

I don't think we can or should disallow this. "indirect-bus" is not particularly special, so the same way one can nest simple-buses one should also be able to nest indirect-buses. The mapping would work based on the address-map property as usual.

Yes if you map the parent, including the entire range of the parent that is a superset, then yes the child is also mapped. The address of peripheral@deadbeef would be translated 3 times:

• ranges of child
• ranges of parent
• address-bus

Yes you can map the child and not the parent too. You can generate an equivalent device tree by creating a parent node empty if not for the child node. Yes the device tree hierarchy should be retained. This would be the same without nesting:

amba {
peripheral@deadbeef {
};
};

in this case if address map is only mapping peripheral@deadbeef and not the whole of amba, should peripheral@deadbeef be moved directly under the root node? The answer is no. The hierarchy should be retained.