On Thu, 23 Aug 2018 12:29:21 +0200
Alban [off-list ref] wrote:

On Tue, 21 Aug 2018 15:01:21 +0200
Boris Brezillon [off-list ref] wrote:

quoted

On Tue, 21 Aug 2018 13:00:04 +0100
Srinivas Kandagatla [off-list ref] wrote:
  

quoted

On 21/08/18 12:39, Alban wrote:    

quoted

However we still have the a potential address space clash between
the nvmem cells and the main device binding.      

Can you elaborate?
    

Yes, I'd be interested in having a real example too, cause I don't see
what this address space clash is.  

Let say I have a device that use the following binding:

device {
    compatible = "example-device";
    #address-cells = <2>;
    #size-cells = <1>;

    child at 0,0 {
        reg = <0x0 0x0>;
        ...
    };

    child at 1,2 {
        reg = <0x1 0x2>;
        ...
    };

};

Now this binding already use the node address space for something,
so putting a nvmem node as direct child would not work. Here it is
quiet clear as we have 2 address cells, however even if the number of
cells and the cells size would match it would still be conceptually
wrong as both bindings then use the same address space for something
different.

When would we end up in this situation? MTD/flash nodes are supposed to
be leaf nodes (if you omit the partitions underneath). What would be
the case for defining subnodes there?

Do you know of any devices that uses such a representation?