On Tue, Aug 04, 2015 at 09:23:27AM -0600, Stephen Warren wrote:
On 08/04/2015 08:26 AM, Thierry Reding wrote: ... [ discussion of new fdtdec_get_addr_size() implementation]
So what this does is really fix parsing of address and size cells in the general case, though it would still fail for values of #address-cells or #size-cells bigger than 2 (because we don't have a datatype that would be able to contain such large values).
Note that there's also still a corner case that this doesn't handle. The DT specification states, if I remember correctly, that #address-cells and #size-cells are inherited. That means with the current code we will wrongly parse something like this:
/ { ... #address-cells = <1>; #size-cells = <1>; ... bus@XXXXXXXX { ... device@XXXXXXXX { ... reg = <0xXXXXXXXX 0x1000>; ... }; ... }; ... };
According to the DT specification the bus@XXXXXXXX node would inherit #address-cells = <1> and #size-cells = <1> from the root node. However with libfdt what really happens is that since bus@XXXXXXXX does not have either property it will default to 2 in both cases. I'm not sure if this really is a problem. Typically nodes are not nested that deeply, or if they are then, typically, they explicitly contain #address-cells and #size-cells properties.
I don't think #address-cells/#size-cells do actually get inherited. Admittedly some other properties (e.g. interrupt-parent) do, but according to:
https://lists.ozlabs.org/pipermail/linuxppc-dev/2008-January/049113.html [PATCH] powerpc: #address-cells & #size-cells properties not inherited
... and my vague memory, these two don't.
You can search Google for e.g. "#address-cells inherited" and find a number of similar assertions.
Okay, that's good. It means there's not even a corner case. =)
Thierry