[PATCH v2 6/7] Holly DTS

[Segher Boessenkool]

>>> easiest is to
>>> replace it with a configurable poison value.  When using asm output,
>>> or with a map file, it might be possible to generate out-of-band
>>> information which we can use to check that the right things are 
>>> filled
>>> in.
>>>
>>> It's not a simply a matter of making the kernel parser recognize the
>>> uninitialized info, because there's no way to encode that in the dtb
>>> itself.
>>
>> There can be, with a new DTB version -- perhaps simply
>> define proplen == -1 to mean "undefined" or similar.
>
> That doesn't quite do it, because I'd prefer ? to cover cases where
> the property's length is known, just not its value.  In particular
> that could be useful for device trees where *only* non-moving edits
> are required by the bootloader, letting us use the asm output mode as
> intended.

[First things first -- it is generally wrong to
require properties to have a certain length, newer
(versions of) bindings can add extra fields to the
end of a property.  It should be okay for this ??
stuff though, if a bootwrapper/firmware fills in
the property it better know its whole semantics,
including the new part.]

Okay, so use negative lengths to represent undefined
properties of a certain length, and use -inf to mean
"completely undefined yet".  Or some other encoding,
there are many ways that would work.

My point is that you need a new DTB version to properly
represent ??.


Segher