On Thu, Sep 14, 2017 at 7:29 AM, Sakari Ailus
[off-list ref] wrote:
Hi everyone,
I recently came across a difference in behaviour of OF phandle parsing and
ACPI reference parsing, both of which can soon be accessed using
fwnode_property_get_reference_args.
The main change in this proposal touches OF, and specifically the change
is about using -ENODATA to tell that the phandle reference list entry that
was accessed does not exist. -ENOENT was used previously, but the same
error code was also used to tell that a phandle was empty, making it
impossible for the caller to figure out which of the two was the case.
I'm sending the set as RFC. In my limited testing I have found no ill
effects.
These patches are on top of linux-next.
Comments on the approach and the changes themselves would be most welcome.
It's not really valid to both have a variable count (and hence need to
retrieve it) and use blank phandle entries. The whole point of blank
entries is to have a fixed length and know what each index corresponds
too. Do you have an example where we hit this?
Rob