On 6/21/24 8:36 AM, Nishanth Menon wrote:
On 11:19-20240619, Andrew Davis wrote:
[...]
This binding is a bit of a mess, the phandle is always a pointer to
a node with the cells length hard-coded to 3. This looks to have been done
to allow the driver to use the function "of_parse_phandle_with_args" which
needs a property name for to find the cell count. But that makes no sense
as the count is always 3, the driver cannot accept any other value. The
driver should have just looped of_get_property() 3 times but wanted to
use the helper. So a silly driver mistake has turned into a binding issue.
We should drop the "pinctrl-single,gpio-range" from the binding and
fix the driver.
My bad, I meant to say drop "#pinctrl-single,gpio-range-cells" and drop
the phandle from "pinctrl-single,gpio-range".
The #-cells is meant for a producer to define args for a consumer phandle.
This is so common in DT there is a helper function for it. The problem
is the author of the driver here wanted to use that helper, even though it
would be backwards in this case (pinmux is the consumer).
So we have a node inside the pinmux node that exists only to expose a fixed
number (3) of cells so that a different node inside the same pinmux node
could point to itself to make a helper function happy (that didn't need to
be used in the first place)..
Andrew
Linus W: pinctrl-single,gpio-range -> any thoughts here? I think it is a
valid (if a bit too flexible design looking at the existing users who
just use a single mux value mapping for all modes)
