On Tue, 16 Aug 2022 at 20:39, Rob Herring <robh@xxxxxxxxxx> wrote:
>
> The Broadcom pinctrl bindings are incomplete for child nodes as they are
> missing 'unevaluatedProperties: false' to prevent unknown properties.
> Fixing this reveals many warnings including having grandchild nodes in some
> cases.
>
> Signed-off-by: Rob Herring <robh@xxxxxxxxxx>
> ---
> This recursive schema requires a fix not yet committed in dtschema.
>
> I'm looking for feedback on whether group->pins or group->groups is the
> right fix here. There's more warnings with this change in the gpio-sysctl
> bindings.
The answer is "yes", though pins is probably the closest for most.
bcm6318 has multiple field-per-pin registers, where each pin is
controlled separately, with more fields than available GPIOs, and the
pins outside the GPIO range controlling other functions, like
switching the second USB port between host and client mode.
bcm6328/6362/6368/63268 have two registers. The first one enables an
alternative function for the first 32 GPIOs, with a 1:1 mapping of
bits to GPIO. The second one enables a function for whole, arbitrary
groups. These groups can overlap, and may also target pins outside the
first 32 GPIOs. The actual pins in use are not documented, and can
sometimes be guessed/inferred by the function names (e.g. "GPIO35"),
sometimes not ("NAND", "UTOPIA").
bcm6358 has only the groups register, which also includes non GPIO
related functions, like inversing the MII clocks for the integrated
macs.
Not supported is bcm6348, which would be the only one where groups
would definitely make more sense: there are 5 groups of 8 GPIOs, where
each group can be set to a certain function (but not all functions are
valid for all groups). E.g. for PCI support, you would need to set the
fields to PCI for groups 0, 1 and 3 (and 2 and 5 could be set to a
different function). You can ignore this though, as bcm6348 is ancient
(doesn't even support ADSL2+).
Hope that helps a bit.
Best regards,
Jonas
