Re: [PATCH 1/4] pinctrl: Introduce generic #pinctrl-cells and pinctrl_parse_index_with_args

From: Linus Walleij
Date: Thu Oct 27 2016 - 10:42:32 EST


On Tue, Oct 25, 2016 at 6:45 PM, Tony Lindgren <tony@xxxxxxxxxxx> wrote:

I need some DT person to take a look at this binding and ACK it.

> +For pin controller hardware with a large number of identical registers naming
> +each bit both can be unmaintainable. Further there can be a large number of similar
> +pinctrl hardware using the same registers for different purposes depending on the
> +packaging. For cases like this, the pinctrl driver may use pinctrl-pin-array helper
> +binding using a hardware based index and a number of configuration values:

Maybe we can reword it a bit so that it is clear that this is an
either-or approach
for the pin controller, either they use the pins/groups/functions scheme
or they use this scheme.

> +pincontroller {
> + ... /* Standard DT properties for the device itself elided */
> + #pinctrl-cells = <2>;
> +
> + state_0_node_a {
> + pinctrl-pin-array = <
> + 0 A_DELAY_PS(0) G_DELAY_PS(120)
> + 4 A_DELAY_PS(0) G_DELAY_PS(360)
> + ...
> + >;
> + };
> + ...
> +};

Looks all right to me. Sad to add to the binding mess, but on the other
hand, in the overall picture this nicely consolidates and structure
pinctrl-single.

> +The index for pinctrl-pin-array must relate to the hardware for the pinctrl
> +registers, and must not be a virtual index of pin instances. The reason for
> +this is to avoid mapping of the index in the dts files and the pin controller
> +driver as it can change.

OK

> And we want to avoid another case of interrupt
> +numbering with pinctrl numbering.

Maybe this file is not a good place for making technical arguments,
more describing what we agreed on, so cut that sentence IMO.

> +/*
> + * For pinctrl binding, typically #pinctrl-cells is for the pin controller
> + * device, so either parent or grandparent. See pinctrl-bindings.txt.
> + */
> +static int pinctrl_find_cells_size(const struct device_node *np,
> + const char *cells_name)
> +{
> + int cells_size, error;
> +
> + error = of_property_read_u32(np->parent, cells_name, &cells_size);
> + if (error) {
> + error = of_property_read_u32(np->parent->parent,
> + cells_name, &cells_size);
> + if (error)
> + return -ENOENT;
> + }
> +
> + return cells_size;
> +}

Can't we just hardcode this to "#pinctrl-cells" and skip the cells_name
parameter? We can parametrize it the day we need it instead.

The rest of the helpers look nice and clean.

Yours,
Linus Walleij