Re: [PATCH v1 2/4] gpio: add parameter to allow the use named gpios

[Alexandre Courbot <gnurou@gmail.com>]

On Wed, Jan 7, 2015 at 6:08 PM, Olliver Schinagl
<oliver+list@schinagl.nl> wrote:
> From: Olliver Schinagl <oliver@schinagl.nl>
>
> The gpio binding document says that new code should always use named
> gpios. Patch 40b73183 added support to parse a list of gpios from child
> nodes, but does not make it possible to use named gpios. This patch adds
> the con_id property and implements it is done in gpiolib.c, where the
> old-style of using unnamed gpios still works.

This is absolutely correct - thanks for spotting this.

>
> Signed-off-by: Olliver Schinagl <oliver@schinagl.nl>
> ---
>  drivers/gpio/devres.c                     | 16 +++++++++++++++-
>  drivers/input/keyboard/gpio_keys_polled.c |  2 +-
>  drivers/leds/leds-gpio.c                  |  2 +-
>  include/linux/gpio/consumer.h             |  1 +
>  4 files changed, 18 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpio/devres.c b/drivers/gpio/devres.c
> index 13dbd3d..b7fbe1c 100644
> --- a/drivers/gpio/devres.c
> +++ b/drivers/gpio/devres.c
> @@ -111,23 +111,37 @@ EXPORT_SYMBOL(__devm_gpiod_get_index);
>  /**
>   * devm_get_gpiod_from_child - get a GPIO descriptor from a device's child node
>   * @dev:       GPIO consumer
> + * @con_id:    function within the GPIO consumer
>   * @child:     firmware node (child of @dev)
>   *
>   * GPIO descriptors returned from this function are automatically disposed on
>   * driver detach.
>   */
>  struct gpio_desc *devm_get_gpiod_from_child(struct device *dev,
> +                                           const char *con_id,
>                                             struct fwnode_handle *child)
>  {
> +       static const char const *suffixes[] = { "gpios", "gpio" };
> +       char prop_name[32]; /* 32 is max size of property name */
>         struct gpio_desc **dr;
>         struct gpio_desc *desc;
> +       unsigned int i;
>
>         dr = devres_alloc(devm_gpiod_release, sizeof(struct gpio_desc *),
>                           GFP_KERNEL);
>         if (!dr)
>                 return ERR_PTR(-ENOMEM);
>
> -       desc = fwnode_get_named_gpiod(child, "gpios");
> +       for (i = 0; i < ARRAY_SIZE(suffixes); i++) {
> +               if (con_id)
> +                       snprintf(prop_name, 32, "%s-%s", con_id, suffixes[i]);
> +               else
> +                       snprintf(prop_name, 32, "%s", suffixes[i]);

Same remark as Dmitry about the hardcoded length of the string, but ...

> +
> +               desc = fwnode_get_named_gpiod(child, prop_name);
> +               if (!IS_ERR(desc) || (PTR_ERR(desc) == -EPROBE_DEFER))
> +                       break;
> +       }

... since it looks like this part has been mostly copy/pasted from
of_find_gpio(), can you add another patch that fixes it there as well?

Also in the case of ACPI this will prove to be an incomplete lookup
since acpi_find_gpio() has an additional fallback if the named lookup
fails.

In that respect, I wonder if it would not be better for
devm_get_gpiod_from_child() to call of_find_gpio() and
acpi_find_gpio() (after making them non-static) followed by
gpiod_request() instead of calling fwnode_get_named_gpiod(). But in
that case it will have to do the OF/ACPI handling by itself.

I'm not really sure about which way is better. I'd appreciate if you
could give a thought to a possible refactoring that would improve the
situation ; otherwise feel free to ignore what I have written above
and to duplicate the property name building code.