Re: [PATCH v2 3/3] pinctrl: da9062: add driver support

[Linus Walleij <linus.walleij@linaro.org>]

Hi Marco,

thanks for your patch!

On Wed, Nov 27, 2019 at 12:56 PM Marco Felsch <m.felsch@pengutronix.de> wrote:

> The DA9062 is a mfd pmic device which supports 5 GPIOs. The GPIOs can
> be used as input, output or have a special use-case.
>
> The patch adds the support for the normal input/output use-case.
>
> Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
(...)

> +config PINCTRL_DA9062
> +       tristate "Dialog Semiconductor DA9062 PMIC pinctrl and GPIO Support"
> +       depends on MFD_DA9062
> +       select GPIOLIB

Hm this would be one of those that could use GENERIC_REGMAP_GPIO
the day we invent it but we haven't invented it yet.

> +#include <../gpio/gpiolib.h>

Put a comment above this telling us why you do this thing.

> +static int da9062_gpio_get(struct gpio_chip *gc, unsigned int offset)
> +{
(...)
> +       return val & BIT(offset);

You should #include <linux/bits.h> since you use BIT()

Usually I clamp it like this:
return !!(val & BIT(offset));

> +static int da9062_gpio_get_direction(struct gpio_chip *gc, unsigned int offset)
> +{
> +       struct da9062_pctl *pctl = gpiochip_get_data(gc);
> +       struct regmap *regmap = pctl->da9062->regmap;
> +       int gpio_mode;
> +
> +       gpio_mode = da9062_pctl_get_pin_mode(regmap, offset);
> +       if (gpio_mode < 0)
> +               return gpio_mode;
> +
> +       switch (gpio_mode) {
> +       case DA9062_PIN_ALTERNATE:
> +               return -ENOTSUPP;
> +       case DA9062_PIN_GPI:
> +               return 1;
> +       case DA9062_PIN_GPO_OD:
> +       case DA9062_PIN_GPO_PP:
> +               return 0;

We recently added defines for these directions in
<linux/gpio/driver.h>:

#define GPIO_LINE_DIRECTION_IN  1
#define GPIO_LINE_DIRECTION_OUT 0

Please use these. (Soon in Torvald's tree, else
in my "devel" branch.)

> +static int da9062_gpio_direction_input(struct gpio_chip *gc,
> +                                      unsigned int offset)
> +{
> +       struct da9062_pctl *pctl = gpiochip_get_data(gc);
> +       struct regmap *regmap = pctl->da9062->regmap;
> +       struct gpio_desc *desc = gpiochip_get_desc(gc, offset);
> +       unsigned int gpi_type;
> +       int ret;
> +
> +       ret = da9062_pctl_set_pin_mode(regmap, offset, DA9062_PIN_GPI);
> +       if (ret)
> +               return ret;
> +
> +       /*
> +        * If the gpio is active low we should set it in hw too. No worries
> +        * about gpio_get() because we read and return the gpio-level. So the
> +        * gpiolib active_low handling is still correct.
> +        *
> +        * 0 - active low, 1 - active high
> +        */
> +       gpi_type = !gpiod_is_active_low(desc);

That's interesting. Correct too, I guess.

> +static int da9062_gpio_direction_output(struct gpio_chip *gc,
> +                                       unsigned int offset, int value)
> +{
> +       /* Push-Pull / Open-Drain options are configured during set_config */
> +       da9062_gpio_set(gc, offset, value);

That looks dangerous. There is no guarantee that .set_config()
always gets called.

Please create a local state container for the mode of each pin in
struct da9062_pctl and set it to push-pull by default at probe, then
set this to whatever the state is here and let the .set_config()
alter it later if need be.

If we don't do that you will get boot-time defaults I think and that
might create bugs.

> +static int da9062_gpio_set_config(struct gpio_chip *gc, unsigned int offset,
> +                                 unsigned long config)
> +{
(...)
> +       case PIN_CONFIG_DRIVE_OPEN_DRAIN:
> +               return da9062_pctl_set_pin_mode(regmap, offset,
> +                                               DA9062_PIN_GPO_OD);
> +       case PIN_CONFIG_DRIVE_PUSH_PULL:
> +               return da9062_pctl_set_pin_mode(regmap, offset,
> +                                               DA9062_PIN_GPO_PP);

So also store this in the per-pin state.

Yours,
Linus Walleij