Re: [PATCH v3 2/3] pinctrl: tps6594: add for TPS6594 PMIC

[Esteban Blanc]

On Fri Apr 21, 2023 at 10:34 AM CEST, Linus Walleij wrote:
> Hi Esteban,
>
> thanks for your patch!
>
> On Fri, Apr 14, 2023 at 12:12 PM Esteban Blanc <eblanc@xxxxxxxxxxxx> wrote:
>
> > TI TPS6594 PMIC has 11 GPIOs which can be used for different
> > functions.
> >
> > This add a pinctrl and pinmux drivers in order to use those functions.
> >
> > Signed-off-by: Esteban Blanc <eblanc@xxxxxxxxxxxx>
> (...)
> > +config PINCTRL_TPS6594
> > +       tristate "Pinctrl and GPIO driver for TI TPS6594 PMIC"
> > +       depends on MFD_TPS6594
> > +       default MFD_TPS6594
> > +       select PINMUX
> > +       select GPIOLIB
>
> select GPIO_REGMAP
> ?
>
> I think this driver can use the GPIO_REGMAP helper library.
>
> Please look into other drivers using this, such as
> drivers/gpio/gpio-sl28cpld.c

I had a look at this driver and gpio-regmap.c. I think I understood
what's going on, but I'm not sure how to handle the
gpio_regmap_set_direction case. It is using the same reg_mask_xlate to
determine the register and value to write as gpio_regmap_set or
gpio_regmap_read. The problem is that this PMIC has 1 register per GPIO
for the configuration (GPIOX_CONF registers with a bit for direction),
while the in and out register are used for 8 pins (GPIO_OUT_1,
GPIO_OUT_2 and GPIO_IN_1, GPIO_IN_2). This means that the register and
mask returned by reg_mask_xlate will be erroneous in one or the other
case.

I noticed that I could override reg_mask_xlate, so I should be able to
"just" match on the base address given as argument to perform a
different computation depending on whether we are using reg_mask_xlate in
a "direction change" or not, but somehow this feels a bit wrong.

Is this the correct solution?
Am I missing something?

Thanks again for your time. Best regards,

-- 
Esteban Blanc
BayLibre