Re: [PATCH v4 3/4] pinctrl: Qualcomm SPMI PMIC GPIO pin controller driver

[Bjorn Andersson <bjorn.andersson@sonymobile.com>]

On Wed 24 Sep 06:09 PDT 2014, Ivan T. Ivanov wrote:

> On Tue, 2014-09-23 at 21:18 -0700, Bjorn Andersson wrote:
> > On Mon 15 Sep 07:44 PDT 2014, Ivan T. Ivanov wrote:
> > 
> 
> <snip>
> 
> > > +static int pmic_gpio_of_xlate(struct gpio_chip *chip,
> > > +                             const struct of_phandle_args *gpio_desc,
> > > +                             u32 *flags)
> > > +{
> > > +       if (chip->of_gpio_n_cells < 2)
> > > +               return -EINVAL;
> > > +
> > > +       if (flags)
> > > +               *flags = gpio_desc->args[1];
> > > +
> > > +       return gpio_desc->args[0] - PMIC_GPIO_PHYSICAL_OFFSET;
> > > +}
> > 
> > If you change:
> >  gpiochip_add_pin_range(&state->chip, dev_name(dev), 0, 0, npins);
> > to:
> >  gpiochip_add_pin_range(&state->chip, dev_name(dev), 1, 0, npins);
> > 
> > And you treat the gpio functions as taking the gpio number instead of pinctrl
> > number (i.e. subtract 1 in those), then gpiolib will provide this function for
> > you.
> > 
> 
> I am unable to make this work. of_gpio_simple_xlate() didn't know that
> GPIO range is offset with 1. Requesting last GPIO return error. And
> debug output looks weird, for example:
> 
> # cat /sys/kernel/debug/gpio
> ...
> GPIOs 220-255, platform/0.c000.gpios, 0.c000.gpios:
> ...
> 
> # cat /sys/kernel/debug/pinctrl/0.c000.gpios/gpio-ranges 
> GPIO ranges handled:
> 1: 0.c000.gpios GPIOS [221 - 256] PINS [0 - 35]
> 
> Advice, please.
> 

Looking at of_gpio_simple_xlate() shows that you're right. It doesn't seem to
be possible to have a gpiochip that is not 0-based. Then I guess you have to
have your own "off-by-one-xlate".

@Linus, any comments on this? All documentation states that we have
gpio1-gpioXX in these chips, so we have to expose it as such or things will be
messy.

> > [..]
> > > +
> > > +static const struct of_device_id pmic_gpio_of_match[] = {
> > > +       { .compatible = "qcom,spmi-pmic-gpio" },
> > 
> > I think this should be more specific, because hopefully the spmi specification
> > will outlive the current pmic gpio block.
> > 
> > So I think you need to list the pmic blocks here (e.g. "qcom,pm8941-gpio").
> 
> I can rename this to lovely "qcom,qpnp-gpio" :-), in this way driver
> can outlive bus on which device is connected this time.
> 

If the mythical qpnp spec actually include a specification for how a gpio block
looks like then sure. Otherwise you would have to list all the pmics that
contain this block.

Regards,
Bjorn