RE: [RFC PATCH 2/6] pinctrl: starfive: jh8100: add pinctrl driver for sys_east domain

[Yuklin Soo]

> -----Original Message-----
> From: Linus Walleij <linus.walleij@xxxxxxxxxx>
> Sent: Sunday, January 28, 2024 7:33 AM
> To: Yuklin Soo <yuklin.soo@xxxxxxxxxxxxxxxx>
> Cc: Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxx>; Hal Feng
> <hal.feng@xxxxxxxxxxxxxxxx>; Leyfoon Tan <leyfoon.tan@xxxxxxxxxxxxxxxx>;
> Jianlong Huang <jianlong.huang@xxxxxxxxxxxxxxxx>; Emil Renner Berthing
> <kernel@xxxxxxxx>; Rob Herring <robh@xxxxxxxxxx>; Krzysztof Kozlowski
> <krzysztof.kozlowski+dt@xxxxxxxxxx>; Conor Dooley <conor+dt@xxxxxxxxxx>;
> Drew Fustini <drew@xxxxxxxxxxxxxxx>; linux-gpio@xxxxxxxxxxxxxxx; linux-
> kernel@xxxxxxxxxxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-
> riscv@xxxxxxxxxxxxxxxxxxx; Paul Walmsley <paul.walmsley@xxxxxxxxxx>;
> Palmer Dabbelt <palmer@xxxxxxxxxxx>; Albert Ou
> <aou@xxxxxxxxxxxxxxxxx>
> Subject: Re: [RFC PATCH 2/6] pinctrl: starfive: jh8100: add pinctrl driver for
> sys_east domain
> 
> Hi Alex,
> 
> thanks for your patch!
> 
> On Thu, Dec 21, 2023 at 9:36 AM Alex Soo <yuklin.soo@xxxxxxxxxxxxxxxx>
> wrote:
> 
> > Add pinctrl driver for sys_east domain.
> 
> This commit message is wrong, it also contains the main driver for jh8100.
> Please add some proper subject and commit message.

Will change the commit log to "add main and sys_east driver" to indicate the commit of both main and sys-east driver.

> 
> > Signed-off-by: Alex Soo <yuklin.soo@xxxxxxxxxxxxxxxx>
> > Reviewed-by: Ley Foon Tan <leyfoon.tan@xxxxxxxxxxxxxxxx>
> (...)
> > +#define pin_to_hwirq(sfp) (((sfp)->wakeup_gpio) - ((sfp)->gc.base))
> 
> Please do not reference gc.base like this, it is a gpio internal detail.
> 
> Also, turn this into a static inline function, the macro is hard to read.

The pin_to_hwirq macro will be converted to to a static inline function to hide gpio
internal detail, and for easier code readability.

> 
> > +/* pad control bits */
> > +#define JH8100_PADCFG_POS      BIT(7)
> > +#define JH8100_PADCFG_SMT      BIT(6)
> > +#define JH8100_PADCFG_SLEW     BIT(5)
> > +#define JH8100_PADCFG_PD       BIT(4)
> > +#define JH8100_PADCFG_PU       BIT(3)
> > +#define JH8100_PADCFG_BIAS     (JH8100_PADCFG_PD |
> JH8100_PADCFG_PU)
> 
> JH8100_PADCFG_BIAS_MASK

Will change to "JH8100_PADCFG_BIAS_MASK" in next version.

> 
> > +#define JH8100_PADCFG_DS_MASK  GENMASK(2, 1)
> > +#define JH8100_PADCFG_DS_2MA   (0U << 1)
> > +#define JH8100_PADCFG_DS_4MA   BIT(1)
> > +#define JH8100_PADCFG_DS_8MA   (2U << 1)
> > +#define JH8100_PADCFG_DS_12MA  (3U << 1)
> 
> Please use (1U << 1) for 4MA, this looks weird otherwise.

Will change to "(1U << 1)" for 4MA in next version.

> 
> > +static const struct pinconf_ops jh8100_pinconf_ops = {
> > +       .pin_config_get         = jh8100_pinconf_get,
> > +       .pin_config_group_get   = jh8100_pinconf_group_get,
> > +       .pin_config_group_set   = jh8100_pinconf_group_set,
> > +       .pin_config_dbg_show    = jh8100_pinconf_dbg_show,
> > +       .is_generic             = true,
> > +};
> > +
> > +static int jh8100_gpio_request(struct gpio_chip *gc, unsigned int
> > +gpio) {
> > +       return pinctrl_gpio_request(gc, gpio); }
> > +
> > +static void jh8100_gpio_free(struct gpio_chip *gc, unsigned int gpio)
> > +{
> > +       pinctrl_gpio_free(gc, gpio);
> > +}
> 
> Skip one level of indirection, just add pinctrl_gpio_request/free directly into
> the vtable.

This will be fixed in next version.

> 
> > +static int jh8100_gpio_set_config(struct gpio_chip *gc,
> > +                                 unsigned int gpio, unsigned long
> > +config) {
> > +       struct jh8100_pinctrl *sfp = container_of(gc,
> > +                       struct jh8100_pinctrl, gc);
> > +       u32 arg = pinconf_to_config_argument(config);
> 
> Please don't reimplement .set_config, just call into the pinctrl backend using
> 
> .set_config = gpiochip_generic_config

Will replace "jh8100_gpio_set_config" by "gpiochip_generic_config" in next version.

> 
> > +static int jh8100_gpio_add_pin_ranges(struct gpio_chip *gc) {
> > +       struct jh8100_pinctrl *sfp = container_of(gc,
> > +                       struct jh8100_pinctrl, gc);
> > +
> > +       sfp->gpios.name = sfp->gc.label;
> > +       sfp->gpios.base = sfp->gc.base;
> > +       sfp->gpios.pin_base = 0;
> > +       sfp->gpios.npins = sfp->gc.ngpio;
> > +       sfp->gpios.gc = &sfp->gc;
> > +       pinctrl_add_gpio_range(sfp->pctl, &sfp->gpios);
> > +       return 0;
> > +}
> 
> Why are you not putting the ranges into the device tree where the GPIO core
> will add them for you?

Will remove the jh8100_gpio_add_pin_ranges function and use gpio-ranges in device tree to
provide information for GPIO core to add pin range for each pin controller.

> 
> > +       if (info->irq_reg) {
> > +               jh8100_irq_chip.name = sfp->gc.label;
> 
> That's not immutable. The struct should be const.
> You have to use .irq_print_chip in the irq_chip.

The struct irq_chip will be used as a constant data structure. 
Will add an irq_print_chip function to display irqchip name to user space.

> 
> > +               gpio_irq_chip_set_chip(&sfp->gc.irq,
> > + &jh8100_irq_chip);
> 
> Use the convention:
> 
> struct gpio_irq_chip *girq;
> 
> girq = &chip->irq;
> gpio_irq_chip_set_chip(girq, &nmk_irq_chip);
> 
> ... and use girq-> in the rest of the assignments.

girq will be used to represent GPIO interrupt controller.

> 
> > +               dev_info(dev, "StarFive GPIO chip registered %d
> > + GPIOs\n", sfp->gc.ngpio);
> 
> StarFive JH8100 (be precise)

"StarFive GPIO" will be changed to "StarFive JH8100 GPIO".

> 
> Yours,
> Linus Walleij