RE: [PATCH v1 2/3] gpio: Add support for TPS68470 GPIOs

From: Mani, Rajmohan
Date: Sun Jun 11 2017 - 01:04:55 EST


Hi Linus,

Thanks for the reviews.

> -----Original Message-----
> From: Linus Walleij [mailto:linus.walleij@xxxxxxxxxx]
> Sent: Friday, June 09, 2017 4:15 AM
> To: Mani, Rajmohan <rajmohan.mani@xxxxxxxxx>
> Cc: linux-kernel@xxxxxxxxxxxxxxx; linux-gpio@xxxxxxxxxxxxxxx; ACPI Devel
> Maling List <linux-acpi@xxxxxxxxxxxxxxx>; Lee Jones <lee.jones@xxxxxxxxxx>;
> Alexandre Courbot <gnurou@xxxxxxxxx>; Rafael J. Wysocki
> <rjw@xxxxxxxxxxxxx>; Len Brown <lenb@xxxxxxxxxx>
> Subject: Re: [PATCH v1 2/3] gpio: Add support for TPS68470 GPIOs
>
> On Tue, Jun 6, 2017 at 1:55 PM, Rajmohan Mani <rajmohan.mani@xxxxxxxxx>
> wrote:
>
> > This patch adds support for TPS68470 GPIOs.
> > There are 7 GPIOs and a few sensor related GPIOs.
> > These GPIOs can be requested and configured as appropriate.
> >
> > Signed-off-by: Rajmohan Mani <rajmohan.mani@xxxxxxxxx>
>
> Same comments as Andy already sent, plus:
>
> > +static inline struct tps68470_gpio_data *to_gpio_data(struct
> > +gpio_chip *gpiochp) {
> > + return container_of(gpiochp, struct tps68470_gpio_data, gc); }
>
> Please use the data pointe inside gpio_chip.
>

This is not clear to me.
The driver already gets the data pointer inside gpio_chip (which is gpiochp)
Is the name gpiochp misleading?
I had to get rid of "i" in gpiochip, to meet 80 char limit.

> struct tps68470_gpio_data *foo = gpiochip_get_data(gc);
>
> > + ret = tps68470_reg_read(tps, reg, &val);
> (...)
> > + tps68470_update_bits(tps, reg, BIT(offset), value ?
> > + BIT(offset) : 0);
> (...)
> > + return tps68470_update_bits(tps, TPS68470_GPIO_CTL_REG_A(offset),
> > + TPS68470_GPIO_MODE_MASK,
> > + TPS68470_GPIO_MODE_OUT_CMOS);
> (...)
> > + return tps68470_update_bits(tps, TPS68470_GPIO_CTL_REG_A(offset),
> > + TPS68470_GPIO_MODE_MASK, 0x00);
>
> This looks like a reimplementation of regmap. Is it not possible to create a
> regmap in the MFD driver and pass that around instead?
>

Ack
Will be addressed in v2 of this patch series.

> > +struct gpiod_lookup_table gpios_table = {
> > + .dev_id = NULL,
> > + .table = {
> > + GPIO_LOOKUP("tps68470-gpio", 0, "gpio.0", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 1, "gpio.1", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 2, "gpio.2", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 3, "gpio.3", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 4, "gpio.4", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 5, "gpio.5", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 6, "gpio.6", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 7, "s_enable",
> GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 8, "s_idle", GPIO_ACTIVE_HIGH),
> > + GPIO_LOOKUP("tps68470-gpio", 9, "s_resetn",
> GPIO_ACTIVE_HIGH),
> > + {},
> > + },
> > +};
>
> Hm that's why you include <linux/gpio/machine.h> I guess.
>

Ack

> This warrants a big comment above it explaining why this is done like that and
> what the use is inside any system with this chip mounted.
>

Ack
I have added comments in the MFD driver, inside which the lookup table has moved.

> > + tps68470_gpio->gc.label = "tps68470-gpio";
> > + tps68470_gpio->gc.owner = THIS_MODULE;
> > + tps68470_gpio->gc.direction_input = tps68470_gpio_input;
> > + tps68470_gpio->gc.direction_output = tps68470_gpio_output;
>
> Please implement .get_direction()
>

Ack

> > + tps68470_gpio->gc.get = tps68470_gpio_get;
> > + tps68470_gpio->gc.set = tps68470_gpio_set;
> > + tps68470_gpio->gc.can_sleep = true;
> > + tps68470_gpio->gc.ngpio = TPS68470_N_GPIO;
> > + tps68470_gpio->gc.base = -1;
> > + tps68470_gpio->gc.parent = &pdev->dev;
>
> Consider assigning the .names() array some sensible line names.
>

Ack

> > +static int tps68470_gpio_remove(struct platform_device *pdev) {
> > + struct tps68470_gpio_data *tps68470_gpio =
> > +platform_get_drvdata(pdev);
> > +
> > + gpiod_remove_lookup_table(&gpios_table);
> > + gpiochip_remove(&tps68470_gpio->gc);
>
> You can't use devm_gpiochip_add()?
>

Originally I couldn't, because the driver can not remove the lookup table upon exit/removal. I moved this code inside the MFD driver, which is modified to use .shutdown() to remove the lookup table, so it can use dev_mfd_* call.