Re: [PATCH v4 07/10] gpio: max7360: Add MAX7360 gpio support

["Mathieu Dubois-Briand" <mathieu.dubois-briand@bootlin.com>]

On Fri Mar 14, 2025 at 9:14 AM CET, Andy Shevchenko wrote:
> Thu, Mar 13, 2025 at 06:07:03PM +0100, Mathieu Dubois-Briand kirjoitti:
> > On Fri Feb 14, 2025 at 4:59 PM CET, Andy Shevchenko wrote:
> > > On Fri, Feb 14, 2025 at 12:49:57PM +0100, Mathieu Dubois-Briand wrote:
> > > > Add driver for Maxim Integrated MAX7360 GPIO/GPO controller.
>
> ...
>
> > > > +	/*
> > > > +	 * MAX7360_REG_DEBOUNCE contains configuration both for keypad debounce
> > > > +	 * timings and gpos/keypad columns repartition. Only the later is
> > > > +	 * modified here.
> > > > +	 */
> > > > +	val = FIELD_PREP(MAX7360_PORTS, ngpios);
> > > > +	ret = regmap_write_bits(regmap, MAX7360_REG_DEBOUNCE, MAX7360_PORTS, val);
> > > > +	if (ret) {
> > > > +		dev_err(dev, "Failed to write max7360 columns/gpos configuration");
> > > > +		return ret;
> > > > +	}
> > >
> > > Shouldn't this be configured via ->set_config() callback?
> > 
> > I believe this comment has been a bit outdated by our discussion on
> > using GPIO valid mask, but I believe we could not use the ->set_config()
> > callback here: this callback is made to configure a single pin while the
> > gpos/keypad columns repartition is global.
>
> Yeah, we have similar desing in Intel Bay Trail (see pinctrl-baytrail.c) and it
> requires some software driven heuristics on how individual setting may affect
> the global one. But the Q here is is the debounce affects only keypad? Then it
> should be configured via keypad matrix driver. Btw, have you checked
> drivers/input/keyboard/matrix_keypad.c? Is there anything that can be useful
> here?
>

Hum, maybe the comment is not clear enough? Not sure, but please tell
me.

So yes, this register is named "debounce" but controls two different
things:
- The keypad debounce: we do not touch it here.
- The partition between keypad columns and gpos. This is the value we do
  modify here.

I've been looking at drivers/input/keyboard/matrix_keypad.c, but I
believe the idea is completely different: it allows to use GPIOs to
create a keypad matrix, without the help of any controller. Here we have
a controller dedicated to keypad matrix, allowing to repurpose unused
columns as GPIOs. So except for some device tree similar properties and
input events, I believe there there isn't much we can reuse.

> ...
>
> > > > +		if (irq < 0)
> > > > +			return dev_err_probe(dev, irq, "Failed to get IRQ\n");
> > > > +
> > > > +		irq_chip = devm_kzalloc(dev, sizeof(*irq_chip), GFP_KERNEL);
> > > > +		if (!irq_chip)
> > > > +			return -ENOMEM;
> > > > +
> > > > +		irq_chip->name = dev_name(dev);
> > > > +		irq_chip->status_base = MAX7360_REG_GPIOIN;
> > > > +		irq_chip->num_regs = 1;
> > > > +		irq_chip->num_irqs = MAX7360_MAX_GPIO;
> > > > +		irq_chip->irqs = max7360_regmap_irqs;
> > > > +		irq_chip->handle_mask_sync = max7360_handle_mask_sync;
> > > > +		irq_chip->status_is_level = true;
> > > > +		irq_chip->irq_drv_data = regmap;
> > > > +
> > > > +		for (unsigned int i = 0; i < MAX7360_MAX_GPIO; i++) {
> > > > +			regmap_write_bits(regmap, MAX7360_REG_PWMCFG(i),
> > > > +					  MAX7360_PORT_CFG_INTERRUPT_EDGES,
> > > > +					  MAX7360_PORT_CFG_INTERRUPT_EDGES);
> > > > +		}
> > > > +
> > > > +		flags = IRQF_TRIGGER_LOW | IRQF_ONESHOT | IRQF_SHARED;
> > > > +		ret = devm_regmap_add_irq_chip_fwnode(dev, dev_fwnode(dev), regmap, irq, flags, 0,
> > > > +						      irq_chip, &irq_chip_data);
> > >
> > > Right.
> > >
> > > What I mean in previous discussion is to update gpio-regmap to call this from inside.
> > > You need to add irq_chip pointer and irq_chip_data pointer to the regmap configuration
> > > and if they are set (or the first one, I dunno if this is supported by IRQ chip core)
> > > call this function and assign domain. This should be called after GPIO chip is
> > > added, but before IRQ domain attachment.
> > >
> > 
> > Ok, this is a bit more clear to me now. So I came up with something, it
> > will be part of the next iteration, probably during the next week.
> > 
> > This required to add a few additional fields to the gpio_regmap_config
> > structure, specifying the IRQ configuration:
> > 
> > + * @regmap_irq_chip:   (Optional) Pointer on an regmap_irq_chip structure. If
> > + *                     set, a regmap-irq device will be created and the IRQ
> > + *                     domain will be set accordingly.
> > + * @regmap_irq_chip_data: (Optional) Pointer on an regmap_irq_chip_data
> > + *                      structure pointer. If set, it will be populated with a
> > + *                      pointer on allocated regmap_irq data.
> > + * @regmap_irq_irqno   (Optional) The IRQ the device uses to signal interrupts.
> > + * @regmap_irq_flags   (Optional) The IRQF_ flags to use for the interrupt.
>
> Okay, just make sure it's guarded by the same ifdeffery as the similar in the
> GPIO:
>
> #ifdef CONFIG_GPIOLIB_IRQCHIP
>

Thanks!

> ...
>
> > > > +
> > > > +		regmap_write(regmap, MAX7360_REG_GPIOOUTM, outconf);
> > > > +	}
> > > > +
> > > > +	/* Add gpio device. */
> > > > +	gpio_config.parent = dev;
> > > > +	gpio_config.regmap = regmap;
> > >
> > > > +	if (gpio_function == MAX7360_GPIO_PORT) {
> > > > +		gpio_config.ngpio = MAX7360_MAX_GPIO;
> > >
> > > Why this case can't be managed also via ngpios property? Maybe at the end of
> > > the day you rather need to have another property to tell where the split is?
> > >
> > > This will help a lot and removes unneeded sharing of ngpios here and there.
> > >
> > > What I read from this code is like you are trying to put _two_in_one_ semantics
> > > on the shoulders of "ngpios".
> > 
> > So as I reworked the keypad columns GPIOs, PORT GPIOs and the COL GPIOs
> > are a bit more similar on this point. So far I now use a constant value
> > assigned in the driver for both, as I believe there is no way the number
> > of GPIOs could be a different. Yet I can easily switch back to a value
> > provided by a device property.
>
> Sounds good as long as ngpios is not overloaded with the additional meanings.

Thanks again for your review.

-- 
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com