Re: [PATCH v3 04/10] gpio: aggregator: refactor the code to add GPIO desc in the forwarder

[Thomas Richard <thomas.richard@bootlin.com>]

Hi Andy,

Thanks a lot for the review !!

On 4/17/25 19:16, Andy Shevchenko wrote:
> On Wed, Apr 16, 2025 at 04:08:12PM +0200, Thomas Richard wrote:
>> Create a dedicated function to add a GPIO desc in the forwarder. Instead of
>> passing an array of GPIO desc, now the GPIO desc are passed on by one to
>> the forwarder.
> 
> ...
> 
>> +static int gpiochip_fwd_add_gpio_desc(struct gpiochip_fwd *fwd,
>> +				      struct gpio_desc *desc,
>> +				      unsigned int offset)
>> +{
>> +	struct gpio_chip *parent = gpiod_to_chip(desc);
>> +	struct gpio_chip *chip = &fwd->chip;
>> +
>> +	if (offset > chip->ngpio)
> 
>> = ?
> 
>> +		return -EINVAL;
> 
>> +	if (fwd->descs[offset])
>> +		return -EEXIST;
> 
> Not sure we need this. I would rather think that something inside struct
> gpiochip_fwd should track this. OTOH, I understand that you want to have
> sparse lists perhaps. I;m wondering why GPIO valid mask can't be used for
> this purposes?

The valid_mask in the gpio_chip is allocated in
gpiochip_add_data_with_key() which is too late for us.

But as you suggested, something (a valid_mask like in gpio_chip) inside
struct gpiochip_fwd should work.

> 
>> +	/*
>> +	 * If any of the GPIO lines are sleeping, then the entire forwarder
>> +	 * will be sleeping.
>> +	 * If any of the chips support .set_config(), then the forwarder will
>> +	 * support setting configs.
>> +	 */
>> +	if (gpiod_cansleep(desc))
>> +		chip->can_sleep = true;
>> +
>> +	if (parent && parent->set_config)
>> +		chip->set_config = gpio_fwd_set_config;
>> +
>> +	fwd->descs[offset] = desc;
>> +
>> +	dev_dbg(chip->parent, "%u => gpio %d irq %d\n", offset,
>> +		desc_to_gpio(desc), gpiod_to_irq(desc));
>> +
>> +	return 0;
>> +}
> 
> The bottom line is that I'm fine with this change without additional checks,
> add them when function will be used not only in the original loop.

Ok so for this patch I do not add checks.
Then I implement a valid_mask (in struct gpiochip_fwd) in patch 08/10
"gpio: aggregator: handle runtime registration of gpio_desc in gpiochip_fwd"

Best Regards,

Thomas