On Fri, Jun 8, 2018 at 2:17 PM, Linus Walleij [off-list ref] wrote:

A second thought:

Second thought is always a good thing :-)

On Fri, Jun 8, 2018 at 1:59 PM, Linus Walleij [off-list ref] wrote:

quoted

quoted

+       select GPIOLIB_IRQCHIP

You are not using this so I guess remove that line.

(...)

quoted

quoted

+static int
+mediatek_gpio_to_irq(struct gpio_chip *chip, unsigned int pin)
+{
+       struct mtk_data *gpio_data = gpiochip_get_data(chip);
+       struct mtk_gc *rg = to_mediatek_gpio(chip);
+
+       return irq_create_mapping(gpio_data->gpio_irq_domain,
+                                 pin + (rg->bank * MTK_BANK_WIDTH));
+}

So this is the result of a custom IRQdomain because you
can't use the generic GPIO IRQ lib.  Oh well, we have to live
with it I guess.

I think maybe you can actually use the generic GPIO IRQCHIP.

It is OK to call gpiochip_set_chained_irqchip() several times.
If you just mark the line with IRQF_SHARED then the handler
will just loop over all three banks until you find a hit, provided
you code it up properly.

Ok, I'll give it a try to use this also.

There were some problems with removing an irqchip like
that but your driver is anyway a bool so I think it might work
just fine.

Yours,
Linus Walleij

Thanks again.

Best regards,
    Sergio Paracuellos