On Wed, Oct 09, 2013 at 02:19:17PM +0200, Linus Walleij wrote:
> On Tue, Oct 8, 2013 at 2:25 PM, Christian Ruppert
> <christian.ruppert@xxxxxxxxxx> wrote:
>
> Overall this driver is looking very nice, we just need to figure out this
> group range concept in the other patch.
>
> > +Example:
> > +
> > + gpioa: gpio@FF140000 {
> > + compatible = "abilis,tb10x-gpio";
> > + interrupt-controller;
> > + #interrupt-cells = <1>;
> > + interrupt-parent = <&tb10x_ictl>;
> > + interrupts = <27 2>;
>
> So this is cascaded off some HWIRQ offset 27 on some interrupt
> controller, OK...
>
> > + reg = <0xFF140000 0x1000>;
> > + gpio-controller;
> > + #gpio-cells = <2>;
> > + abilis,ngpio = <3>;
> > + gpio-ranges = <&iomux 0 0 0>;
> > + gpio-ranges-group-names = "gpioa_pins";
> > + };
>
>
> But this thing:
>
> (...)
> > +static irqreturn_t tb10x_gpio_irq_cascade(int irq, void *data)
> > +{
> > + struct tb10x_gpio *tb10x_gpio = data;
> > + u32 r = tb10x_reg_read(tb10x_gpio, OFFSET_TO_REG_CHANGE);
> > + u32 m = tb10x_reg_read(tb10x_gpio, OFFSET_TO_REG_INT_EN);
> > + const unsigned long bits = r & m;
> > + int i;
> > +
> > + for_each_set_bit(i, &bits, 32)
> > + generic_handle_irq(irq_find_mapping(tb10x_gpio->domain, i));
> > +
> > + return IRQ_HANDLED;
> > +}
>
> (...)
> > + ret = platform_get_irq(pdev, 0);
> > + if (ret < 0) {
> > + dev_err(&pdev->dev, "No interrupt specified.\n");
> > + goto fail_get_irq;
> > + }
> > +
> > + tb10x_gpio->gc.to_irq = tb10x_gpio_to_irq;
> > + tb10x_gpio->irq = ret;
> > +
> > + ret = devm_request_irq(&pdev->dev, ret, tb10x_gpio_irq_cascade,
> > + IRQF_TRIGGER_NONE | IRQF_SHARED,
> > + dev_name(&pdev->dev), tb10x_gpio);
>
> Why aren't you simply using
>
> irq_set_chained_handler()
> irq_set_handler_data(tb10x_gpio);
>
> And in the handler function that need a signature like
> this:
>
> static void tb10x_gpio_irq_handler(unsigned int irq, struct irq_desc *desc)
> {
> struct tb10x_gpio *tb10x_gpio = = irq_get_handler_data(irq);
> struct irq_chip *host_chip = irq_get_chip(irq);
> chained_irq_enter(host_chip, desc);
> (...)
> chained_irq_exit(host_chip, desc);
> }
>
> ?
>
> It's not like I'm 100% certain on where to use one or the other
> construct (a mechanism like the above is needed for threaded
> IRQs I've noticed) but the chained handler seems more to the
> point does it not?
>
> The only downside I've seen is that the parent IRQ does not get
> a name and the accumulated IRQ stats in /proc/interrupts but
> surely we can live without that (or fix it).
>
> Since I'm a bit rusty on chained IRQs correct me if I'm wrong...
OK, it took me a while to figure this back out again because as far as
I'm familiar with the IRQ framework you're right. The reason I'm not
using irq_set_chained_handler is that we have one driver instance per
GPIO bank and all GPIO banks share the same interrupt line. This means
every driver instance needs its own (different) user data and a simple
call to irq_set_handler_data(tb10x_gpio) won't suffice. I'm not aware of
any mechanism that allows interrupt sharing with the
irq_set_chained_handler() mechanism.
Greetings,
Christian
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
