[PATCH 5/6] gpio: tegra: Dynamically allocate IRQ base, and support DT

[Stephen Warren]

Thierry Reding wrote at Thursday, January 05, 2012 12:23 AM:
> * Stephen Warren wrote:
> > @@ -343,6 +344,16 @@ static int __devinit tegra_gpio_probe(struct platform_device *pdev)
> >  	int i;
> >  	int j;
> >
> > +	irq_domain.irq_base = irq_alloc_descs(-1, 0, TEGRA_NR_GPIOS, 0);
> > +	if (irq_domain.irq_base < 0) {
> > +		dev_err(&pdev->dev, "Couldn't allocate IRQ numbers\n");
> > +		return -ENODEV;
> > +	}
> > +	irq_domain.nr_irq = TEGRA_NR_GPIOS;
> > +	irq_domain.ops = &irq_domain_simple_ops;
> > +	irq_domain.of_node = pdev->dev.of_node;
> > +	irq_domain_add(&irq_domain);
> 
> I was wondering: except for setting the nr_irq field this is pretty much what
> irq_domain_add_simple() does. While I get that you need access to the domain
> later on and cannot use the helper, I'm still wondering why the nr_irq field
> is not initialized by irq_domain_add_simple().

I'm not completely sure, but I think irq_domain_add_simple() was initially
added as a transition measure; it looks like all the current users are for
a single top-level interrupt controller where the Linux IRQ numbers are
used directly in the .dts files. Once you add other interrupt controllers
into the mix, the API as-is starts to make less sense.

> I have a driver for a GPIO/IRQ expander that does exactly the same and was
> always wondering why the irq_data.hwirq field isn't properly setup, and
> irq_domain.nr_irq being 0 seems to be exactly the reason. So am I supposed to
> not use irq_domain_add_simple() in this case or should we rather update the
> helper to take a nr_irq parameter that can be used to initialize the nr_irq
> field?

I think updating the helper like that makes sense, and also have it return
the IRQ domain object. Grant, do you agree?

-- 
nvpublic