Re: [PATCH 2/2] gpio/tegra: Make it a DT interrupt controller

[Rob Herring <robherring2-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

On 11/30/2011 11:19 AM, Stephen Warren wrote:
> Rob Herring wrote at Tuesday, November 29, 2011 8:28 PM:
>> On 11/29/2011 07:32 PM, Stephen Warren wrote:
>>> Fix the DT binding documentation to describe interrupt-related properties,
>>> and the contents of "child" node interrupts property.
>>>
>>> Update tegra20.dtsi to specify the required interrupt-related properties.
>>>
>>> Fix the driver to creating an IRQ domain for itself, so that child node
>>> interrupts properties are correctly parsed.
> 
>>> diff --git a/Documentation/devicetree/bindings/gpio/gpio_nvidia.txt b/Documentation/devicetree/bindings/gpio/gpio_nvidia.txt
> ...
>>> +- #interrupt-cells : Should be 3.
>>
>> Should be 2??
> 
> Yes, typo.
> 
>>> diff --git a/drivers/gpio/gpio-tegra.c b/drivers/gpio/gpio-tegra.c
> 
>>> +	if (pdev->dev.of_node)
>>> +		irq_domain_add_simple(pdev->dev.of_node, TEGRA_GPIO_TO_IRQ(0));
>>> +
>>
>> This is a bit of a temporary solution. Really, you want to end up with
>> something like this to eliminate INT_GPIO_BASE and dynamically assign
>> the irqbase:
> 
>>  static int tegra_gpio_to_irq(struct gpio_chip *chip, unsigned offset)
>>  {
>> -       return TEGRA_GPIO_TO_IRQ(offset);
>> +       return irq_domain_to_irq(chip->domain, offset);
>>
>>>>> This is a bit wrong. You need a tegra_gpio_chip containing a domain
>> and gpio_chip.
> 
> That implies there's always an IRQ domain for this chip. Right now, my
> patch only creates an IRQ domain when the GPIO controller is instantiated
> from DT; there is no domain when instantiated as a platform device from a
> board file. I suppose that really should change, but:
> 
> Some board files initialize platform devices as follows:
> 
> ./arch/arm/mach-tegra/board-seaboard.c:162:	.irq = TEGRA_GPIO_TO_IRQ(TEGRA_GPIO_ISL29018_IRQ),
> ./arch/arm/mach-tegra/board-seaboard.c:186:	.irq = TEGRA_GPIO_TO_IRQ(TEGRA_GPIO_CDC_IRQ),
> ./arch/arm/mach-tegra/board-harmony.c:104:	.irq = TEGRA_GPIO_TO_IRQ(TEGRA_GPIO_CDC_IRQ),
> 
> Is there a way to make that work if there is no static IRQ numbering and
> hence the compile-time macro TEGRA_GPIO_TO_IRQ() doesn't exist?
> 
> For Tegra, IRQ domain support (CONFIG_IRQ_DOMAIN) is only enabled via
> CONFIG_USE_OF, so for non-DT builds isn't enabled. I guess we could
> easily make CONFIG_ARCH_TEGRA select CONFIG_IRQ_DOMAIN though.
> 

No, the GIC selects IRQ_DOMAIN now.

> The sooner we switch to DT-only the better!

Dynamic irq assignment doesn't require DT. If you can change the .irq
init above to run-time, I think you can just call gpio_to_irq. You could
have some init ordering issues though.

You could leave the base hard-coded for now, but limit it's use to
domain.irqbase. That's at least much better INT_GPIO_BASE used all over
the place.

> 
>>  static void tegra_gpio_irq_ack(struct irq_data *d)
>>  {
>> -       int gpio = d->irq - INT_GPIO_BASE;
>> +       int gpio = d->hwirq;
> 
> Is the hwirq field valid when there isn't an IRQ domain?

No. irq_domain_add sets it up.

Rob