[PATCH 2/2] Documentation: gpio: Update description for X-Gene standby GPIO controller DTS binding

[Y Vo]

On Mon, Sep 14, 2015 at 4:11 PM, Arnd Bergmann <arnd at arndb.de> wrote:
> On Saturday 12 September 2015 12:55:55 Y Vo wrote:
>> On Fri, Sep 11, 2015 at 11:45 PM, Arnd Bergmann <arnd at arndb.de> wrote:
>> > On Friday 11 September 2015 22:06:58 Y Vo wrote:
>> >> On Fri, Sep 11, 2015 at 9:47 PM, Arnd Bergmann <arnd at arndb.de> wrote:
>> >> > On Friday 11 September 2015 21:23:31 Y Vo wrote:
>> >> >> Is that possible if we change to this approach:
>> >> >> - Keep all GPIO as GPIO mode by default.
>> >> >> - If anyone want to use them as interrupt, they must add the interrupt
>> >> >> node to DT specific-board below:
>> >> >>
>> >> >> Example to configure GPIO_DS11, 12, 13 as external irq:
>> >> >>
>> >> >> &sbgpio {
>> >> >>         interrupts = <0x0 0x2b 0x1>,         /* GPIO_DS11 */
>> >> >>                         <0x0 0x2c 0x1>,         /* GPIO_DS12 */
>> >> >>                         <0x0 0x2d 0x1>;         /* GPIO_DS13 */
>> >> >
>> >> > That is exactly the same as you have in your patch. You should
>> >> > always list all interrupts that the gpio controller is wired
>> >> > to, whether they are used or not.
>> >> >
>> >> > Ideally, the driver just decides at runtime how to wire that
>> >> > mux, based on how the gpio is configured by the client driver.
>> >> > If the client asks for an interrupt (with gpio_to_irq) and is
>> >> > it is configured as an input, then you connect it to the GIC,
>> >> > otherwise you leave it to be handled by the gpio chip itself.
>> >> > Any reason you can't do that?
>> >> >
>> >> As I just explained, the GPIO Standby doesn't support the interrupt.
>> >> Both interrupts are came from GIC, so if the client asks for an
>> >> interrupt, it will go to the GIC driver (not GPIO Standby driver), so
>> >> that's my trouble. We can not implement the code to configured the
>> >> GPIO standby at GIC driver.
>> >
>> > So you are saying we merged a binding that is incapable of describing
>> > the hardware? If that is the case, we may have to go back to
>> > the drawing board and come up with a working binding.
>> >
>> > One part that I don't understand yet is how the driver handles
>> > irqs to start with, as it does not register a nested irq domain
>> > or irqchip, so I'm guessing that client drivers cannot actually
>> > use the IRQ functionality. Is that right?
>> For our case the irq_domain is in irq-gic driver, the GPIO driver only
>> is responsible to wire 2 the external irq of GIC to GPIO pin.
>>
>> Example for configure GPIO_DS13 as interrupt and use as button with
>> the current gpio driver:
>>         gpio-keys {
>>                 compatible = "gpio-keys";
>>                 button at 1 {
>>                         label = "POWER";
>>                         linux,code = <116>;
>>                         linux,input-type = <0x1>;
>>                         interrupts = <0x0 0x2d 0x1>;
>>                 };
>>         };
>
> Wait, this looks wrong: the gpio driver doesn't actually see
> the connection here and won't be able to configure the interrupt
> correctly. The interrupt is already owned by the gpio driver, so
> you cannot use it in the button node.

In summary:
- Our GPIO doesn't support interrupt controller.
- There are 6 pins which used the external interrupt from GIC, so all
setup for those irqs are from gic driver. The GPIO driver only
configure to wire those lines.

For your concern:
- That's correct: if we use that defined, the gpio driver never saw
the connection here (That's why it already is configued at the
beginning).
- At the first time, we tried to use the define: <&sbgpio 13 1>, it
means using the GPIO_DS13, it will go into the GPIO driver to setup,
but there is another problem which I have sent out to all of you:
+ It will go into gpio_keys_setup_key (gpio_keys.c driver) function,
then set the irqflags = IRQF_TRIGGER_RISING | IRQF_TRIGGER_FALLING,
but the gic only support IRQ_TYPE_LEVEL_HIGH && IRQF_TRIGGER_RISING,
so it always returns failed at gpio_keys_setup_key function. Please
see the gic_set_type at gic driver.

static int gic_set_type(struct irq_data *d, unsigned int type)
{
        void __iomem *base = gic_dist_base(d);
        unsigned int gicirq = gic_irq(d);

        /* Interrupt configuration for SGIs can't be changed */
        if (gicirq < 16)
                return -EINVAL;

        /* SPIs have restrictions on the supported types */
        if (gicirq >= 32 && type != IRQ_TYPE_LEVEL_HIGH &&
                            type != IRQ_TYPE_EDGE_RISING)
                return -EINVAL;

        return gic_configure_irq(gicirq, type, base, NULL);
}
+ Another issue: in order gpio_key works it needs the status of GPIO.
For our chip, when the GPIO is configued as interrupt, we need to
access to GIC register to read the real status, it is not acceptable
to implement accessing GIC registers at gpio driver. The function
irq_get_irqchip_state(..) also doesn't work in our chip too. Because
it needs to access different offset.

>
>> >
>> > It also seems to me that the binding cannot distinguish between a
>> > line configured as an input and one that is configured as an
>> > interrupt, which are for other gpio chips the same thing, but
>> > not on this one.  Could this be rectified by using another bit
>> > of the second gpio cell? The low bit is used for active-high/active-low,
>> > so you could use the second bit for irq/input.
>> >
>> Do you mean #gpio-cells property and using the high bit of the second
>> bit for irq/input  ?
>
> Yes, that would be an option.
I will look into it.

Is there possible if:
- Keep GPIO8..GPIO as interrupt by default.
- Anyone want to use these GPIO pins as GPIO, we will re-configure
them to GPIO mode ?


>
>         Arnd