[PATCH] gpio/at91: free GPIO after configuring as input

[Javier Martinez Canillas]

On Fri, Apr 11, 2014 at 5:45 PM, Alexander Stein <alexanders83 at web.de> wrote:
> I'm CC'ing some input guys if this can't be solved in the pinctrl/irq side.
>
> On Friday 11 April 2014, 23:30:33 wrote Jean-Christophe PLAGNIOL-VILLARD:
>>
>> On Apr 11, 2014, at 10:24 PM, Alexander Stein <alexanders83 at web.de> wrote:
>>
>> >
>> > If the GPIO stays requested a device driver can't request it again.
>> > e.g. Without this patch the ads7846 driver returns the following error:
>> > ads7846 spi32766.3: failed to request/setup pendown GPIO15: -16
>> > ads7846: probe of spi32766.3 failed with error -16
>> >
>> > /sys/kernel/debug/gpio shows this:
>> > GPIOs 0-31, platform/fffff200.gpio, fffff200.gpio:
>> > [/ahb/apb/pinctrl at fffff200/gpio at fffff200] GPIOfffff200.gpio15: [gpio] set
>> >
>> > Signed-off-by: Alexander Stein <alexanders83 at web.de>
>> > ---
>> > I'm aware that it makes sense this GPIO is/stays requested, but either the
>> > pinctl or device driver have to be adjusted as both can't request this GPIO.
>> > I think the latter shouldn't change.
>> >
>> > drivers/pinctrl/pinctrl-at91.c | 2 ++
>> > 1 file changed, 2 insertions(+)
>> >
>> > diff --git a/drivers/pinctrl/pinctrl-at91.c b/drivers/pinctrl/pinctrl-at91.c
>> > index d990e33..63176f2 100644
>> > --- a/drivers/pinctrl/pinctrl-at91.c
>> > +++ b/drivers/pinctrl/pinctrl-at91.c
>> > @@ -1493,6 +1493,8 @@ static int at91_gpio_irq_domain_xlate(struct irq_domain *d,
>> >     if (ret)
>> >             return ret;
>> >
>> > +   gpio_free(pin);
>> > +
>>
>> NACK it the whole key point the gpio use as a IRQ so the irq generic code request it
>> >     return 0;
>> > }
>> >
>>

The problem is that the GPIO and IRQ subsystems are somehow related
but completely independent. A GPIO pin used as an IRQ line is
completely orthogonal to requesting a GPIO pin.

Is true that here is a common pattern in the kernel that is:

gpio_request(gpio,...);
gpio_direction_input()
request[_threaded]_irq(gpio_to_irq(gpio), ...);

But one should not assume that requesting a GPIO to be used as an IRQ
line implies requesting the GPIO first. It is completely legal to only
request the IRQ without requesting the GPIO before.

Of course at a hardware level the pin has to be configured as input if
needed by the chip controller and also a following call to
gpio_request() as output should not be supported.

After a long discussion the agreement is that the driver should be
able to setup as a hw level if another driver wants to use a GPIO pin
as an IRQ line but not requesting the GPIO. So, I think that is wrong
to call gpio_request() and gpio_direction_input() on a irq_domain_ops
.xlate() function handler since by doing that a subsequent call to
gpio_request() is failing like you are reporting.

The GPIO subsystem has a new gpio_lock_as_irq() helper function to
mark a GPIO pin as already used as a IRQ line and only allowing to
request a GPIO as input if is already marked as an IRQ.

Please take a look to commit 2f56e0a ("gpio/omap: use gpiolib API to
mark a GPIO used as an IRQ") for an example on how to use this helper
function.

Also, there is a new GPIOLIB_IRQCHIP infrastructure in gpiolib that
are a set of helper functions to associate an IRQ chip to a GPIO
controller. This should cover the most common use case and most
probably this driver and can be converted to use it.

Please refer to commits:

1425052 ("gpio: add IRQ chip helpers in gpiolib")
e0bc34a ("pinctrl: nomadik: convert driver to use gpiolib irqchip")

for an example on how to use it.

Thanks a lot and best regards,
Javier