On Thu, Mar 26, 2015 at 5:00 PM, Octavian Purdila
<octavian.purdila@xxxxxxxxx> wrote:
> On Thu, Mar 26, 2015 at 4:47 PM, Mika Westerberg
> <mika.westerberg@xxxxxxxxxxxxxxx> wrote:
>> On Thu, Mar 26, 2015 at 04:37:39PM +0200, Octavian Purdila wrote:
>>> On Thu, Mar 26, 2015 at 4:04 PM, Mika Westerberg
>>> <mika.westerberg@xxxxxxxxxxxxxxx> wrote:
>>> > On Thu, Mar 26, 2015 at 02:04:35PM +0200, Octavian Purdila wrote:
>>> >> On Thu, Mar 26, 2015 at 12:16 PM, Mika Westerberg
>>> >> <mika.westerberg@xxxxxxxxxxxxxxx> wrote:
>>> >> > On Wed, Mar 25, 2015 at 11:12:16PM +0200, Octavian Purdila wrote:
>>> >> >> On Wed, Mar 25, 2015 at 3:21 PM, Mika Westerberg
>>> >> >> <mika.westerberg@xxxxxxxxxxxxxxx> wrote:
>>> >> >> > On Wed, Mar 25, 2015 at 02:25:05PM +0200, Mika Westerberg wrote:
>>> >> >> >> I think we can do the same for ACPI GpioInts so that we introduce
>>> >> >> >> acpi_gpio_irq_get() that translates from GpioInt to Linux IRQ
>>> >> >> >> numberspace. Then we can do something like below in I2C core:
>>> >> >> >>
>>> >> >> >> if (client->irq <= 0) {
>>> >> >> >> int irq = -ENOENT;
>>> >> >> >>
>>> >> >> >> if (dev->of_node)
>>> >> >> >> irq = of_irq_get(dev->of_node, 0);
>>> >> >> >> else if (ACPI_COMPANION(dev))
>>> >> >> >> irq = acpi_gpio_irq_get(ACPI_COMPANION(dev), 0);
>>> >> >> >>
>>> >> >> >> if (irq == -EPROBE_DEFER)
>>> >> >> >> return irq;
>>> >> >> >> if (irq < 0)
>>> >> >> >> irq = 0;
>>> >> >> >>
>>> >> >> >> client->irq = irq;
>>> >> >> >> }
>>> >> >> >>
>>> >> >> >> Now it has the drawback that the first GpioInt will not be available to
>>> >> >> >> the driver anymore (as a GPIO since it is locked) but if DT already does
>>> >> >> >> the same we should be fine.
>>> >> >> >
>>> >> >> > Below patch should take care of this.
>>> >> >> >
>>> >> >>
>>> >> >> One issue we noticed is that now the gpio request and set input
>>> >> >> directions operations are not called anymore. Some gpio controller
>>> >> >> drivers (dln2, adnp, lynx_point from quickly browsing the code) do not
>>> >> >> explicitly enable the GPIO pin nor set direction to input when the
>>> >> >> interrupt is enabled. Depending on hardware this may be an issue - it
>>> >> >> is on dln2 for example.
>>> >> >>
>>> >> >> Should the gpio controllers enable and set to input in irq_enable,
>>> >> >> irq_bus_sync_unlock, etc.? Or should this be done in gpiolib?
>>> >> >
>>> >> > Good question.
>>> >> >
>>> >> > In general I think that it is assumed that the boot firmware configures
>>> >> > the pin upfront. However, we have seen too many times that it actually
>>> >> > doesn't happen or it is configured wrong.
>>> >> >
>>> >> > Perhaps we could do this in GPIO core, for example in
>>> >> > gpiochip_irq_reqres/gpiochip_irq_map or so.
>>> >> >
>>> >>
>>> >> That sounds good to me. We tested your patch with the patch below and
>>> >> we can now directly use client->irq:
>>> >>
>>> >> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
>>> >> index 568aa2b..9865627 100644
>>> >> --- a/drivers/gpio/gpiolib.c
>>> >> +++ b/drivers/gpio/gpiolib.c
>>> >> @@ -511,6 +511,19 @@ static const struct irq_domain_ops gpiochip_domain_ops = {
>>> >> static int gpiochip_irq_reqres(struct irq_data *d)
>>> >> {
>>> >> struct gpio_chip *chip = irq_data_get_irq_chip_data(d);
>>> >> + int ret;
>>> >> +
>>> >> + ret = gpiod_request(&chip->desc[d->hwirq], "IRQ");
>>> >> + if (ret) {
>>> >> + chip_err(chip, "unable to request %lu for IRQ\n", d->hwirq);
>>> >> + return ret;
>>> >> + }
>>> >
>>> > What if the driver has already requested the GPIO?
>>> >
>>>
>>> Initially I implemented the above to take that into account, e.g. if
>>> (test_and_set_bit(FLAG_REQUESTED, &desc->flags) ...
>>>
>>> But than I thought that we can't mess up with the GPIO anyway while
>>> the interrupt is in use.
>>
>> That's right but then the above will fail also normal cases. For example
>> if the driver gets the irq like:
>>
>> desc = devm_gpiod_get(dev, ..);
>> gpiod_direction_input(desc);
>> irq = gpiod_to_irq(desc);
>>
>> ret = request_irq(irq, ...)
>>
>> at this point we end up calling gpiochip_irq_reqres() which cannot
>> request the GPIO again and fails.
>>
>
> Good point, let me add back that check then :)
>
I just realized that there is another issue: gpiochip_irq_reqres() is
called under a spinlock, so we can call gpiod_request() only if the
gpio controller does not sleep.
For the sleep case I think the GPIO controller needs to do the pin
enable and set input direction operation in it's irq_bus_sync_unlock.
>>> One case I missed was if the user wants to read the GPIO while using
>>> it as an interrupt which seems to be possible...
>>
>> While the GPIO is locked as IRQ it cannot be done as far as I can tell
>> but you can work it around by calling free_irq() first.
>>
>
> AFAICS you can not set the direction to output but you can still read
> values while the interrupt is active.
>
>>>
>>> >> +
>>> >> + ret = gpiod_direction_input(&chip->desc[d->hwirq]);
>>> >> + if (ret) {
>>> >> + chip_err(chip, "unable to set HW IRQ %lu as input\n", d->hwirq);
>>> >> + return ret;
>>> >> + }
>>> >>
>>> >> if (gpiochip_lock_as_irq(chip, d->hwirq)) {
>>> >> chip_err(chip,
--
To unsubscribe from this list: send the line "unsubscribe linux-iio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
