Re: [PATCH] IIO: Adds ACPI support for ST gyroscopes

[Robert Dolca <robert.dolca@gmail.com>]

On Tue, Mar 24, 2015 at 2:17 PM, Lars-Peter Clausen <lars@metafoo.de> wrote:
> [...]
>>
>> +int st_sensors_acpi_i2c_probe(struct i2c_client *client,
>> +                              const struct acpi_device_id *match)
>> +{
>> +       const struct acpi_device_id *id;
>> +       struct gpio_desc *gpiod_irq;
>> +       int ret;
>> +
>> +       id = acpi_match_device(match, &client->dev);
>> +       if (!id)
>> +               return -ENODEV;
>> +
>> +       /* Get IRQ GPIO */
>> +       gpiod_irq = devm_gpiod_get_index(&client->dev, 0, 0);
>> +       if (IS_ERR(gpiod_irq))
>> +               return -ENODEV;
>> +
>> +       /* Configure IRQ GPIO */
>> +       ret = gpiod_direction_input(gpiod_irq);
>> +       if (ret)
>> +               return ret;
>
>
> How exactly does this whole GPIO IRQ thing work with ACPI. Does the ACPI
> description just specify the GPIOs and the driver needs to know which GPIO
> is the is used for the IRQ or does the description indicate that a certain
> GPIO should be used as a IRQ. The reason why I'm asking is that same code
> pops up in pretty much every ACPI I2C sensor driver now. Which suggests that
> this should be factored out into common infrastructure.
>
> And especially the requesting and the setting of the direction of the GPIO
> should not be necessary if the GPIO controller implements interrupt handling
> correctly as this is something that is, as far as I understand, taken care
> of by the GPIO framework when the IRQ is requested.

Hi Lars,

In the ACPI description you specify one or more IRQ GPIO pins. In the
driver you request the GPIO pin using the index. In the ACPI 5.1
specification you can use named GPIOs instead of index.
As far as I know you can not specify the direction in the ACPI
description and I am not sure if we can rely on the fact that the GPIO
pins are in input mode by default.

>
>> +
>> +       /* Map the pin to an IRQ */
>> +       client->irq = gpiod_to_irq(gpiod_irq);
>> +
>> +       /* The name from the ACPI match takes precedence if present */
>> +       memset(client->name, 0, sizeof(client->name));
>> +       strncpy(client->name, (char *) id->driver_data,
>> +               min(sizeof(client->name), strlen((char *)
>> id->driver_data)));
>
>
> Both client->irq and client->name should not be modified by the driver,
> these are only supposed to be set by the I2C framework. Modifying them in
> the driver can result in undefined behavior.

I understand that modifying the name is not a good approach. Doing it
for device tree and not for ACPI can also result in a unknown
behavior. Seems like the best approach here is to remove that name
overwriting from both device tree and acpi probe after becomes clear
why was it there in the 1st place.

Robert