gpio_request() vs __gpio_get()

gpio_request() vs __gpio_get()

[Lucas De Marchi]

Hi,

I was looking at drivers/leds/ in how it interacts with gpiolib. I see
that there's a new gpiod_* API.
"git grep gpio_request -- drivers/leds/" reveals quite a number of
users of gpio_request() and gpio_request_one().

Am I right that these are going to be ported to the new gpiod_* API?
If yes, what would be the equivalent calls?  I thought it would be one
of the __gpiod_get() functions, but these would invariably fail if I
passed dev == NULL which is my case.


thanks

-- 
Lucas De Marchi

Re: gpio_request() vs __gpio_get()

[Alexandre Courbot]

Hi,

On Mon, Mar 9, 2015 at 5:14 AM, Lucas De Marchi
<lucas.de.marchi@gmail.com> wrote:
> I was looking at drivers/leds/ in how it interacts with gpiolib. I see
> that there's a new gpiod_* API.
> "git grep gpio_request -- drivers/leds/" reveals quite a number of
> users of gpio_request() and gpio_request_one().
>
> Am I right that these are going to be ported to the new gpiod_* API?
> If yes, what would be the equivalent calls?  I thought it would be one
> of the __gpiod_get() functions, but these would invariably fail if I
> passed dev == NULL which is my case.

Any new code should use the gpiod API. gpio_* are considered
deprecated and unsafe, and users are encouraged to switch to gpiod
(which is not always easy if they use GPIOs in unconventional ways).

With gpiod, GPIO descriptors are obtained using one of the
gpiod_get*() functions (do not call __gpiod_get directly) and must be
assigned to the device and function using either Device Tree, ACPI, or
platform tables. See Documentation/gpio/board.txt for how GPIOs are
assigned, and Documentation/gpio/consumer.txt for how to use the gpiod
API.

Re: gpio_request() vs __gpio_get()

[Lucas De Marchi]

On Mon, Mar 9, 2015 at 6:21 AM, Alexandre Courbot <gnurou@gmail.com> wrote:
> Hi,
>
> On Mon, Mar 9, 2015 at 5:14 AM, Lucas De Marchi
> <lucas.de.marchi@gmail.com> wrote:
>> I was looking at drivers/leds/ in how it interacts with gpiolib. I see
>> that there's a new gpiod_* API.
>> "git grep gpio_request -- drivers/leds/" reveals quite a number of
>> users of gpio_request() and gpio_request_one().
>>
>> Am I right that these are going to be ported to the new gpiod_* API?
>> If yes, what would be the equivalent calls?  I thought it would be one
>> of the __gpiod_get() functions, but these would invariably fail if I
>> passed dev == NULL which is my case.
>
> Any new code should use the gpiod API. gpio_* are considered
> deprecated and unsafe, and users are encouraged to switch to gpiod
> (which is not always easy if they use GPIOs in unconventional ways).
>
> With gpiod, GPIO descriptors are obtained using one of the
> gpiod_get*() functions (do not call __gpiod_get directly) and must be
> assigned to the device and function using either Device Tree, ACPI, or
> platform tables. See Documentation/gpio/board.txt for how GPIOs are
> assigned, and Documentation/gpio/consumer.txt for how to use the gpiod
> API.

Ahhn. Ok. I'm converting a "userspace driver" that used to use the
sysfs interface to be a proper driver in kernel. The use case is for
boards with expansion headers like minnow, beaglebone etc.  In
userspace it would refer to a certain gpio index without any
"function" from DT/ACPI.

My doubt is how to map from this number to gpio_desc. I see there's
gpio_to_desc(), but then I'd need to use gpiod_request() (like
gpiolib-sysfs.c does) which is not exported for users outside of gpio
subsystem.

thanks

-- 
Lucas De Marchi