GPIO controllers may have some pins which can be excluded from the GPIO
usage on certain hardware configurations. The valid_mask member of the
struct gpio_chip has been used to denote usable pins if some pins should
be excluded.

The GPIO request should fail for GPIOs which are masked. Under certain
conditions this was only done when GPIO chip provided the 'request'
callback. We fix this to be always done.

The valid_mask member of the gpio_chip should no longer be directly
populated by the drivers but GPIO core does this (unconditionally,
overwriting any mask set directly by the drivers). Drivers are intended
to populate the valid_mask using the init_valid_mask -callback.

This series enforces using the init_valid_mask by hiding the valid_mask
in structure which is internal to the GPIO core. A single in-tree driver
was found to access the valid_mask directly. This series also removes
those direct accesses as has been discussed [1]. Additionally, we
introduce a getter-function which can be used to obtain the valid_mask.

[1]: https://lore.kernel.org/all/TY3PR01MB11346EC54C8672C4D28F931F686CC2@TY3PR01MB11346.jpnprd01.prod.outlook.com/

Revision history:
v1 => v2:
 - Hide the 'valid_mask' instead of documenting it to be internal
 - Make the gpio_request() to obey the valid_mask whether the gpio_chip
   has the 'request' -callback populated or not.

---

Matti Vaittinen (4):
  gpio: Respect valid_mask when requesting GPIOs
  gpio: Add a valid_mask getter
  gpio: gpio-rcar: Drop direct use of valid_mask
  gpio: Hide valid_mask from direct assignments

 drivers/gpio/gpio-rcar.c    | 13 +++++-------
 drivers/gpio/gpiolib.c      | 40 ++++++++++++++++++++++++++-----------
 drivers/gpio/gpiolib.h      |  3 +++
 include/linux/gpio/driver.h |  9 +--------
 4 files changed, 37 insertions(+), 28 deletions(-)


base-commit: d082ecbc71e9e0bf49883ee4afd435a77a5101b6
-- 
2.48.1

On Fri, Feb 28, 2025 at 1:35 PM Matti Vaittinen
<mazziesaccount@gmail.com> wrote:
>
> GPIO controllers may have some pins which can be excluded from the GPIO
> usage on certain hardware configurations. The valid_mask member of the
> struct gpio_chip has been used to denote usable pins if some pins should
> be excluded.
>
> The GPIO request should fail for GPIOs which are masked. Under certain
> conditions this was only done when GPIO chip provided the 'request'
> callback. We fix this to be always done.
>
> The valid_mask member of the gpio_chip should no longer be directly
> populated by the drivers but GPIO core does this (unconditionally,
> overwriting any mask set directly by the drivers). Drivers are intended
> to populate the valid_mask using the init_valid_mask -callback.
>
> This series enforces using the init_valid_mask by hiding the valid_mask
> in structure which is internal to the GPIO core. A single in-tree driver
> was found to access the valid_mask directly. This series also removes
> those direct accesses as has been discussed [1]. Additionally, we
> introduce a getter-function which can be used to obtain the valid_mask.
>
> [1]: https://lore.kernel.org/all/TY3PR01MB11346EC54C8672C4D28F931F686CC2@TY3PR01MB11346.jpnprd01.prod.outlook.com/
>

Hey! This no longer applies to my gpio/for-next branch, would you mind
rebasing and resending?

Bart