On Mon Feb 17, 2025 at 9:08 PM CET, Andy Shevchenko wrote: > On Mon, Feb 17, 2025 at 12:20:13PM +0100, Mathieu Dubois-Briand wrote: > > To provide a bit more details, there is basically two set of pins usable > > as GPIOs. > > > > On one side we have what I refer to as GPIOs: > > - PORT0 to PORT7 pins of the chip. > > - Shared with PWM and rotary encoder functionalities. Functionality > > selection can be made independently for each pin. We have to ensure > > the same pin is not used by two drivers at the same time. E.g. we > > cannot have at the same time GPIO4 and PWM4. > > - Supports input and interrupts. > > - Outputs may be configured as constant current. > > - 8 GPIOS supported, so ngpios is fixed to MAX7360_MAX_GPIO. > > - maxim,max7360-gpio compatible, gpio_function == MAX7360_GPIO_PORT. > > > > On the other side, we have what I refer to as GPOs: > > - COL2 to COL7 pins of the chip. > > - Shared with the keypad functionality. Selections is made by > > partitioning the pins: first pins for keypad columns, last pins for > > GPOs. Partition is described by the ngpios property. > > - Only support outputs. > > - maxim,max7360-gpo compatible, gpio_function == MAX7360_GPIO_COL. > > > > > Or you mean that there output only GPIO lines in HW after all? > > > Is there a link to the datasheet? > > > > A datasheet is available on https://www.analog.com/en/products/max7360.html > > Thank you for this good elaboration! > I will check on the datasheet later on, having one week off. > Thanks for your feedback! Sorry I haven't been able to work on this series for the last few weeks, but I finally had the opportunity to integrate your comments. > But what I have read above sounds to me like the following: > > 1) the PORT0-PORT7 should be just a regular pin control with the respective > function being provided (see pinctrl-cy8c95x0.c as an example); > Ok, so I created a pin control driver for the PORT pins. This will effectively help to prevent concurrent use of pins in place of the request()/free() callbacks. My only concern is: as there is no real pin muxing on the chip, my .set_mux callabck in pinmux_ops structure is not doing anything. It looks like I'm not the only one (drivers/pinctrl/pinctrl-microchip-sgpio.c does the same thing), but I hope this is OK. > 2) the COL2 COL7 case can be modeled as a simplest GPIO (GPO) driver with > reserved lines property (this will set valid mask and let GPIOLIB to refuse any > use of the keypad connected pins. > I mostly went that way, just a few notes. I chose to not use the reserved lines property in the device tree, but instead implemented a gpiolib init_valid_mask() callback. In believe this is better, as: - We can automatically generate the valid gpios mask, based on the number of columns used. - It allows to get rid of the compatibility check between the number of columns and the number of GPIOs provided by the device tree: DT provides the number of columns, we deduct the number of GPIOs. I chose to number GPIOs from 0 to 7. - This might be a bit questionable, as GPIO 0 and 1 will always be invalid: pins 0 and 1 of the chip cannot be used as GPIOs. I'm definitely open to discussion on this point. - Yet I believe it simplifies everything for the user: pin numbers and GPIO numbers are the same instead of having an offset of 2. - It also simplifies a bit the GPIO driver code. -- Mathieu Dubois-Briand, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
