On Wed, Apr 5, 2023 at 3:57 PM Michael Walle <michael@xxxxxxxx> wrote: > OTOH I'm not sure the driver is doing it correctly, because it also > seems to switch the pullup resisters together with the direction. > I'm not sure that is correct. So there might be just one register > involved after all and the GPIO_REGMAP should work again. I'm pretty sure that should be in the .set_config() callback. > Also, according to the datasheet this has some nv memory (to set the > initial state of the GPIOs [?]). So it should really be a multi-function > device. I'm not sure if this has to be considered right from the > beginning or if the device support can start with GPIO only and later > be transitioned to a full featured MFD (probably with nvmem support). That's a bit of a soft definition. If the chip is *only* doing GPIO and nvram it can be a GPIO-only device I think. The precedent is a ton of ethernet drivers with nvram for storing e.g. the MAC address. We don't make all of those into MFDs, as the nvram is closely tied to the one and only function of the block. Yours, Linus Walleij
