On Thu, 2023-11-30 at 15:36 +0200, Andy Shevchenko wrote: > On Thu, Nov 30, 2023 at 11:20:32AM +0100, Nuno Sá wrote: > > On Wed, 2023-11-29 at 21:55 +0100, Linus Walleij wrote: > > > On Wed, Nov 29, 2023 at 5:08 PM Nuno Sá <noname.nuno@xxxxxxxxx> wrote: > > > > > > > Cool, I actually thought that having the direction + get/set stuff would be > > > > weird > > > > given the fact that we can only PULL_LOW or HIGH_Z the pins. > > > > > > There are several drivers in the kernel that implement .set_config(), > > > it's existing and should be enabled if it has uses. > > > > Yeah, it might make sense to support it specially for the input case. AFAICT, if > > I > > use the .set_config() (but from a quick look I think we will need to add support > > for > > it in gpiolib for the high-z configuration), then I can't use the gpio_regmap > > stuff. > > As the driver stands I don't think I could do it anyways because setting gpio2-3 > > and > > alert requires to write 0 on the register rather than 1. But again, I'm still > > very > > suspicious about the whole thing. The datasheet states: > > > > "GPIO1-GPIO3 and ALERT all have comparators monitoring > > the voltage on these pins with a threshold of 1.28V even when > > the pins are configured as outputs." > > > > But we can't really set the direction for gpio2-3 and the alert pins (only > > getting > > the level and setting it as PULL_LOW or HIGH_Z. gpio1 is the only one where we > > can > > configure it as input or open drain ouput. Bah, I'll try to see if someone > > internally > > can shed some light on this. > > I have better proposal. If these GPIOs are not needed for the main > functionality of the hardware, can we just left it out for now and implement > later if required? > Well, I did spent some time on the gpio thing so I would like to have it in but yeah, no hard feelings if it does not go in. So, I actually talk with some hw guys and the pull_low is not really like a pull_low resistor. These pins are effectively an open drain. Which means, setting them as input means setting them in high-z (turning off the mosffet) - and I do have a bug in my code regarding this - Or if you want them as outputs you can set the level low (and it will always be low - just turn on the mosffet) or you can also set high-z which means it will be either low or high depending on your external circuitry. The point is, you can still have your pin acting as a normal gpo if you accommodate your circuitry for it (can also use these pins for things like buses). Also got me thinking if a gpi vs gpo devicetree property would make sense. But I would likely leave it more generic/relaxed for now (even though I think you would need to be creative and actually use more HW to have the possibility of using these pins as GPIs and GPOs at the same time). - Nuno Sá
