Add reference counting to GPIO pins to init/deinit clock control to port

[tullom]

Resolves #61
GPIO port clocks are now disabled if no pin from that port is active.

gpio::init() has been changed to disable all clocks and only explicitly enable the port clock when a gpio pin (Input, Output, Flex) is instantiated.

[tullom]

Some kicking-off questions

[felipebalbi]

I think as an overall comment, since you're refcounting the clocks for GPIO, why isn't this done as part of the clock driver? We already have a good trait, IIRC, for modelling syscon peripherals. Could that be used together with refcounting to have globally refcounted clocks?

[felipebalbi]

why do you disable the peripheral during initialization?

[tullom]

I did some testing and it looks like the gpio clocks are being enabled somewhere in our embassy_imxrt::init() function, so enable_and_resetting() here technically does nothing. I double checked this with asserting cc1.pscctl1().read().hsgpio0_clk().is_disable_clock() before instantiating the first pin. Since we want to save power, i think these should be off by default. But to your general comment, if we move this to the clock code then this will probably disappear as we can (maybe) disable the clock there?

[tullom]

Update: see comment below for justification for why i don't think moving to the clock driver is correct here.

[felipebalbi]

clocks are enabled as part of gpio::init() which is called by the application (should it be, though?). Either way, disabling the clocks here is the wrong thing to do. Rather, you should review the current clock usage for this driver and make sure it matches what you're trying to achieve, fixing any discrepancies along the way.

[jerrysxie]

Yeah, we should just remove the gpio::init() from embassy_imxrt::init() if now it is automatically enabled and disabled with your changes.

[tullom]

Can the interrupt logic in the artifacts of init() be moved into the refcounting logic? aka, calling unpend() and enable() when initing and pend() and disable() when deiniting?

[tullom]

To @felipebalbi's point as to why this isn't done as part of the clock driver, i think that implementing this on the peripheral level allows for more flexibility since we are using the clocks::enable_and_reset() clock fn's in the new() function of the peripherals. In order to set up refcounting for those, we would need to implement drop for those peripherals as well (to call clocks::disable()), which i think is beyond the scope of this PR

[felipebalbi]

To @felipebalbi's point as to why this isn't done as part of the clock driver, i think that implementing this on the peripheral level allows for more flexibility since we are using the clocks::enable_and_reset() clock fn's in the new() function of the peripherals. In order to set up refcounting for those, we would need to implement drop for those peripherals as well (to call clocks::disable()), which i think is beyond the scope of this PR

once could argue that a PR implementing refcounting always maintains it balanced, regardless of the method used for balacing those calls. Rust just happens to provide this nice Drop trait which can be used for this. Additionally, we should disable clocks when the peripheral is dropped (or, decrement the refcount).

[jerrysxie]

We will need a rebase on main as there is conflicts.

[jerrysxie]

Yeah, we should just remove the gpio::init() from embassy_imxrt::init() if now it is automatically enabled and disabled with your changes.

[tullom]

Yeah, we should just remove the gpio::init() from embassy_imxrt::init() if now it is automatically enabled and disabled with your changes.

Can the interrupt logic in the artifacts of init() be moved into the refcounting logic? aka, calling unpend() and enable() when initing and pend() and disable() when deiniting?

[jerrysxie]

Looks good to me. Please coordinate with @madeleyneVaca to have them test their ADO SoC repos against this change before uptake this. This introduces the drop semantic for gpio so if they are not holding onto the ownership of the pins. It could produce bugs.