Race condition in gpio

[usbalbin]

I hope I am wrong, but I think there might be a race condition in the gpio's into_{x} methods.

stm32g4xx-hal/src/gpio.rs

Line 386 in f8b6ff8

gpio.pupdr.modify(|r, w| {

Notice how we perform a read-modify-write sequence without anything preventing the same thing being done to another pin in the same port.

As far as I can see, I believe we would need either a critical section or exclusive access to some sort of token to guarantee exclusive access. A critical section would cause a slight bit of overhead but would otherwise be non-breaking API-wise, I think.

[dtjones190]

I think that you are correct, there is a potential race condition here.
It looks like the stm32g4 supports bit-banding on peripherals, so that would give you atomic writes of single bits without needing critical sections or API changes.

[usbalbin]

But writing the bits one at a time would on the other hand cause potentialy bad things to happen since the write would be split up into multiple writes.

For example looking at PUPD[15:0][1:0] in the RM, changing from 01: Pull-up to 10: Pull-down would be

1. Store 1 to left bit at bit banding address <-- We now have 11 which is reserved
2. Store 0 to right bit

[usbalbin]

However i suppose it would work very well for things that are single bit like OT[15:0] in GPIOx_OTYPER

[usbalbin]

Related issue stm32-rs/stm32f3xx-hal#37