Remove references to static mut

[richardeoin]

rust/#114447 made &mut MY_STATIC_MUT a warning, with the intention of disallowing it entirely in the 2024 edition. We never used this in the crate itself, but it is used in many examples.

This commit removes these references from the examples. In general the new approach is to change static mut of type X to type MaybeUninit<X>. These can then cleanly be accessed with MY_STATIC_MUT.write() followed by MY_STATIC_MUT.assume_init_mut().

Where we need to cast a (reference to a)MaybeUninit<[X; N]> to a [MaybeUninit<X>; N] in order to initialise it element-by-element, the &mut *ptr::addr_of_mut!(MY_STATIC_MUT) construction is still used. After initialisation, we can then access it with MY_STATIC_MUT.assume_init_mut().

In a few DMA Examples we were taking a reference &mut TARGET_BUFFER without initialising TARGET_BUFFER. This commit changes this to TARGET_BUFFER.assume_init_mut() without initialising TARGET_BUFFER. This was and remains UB.

• USB Examples: static mut EP_MEMORY now has type MaybeUninit<[u32; 1024]>
• Ethernet Examples: static mut DES_RING now has type MaybeUninit<ethernet::DesRing<4, 4>>

[richardeoin]

I re-tested the following examples and they are still working:

• examples/ethernet-rtic-stm32h747i-disco.rs
• examples/mdma.rs
• examples/usb_serial.rs

I'm going to tag a few of the original authors by their examples below. If you still have access to some suitable hardware it would be really awesome if you can re-test these examples in the next week or two(!)

examples/ethernet-rtic-nucleo-h723zg.rs @jlogan03
examples/sai_dma_passthru.rs @antoinevg
examples/serial-dma.rs @mattico
examples/spi-dma-rtic.rs @ryan-summers

[jlogan03]

Hi! Thanks for doing this - I'm traveling this week, but I'll test it as soon as I'm back to my hardware

[mattico]

@richardeoin thanks again for all the work you do maintaining this!

I tested examples/serial-dma.rs on an ST NUCLEO-H743ZI2 board and it worked, I checked the transmitted data.

I also tested examples/spi-dma-rtic.rs on the same board and checked the transmitted data. Well the first few bytes at least, I had to decode manually because my scope's SPI decoder wasn't working. It's been a while.

---

Rather than doing TARGET_BUFFER.assume_init_mut() on an uninitialized buffer, what would the "correct" solution be? Creating a wrapper struct containing a pointer and implementing embedded_dma::WriteBuffer on it? I won't ask you to implement that in this PR but maybe you could add a comment saying it's UB and pointing in the right direction, wherever that may be.

[antoinevg]

Can confirm examples/sai_dma_passthru.rs is still working fine.

[richardeoin]

Rather than doing TARGET_BUFFER.assume_init_mut() on an uninitialized buffer, what would the "correct" solution be? Creating a wrapper struct containing a pointer and implementing embedded_dma::WriteBuffer on it? I won't ask you to implement that in this PR but maybe you could add a comment saying it's UB and pointing in the right direction, wherever that may be.

Creating a mutable reference like &mut [u8] to uninitialised memory is UB. The solutions would be:

• Initialise the memory first
• Or yes have a wrapper type around a raw pointer that implements embedded_dma::WriteBuffer without creating that &mut reference

[richardeoin]

Rebased onto master

[jlogan03]

Confirmed ethernet-rtic-nucleo-h723zg still runs