Acquire and release semantics affect all memory, if I understand right. But wouldn't (in theory) a mutex just need to synchronize its contents (in case of a mutex that contains a value, which is the case in Rust, but isn't the case in C)?
For example, could a Mutex<u8> (in theory) be implemented with relaxed memory order and yet fulfill the API specifications? For example, aquiring a lock could set an AtomicI16 to -1 while the lock is held and releasing the lock would set it to a value between 0 and 255 (indicating the stored value). Using Ordering::Relaxed could still ensure that only one thread accesses this "mutex" at a time. Or does the terminology "mutex" imply that it also synchronizes accesses outside the mutex? And if so (or not), where is this documented/specified? Or is this common knowledge?
Sorry to ask such (possibly) "dumb" questions, but for me, synchronization is a difficult topic to understand, and I would like to get a deeper understanding. I might solve my broadcast channel problem with a less-efficient yet safe variant, but ultimately, I would like to understand how to implement locks and channels on my own (and I feel like it will require a lot of understanding to not run into soundness issues or other errors).