nRF52 serial: Tighten/simplify atomics

[kjbracey]

Description

Use new atomics (exchange, load, store and bool types) to simplify
and improve the atomics in the nRF52 serial HAL.

• Ensure mutexes are released last and atomically when done
  done inside a critical section.
• Compare-and-swap is not required for the spinlock - exchange is
  sufficient. (Not clear a spinlock is needed anyway, but left in).
• Remove unneeded volatile, and make mutexes bool.

Requires preceding PR #9600.

Pull request type

[ ] Fix
[ ] Refactor
[x] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

@marcuschangarm

[ciarmcom]

@kjbracey-arm, thank you for your changes.
@marcuschangarm @ARMmbed/mbed-os-maintainers please review.

[pan-]

Changes looks good however I'm concern about the flags tx_async and rx_asynch which are used to determine if a transaction is in progress.

[kjbracey]

I see what you mean. I was under the impression that the async variables were protected by the in_progress atomic, but that's not really true (any more?), is it? The in_progress is actually only set when it's async anyway.

It might be okay - it's only read by the interrupt handler, and it's written by the async setup code to make it respond differently. As long as it writes to it before enabling interrupts, and the interrupt enable is a memory barrier, you're fine.

But interrupt enables might not be memory barriers: patch coming for that here - ARM-software/CMSIS_5#510

I'm not really sure how much any of this is required - most HALs don't have any sort of protection in this area, and just assume you don't mix async and sync operations, or attempt overlapping async.

[0xc0170]

CI started

[mbed-ci]

Test run: SUCCESS

Summary: 12 of 12 test jobs passed
Build number : 1
Build artifacts

[cmonr]

cmonr added release-version: 5.12.0-rc1 release-version: 5.11.5 and removed release-version: 5.12.0-rc1 labels a day ago

Ha. Should've known. Preceeding PR was targeting 5.12, so this needs to as well.