[WIP] Synchronization primitives #11
Conversation
|
7408c0e sketches what I mean by "reference counting" of critical sections (a poor choice of words). |
|
a53c68b is another refinement. |
| static mut irq_level : uint = 0; | ||
|
|
||
| #[inline(always)] | ||
| pub fn disable_irqs() { |
There was a problem hiding this comment.
Can you please add some documentation here about which irqs are/aren't disabled. cpsid i doesn't disable Reset, HardFault or NMI. cpsid f doesn't disable NMI.
|
The cortex M3/M4 should be using LDREX to get the lock. Only the M0/M0+/M1 need to disable/enable interrupts. Also thinking this would be more flexible (and easier to review) if you start with a lock primitive in cortex_m3 and build the mutex/condition variable on top of that. Also needs a rebase/squash. |
|
@bharrisau, thanks! I'll rework things a bit more and try my hand at using LDREX. |
|
@bharrisau, how does this look? As far as I can tell, |
|
That one looks nice, I'll drop a few style comments in a moment. |
| #[must_use] | ||
| pub struct Guard<'a>(&'a Lock); | ||
|
|
||
| pub static STATIC_Lock: Lock = Lock { locked: Unsafe { value: false, marker1: InvariantType } }; |
|
Also, a few generic comments: use 2-space idents and make sure you have licence headers in all new files. Would it be reasonable to make |
| */ | ||
| pub fn lock<'a>(&'a self) -> Guard<'a> { | ||
| unsafe { | ||
| // we need the critical section until the end of this function |
This is a basic locking primitive for buillding synchronization primitives.
This implements a simple abstraction for critical sections where interrupts must be disabled.
This is a simple container for thread-safe state with access guarded by a CritSection.
This implements a simple thread-safe queue.
| pub fn try_lock<'a>(&'a self) -> Option<Guard<'a>> { | ||
| unsafe { | ||
| match *self.owner.get() { | ||
| None => { |
|
LGTM |
Added a few synchronization primitives.
|
Good job. Once we get a solution for bit-banded access we can add a third method for handling locks - bit band writes are atomic so you can do a 32 position lock with low overhead. |
Here is the current state of my mutex support. Progress on this has been pretty sporatic and things are still in a state of minor disarray but I've cleaned things up the best I can and rebased against master for reference.
There are a few things I should point out there,
IrqDisabledtoken is reasonable given how it is abused inMutex::lock. The idea here is that operations onQueue(which is ideally thread-safe) require that you prove that the caller is in a critical section with interrupts disabled. It might be better to instead use a reference-counting approach here instead.taskcould use some cleaning up. At the moment I've just marked a few things aspubto get things to compile but things could be a bit cleaner.