Merge arbiter and channel into sync

6 files changed, 827 insertions, 0 deletions

diff --git a/rtic-sync/.gitignore b/rtic-sync/.gitignore
new file mode 100644
index 0000000..1e7caa9
--- /dev/null
+++ b/rtic-sync/.gitignore
@@ -0,0 +1,2 @@
+Cargo.lock
+target/
diff --git a/rtic-sync/CHANGELOG.md b/rtic-sync/CHANGELOG.md
new file mode 100644
index 0000000..d3a9d84
--- /dev/null
+++ b/rtic-sync/CHANGELOG.md
@@ -0,0 +1,16 @@
+# Change Log
+
+All notable changes to this project will be documented in this file.
+This project adheres to [Semantic Versioning](http://semver.org/).
+
+For each category, *Added*, *Changed*, *Fixed* add new entries at the top!
+
+## [Unreleased]
+
+### Added
+
+### Changed
+
+### Fixed
+
+## [v1.0.0] - 2023-xx-xx
diff --git a/rtic-sync/Cargo.toml b/rtic-sync/Cargo.toml
new file mode 100644
index 0000000..440a6be
--- /dev/null
+++ b/rtic-sync/Cargo.toml
@@ -0,0 +1,30 @@
+[package]
+name = "rtic-sync"
+version = "1.0.0-alpha.0"
+
+edition = "2021"
+authors = [
+  "The Real-Time Interrupt-driven Concurrency developers",
+  "Emil Fresk <emil.fresk@gmail.com>",
+  "Henrik Tjäder <henrik@tjaders.com>",
+  "Jorge Aparicio <jorge@japaric.io>",
+  "Per Lindgren <per.lindgren@ltu.se>",
+]
+categories = ["concurrency", "embedded", "no-std", "asynchronous"]
+description = "Synchronization primitives for asynchronous contexts"
+license = "MIT OR Apache-2.0"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[dependencies]
+heapless = "0.7"
+critical-section = "1"
+rtic-common = { version = "1.0.0-alpha.0", path = "../rtic-common" }
+
+[dev-dependencies]
+tokio = { version = "1", features = ["rt", "macros", "time"] }
+
+
+[features]
+default = []
+testing = ["critical-section/std", "rtic-common/testing"]
diff --git a/rtic-sync/src/arbiter.rs b/rtic-sync/src/arbiter.rs
new file mode 100644
index 0000000..d50b1ea
--- /dev/null
+++ b/rtic-sync/src/arbiter.rs
@@ -0,0 +1,194 @@
+//! Crate
+
+use core::cell::UnsafeCell;
+use core::future::poll_fn;
+use core::ops::{Deref, DerefMut};
+use core::pin::Pin;
+use core::sync::atomic::{fence, AtomicBool, Ordering};
+use core::task::{Poll, Waker};
+
+use rtic_common::dropper::OnDrop;
+use rtic_common::wait_queue::{Link, WaitQueue};
+
+/// This is needed to make the async closure in `send` accept that we "share"
+/// the link possible between threads.
+#[derive(Clone)]
+struct LinkPtr(*mut Option<Link<Waker>>);
+
+impl LinkPtr {
+    /// This will dereference the pointer stored within and give out an `&mut`.
+    unsafe fn get(&mut self) -> &mut Option<Link<Waker>> {
+        &mut *self.0
+    }
+}
+
+unsafe impl Send for LinkPtr {}
+unsafe impl Sync for LinkPtr {}
+
+/// An FIFO waitqueue for use in shared bus usecases.
+pub struct Arbiter<T> {
+    wait_queue: WaitQueue,
+    inner: UnsafeCell<T>,
+    taken: AtomicBool,
+}
+
+unsafe impl<T> Send for Arbiter<T> {}
+unsafe impl<T> Sync for Arbiter<T> {}
+
+impl<T> Arbiter<T> {
+    /// Create a new arbiter.
+    pub const fn new(inner: T) -> Self {
+        Self {
+            wait_queue: WaitQueue::new(),
+            inner: UnsafeCell::new(inner),
+            taken: AtomicBool::new(false),
+        }
+    }
+
+    /// Get access to the inner value in the `Arbiter`. This will wait until access is granted,
+    /// for non-blocking access use `try_access`.
+    pub async fn access(&self) -> ExclusiveAccess<'_, T> {
+        let mut link_ptr: Option<Link<Waker>> = None;
+
+        // Make this future `Drop`-safe.
+        // SAFETY(link_ptr): Shadow the original definition of `link_ptr` so we can't abuse it.
+        let mut link_ptr = LinkPtr(&mut link_ptr as *mut Option<Link<Waker>>);
+
+        let mut link_ptr2 = link_ptr.clone();
+        let dropper = OnDrop::new(|| {
+            // SAFETY: We only run this closure and dereference the pointer if we have
+            // exited the `poll_fn` below in the `drop(dropper)` call. The other dereference
+            // of this pointer is in the `poll_fn`.
+            if let Some(link) = unsafe { link_ptr2.get() } {
+                link.remove_from_list(&self.wait_queue);
+            }
+        });
+
+        poll_fn(|cx| {
+            critical_section::with(|_| {
+                fence(Ordering::SeqCst);
+
+                // The queue is empty and noone has taken the value.
+                if self.wait_queue.is_empty() && !self.taken.load(Ordering::Relaxed) {
+                    self.taken.store(true, Ordering::Relaxed);
+
+                    return Poll::Ready(());
+                }
+
+                // SAFETY: This pointer is only dereferenced here and on drop of the future
+                // which happens outside this `poll_fn`'s stack frame.
+                let link = unsafe { link_ptr.get() };
+                if let Some(link) = link {
+                    if link.is_popped() {
+                        return Poll::Ready(());
+                    }
+                } else {
+                    // Place the link in the wait queue on first run.
+                    let link_ref = link.insert(Link::new(cx.waker().clone()));
+
+                    // SAFETY(new_unchecked): The address to the link is stable as it is defined
+                    // outside this stack frame.
+                    // SAFETY(push): `link_ref` lifetime comes from `link_ptr` that is shadowed,
+                    // and  we make sure in `dropper` that the link is removed from the queue
+                    // before dropping `link_ptr` AND `dropper` makes sure that the shadowed
+                    // `link_ptr` lives until the end of the stack frame.
+                    unsafe { self.wait_queue.push(Pin::new_unchecked(link_ref)) };
+                }
+
+                Poll::Pending
+            })
+        })
+        .await;
+
+        // Make sure the link is removed from the queue.
+        drop(dropper);
+
+        // SAFETY: One only gets here if there is exlusive access.
+        ExclusiveAccess {
+            arbiter: self,
+            inner: unsafe { &mut *self.inner.get() },
+        }
+    }
+
+    /// Non-blockingly tries to access the underlying value.
+    /// If someone is in queue to get it, this will return `None`.
+    pub fn try_access(&self) -> Option<ExclusiveAccess<'_, T>> {
+        critical_section::with(|_| {
+            fence(Ordering::SeqCst);
+
+            // The queue is empty and noone has taken the value.
+            if self.wait_queue.is_empty() && !self.taken.load(Ordering::Relaxed) {
+                self.taken.store(true, Ordering::Relaxed);
+
+                // SAFETY: One only gets here if there is exlusive access.
+                Some(ExclusiveAccess {
+                    arbiter: self,
+                    inner: unsafe { &mut *self.inner.get() },
+                })
+            } else {
+                None
+            }
+        })
+    }
+}
+
+/// This token represents exclusive access to the value protected by the `Arbiter`.
+pub struct ExclusiveAccess<'a, T> {
+    arbiter: &'a Arbiter<T>,
+    inner: &'a mut T,
+}
+
+impl<'a, T> Drop for ExclusiveAccess<'a, T> {
+    fn drop(&mut self) {
+        critical_section::with(|_| {
+            fence(Ordering::SeqCst);
+
+            if self.arbiter.wait_queue.is_empty() {
+                // If noone is in queue and we release exclusive access, reset `taken`.
+                self.arbiter.taken.store(false, Ordering::Relaxed);
+            } else if let Some(next) = self.arbiter.wait_queue.pop() {
+                // Wake the next one in queue.
+                next.wake();
+            }
+        })
+    }
+}
+
+impl<'a, T> Deref for ExclusiveAccess<'a, T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        self.inner
+    }
+}
+
+impl<'a, T> DerefMut for ExclusiveAccess<'a, T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        self.inner
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[tokio::test]
+    async fn stress_channel() {
+        const NUM_RUNS: usize = 100_000;
+
+        static ARB: Arbiter<usize> = Arbiter::new(0);
+        let mut v = std::vec::Vec::new();
+
+        for _ in 0..NUM_RUNS {
+            v.push(tokio::spawn(async move {
+                *ARB.access().await += 1;
+            }));
+        }
+
+        for v in v {
+            v.await.unwrap();
+        }
+
+        assert_eq!(*ARB.access().await, NUM_RUNS)
+    }
+}
diff --git a/rtic-sync/src/channel.rs b/rtic-sync/src/channel.rs
new file mode 100644
index 0000000..8a817c8
--- /dev/null
+++ b/rtic-sync/src/channel.rs
@@ -0,0 +1,573 @@
+//! Crate
+
+use core::{
+    cell::UnsafeCell,
+    future::poll_fn,
+    mem::MaybeUninit,
+    pin::Pin,
+    ptr,
+    sync::atomic::{fence, Ordering},
+    task::{Poll, Waker},
+};
+use heapless::Deque;
+use rtic_common::waker_registration::CriticalSectionWakerRegistration as WakerRegistration;
+use rtic_common::{
+    dropper::OnDrop,
+    wait_queue::{Link, WaitQueue},
+};
+
+/// An MPSC channel for use in no-alloc systems. `N` sets the size of the queue.
+///
+/// This channel uses critical sections, however there are extremely small and all `memcpy`
+/// operations of `T` are done without critical sections.
+pub struct Channel<T, const N: usize> {
+    // Here are all indexes that are not used in `slots` and ready to be allocated.
+    freeq: UnsafeCell<Deque<u8, N>>,
+    // Here are wakers and indexes to slots that are ready to be dequeued by the receiver.
+    readyq: UnsafeCell<Deque<u8, N>>,
+    // Waker for the receiver.
+    receiver_waker: WakerRegistration,
+    // Storage for N `T`s, so we don't memcpy around a lot of `T`s.
+    slots: [UnsafeCell<MaybeUninit<T>>; N],
+    // If there is no room in the queue a `Sender`s can wait for there to be place in the queue.
+    wait_queue: WaitQueue,
+    // Keep track of the receiver.
+    receiver_dropped: UnsafeCell<bool>,
+    // Keep track of the number of senders.
+    num_senders: UnsafeCell<usize>,
+}
+
+unsafe impl<T, const N: usize> Send for Channel<T, N> {}
+unsafe impl<T, const N: usize> Sync for Channel<T, N> {}
+
+struct UnsafeAccess<'a, const N: usize> {
+    freeq: &'a mut Deque<u8, N>,
+    readyq: &'a mut Deque<u8, N>,
+    receiver_dropped: &'a mut bool,
+    num_senders: &'a mut usize,
+}
+
+impl<T, const N: usize> Channel<T, N> {
+    const _CHECK: () = assert!(N < 256, "This queue support a maximum of 255 entries");
+
+    const INIT_SLOTS: UnsafeCell<MaybeUninit<T>> = UnsafeCell::new(MaybeUninit::uninit());
+
+    /// Create a new channel.
+    pub const fn new() -> Self {
+        Self {
+            freeq: UnsafeCell::new(Deque::new()),
+            readyq: UnsafeCell::new(Deque::new()),
+            receiver_waker: WakerRegistration::new(),
+            slots: [Self::INIT_SLOTS; N],
+            wait_queue: WaitQueue::new(),
+            receiver_dropped: UnsafeCell::new(false),
+            num_senders: UnsafeCell::new(0),
+        }
+    }
+
+    /// Split the queue into a `Sender`/`Receiver` pair.
+    pub fn split(&mut self) -> (Sender<'_, T, N>, Receiver<'_, T, N>) {
+        // Fill free queue
+        for idx in 0..N as u8 {
+            debug_assert!(!self.freeq.get_mut().is_full());
+
+            // SAFETY: This safe as the loop goes from 0 to the capacity of the underlying queue.
+            unsafe {
+                self.freeq.get_mut().push_back_unchecked(idx);
+            }
+        }
+
+        debug_assert!(self.freeq.get_mut().is_full());
+
+        // There is now 1 sender
+        *self.num_senders.get_mut() = 1;
+
+        (Sender(self), Receiver(self))
+    }
+
+    fn access<'a>(&'a self, _cs: critical_section::CriticalSection) -> UnsafeAccess<'a, N> {
+        // SAFETY: This is safe as are in a critical section.
+        unsafe {
+            UnsafeAccess {
+                freeq: &mut *self.freeq.get(),
+                readyq: &mut *self.readyq.get(),
+                receiver_dropped: &mut *self.receiver_dropped.get(),
+                num_senders: &mut *self.num_senders.get(),
+            }
+        }
+    }
+}
+
+/// Creates a split channel with `'static` lifetime.
+#[macro_export]
+macro_rules! make_channel {
+    ($type:path, $size:expr) => {{
+        static mut CHANNEL: Channel<$type, $size> = Channel::new();
+
+        // SAFETY: This is safe as we hide the static mut from others to access it.
+        // Only this point is where the mutable access happens.
+        unsafe { CHANNEL.split() }
+    }};
+}
+
+// -------- Sender
+
+/// Error state for when the receiver has been dropped.
+pub struct NoReceiver<T>(pub T);
+
+/// Errors that 'try_send` can have.
+pub enum TrySendError<T> {
+    /// Error state for when the receiver has been dropped.
+    NoReceiver(T),
+    /// Error state when the queue is full.
+    Full(T),
+}
+
+impl<T> core::fmt::Debug for NoReceiver<T>
+where
+    T: core::fmt::Debug,
+{
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        write!(f, "NoReceiver({:?})", self.0)
+    }
+}
+
+impl<T> core::fmt::Debug for TrySendError<T>
+where
+    T: core::fmt::Debug,
+{
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        match self {
+            TrySendError::NoReceiver(v) => write!(f, "NoReceiver({v:?})"),
+            TrySendError::Full(v) => write!(f, "Full({v:?})"),
+        }
+    }
+}
+
+impl<T> PartialEq for TrySendError<T>
+where
+    T: PartialEq,
+{
+    fn eq(&self, other: &Self) -> bool {
+        match (self, other) {
+            (TrySendError::NoReceiver(v1), TrySendError::NoReceiver(v2)) => v1.eq(v2),
+            (TrySendError::NoReceiver(_), TrySendError::Full(_)) => false,
+            (TrySendError::Full(_), TrySendError::NoReceiver(_)) => false,
+            (TrySendError::Full(v1), TrySendError::Full(v2)) => v1.eq(v2),
+        }
+    }
+}
+
+/// A `Sender` can send to the channel and can be cloned.
+pub struct Sender<'a, T, const N: usize>(&'a Channel<T, N>);
+
+unsafe impl<'a, T, const N: usize> Send for Sender<'a, T, N> {}
+
+/// This is needed to make the async closure in `send` accept that we "share"
+/// the link possible between threads.
+#[derive(Clone)]
+struct LinkPtr(*mut Option<Link<Waker>>);
+
+impl LinkPtr {
+    /// This will dereference the pointer stored within and give out an `&mut`.
+    unsafe fn get(&mut self) -> &mut Option<Link<Waker>> {
+        &mut *self.0
+    }
+}
+
+unsafe impl Send for LinkPtr {}
+unsafe impl Sync for LinkPtr {}
+
+impl<'a, T, const N: usize> core::fmt::Debug for Sender<'a, T, N> {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        write!(f, "Sender")
+    }
+}
+
+impl<'a, T, const N: usize> Sender<'a, T, N> {
+    #[inline(always)]
+    fn send_footer(&mut self, idx: u8, val: T) {
+        // Write the value to the slots, note; this memcpy is not under a critical section.
+        unsafe {
+            ptr::write(
+                self.0.slots.get_unchecked(idx as usize).get() as *mut T,
+                val,
+            )
+        }
+
+        // Write the value into the ready queue.
+        critical_section::with(|cs| {
+            debug_assert!(!self.0.access(cs).readyq.is_full());
+            unsafe { self.0.access(cs).readyq.push_back_unchecked(idx) }
+        });
+
+        fence(Ordering::SeqCst);
+
+        // If there is a receiver waker, wake it.
+        self.0.receiver_waker.wake();
+    }
+
+    /// Try to send a value, non-blocking. If the channel is full this will return an error.
+    pub fn try_send(&mut self, val: T) -> Result<(), TrySendError<T>> {
+        // If the wait queue is not empty, we can't try to push into the queue.
+        if !self.0.wait_queue.is_empty() {
+            return Err(TrySendError::Full(val));
+        }
+
+        // No receiver available.
+        if self.is_closed() {
+            return Err(TrySendError::NoReceiver(val));
+        }
+
+        let idx =
+            if let Some(idx) = critical_section::with(|cs| self.0.access(cs).freeq.pop_front()) {
+                idx
+            } else {
+                return Err(TrySendError::Full(val));
+            };
+
+        self.send_footer(idx, val);
+
+        Ok(())
+    }
+
+    /// Send a value. If there is no place left in the queue this will wait until there is.
+    /// If the receiver does not exist this will return an error.
+    pub async fn send(&mut self, val: T) -> Result<(), NoReceiver<T>> {
+        let mut link_ptr: Option<Link<Waker>> = None;
+
+        // Make this future `Drop`-safe.
+        // SAFETY(link_ptr): Shadow the original definition of `link_ptr` so we can't abuse it.
+        let mut link_ptr = LinkPtr(&mut link_ptr as *mut Option<Link<Waker>>);
+
+        let mut link_ptr2 = link_ptr.clone();
+        let dropper = OnDrop::new(|| {
+            // SAFETY: We only run this closure and dereference the pointer if we have
+            // exited the `poll_fn` below in the `drop(dropper)` call. The other dereference
+            // of this pointer is in the `poll_fn`.
+            if let Some(link) = unsafe { link_ptr2.get() } {
+                link.remove_from_list(&self.0.wait_queue);
+            }
+        });
+
+        let idx = poll_fn(|cx| {
+            if self.is_closed() {
+                return Poll::Ready(Err(()));
+            }
+
+            //  Do all this in one critical section, else there can be race conditions
+            let queue_idx = critical_section::with(|cs| {
+                let wq_empty = self.0.wait_queue.is_empty();
+                let fq_empty = self.0.access(cs).freeq.is_empty();
+                if !wq_empty || fq_empty {
+                    // SAFETY: This pointer is only dereferenced here and on drop of the future
+                    // which happens outside this `poll_fn`'s stack frame.
+                    let link = unsafe { link_ptr.get() };
+                    if let Some(link) = link {
+                        if !link.is_popped() {
+                            return None;
+                        } else {
+                            // Fall through to dequeue
+                        }
+                    } else {
+                        // Place the link in the wait queue on first run.
+                        let link_ref = link.insert(Link::new(cx.waker().clone()));
+
+                        // SAFETY(new_unchecked): The address to the link is stable as it is defined
+                        // outside this stack frame.
+                        // SAFETY(push): `link_ref` lifetime comes from `link_ptr` that is shadowed,
+                        // and  we make sure in `dropper` that the link is removed from the queue
+                        // before dropping `link_ptr` AND `dropper` makes sure that the shadowed
+                        // `link_ptr` lives until the end of the stack frame.
+                        unsafe { self.0.wait_queue.push(Pin::new_unchecked(link_ref)) };
+
+                        return None;
+                    }
+                }
+
+                debug_assert!(!self.0.access(cs).freeq.is_empty());
+                // Get index as the queue is guaranteed not empty and the wait queue is empty
+                let idx = unsafe { self.0.access(cs).freeq.pop_front_unchecked() };
+
+                Some(idx)
+            });
+
+            if let Some(idx) = queue_idx {
+                // Return the index
+                Poll::Ready(Ok(idx))
+            } else {
+                Poll::Pending
+            }
+        })
+        .await;
+
+        // Make sure the link is removed from the queue.
+        drop(dropper);
+
+        if let Ok(idx) = idx {
+            self.send_footer(idx, val);
+
+            Ok(())
+        } else {
+            Err(NoReceiver(val))
+        }
+    }
+
+    /// Returns true if there is no `Receiver`s.
+    pub fn is_closed(&self) -> bool {
+        critical_section::with(|cs| *self.0.access(cs).receiver_dropped)
+    }
+
+    /// Is the queue full.
+    pub fn is_full(&self) -> bool {
+        critical_section::with(|cs| self.0.access(cs).freeq.is_empty())
+    }
+
+    /// Is the queue empty.
+    pub fn is_empty(&self) -> bool {
+        critical_section::with(|cs| self.0.access(cs).freeq.is_full())
+    }
+}
+
+impl<'a, T, const N: usize> Drop for Sender<'a, T, N> {
+    fn drop(&mut self) {
+        // Count down the reference counter
+        let num_senders = critical_section::with(|cs| {
+            *self.0.access(cs).num_senders -= 1;
+
+            *self.0.access(cs).num_senders
+        });
+
+        // If there are no senders, wake the receiver to do error handling.
+        if num_senders == 0 {
+            self.0.receiver_waker.wake();
+        }
+    }
+}
+
+impl<'a, T, const N: usize> Clone for Sender<'a, T, N> {
+    fn clone(&self) -> Self {
+        // Count up the reference counter
+        critical_section::with(|cs| *self.0.access(cs).num_senders += 1);
+
+        Self(self.0)
+    }
+}
+
+// -------- Receiver
+
+/// A receiver of the channel. There can only be one receiver at any time.
+pub struct Receiver<'a, T, const N: usize>(&'a Channel<T, N>);
+
+unsafe impl<'a, T, const N: usize> Send for Receiver<'a, T, N> {}
+
+impl<'a, T, const N: usize> core::fmt::Debug for Receiver<'a, T, N> {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        write!(f, "Receiver")
+    }
+}
+
+/// Possible receive errors.
+#[derive(Debug, PartialEq, Eq)]
+pub enum ReceiveError {
+    /// Error state for when all senders has been dropped.
+    NoSender,
+    /// Error state for when the queue is empty.
+    Empty,
+}
+
+impl<'a, T, const N: usize> Receiver<'a, T, N> {
+    /// Receives a value if there is one in the channel, non-blocking.
+    pub fn try_recv(&mut self) -> Result<T, ReceiveError> {
+        // Try to get a ready slot.
+        let ready_slot = critical_section::with(|cs| self.0.access(cs).readyq.pop_front());
+
+        if let Some(rs) = ready_slot {
+            // Read the value from the slots, note; this memcpy is not under a critical section.
+            let r = unsafe { ptr::read(self.0.slots.get_unchecked(rs as usize).get() as *const T) };
+
+            // Return the index to the free queue after we've read the value.
+            critical_section::with(|cs| {
+                debug_assert!(!self.0.access(cs).freeq.is_full());
+                unsafe { self.0.access(cs).freeq.push_back_unchecked(rs) }
+            });
+
+            fence(Ordering::SeqCst);
+
+            // If someone is waiting in the WaiterQueue, wake the first one up.
+            if let Some(wait_head) = self.0.wait_queue.pop() {
+                wait_head.wake();
+            }
+
+            Ok(r)
+        } else if self.is_closed() {
+            Err(ReceiveError::NoSender)
+        } else {
+            Err(ReceiveError::Empty)
+        }
+    }
+
+    /// Receives a value, waiting if the queue is empty.
+    /// If all senders are dropped this will error with `NoSender`.
+    pub async fn recv(&mut self) -> Result<T, ReceiveError> {
+        // There was nothing in the queue, setup the waiting.
+        poll_fn(|cx| {
+            // Register waker.
+            // TODO: Should it happen here or after the if? This might cause a spurious wake.
+            self.0.receiver_waker.register(cx.waker());
+
+            // Try to dequeue.
+            match self.try_recv() {
+                Ok(val) => {
+                    return Poll::Ready(Ok(val));
+                }
+                Err(ReceiveError::NoSender) => {
+                    return Poll::Ready(Err(ReceiveError::NoSender));
+                }
+                _ => {}
+            }
+
+            Poll::Pending
+        })
+        .await
+    }
+
+    /// Returns true if there are no `Sender`s.
+    pub fn is_closed(&self) -> bool {
+        critical_section::with(|cs| *self.0.access(cs).num_senders == 0)
+    }
+
+    /// Is the queue full.
+    pub fn is_full(&self) -> bool {
+        critical_section::with(|cs| self.0.access(cs).readyq.is_full())
+    }
+
+    /// Is the queue empty.
+    pub fn is_empty(&self) -> bool {
+        critical_section::with(|cs| self.0.access(cs).readyq.is_empty())
+    }
+}
+
+impl<'a, T, const N: usize> Drop for Receiver<'a, T, N> {
+    fn drop(&mut self) {
+        // Mark the receiver as dropped and wake all waiters
+        critical_section::with(|cs| *self.0.access(cs).receiver_dropped = true);
+
+        while let Some(waker) = self.0.wait_queue.pop() {
+            waker.wake();
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn empty() {
+        let (mut s, mut r) = make_channel!(u32, 10);
+
+        assert!(s.is_empty());
+        assert!(r.is_empty());
+
+        s.try_send(1).unwrap();
+
+        assert!(!s.is_empty());
+        assert!(!r.is_empty());
+
+        r.try_recv().unwrap();
+
+        assert!(s.is_empty());
+        assert!(r.is_empty());
+    }
+
+    #[test]
+    fn full() {
+        let (mut s, mut r) = make_channel!(u32, 3);
+
+        for _ in 0..3 {
+            assert!(!s.is_full());
+            assert!(!r.is_full());
+
+            s.try_send(1).unwrap();
+        }
+
+        assert!(s.is_full());
+        assert!(r.is_full());
+
+        for _ in 0..3 {
+            r.try_recv().unwrap();
+
+            assert!(!s.is_full());
+            assert!(!r.is_full());
+        }
+    }
+
+    #[test]
+    fn send_recieve() {
+        let (mut s, mut r) = make_channel!(u32, 10);
+
+        for i in 0..10 {
+            s.try_send(i).unwrap();
+        }
+
+        assert_eq!(s.try_send(11), Err(TrySendError::Full(11)));
+
+        for i in 0..10 {
+            assert_eq!(r.try_recv().unwrap(), i);
+        }
+
+        assert_eq!(r.try_recv(), Err(ReceiveError::Empty));
+    }
+
+    #[test]
+    fn closed_recv() {
+        let (s, mut r) = make_channel!(u32, 10);
+
+        drop(s);
+
+        assert!(r.is_closed());
+
+        assert_eq!(r.try_recv(), Err(ReceiveError::NoSender));
+    }
+
+    #[test]
+    fn closed_sender() {
+        let (mut s, r) = make_channel!(u32, 10);
+
+        drop(r);
+
+        assert!(s.is_closed());
+
+        assert_eq!(s.try_send(11), Err(TrySendError::NoReceiver(11)));
+    }
+
+    #[tokio::test]
+    async fn stress_channel() {
+        const NUM_RUNS: usize = 1_000;
+        const QUEUE_SIZE: usize = 10;
+
+        let (s, mut r) = make_channel!(u32, QUEUE_SIZE);
+        let mut v = std::vec::Vec::new();
+
+        for i in 0..NUM_RUNS {
+            let mut s = s.clone();
+
+            v.push(tokio::spawn(async move {
+                s.send(i as _).await.unwrap();
+            }));
+        }
+
+        let mut map = std::collections::BTreeSet::new();
+
+        for _ in 0..NUM_RUNS {
+            map.insert(r.recv().await.unwrap());
+        }
+
+        assert_eq!(map.len(), NUM_RUNS);
+
+        for v in v {
+            v.await.unwrap();
+        }
+    }
+}
diff --git a/rtic-sync/src/lib.rs b/rtic-sync/src/lib.rs
new file mode 100644
index 0000000..ac06220
--- /dev/null
+++ b/rtic-sync/src/lib.rs
@@ -0,0 +1,12 @@
+//! Synchronization primitives for asynchronous contexts.
+
+#![no_std]
+#![deny(missing_docs)]
+//deny_warnings_placeholder_for_ci
+
+pub mod arbiter;
+pub mod channel;
+
+#[cfg(test)]
+#[macro_use]
+extern crate std;