diff --git a/cyw43/src/ioctl.rs b/cyw43/src/ioctl.rs
index f8b2d9aba..af8bb695b 100644
--- a/cyw43/src/ioctl.rs
+++ b/cyw43/src/ioctl.rs
@@ -1,5 +1,5 @@
 use core::cell::{Cell, RefCell};
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::task::{Poll, Waker};
 
 use embassy_sync::waitqueue::WakerRegistration;
@@ -71,7 +71,7 @@ impl IoctlState {
         self.wakers.borrow_mut().runner.register(waker);
     }
 
-    pub async fn wait_complete(&self) -> usize {
+    pub fn wait_complete(&self) -> impl Future<Output = usize> + '_ {
         poll_fn(|cx| {
             if let IoctlStateInner::Done { resp_len } = self.state.get() {
                 Poll::Ready(resp_len)
@@ -80,22 +80,18 @@ impl IoctlState {
                 Poll::Pending
             }
         })
-        .await
     }
 
-    pub async fn wait_pending(&self) -> PendingIoctl {
-        let pending = poll_fn(|cx| {
+    pub fn wait_pending(&self) -> impl Future<Output = PendingIoctl> + '_ {
+        poll_fn(|cx| {
             if let IoctlStateInner::Pending(pending) = self.state.get() {
+                self.state.set(IoctlStateInner::Sent { buf: pending.buf });
                 Poll::Ready(pending)
             } else {
                 self.register_runner(cx.waker());
                 Poll::Pending
             }
         })
-        .await;
-
-        self.state.set(IoctlStateInner::Sent { buf: pending.buf });
-        pending
     }
 
     pub fn cancel_ioctl(&self) {
diff --git a/embassy-executor/src/spawner.rs b/embassy-executor/src/spawner.rs
index 41320d4c3..ce24589bf 100644
--- a/embassy-executor/src/spawner.rs
+++ b/embassy-executor/src/spawner.rs
@@ -1,4 +1,4 @@
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::mem;
 use core::sync::atomic::Ordering;
@@ -100,7 +100,7 @@ impl Spawner {
     /// # Panics
     ///
     /// Panics if the current executor is not an Embassy executor.
-    pub async fn for_current_executor() -> Self {
+    pub fn for_current_executor() -> impl Future<Output = Self> {
         poll_fn(|cx| {
             let task = raw::task_from_waker(cx.waker());
             let executor = unsafe {
@@ -113,7 +113,6 @@ impl Spawner {
             let executor = unsafe { raw::Executor::wrap(executor) };
             Poll::Ready(Self::new(executor))
         })
-        .await
     }
 
     /// Spawn a task into an executor.
@@ -178,7 +177,7 @@ impl SendSpawner {
     /// # Panics
     ///
     /// Panics if the current executor is not an Embassy executor.
-    pub async fn for_current_executor() -> Self {
+    pub fn for_current_executor() -> impl Future<Output = Self> {
         poll_fn(|cx| {
             let task = raw::task_from_waker(cx.waker());
             let executor = unsafe {
@@ -190,7 +189,6 @@ impl SendSpawner {
             };
             Poll::Ready(Self::new(executor))
         })
-        .await
     }
 
     /// Spawn a task into an executor.
diff --git a/embassy-net-esp-hosted/src/ioctl.rs b/embassy-net-esp-hosted/src/ioctl.rs
index e2a6815aa..512023206 100644
--- a/embassy-net-esp-hosted/src/ioctl.rs
+++ b/embassy-net-esp-hosted/src/ioctl.rs
@@ -1,5 +1,5 @@
 use core::cell::RefCell;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::task::Poll;
 
 use embassy_sync::waitqueue::WakerRegistration;
@@ -38,7 +38,7 @@ impl Shared {
         }))
     }
 
-    pub async fn ioctl_wait_complete(&self) -> usize {
+    pub fn ioctl_wait_complete(&self) -> impl Future<Output = usize> + '_ {
         poll_fn(|cx| {
             let mut this = self.0.borrow_mut();
             if let IoctlState::Done { resp_len } = this.ioctl {
@@ -48,7 +48,6 @@ impl Shared {
                 Poll::Pending
             }
         })
-        .await
     }
 
     pub async fn ioctl_wait_pending(&self) -> PendingIoctl {
@@ -108,7 +107,7 @@ impl Shared {
         this.control_waker.wake();
     }
 
-    pub async fn init_wait(&self) {
+    pub fn init_wait(&self) -> impl Future<Output = ()> + '_ {
         poll_fn(|cx| {
             let mut this = self.0.borrow_mut();
             if this.is_init {
@@ -118,6 +117,5 @@ impl Shared {
                 Poll::Pending
             }
         })
-        .await
     }
 }
diff --git a/embassy-net-nrf91/src/lib.rs b/embassy-net-nrf91/src/lib.rs
index 3abe2c766..61fcaea1f 100644
--- a/embassy-net-nrf91/src/lib.rs
+++ b/embassy-net-nrf91/src/lib.rs
@@ -9,7 +9,7 @@ mod fmt;
 pub mod context;
 
 use core::cell::RefCell;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::mem::{self, MaybeUninit};
 use core::ptr::{self, addr_of, addr_of_mut, copy_nonoverlapping};
@@ -737,7 +737,7 @@ pub struct Control<'a> {
 
 impl<'a> Control<'a> {
     /// Wait for modem IPC to be initialized.
-    pub async fn wait_init(&self) {
+    pub fn wait_init(&self) -> impl Future<Output = ()> + '_ {
         poll_fn(|cx| {
             let mut state = self.state.borrow_mut();
             if state.init {
@@ -746,7 +746,6 @@ impl<'a> Control<'a> {
             state.init_waker.register(cx.waker());
             Poll::Pending
         })
-        .await
     }
 
     async fn request(&self, msg: &mut Message, req_data: &[u8], resp_data: &mut [u8]) -> usize {
diff --git a/embassy-net/src/raw.rs b/embassy-net/src/raw.rs
index a88bcc458..c9f753f13 100644
--- a/embassy-net/src/raw.rs
+++ b/embassy-net/src/raw.rs
@@ -1,6 +1,6 @@
 //! Raw sockets.
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::mem;
 use core::task::{Context, Poll};
 
@@ -66,8 +66,8 @@ impl<'a> RawSocket<'a> {
     ///
     /// A socket is readable when a packet has been received, or when there are queued packets in
     /// the buffer.
-    pub async fn wait_recv_ready(&self) {
-        poll_fn(move |cx| self.poll_recv_ready(cx)).await
+    pub fn wait_recv_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.poll_recv_ready(cx))
     }
 
     /// Receive a datagram.
@@ -115,8 +115,8 @@ impl<'a> RawSocket<'a> {
     ///
     /// A socket becomes writable when there is space in the buffer, from initial memory or after
     /// dispatching datagrams on a full buffer.
-    pub async fn wait_send_ready(&self) {
-        poll_fn(move |cx| self.poll_send_ready(cx)).await
+    pub fn wait_send_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.poll_send_ready(cx))
     }
 
     /// Wait until a datagram can be sent.
@@ -141,8 +141,8 @@ impl<'a> RawSocket<'a> {
     /// Send a datagram.
     ///
     /// This method will wait until the datagram has been sent.`
-    pub async fn send(&self, buf: &[u8]) {
-        poll_fn(move |cx| self.poll_send(buf, cx)).await
+    pub fn send<'s>(&'s self, buf: &'s [u8]) -> impl Future<Output = ()> + 's {
+        poll_fn(|cx| self.poll_send(buf, cx))
     }
 
     /// Send a datagram.
@@ -165,8 +165,8 @@ impl<'a> RawSocket<'a> {
     /// Flush the socket.
     ///
     /// This method will wait until the socket is flushed.
-    pub async fn flush(&mut self) {
-        poll_fn(move |cx| {
+    pub fn flush(&mut self) -> impl Future<Output = ()> + '_ {
+        poll_fn(|cx| {
             self.with_mut(|s, _| {
                 if s.send_queue() == 0 {
                     Poll::Ready(())
@@ -176,7 +176,6 @@ impl<'a> RawSocket<'a> {
                 }
             })
         })
-        .await
     }
 }
 
diff --git a/embassy-net/src/tcp.rs b/embassy-net/src/tcp.rs
index 32d374064..d0230b581 100644
--- a/embassy-net/src/tcp.rs
+++ b/embassy-net/src/tcp.rs
@@ -8,7 +8,7 @@
 //! Incoming connections when no socket is listening are rejected. To accept many incoming
 //! connections, create many sockets and put them all into listening mode.
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::mem;
 use core::task::{Context, Poll};
 
@@ -79,8 +79,8 @@ impl<'a> TcpReader<'a> {
     /// (see [`may_recv()`](TcpSocket::may_recv)), and there is some pending data in the receive buffer.
     ///
     /// This is the equivalent of [read](#method.read), without buffering any data.
-    pub async fn wait_read_ready(&self) {
-        poll_fn(move |cx| self.io.poll_read_ready(cx)).await
+    pub fn wait_read_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.io.poll_read_ready(cx))
     }
 
     /// Read data from the socket.
@@ -131,24 +131,24 @@ impl<'a> TcpWriter<'a> {
     /// (see [`may_send()`](TcpSocket::may_send)), and the transmit buffer is not full.
     ///
     /// This is the equivalent of [write](#method.write), without sending any data.
-    pub async fn wait_write_ready(&self) {
-        poll_fn(move |cx| self.io.poll_write_ready(cx)).await
+    pub fn wait_write_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.io.poll_write_ready(cx))
     }
 
     /// Write data to the socket.
     ///
     /// Returns how many bytes were written, or an error. If the socket is not ready to
     /// accept data, it waits until it is.
-    pub async fn write(&mut self, buf: &[u8]) -> Result<usize, Error> {
-        self.io.write(buf).await
+    pub fn write<'s>(&'s mut self, buf: &'s [u8]) -> impl Future<Output = Result<usize, Error>> + 's {
+        self.io.write(buf)
     }
 
     /// Flushes the written data to the socket.
     ///
     /// This waits until all data has been sent, and ACKed by the remote host. For a connection
     /// closed with [`abort()`](TcpSocket::abort) it will wait for the TCP RST packet to be sent.
-    pub async fn flush(&mut self) -> Result<(), Error> {
-        self.io.flush().await
+    pub fn flush(&mut self) -> impl Future<Output = Result<(), Error>> + '_ {
+        self.io.flush()
     }
 
     /// Call `f` with the largest contiguous slice of octets in the transmit buffer,
@@ -300,8 +300,8 @@ impl<'a> TcpSocket<'a> {
     /// (see [may_recv](#method.may_recv)), and there is some pending data in the receive buffer.
     ///
     /// This is the equivalent of [read](#method.read), without buffering any data.
-    pub async fn wait_read_ready(&self) {
-        poll_fn(move |cx| self.io.poll_read_ready(cx)).await
+    pub fn wait_read_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.io.poll_read_ready(cx))
     }
 
     /// Read data from the socket.
@@ -311,8 +311,8 @@ impl<'a> TcpSocket<'a> {
     ///
     /// A return value of Ok(0) means that the socket was closed and is longer
     /// able to receive any data.
-    pub async fn read(&mut self, buf: &mut [u8]) -> Result<usize, Error> {
-        self.io.read(buf).await
+    pub fn read<'s>(&'s mut self, buf: &'s mut [u8]) -> impl Future<Output = Result<usize, Error>> + 's {
+        self.io.read(buf)
     }
 
     /// Wait until the socket becomes writable.
@@ -321,24 +321,24 @@ impl<'a> TcpSocket<'a> {
     /// (see [may_send](#method.may_send)), and the transmit buffer is not full.
     ///
     /// This is the equivalent of [write](#method.write), without sending any data.
-    pub async fn wait_write_ready(&self) {
-        poll_fn(move |cx| self.io.poll_write_ready(cx)).await
+    pub fn wait_write_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.io.poll_write_ready(cx))
     }
 
     /// Write data to the socket.
     ///
     /// Returns how many bytes were written, or an error. If the socket is not ready to
     /// accept data, it waits until it is.
-    pub async fn write(&mut self, buf: &[u8]) -> Result<usize, Error> {
-        self.io.write(buf).await
+    pub fn write<'s>(&'s mut self, buf: &'s [u8]) -> impl Future<Output = Result<usize, Error>> + 's {
+        self.io.write(buf)
     }
 
     /// Flushes the written data to the socket.
     ///
     /// This waits until all data has been sent, and ACKed by the remote host. For a connection
     /// closed with [`abort()`](TcpSocket::abort) it will wait for the TCP RST packet to be sent.
-    pub async fn flush(&mut self) -> Result<(), Error> {
-        self.io.flush().await
+    pub fn flush(&mut self) -> impl Future<Output = Result<(), Error>> + '_ {
+        self.io.flush()
     }
 
     /// Set the timeout for the socket.
@@ -501,8 +501,8 @@ impl<'d> TcpIo<'d> {
         })
     }
 
-    async fn read(&mut self, buf: &mut [u8]) -> Result<usize, Error> {
-        poll_fn(move |cx| {
+    fn read<'s>(&'s mut self, buf: &'s mut [u8]) -> impl Future<Output = Result<usize, Error>> + 's {
+        poll_fn(|cx| {
             // CAUTION: smoltcp semantics around EOF are different to what you'd expect
             // from posix-like IO, so we have to tweak things here.
             self.with_mut(|s, _| match s.recv_slice(buf) {
@@ -526,7 +526,6 @@ impl<'d> TcpIo<'d> {
                 Err(tcp::RecvError::InvalidState) => Poll::Ready(Err(Error::ConnectionReset)),
             })
         })
-        .await
     }
 
     fn poll_write_ready(&self, cx: &mut Context<'_>) -> Poll<()> {
@@ -540,8 +539,8 @@ impl<'d> TcpIo<'d> {
         })
     }
 
-    async fn write(&mut self, buf: &[u8]) -> Result<usize, Error> {
-        poll_fn(move |cx| {
+    fn write<'s>(&'s mut self, buf: &'s [u8]) -> impl Future<Output = Result<usize, Error>> + 's {
+        poll_fn(|cx| {
             self.with_mut(|s, _| match s.send_slice(buf) {
                 // Not ready to send (no space in the tx buffer)
                 Ok(0) => {
@@ -554,7 +553,6 @@ impl<'d> TcpIo<'d> {
                 Err(tcp::SendError::InvalidState) => Poll::Ready(Err(Error::ConnectionReset)),
             })
         })
-        .await
     }
 
     async fn write_with<F, R>(&mut self, f: F) -> Result<R, Error>
@@ -615,8 +613,8 @@ impl<'d> TcpIo<'d> {
         .await
     }
 
-    async fn flush(&mut self) -> Result<(), Error> {
-        poll_fn(move |cx| {
+    fn flush(&mut self) -> impl Future<Output = Result<(), Error>> + '_ {
+        poll_fn(|cx| {
             self.with_mut(|s, _| {
                 let data_pending = (s.send_queue() > 0) && s.state() != tcp::State::Closed;
                 let fin_pending = matches!(
@@ -636,7 +634,6 @@ impl<'d> TcpIo<'d> {
                 }
             })
         })
-        .await
     }
 
     fn recv_capacity(&self) -> usize {
diff --git a/embassy-net/src/udp.rs b/embassy-net/src/udp.rs
index 76602edc2..64a22d45b 100644
--- a/embassy-net/src/udp.rs
+++ b/embassy-net/src/udp.rs
@@ -1,6 +1,6 @@
 //! UDP sockets.
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::mem;
 use core::task::{Context, Poll};
 
@@ -107,8 +107,8 @@ impl<'a> UdpSocket<'a> {
     ///
     /// A socket is readable when a packet has been received, or when there are queued packets in
     /// the buffer.
-    pub async fn wait_recv_ready(&self) {
-        poll_fn(move |cx| self.poll_recv_ready(cx)).await
+    pub fn wait_recv_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(move |cx| self.poll_recv_ready(cx))
     }
 
     /// Wait until a datagram can be read.
@@ -134,8 +134,11 @@ impl<'a> UdpSocket<'a> {
     /// This method will wait until a datagram is received.
     ///
     /// Returns the number of bytes received and the remote endpoint.
-    pub async fn recv_from(&self, buf: &mut [u8]) -> Result<(usize, UdpMetadata), RecvError> {
-        poll_fn(move |cx| self.poll_recv_from(buf, cx)).await
+    pub fn recv_from<'s>(
+        &'s self,
+        buf: &'s mut [u8],
+    ) -> impl Future<Output = Result<(usize, UdpMetadata), RecvError>> + 's {
+        poll_fn(|cx| self.poll_recv_from(buf, cx))
     }
 
     /// Receive a datagram.
@@ -194,8 +197,8 @@ impl<'a> UdpSocket<'a> {
     ///
     /// A socket becomes writable when there is space in the buffer, from initial memory or after
     /// dispatching datagrams on a full buffer.
-    pub async fn wait_send_ready(&self) {
-        poll_fn(move |cx| self.poll_send_ready(cx)).await
+    pub fn wait_send_ready(&self) -> impl Future<Output = ()> + '_ {
+        poll_fn(|cx| self.poll_send_ready(cx))
     }
 
     /// Wait until a datagram can be sent.
@@ -297,8 +300,8 @@ impl<'a> UdpSocket<'a> {
     /// Flush the socket.
     ///
     /// This method will wait until the socket is flushed.
-    pub async fn flush(&mut self) {
-        poll_fn(move |cx| {
+    pub fn flush(&mut self) -> impl Future<Output = ()> + '_ {
+        poll_fn(|cx| {
             self.with_mut(|s, _| {
                 if s.send_queue() == 0 {
                     Poll::Ready(())
@@ -308,7 +311,6 @@ impl<'a> UdpSocket<'a> {
                 }
             })
         })
-        .await
     }
 
     /// Returns the local endpoint of the socket.
diff --git a/embassy-nrf/src/buffered_uarte.rs b/embassy-nrf/src/buffered_uarte.rs
index b55e70a36..c3fcfd06e 100644
--- a/embassy-nrf/src/buffered_uarte.rs
+++ b/embassy-nrf/src/buffered_uarte.rs
@@ -9,7 +9,7 @@
 //! Please also see [crate::uarte] to understand when [BufferedUarte] should be used.
 
 use core::cmp::min;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::slice;
 use core::sync::atomic::{compiler_fence, AtomicBool, AtomicU8, AtomicUsize, Ordering};
@@ -452,7 +452,7 @@ impl<'d, U: UarteInstance> BufferedUarteTx<'d, U> {
     }
 
     /// Write a buffer into this writer, returning how many bytes were written.
-    pub async fn write(&mut self, buf: &[u8]) -> Result<usize, Error> {
+    pub fn write<'a>(&'a mut self, buf: &'a [u8]) -> impl Future<Output = Result<usize, Error>> + 'a {
         poll_fn(move |cx| {
             //trace!("poll_write: {:?}", buf.len());
             let ss = U::state();
@@ -477,7 +477,6 @@ impl<'d, U: UarteInstance> BufferedUarteTx<'d, U> {
 
             Poll::Ready(Ok(n))
         })
-        .await
     }
 
     /// Try writing a buffer without waiting, returning how many bytes were written.
@@ -504,7 +503,7 @@ impl<'d, U: UarteInstance> BufferedUarteTx<'d, U> {
     }
 
     /// Flush this output stream, ensuring that all intermediately buffered contents reach their destination.
-    pub async fn flush(&mut self) -> Result<(), Error> {
+    pub fn flush(&mut self) -> impl Future<Output = Result<(), Error>> + '_ {
         poll_fn(move |cx| {
             //trace!("poll_flush");
             let ss = U::state();
@@ -517,7 +516,6 @@ impl<'d, U: UarteInstance> BufferedUarteTx<'d, U> {
 
             Poll::Ready(Ok(()))
         })
-        .await
     }
 }
 
@@ -721,7 +719,7 @@ impl<'d, U: UarteInstance, T: TimerInstance> BufferedUarteRx<'d, U, T> {
     }
 
     /// Return the contents of the internal buffer, filling it with more data from the inner reader if it is empty.
-    pub async fn fill_buf(&mut self) -> Result<&[u8], Error> {
+    pub fn fill_buf(&mut self) -> impl Future<Output = Result<&'_ [u8], Error>> {
         poll_fn(move |cx| {
             compiler_fence(Ordering::SeqCst);
             //trace!("poll_read");
@@ -771,7 +769,6 @@ impl<'d, U: UarteInstance, T: TimerInstance> BufferedUarteRx<'d, U, T> {
             let buf = s.rx_buf.buf.load(Ordering::Relaxed);
             Poll::Ready(Ok(unsafe { slice::from_raw_parts(buf.add(start), n) }))
         })
-        .await
     }
 
     /// Tell this buffer that `amt` bytes have been consumed from the buffer, so they should no longer be returned in calls to `fill_buf`.
diff --git a/embassy-nrf/src/qspi.rs b/embassy-nrf/src/qspi.rs
index 255b43c33..17e127700 100755
--- a/embassy-nrf/src/qspi.rs
+++ b/embassy-nrf/src/qspi.rs
@@ -2,7 +2,7 @@
 
 #![macro_use]
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::ptr;
 use core::task::Poll;
@@ -314,7 +314,7 @@ impl<'d, T: Instance> Qspi<'d, T> {
         Ok(())
     }
 
-    async fn wait_ready(&mut self) {
+    fn wait_ready(&mut self) -> impl Future<Output = ()> {
         poll_fn(move |cx| {
             let r = T::regs();
             let s = T::state();
@@ -324,7 +324,6 @@ impl<'d, T: Instance> Qspi<'d, T> {
             }
             Poll::Pending
         })
-        .await
     }
 
     fn blocking_wait_ready() {
diff --git a/embassy-nrf/src/usb/mod.rs b/embassy-nrf/src/usb/mod.rs
index a9bf16708..06dae694b 100644
--- a/embassy-nrf/src/usb/mod.rs
+++ b/embassy-nrf/src/usb/mod.rs
@@ -4,7 +4,7 @@
 
 pub mod vbus_detect;
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::mem::MaybeUninit;
 use core::sync::atomic::{compiler_fence, AtomicU32, Ordering};
@@ -219,8 +219,8 @@ impl<'d, T: Instance, V: VbusDetect> driver::Bus for Bus<'d, T, V> {
         regs.enable().write(|x| x.set_enable(false));
     }
 
-    async fn poll(&mut self) -> Event {
-        poll_fn(move |cx| {
+    fn poll(&mut self) -> impl Future<Output = Event> {
+        poll_fn(|cx| {
             BUS_WAKER.register(cx.waker());
             let regs = T::regs();
 
@@ -277,7 +277,6 @@ impl<'d, T: Instance, V: VbusDetect> driver::Bus for Bus<'d, T, V> {
 
             Poll::Pending
         })
-        .await
     }
 
     fn endpoint_set_stalled(&mut self, ep_addr: EndpointAddress, stalled: bool) {
@@ -468,7 +467,7 @@ impl<'d, T: Instance, Dir: EndpointDir> driver::Endpoint for Endpoint<'d, T, Dir
 
 #[allow(private_bounds)]
 impl<'d, T: Instance, Dir: EndpointDir> Endpoint<'d, T, Dir> {
-    async fn wait_enabled_state(&mut self, state: bool) {
+    fn wait_enabled_state(&mut self, state: bool) -> impl Future<Output = ()> {
         let i = self.info.addr.index();
         assert!(i != 0);
 
@@ -480,12 +479,11 @@ impl<'d, T: Instance, Dir: EndpointDir> Endpoint<'d, T, Dir> {
                 Poll::Pending
             }
         })
-        .await
     }
 
     /// Wait for the endpoint to be disabled
-    pub async fn wait_disabled(&mut self) {
-        self.wait_enabled_state(false).await
+    pub fn wait_disabled(&mut self) -> impl Future<Output = ()> {
+        self.wait_enabled_state(false)
     }
 }
 
diff --git a/embassy-nrf/src/usb/vbus_detect.rs b/embassy-nrf/src/usb/vbus_detect.rs
index bdc088dcb..8794beb2d 100644
--- a/embassy-nrf/src/usb/vbus_detect.rs
+++ b/embassy-nrf/src/usb/vbus_detect.rs
@@ -1,6 +1,6 @@
 //! Trait and implementations for performing VBUS detection.
 
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::sync::atomic::{AtomicBool, Ordering};
 use core::task::Poll;
 
@@ -99,8 +99,8 @@ impl VbusDetect for HardwareVbusDetect {
         regs.usbregstatus().read().vbusdetect()
     }
 
-    async fn wait_power_ready(&mut self) -> Result<(), ()> {
-        poll_fn(move |cx| {
+    fn wait_power_ready(&mut self) -> impl Future<Output = Result<(), ()>> {
+        poll_fn(|cx| {
             POWER_WAKER.register(cx.waker());
             let regs = USB_REG_PERI;
 
@@ -112,7 +112,6 @@ impl VbusDetect for HardwareVbusDetect {
                 Poll::Pending
             }
         })
-        .await
     }
 }
 
@@ -163,7 +162,7 @@ impl VbusDetect for &SoftwareVbusDetect {
         self.usb_detected.load(Ordering::Relaxed)
     }
 
-    async fn wait_power_ready(&mut self) -> Result<(), ()> {
+    fn wait_power_ready(&mut self) -> impl Future<Output = Result<(), ()>> {
         poll_fn(move |cx| {
             POWER_WAKER.register(cx.waker());
 
@@ -175,6 +174,5 @@ impl VbusDetect for &SoftwareVbusDetect {
                 Poll::Pending
             }
         })
-        .await
     }
 }
diff --git a/embassy-rp/src/adc.rs b/embassy-rp/src/adc.rs
index 9582e43c8..19441f194 100644
--- a/embassy-rp/src/adc.rs
+++ b/embassy-rp/src/adc.rs
@@ -1,5 +1,5 @@
 //! ADC driver.
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::mem;
 use core::sync::atomic::{compiler_fence, Ordering};
@@ -193,18 +193,18 @@ impl<'d> Adc<'d, Async> {
         Self { phantom: PhantomData }
     }
 
-    async fn wait_for_ready() {
+    fn wait_for_ready() -> impl Future<Output = ()> {
         let r = Self::regs();
         r.inte().write(|w| w.set_fifo(true));
         compiler_fence(Ordering::SeqCst);
-        poll_fn(|cx| {
+
+        poll_fn(move |cx| {
             WAKER.register(cx.waker());
             if r.cs().read().ready() {
                 return Poll::Ready(());
             }
             Poll::Pending
         })
-        .await;
     }
 
     /// Sample a value from a channel until completed.
diff --git a/embassy-sync/src/mutex.rs b/embassy-sync/src/mutex.rs
index 08f66e374..f25f74336 100644
--- a/embassy-sync/src/mutex.rs
+++ b/embassy-sync/src/mutex.rs
@@ -2,7 +2,7 @@
 //!
 //! This module provides a mutex that can be used to synchronize data between asynchronous tasks.
 use core::cell::{RefCell, UnsafeCell};
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::ops::{Deref, DerefMut};
 use core::task::Poll;
 use core::{fmt, mem};
@@ -73,7 +73,7 @@ where
     /// Lock the mutex.
     ///
     /// This will wait for the mutex to be unlocked if it's already locked.
-    pub async fn lock(&self) -> MutexGuard<'_, M, T> {
+    pub fn lock(&self) -> impl Future<Output = MutexGuard<'_, M, T>> {
         poll_fn(|cx| {
             let ready = self.state.lock(|s| {
                 let mut s = s.borrow_mut();
@@ -92,7 +92,6 @@ where
                 Poll::Pending
             }
         })
-        .await
     }
 
     /// Attempt to immediately lock the mutex.
diff --git a/embassy-sync/src/once_lock.rs b/embassy-sync/src/once_lock.rs
index 55608ba32..cd05b986d 100644
--- a/embassy-sync/src/once_lock.rs
+++ b/embassy-sync/src/once_lock.rs
@@ -1,7 +1,7 @@
 //! Synchronization primitive for initializing a value once, allowing others to await a reference to the value.
 
 use core::cell::Cell;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::mem::MaybeUninit;
 use core::sync::atomic::{AtomicBool, Ordering};
 use core::task::Poll;
@@ -55,7 +55,7 @@ impl<T> OnceLock<T> {
 
     /// Get a reference to the underlying value, waiting for it to be set.
     /// If the value is already set, this will return immediately.
-    pub async fn get(&self) -> &T {
+    pub fn get(&self) -> impl Future<Output = &T> {
         poll_fn(|cx| match self.try_get() {
             Some(data) => Poll::Ready(data),
             None => {
@@ -63,7 +63,6 @@ impl<T> OnceLock<T> {
                 Poll::Pending
             }
         })
-        .await
     }
 
     /// Try to get a reference to the underlying value if it exists.
diff --git a/embassy-sync/src/watch.rs b/embassy-sync/src/watch.rs
index 404e31714..e76646c0b 100644
--- a/embassy-sync/src/watch.rs
+++ b/embassy-sync/src/watch.rs
@@ -1,7 +1,7 @@
 //! A synchronization primitive for passing the latest value to **multiple** receivers.
 
 use core::cell::RefCell;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::ops::{Deref, DerefMut};
 use core::task::{Context, Poll};
@@ -547,8 +547,8 @@ impl<'a, T: Clone, W: WatchBehavior<T> + ?Sized> Rcv<'a, T, W> {
     /// Returns the current value of the `Watch` once it is initialized, marking it as seen.
     ///
     /// **Note**: Futures do nothing unless you `.await` or poll them.
-    pub async fn get(&mut self) -> T {
-        poll_fn(|cx| self.watch.poll_get(&mut self.at_id, cx)).await
+    pub fn get(&mut self) -> impl Future<Output = T> + '_ {
+        poll_fn(|cx| self.watch.poll_get(&mut self.at_id, cx))
     }
 
     /// Tries to get the current value of the `Watch` without waiting, marking it as seen.
diff --git a/embassy-sync/src/zerocopy_channel.rs b/embassy-sync/src/zerocopy_channel.rs
index fabb69bf6..56433cd8a 100644
--- a/embassy-sync/src/zerocopy_channel.rs
+++ b/embassy-sync/src/zerocopy_channel.rs
@@ -15,7 +15,7 @@
 //! another message will result in an error being returned.
 
 use core::cell::RefCell;
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::task::{Context, Poll};
 
@@ -131,12 +131,15 @@ impl<'a, M: RawMutex, T> Sender<'a, M, T> {
     }
 
     /// Asynchronously send a value over the channel.
-    pub async fn send(&mut self) -> &mut T {
-        let i = poll_fn(|cx| {
+    pub fn send(&mut self) -> impl Future<Output = &mut T> {
+        poll_fn(|cx| {
             self.channel.state.lock(|s| {
                 let s = &mut *s.borrow_mut();
                 match s.push_index() {
-                    Some(i) => Poll::Ready(i),
+                    Some(i) => {
+                        let r = unsafe { &mut *self.channel.buf.add(i) };
+                        Poll::Ready(r)
+                    }
                     None => {
                         s.receive_waker.register(cx.waker());
                         Poll::Pending
@@ -144,8 +147,6 @@ impl<'a, M: RawMutex, T> Sender<'a, M, T> {
                 }
             })
         })
-        .await;
-        unsafe { &mut *self.channel.buf.add(i) }
     }
 
     /// Notify the channel that the sending of the value has been finalized.
@@ -213,12 +214,15 @@ impl<'a, M: RawMutex, T> Receiver<'a, M, T> {
     }
 
     /// Asynchronously receive a value over the channel.
-    pub async fn receive(&mut self) -> &mut T {
-        let i = poll_fn(|cx| {
+    pub fn receive(&mut self) -> impl Future<Output = &mut T> {
+        poll_fn(|cx| {
             self.channel.state.lock(|s| {
                 let s = &mut *s.borrow_mut();
                 match s.pop_index() {
-                    Some(i) => Poll::Ready(i),
+                    Some(i) => {
+                        let r = unsafe { &mut *self.channel.buf.add(i) };
+                        Poll::Ready(r)
+                    }
                     None => {
                         s.send_waker.register(cx.waker());
                         Poll::Pending
@@ -226,8 +230,6 @@ impl<'a, M: RawMutex, T> Receiver<'a, M, T> {
                 }
             })
         })
-        .await;
-        unsafe { &mut *self.channel.buf.add(i) }
     }
 
     /// Notify the channel that the receiving of the value has been finalized.
diff --git a/embassy-usb/src/class/cdc_acm.rs b/embassy-usb/src/class/cdc_acm.rs
index 2823e522e..c5b1a56fe 100644
--- a/embassy-usb/src/class/cdc_acm.rs
+++ b/embassy-usb/src/class/cdc_acm.rs
@@ -1,7 +1,7 @@
 //! CDC-ACM class implementation, aka Serial over USB.
 
 use core::cell::{Cell, RefCell};
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::mem::{self, MaybeUninit};
 use core::sync::atomic::{AtomicBool, Ordering};
 use core::task::Poll;
@@ -108,7 +108,7 @@ impl Default for ControlShared {
 }
 
 impl ControlShared {
-    async fn changed(&self) {
+    fn changed(&self) -> impl Future<Output = ()> + '_ {
         poll_fn(|cx| {
             if self.changed.load(Ordering::Relaxed) {
                 self.changed.store(false, Ordering::Relaxed);
@@ -118,7 +118,6 @@ impl ControlShared {
                 Poll::Pending
             }
         })
-        .await;
     }
 }
 
diff --git a/embassy-usb/src/class/uac1/speaker.rs b/embassy-usb/src/class/uac1/speaker.rs
index 6c3a4e378..25de25d9c 100644
--- a/embassy-usb/src/class/uac1/speaker.rs
+++ b/embassy-usb/src/class/uac1/speaker.rs
@@ -11,7 +11,7 @@
 //! The class provides volume and mute controls for each channel.
 
 use core::cell::{Cell, RefCell};
-use core::future::poll_fn;
+use core::future::{poll_fn, Future};
 use core::marker::PhantomData;
 use core::sync::atomic::{AtomicBool, AtomicU32, Ordering};
 use core::task::Poll;
@@ -389,7 +389,7 @@ impl<'d> Default for SharedControl<'d> {
 }
 
 impl<'d> SharedControl<'d> {
-    async fn changed(&self) {
+    fn changed(&self) -> impl Future<Output = ()> + '_ {
         poll_fn(|context| {
             if self.changed.load(Ordering::Relaxed) {
                 self.changed.store(false, Ordering::Relaxed);
@@ -399,7 +399,6 @@ impl<'d> SharedControl<'d> {
                 Poll::Pending
             }
         })
-        .await;
     }
 }