65 files changed, 1741 insertions, 996 deletions
diff --git a/docs/modules/ROOT/examples/basic/Cargo.toml b/docs/modules/ROOT/examples/basic/Cargo.toml
index d9f8a285a..e3e446e63 100644
--- a/docs/modules/ROOT/examples/basic/Cargo.toml
+++ b/docs/modules/ROOT/examples/basic/Cargo.toml
@@ -6,7 +6,7 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 [dependencies]
-embassy-executor = { version = "0.1.0", path = "../../../../../embassy-executor", features = ["defmt", "nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../../embassy-executor", features = ["defmt", "nightly", "integrated-timers", "arch-cortex-m", "executor-thread"] }
 embassy-time = { version = "0.1.0", path = "../../../../../embassy-time", features = ["defmt", "nightly"] }
 embassy-nrf = { version = "0.1.0", path = "../../../../../embassy-nrf", features = ["defmt", "nrf52840", "time-driver-rtc1", "gpiote", "nightly"] }
diff --git a/docs/modules/ROOT/examples/layer-by-layer/blinky-async/Cargo.toml b/docs/modules/ROOT/examples/layer-by-layer/blinky-async/Cargo.toml
index c9a963d4d..a11a7e0ba 100644
--- a/docs/modules/ROOT/examples/layer-by-layer/blinky-async/Cargo.toml
+++ b/docs/modules/ROOT/examples/layer-by-layer/blinky-async/Cargo.toml
@@ -8,7 +8,7 @@ license = "MIT OR Apache-2.0"
 cortex-m = "0.7"
 cortex-m-rt = "0.7"
 embassy-stm32 = { version = "0.1.0", features = ["stm32l475vg", "memory-x", "exti"], default-features = false  }
-embassy-executor = { version = "0.1.0", default-features = false, features = ["nightly"] }
+embassy-executor = { version = "0.1.0", default-features = false, features = ["nightly", "arch-cortex-m", "executor-thread"] }
 defmt = "0.3.0"
 defmt-rtt = "0.3.0"
diff --git a/embassy-boot/boot/src/boot_loader.rs b/embassy-boot/boot/src/boot_loader.rs
index e2e361e3c..9d047f778 100644
--- a/embassy-boot/boot/src/boot_loader.rs
+++ b/embassy-boot/boot/src/boot_loader.rs
@@ -31,7 +31,7 @@ where
 }
 /// Extension of the embedded-storage flash type information with block size and erase value.
-pub trait Flash: NorFlash + ReadNorFlash {
+pub trait Flash: NorFlash {
    /// The block size that should be used when writing to flash. For most builtin flashes, this is the same as the erase
    /// size of the flash, but for external QSPI flash modules, this can be lower.
    const BLOCK_SIZE: usize;
@@ -60,9 +60,11 @@ pub trait FlashConfig {
 /// different page sizes and flash write sizes.
 pub struct BootLoader {
    // Page with current state of bootloader. The state partition has the following format:
-    // | Range          | Description                                                                      |
+    // All ranges are in multiples of WRITE_SIZE bytes.
-    // | 0 - WRITE_SIZE | Magic indicating bootloader state. BOOT_MAGIC means boot, SWAP_MAGIC means swap. |
+    // | Range    | Description                                                                      |
-    // | WRITE_SIZE - N | Progress index used while swapping or reverting                                  |
+    // | 0..1     | Magic indicating bootloader state. BOOT_MAGIC means boot, SWAP_MAGIC means swap. |
+    // | 1..2     | Progress validity. ERASE_VALUE means valid, !ERASE_VALUE means invalid.          |
+    // | 2..2 + N | Progress index used while swapping or reverting                                  |
    state: Partition,
    // Location of the partition which will be booted from
    active: Partition,
@@ -192,12 +194,17 @@ impl BootLoader {
                trace!("Reverting");
                self.revert(p, magic, page)?;
-                // Overwrite magic and reset progress
                let state_flash = p.state();
+                // Invalidate progress
                magic.fill(!P::STATE::ERASE_VALUE);
-                self.state.write_blocking(state_flash, 0, magic)?;
+                self.state
+                    .write_blocking(state_flash, P::STATE::WRITE_SIZE as u32, magic)?;
+                // Clear magic and progress
                self.state.wipe_blocking(state_flash)?;
+                // Set magic
                magic.fill(BOOT_MAGIC);
                self.state.write_blocking(state_flash, 0, magic)?;
            }
@@ -215,28 +222,34 @@ impl BootLoader {
    fn current_progress<P: FlashConfig>(&mut self, config: &mut P, aligned: &mut [u8]) -> Result<usize, BootError> {
        let write_size = aligned.len();
-        let max_index = ((self.state.len() - write_size) / write_size) - 1;
+        let max_index = ((self.state.len() - write_size) / write_size) - 2;
        aligned.fill(!P::STATE::ERASE_VALUE);
        let state_flash = config.state();
-        for i in 0..max_index {
+        self.state
+            .read_blocking(state_flash, P::STATE::WRITE_SIZE as u32, aligned)?;
+        if aligned.iter().any(|&b| b != P::STATE::ERASE_VALUE) {
+            // Progress is invalid
+            return Ok(max_index);
+        }
+        for index in 0..max_index {
            self.state
-                .read_blocking(state_flash, (write_size + i * write_size) as u32, aligned)?;
+                .read_blocking(state_flash, (2 + index) as u32 * P::STATE::WRITE_SIZE as u32, aligned)?;
            if aligned.iter().any(|&b| b == P::STATE::ERASE_VALUE) {
-                return Ok(i);
+                return Ok(index);
            }
        }
        Ok(max_index)
    }
-    fn update_progress<P: FlashConfig>(&mut self, idx: usize, p: &mut P, magic: &mut [u8]) -> Result<(), BootError> {
+    fn update_progress<P: FlashConfig>(&mut self, index: usize, p: &mut P, magic: &mut [u8]) -> Result<(), BootError> {
-        let write_size = magic.len();
        let aligned = magic;
        aligned.fill(!P::STATE::ERASE_VALUE);
        self.state
-            .write_blocking(p.state(), (write_size + idx * write_size) as u32, aligned)?;
+            .write_blocking(p.state(), (2 + index) as u32 * P::STATE::WRITE_SIZE as u32, aligned)?;
        Ok(())
    }
@@ -360,7 +373,7 @@ fn assert_partitions(active: Partition, dfu: Partition, state: Partition, page_s
    assert_eq!(active.len() % page_size, 0);
    assert_eq!(dfu.len() % page_size, 0);
    assert!(dfu.len() - active.len() >= page_size);
-    assert!(2 * (active.len() / page_size) <= (state.len() - write_size) / write_size);
+    assert!(2 + 2 * (active.len() / page_size) <= state.len() / write_size);
 }
 /// A flash wrapper implementing the Flash and embedded_storage traits.
diff --git a/embassy-boot/boot/src/firmware_updater.rs b/embassy-boot/boot/src/firmware_updater.rs
index 2d1b26980..819e20201 100644
--- a/embassy-boot/boot/src/firmware_updater.rs
+++ b/embassy-boot/boot/src/firmware_updater.rs
@@ -234,11 +234,24 @@ impl FirmwareUpdater {
        self.state.read(state_flash, 0, aligned).await?;
        if aligned.iter().any(|&b| b != magic) {
-            aligned.fill(0);
+            // Read progress validity
+            self.state.read(state_flash, F::WRITE_SIZE as u32, aligned).await?;
+            // FIXME: Do not make this assumption.
+            const STATE_ERASE_VALUE: u8 = 0xFF;
+            if aligned.iter().any(|&b| b != STATE_ERASE_VALUE) {
+                // The current progress validity marker is invalid
+            } else {
+                // Invalidate progress
+                aligned.fill(!STATE_ERASE_VALUE);
+                self.state.write(state_flash, F::WRITE_SIZE as u32, aligned).await?;
+            }
-            self.state.write(state_flash, 0, aligned).await?;
+            // Clear magic and progress
            self.state.wipe(state_flash).await?;
+            // Set magic
            aligned.fill(magic);
            self.state.write(state_flash, 0, aligned).await?;
        }
@@ -441,11 +454,24 @@ impl FirmwareUpdater {
        self.state.read_blocking(state_flash, 0, aligned)?;
        if aligned.iter().any(|&b| b != magic) {
-            aligned.fill(0);
+            // Read progress validity
+            self.state.read_blocking(state_flash, F::WRITE_SIZE as u32, aligned)?;
+            // FIXME: Do not make this assumption.
+            const STATE_ERASE_VALUE: u8 = 0xFF;
+            if aligned.iter().any(|&b| b != STATE_ERASE_VALUE) {
+                // The current progress validity marker is invalid
+            } else {
+                // Invalidate progress
+                aligned.fill(!STATE_ERASE_VALUE);
+                self.state.write_blocking(state_flash, F::WRITE_SIZE as u32, aligned)?;
+            }
-            self.state.write_blocking(state_flash, 0, aligned)?;
+            // Clear magic and progress
            self.state.wipe_blocking(state_flash)?;
+            // Set magic
            aligned.fill(magic);
            self.state.write_blocking(state_flash, 0, aligned)?;
        }
diff --git a/embassy-boot/boot/src/lib.rs b/embassy-boot/boot/src/lib.rs
index da9055476..d4078f1cb 100644
--- a/embassy-boot/boot/src/lib.rs
+++ b/embassy-boot/boot/src/lib.rs
@@ -8,6 +8,7 @@ mod fmt;
 mod boot_loader;
 mod digest_adapters;
 mod firmware_updater;
+mod mem_flash;
 mod partition;
 pub use boot_loader::{BootError, BootFlash, BootLoader, Flash, FlashConfig, MultiFlashConfig, SingleFlashConfig};
@@ -45,13 +46,10 @@ impl<const N: usize> AsMut<[u8]> for AlignedBuffer<N> {
 #[cfg(test)]
 mod tests {
-    use core::convert::Infallible;
-    use embedded_storage::nor_flash::{ErrorType, NorFlash, ReadNorFlash};
-    use embedded_storage_async::nor_flash::{NorFlash as AsyncNorFlash, ReadNorFlash as AsyncReadNorFlash};
    use futures::executor::block_on;
    use super::*;
+    use crate::mem_flash::MemFlash;
    /*
    #[test]
@@ -74,8 +72,8 @@ mod tests {
        const ACTIVE: Partition = Partition::new(4096, 61440);
        const DFU: Partition = Partition::new(61440, 122880);
-        let mut flash = MemFlash::<131072, 4096, 4>([0xff; 131072]);
+        let mut flash = MemFlash::<131072, 4096, 4>::default();
-        flash.0[0..4].copy_from_slice(&[BOOT_MAGIC; 4]);
+        flash.mem[0..4].copy_from_slice(&[BOOT_MAGIC; 4]);
        let mut flash = SingleFlashConfig::new(&mut flash);
        let mut bootloader: BootLoader = BootLoader::new(ACTIVE, DFU, STATE);
@@ -94,14 +92,14 @@ mod tests {
        const STATE: Partition = Partition::new(0, 4096);
        const ACTIVE: Partition = Partition::new(4096, 61440);
        const DFU: Partition = Partition::new(61440, 122880);
-        let mut flash = MemFlash::<131072, 4096, 4>([0xff; 131072]);
+        let mut flash = MemFlash::<131072, 4096, 4>::random();
        let original: [u8; ACTIVE.len()] = [rand::random::<u8>(); ACTIVE.len()];
        let update: [u8; DFU.len()] = [rand::random::<u8>(); DFU.len()];
        let mut aligned = [0; 4];
        for i in ACTIVE.from..ACTIVE.to {
-            flash.0[i] = original[i - ACTIVE.from];
+            flash.mem[i] = original[i - ACTIVE.from];
        }
        let mut bootloader: BootLoader = BootLoader::new(ACTIVE, DFU, STATE);
@@ -123,12 +121,12 @@ mod tests {
        );
        for i in ACTIVE.from..ACTIVE.to {
-            assert_eq!(flash.0[i], update[i - ACTIVE.from], "Index {}", i);
+            assert_eq!(flash.mem[i], update[i - ACTIVE.from], "Index {}", i);
        }
        // First DFU page is untouched
        for i in DFU.from + 4096..DFU.to {
-            assert_eq!(flash.0[i], original[i - DFU.from - 4096], "Index {}", i);
+            assert_eq!(flash.mem[i], original[i - DFU.from - 4096], "Index {}", i);
        }
        // Running again should cause a revert
@@ -140,12 +138,12 @@ mod tests {
        );
        for i in ACTIVE.from..ACTIVE.to {
-            assert_eq!(flash.0[i], original[i - ACTIVE.from], "Index {}", i);
+            assert_eq!(flash.mem[i], original[i - ACTIVE.from], "Index {}", i);
        }
        // Last page is untouched
        for i in DFU.from..DFU.to - 4096 {
-            assert_eq!(flash.0[i], update[i - DFU.from], "Index {}", i);
+            assert_eq!(flash.mem[i], update[i - DFU.from], "Index {}", i);
        }
        // Mark as booted
@@ -165,16 +163,16 @@ mod tests {
        const ACTIVE: Partition = Partition::new(4096, 16384);
        const DFU: Partition = Partition::new(0, 16384);
-        let mut active = MemFlash::<16384, 4096, 8>([0xff; 16384]);
+        let mut active = MemFlash::<16384, 4096, 8>::random();
-        let mut dfu = MemFlash::<16384, 2048, 8>([0xff; 16384]);
+        let mut dfu = MemFlash::<16384, 2048, 8>::random();
-        let mut state = MemFlash::<4096, 128, 4>([0xff; 4096]);
+        let mut state = MemFlash::<4096, 128, 4>::random();
        let mut aligned = [0; 4];
        let original: [u8; ACTIVE.len()] = [rand::random::<u8>(); ACTIVE.len()];
        let update: [u8; DFU.len()] = [rand::random::<u8>(); DFU.len()];
        for i in ACTIVE.from..ACTIVE.to {
-            active.0[i] = original[i - ACTIVE.from];
+            active.mem[i] = original[i - ACTIVE.from];
        }
        let mut updater = FirmwareUpdater::new(DFU, STATE);
@@ -202,12 +200,12 @@ mod tests {
        );
        for i in ACTIVE.from..ACTIVE.to {
-            assert_eq!(active.0[i], update[i - ACTIVE.from], "Index {}", i);
+            assert_eq!(active.mem[i], update[i - ACTIVE.from], "Index {}", i);
        }
        // First DFU page is untouched
        for i in DFU.from + 4096..DFU.to {
-            assert_eq!(dfu.0[i], original[i - DFU.from - 4096], "Index {}", i);
+            assert_eq!(dfu.mem[i], original[i - DFU.from - 4096], "Index {}", i);
        }
    }
@@ -219,15 +217,15 @@ mod tests {
        const DFU: Partition = Partition::new(0, 16384);
        let mut aligned = [0; 4];
-        let mut active = MemFlash::<16384, 2048, 4>([0xff; 16384]);
+        let mut active = MemFlash::<16384, 2048, 4>::random();
-        let mut dfu = MemFlash::<16384, 4096, 8>([0xff; 16384]);
+        let mut dfu = MemFlash::<16384, 4096, 8>::random();
-        let mut state = MemFlash::<4096, 128, 4>([0xff; 4096]);
+        let mut state = MemFlash::<4096, 128, 4>::random();
        let original: [u8; ACTIVE.len()] = [rand::random::<u8>(); ACTIVE.len()];
        let update: [u8; DFU.len()] = [rand::random::<u8>(); DFU.len()];
        for i in ACTIVE.from..ACTIVE.to {
-            active.0[i] = original[i - ACTIVE.from];
+            active.mem[i] = original[i - ACTIVE.from];
        }
        let mut updater = FirmwareUpdater::new(DFU, STATE);
@@ -254,12 +252,12 @@ mod tests {
        );
        for i in ACTIVE.from..ACTIVE.to {
-            assert_eq!(active.0[i], update[i - ACTIVE.from], "Index {}", i);
+            assert_eq!(active.mem[i], update[i - ACTIVE.from], "Index {}", i);
        }
        // First DFU page is untouched
        for i in DFU.from + 4096..DFU.to {
-            assert_eq!(dfu.0[i], original[i - DFU.from - 4096], "Index {}", i);
+            assert_eq!(dfu.mem[i], original[i - DFU.from - 4096], "Index {}", i);
        }
    }
@@ -289,13 +287,13 @@ mod tests {
        const STATE: Partition = Partition::new(0, 4096);
        const DFU: Partition = Partition::new(4096, 8192);
-        let mut flash = MemFlash::<8192, 4096, 4>([0xff; 8192]);
+        let mut flash = MemFlash::<8192, 4096, 4>::default();
        let firmware_len = firmware.len();
        let mut write_buf = [0; 4096];
        write_buf[0..firmware_len].copy_from_slice(firmware);
-        NorFlash::write(&mut flash, DFU.from as u32, &write_buf).unwrap();
+        DFU.write_blocking(&mut flash, 0, &write_buf).unwrap();
        // On with the test
@@ -312,113 +310,4 @@ mod tests {
        ))
        .is_ok());
    }
-    pub struct MemFlash<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize>(pub [u8; SIZE]);
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> NorFlash
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        const WRITE_SIZE: usize = WRITE_SIZE;
-        const ERASE_SIZE: usize = ERASE_SIZE;
-        fn erase(&mut self, from: u32, to: u32) -> Result<(), Self::Error> {
-            let from = from as usize;
-            let to = to as usize;
-            assert!(from % ERASE_SIZE == 0);
-            assert!(to % ERASE_SIZE == 0, "To: {}, erase size: {}", to, ERASE_SIZE);
-            for i in from..to {
-                self.0[i] = 0xFF;
-            }
-            Ok(())
-        }
-        fn write(&mut self, offset: u32, data: &[u8]) -> Result<(), Self::Error> {
-            assert!(data.len() % WRITE_SIZE == 0);
-            assert!(offset as usize % WRITE_SIZE == 0);
-            assert!(offset as usize + data.len() <= SIZE);
-            self.0[offset as usize..offset as usize + data.len()].copy_from_slice(data);
-            Ok(())
-        }
-    }
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> ErrorType
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        type Error = Infallible;
-    }
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> ReadNorFlash
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        const READ_SIZE: usize = 1;
-        fn read(&mut self, offset: u32, buf: &mut [u8]) -> Result<(), Self::Error> {
-            let len = buf.len();
-            buf[..].copy_from_slice(&self.0[offset as usize..offset as usize + len]);
-            Ok(())
-        }
-        fn capacity(&self) -> usize {
-            SIZE
-        }
-    }
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> super::Flash
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        const BLOCK_SIZE: usize = ERASE_SIZE;
-        const ERASE_VALUE: u8 = 0xFF;
-    }
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> AsyncReadNorFlash
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        const READ_SIZE: usize = 1;
-        async fn read(&mut self, offset: u32, buf: &mut [u8]) -> Result<(), Self::Error> {
-            let len = buf.len();
-            buf[..].copy_from_slice(&self.0[offset as usize..offset as usize + len]);
-            Ok(())
-        }
-        fn capacity(&self) -> usize {
-            SIZE
-        }
-    }
-    impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> AsyncNorFlash
-        for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
-    {
-        const WRITE_SIZE: usize = WRITE_SIZE;
-        const ERASE_SIZE: usize = ERASE_SIZE;
-        async fn erase(&mut self, from: u32, to: u32) -> Result<(), Self::Error> {
-            let from = from as usize;
-            let to = to as usize;
-            assert!(from % ERASE_SIZE == 0);
-            assert!(to % ERASE_SIZE == 0);
-            for i in from..to {
-                self.0[i] = 0xFF;
-            }
-            Ok(())
-        }
-        async fn write(&mut self, offset: u32, data: &[u8]) -> Result<(), Self::Error> {
-            info!("Writing {} bytes to 0x{:x}", data.len(), offset);
-            assert!(data.len() % WRITE_SIZE == 0);
-            assert!(offset as usize % WRITE_SIZE == 0);
-            assert!(
-                offset as usize + data.len() <= SIZE,
-                "OFFSET: {}, LEN: {}, FLASH SIZE: {}",
-                offset,
-                data.len(),
-                SIZE
-            );
-            self.0[offset as usize..offset as usize + data.len()].copy_from_slice(data);
-            Ok(())
-        }
-    }
 }
diff --git a/embassy-boot/boot/src/mem_flash.rs b/embassy-boot/boot/src/mem_flash.rs
new file mode 100644
index 000000000..828aad9d9
--- /dev/null
+++ b/embassy-boot/boot/src/mem_flash.rs
@@ -0,0 +1,156 @@
+#![allow(unused)]
+use core::ops::{Bound, Range, RangeBounds};
+use embedded_storage::nor_flash::{ErrorType, NorFlash, NorFlashError, NorFlashErrorKind, ReadNorFlash};
+use embedded_storage_async::nor_flash::{NorFlash as AsyncNorFlash, ReadNorFlash as AsyncReadNorFlash};
+use crate::Flash;
+pub struct MemFlash<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> {
+    pub mem: [u8; SIZE],
+    pub pending_write_successes: Option<usize>,
+}
+#[derive(Debug)]
+pub struct MemFlashError;
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE> {
+    pub const fn new(fill: u8) -> Self {
+        Self {
+            mem: [fill; SIZE],
+            pending_write_successes: None,
+        }
+    }
+    #[cfg(test)]
+    pub fn random() -> Self {
+        let mut mem = [0; SIZE];
+        for byte in mem.iter_mut() {
+            *byte = rand::random::<u8>();
+        }
+        Self {
+            mem,
+            pending_write_successes: None,
+        }
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> Default
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    fn default() -> Self {
+        Self::new(0xFF)
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> Flash
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    const BLOCK_SIZE: usize = ERASE_SIZE;
+    const ERASE_VALUE: u8 = 0xFF;
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> ErrorType
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    type Error = MemFlashError;
+}
+impl NorFlashError for MemFlashError {
+    fn kind(&self) -> NorFlashErrorKind {
+        NorFlashErrorKind::Other
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> ReadNorFlash
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    const READ_SIZE: usize = 1;
+    fn read(&mut self, offset: u32, bytes: &mut [u8]) -> Result<(), Self::Error> {
+        let len = bytes.len();
+        bytes.copy_from_slice(&self.mem[offset as usize..offset as usize + len]);
+        Ok(())
+    }
+    fn capacity(&self) -> usize {
+        SIZE
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> NorFlash
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    const WRITE_SIZE: usize = WRITE_SIZE;
+    const ERASE_SIZE: usize = ERASE_SIZE;
+    fn erase(&mut self, from: u32, to: u32) -> Result<(), Self::Error> {
+        let from = from as usize;
+        let to = to as usize;
+        assert!(from % ERASE_SIZE == 0);
+        assert!(to % ERASE_SIZE == 0, "To: {}, erase size: {}", to, ERASE_SIZE);
+        for i in from..to {
+            self.mem[i] = 0xFF;
+        }
+        Ok(())
+    }
+    fn write(&mut self, offset: u32, bytes: &[u8]) -> Result<(), Self::Error> {
+        let offset = offset as usize;
+        assert!(bytes.len() % WRITE_SIZE == 0);
+        assert!(offset % WRITE_SIZE == 0);
+        assert!(offset + bytes.len() <= SIZE);
+        if let Some(pending_successes) = self.pending_write_successes {
+            if pending_successes > 0 {
+                self.pending_write_successes = Some(pending_successes - 1);
+            } else {
+                return Err(MemFlashError);
+            }
+        }
+        for ((offset, mem_byte), new_byte) in self
+            .mem
+            .iter_mut()
+            .enumerate()
+            .skip(offset)
+            .take(bytes.len())
+            .zip(bytes)
+        {
+            assert_eq!(0xFF, *mem_byte, "Offset {} is not erased", offset);
+            *mem_byte = *new_byte;
+        }
+        Ok(())
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> AsyncReadNorFlash
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    const READ_SIZE: usize = 1;
+    async fn read(&mut self, offset: u32, bytes: &mut [u8]) -> Result<(), Self::Error> {
+        <Self as ReadNorFlash>::read(self, offset, bytes)
+    }
+    fn capacity(&self) -> usize {
+        <Self as ReadNorFlash>::capacity(self)
+    }
+}
+impl<const SIZE: usize, const ERASE_SIZE: usize, const WRITE_SIZE: usize> AsyncNorFlash
+    for MemFlash<SIZE, ERASE_SIZE, WRITE_SIZE>
+{
+    const WRITE_SIZE: usize = WRITE_SIZE;
+    const ERASE_SIZE: usize = ERASE_SIZE;
+    async fn erase(&mut self, from: u32, to: u32) -> Result<(), Self::Error> {
+        <Self as NorFlash>::erase(self, from, to)
+    }
+    async fn write(&mut self, offset: u32, bytes: &[u8]) -> Result<(), Self::Error> {
+        <Self as NorFlash>::write(self, offset, bytes)
+    }
+}
diff --git a/embassy-boot/boot/src/partition.rs b/embassy-boot/boot/src/partition.rs
index 3ccd4dd76..ac6b0ed0f 100644
--- a/embassy-boot/boot/src/partition.rs
+++ b/embassy-boot/boot/src/partition.rs
@@ -24,7 +24,7 @@ impl Partition {
    }
    /// Read from the partition on the provided flash
-    pub(crate) async fn read<F: AsyncReadNorFlash>(
+    pub async fn read<F: AsyncReadNorFlash>(
        &self,
        flash: &mut F,
        offset: u32,
@@ -35,12 +35,7 @@ impl Partition {
    }
    /// Write to the partition on the provided flash
-    pub(crate) async fn write<F: AsyncNorFlash>(
+    pub async fn write<F: AsyncNorFlash>(&self, flash: &mut F, offset: u32, bytes: &[u8]) -> Result<(), F::Error> {
-        &self,
-        flash: &mut F,
-        offset: u32,
-        bytes: &[u8],
-    ) -> Result<(), F::Error> {
        let offset = self.from as u32 + offset;
        flash.write(offset, bytes).await?;
        trace!("Wrote from 0x{:x} len {}", offset, bytes.len());
@@ -48,7 +43,7 @@ impl Partition {
    }
    /// Erase part of the partition on the provided flash
-    pub(crate) async fn erase<F: AsyncNorFlash>(&self, flash: &mut F, from: u32, to: u32) -> Result<(), F::Error> {
+    pub async fn erase<F: AsyncNorFlash>(&self, flash: &mut F, from: u32, to: u32) -> Result<(), F::Error> {
        let from = self.from as u32 + from;
        let to = self.from as u32 + to;
        flash.erase(from, to).await?;
@@ -66,18 +61,13 @@ impl Partition {
    }
    /// Read from the partition on the provided flash
-    pub(crate) fn read_blocking<F: ReadNorFlash>(
+    pub fn read_blocking<F: ReadNorFlash>(&self, flash: &mut F, offset: u32, bytes: &mut [u8]) -> Result<(), F::Error> {
-        &self,
-        flash: &mut F,
-        offset: u32,
-        bytes: &mut [u8],
-    ) -> Result<(), F::Error> {
        let offset = self.from as u32 + offset;
        flash.read(offset, bytes)
    }
    /// Write to the partition on the provided flash
-    pub(crate) fn write_blocking<F: NorFlash>(&self, flash: &mut F, offset: u32, bytes: &[u8]) -> Result<(), F::Error> {
+    pub fn write_blocking<F: NorFlash>(&self, flash: &mut F, offset: u32, bytes: &[u8]) -> Result<(), F::Error> {
        let offset = self.from as u32 + offset;
        flash.write(offset, bytes)?;
        trace!("Wrote from 0x{:x} len {}", offset, bytes.len());
@@ -85,7 +75,7 @@ impl Partition {
    }
    /// Erase part of the partition on the provided flash
-    pub(crate) fn erase_blocking<F: NorFlash>(&self, flash: &mut F, from: u32, to: u32) -> Result<(), F::Error> {
+    pub fn erase_blocking<F: NorFlash>(&self, flash: &mut F, from: u32, to: u32) -> Result<(), F::Error> {
        let from = self.from as u32 + from;
        let to = self.from as u32 + to;
        flash.erase(from, to)?;
@@ -105,45 +95,45 @@ impl Partition {
 #[cfg(test)]
 mod tests {
-    use crate::tests::MemFlash;
+    use crate::mem_flash::MemFlash;
    use crate::Partition;
    #[test]
    fn can_erase() {
-        let mut flash = MemFlash::<1024, 64, 4>([0x00; 1024]);
+        let mut flash = MemFlash::<1024, 64, 4>::new(0x00);
        let partition = Partition::new(256, 512);
        partition.erase_blocking(&mut flash, 64, 192).unwrap();
-        for (index, byte) in flash.0.iter().copied().enumerate().take(256 + 64) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().take(256 + 64) {
            assert_eq!(0x00, byte, "Index {}", index);
        }
-        for (index, byte) in flash.0.iter().copied().enumerate().skip(256 + 64).take(128) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().skip(256 + 64).take(128) {
            assert_eq!(0xFF, byte, "Index {}", index);
        }
-        for (index, byte) in flash.0.iter().copied().enumerate().skip(256 + 64 + 128) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().skip(256 + 64 + 128) {
            assert_eq!(0x00, byte, "Index {}", index);
        }
    }
    #[test]
    fn can_wipe() {
-        let mut flash = MemFlash::<1024, 64, 4>([0x00; 1024]);
+        let mut flash = MemFlash::<1024, 64, 4>::new(0x00);
        let partition = Partition::new(256, 512);
        partition.wipe_blocking(&mut flash).unwrap();
-        for (index, byte) in flash.0.iter().copied().enumerate().take(256) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().take(256) {
            assert_eq!(0x00, byte, "Index {}", index);
        }
-        for (index, byte) in flash.0.iter().copied().enumerate().skip(256).take(256) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().skip(256).take(256) {
            assert_eq!(0xFF, byte, "Index {}", index);
        }
-        for (index, byte) in flash.0.iter().copied().enumerate().skip(512) {
+        for (index, byte) in flash.mem.iter().copied().enumerate().skip(512) {
            assert_eq!(0x00, byte, "Index {}", index);
        }
    }
diff --git a/embassy-cortex-m/src/executor.rs b/embassy-cortex-m/src/executor.rs
deleted file mode 100644
index 558539e73..000000000
--- a/embassy-cortex-m/src/executor.rs
+++ /dev/null
@@ -1,116 +0,0 @@
-//! Executor specific to cortex-m devices.
-use core::cell::UnsafeCell;
-use core::mem::MaybeUninit;
-use atomic_polyfill::{AtomicBool, Ordering};
-use cortex_m::interrupt::InterruptNumber;
-use cortex_m::peripheral::NVIC;
-pub use embassy_executor::*;
-#[derive(Clone, Copy)]
-struct N(u16);
-unsafe impl cortex_m::interrupt::InterruptNumber for N {
-    fn number(self) -> u16 {
-        self.0
-    }
-}
-fn pend_by_number(n: u16) {
-    cortex_m::peripheral::NVIC::pend(N(n))
-}
-/// Interrupt mode executor.
-///
-/// This executor runs tasks in interrupt mode. The interrupt handler is set up
-/// to poll tasks, and when a task is woken the interrupt is pended from software.
-///
-/// This allows running async tasks at a priority higher than thread mode. One
-/// use case is to leave thread mode free for non-async tasks. Another use case is
-/// to run multiple executors: one in thread mode for low priority tasks and another in
-/// interrupt mode for higher priority tasks. Higher priority tasks will preempt lower
-/// priority ones.
-///
-/// It is even possible to run multiple interrupt mode executors at different priorities,
-/// by assigning different priorities to the interrupts. For an example on how to do this,
-/// See the 'multiprio' example for 'embassy-nrf'.
-///
-/// To use it, you have to pick an interrupt that won't be used by the hardware.
-/// Some chips reserve some interrupts for this purpose, sometimes named "software interrupts" (SWI).
-/// If this is not the case, you may use an interrupt from any unused peripheral.
-///
-/// It is somewhat more complex to use, it's recommended to use the thread-mode
-/// [`Executor`] instead, if it works for your use case.
-pub struct InterruptExecutor {
-    started: AtomicBool,
-    executor: UnsafeCell<MaybeUninit<raw::Executor>>,
-}
-unsafe impl Send for InterruptExecutor {}
-unsafe impl Sync for InterruptExecutor {}
-impl InterruptExecutor {
-    /// Create a new, not started `InterruptExecutor`.
-    #[inline]
-    pub const fn new() -> Self {
-        Self {
-            started: AtomicBool::new(false),
-            executor: UnsafeCell::new(MaybeUninit::uninit()),
-        }
-    }
-    /// Executor interrupt callback.
-    ///
-    /// # Safety
-    ///
-    /// You MUST call this from the interrupt handler, and from nowhere else.
-    pub unsafe fn on_interrupt(&'static self) {
-        let executor = unsafe { (&*self.executor.get()).assume_init_ref() };
-        executor.poll();
-    }
-    /// Start the executor.
-    ///
-    /// This initializes the executor, enables the interrupt, and returns.
-    /// The executor keeps running in the background through the interrupt.
-    ///
-    /// This returns a [`SendSpawner`] you can use to spawn tasks on it. A [`SendSpawner`]
-    /// is returned instead of a [`Spawner`](embassy_executor::Spawner) because the executor effectively runs in a
-    /// different "thread" (the interrupt), so spawning tasks on it is effectively
-    /// sending them.
-    ///
-    /// To obtain a [`Spawner`](embassy_executor::Spawner) for this executor, use [`Spawner::for_current_executor()`](embassy_executor::Spawner::for_current_executor()) from
-    /// a task running in it.
-    ///
-    /// # Interrupt requirements
-    ///
-    /// You must write the interrupt handler yourself, and make it call [`on_interrupt()`](Self::on_interrupt).
-    ///
-    /// This method already enables (unmasks) the interrupt, you must NOT do it yourself.
-    ///
-    /// You must set the interrupt priority before calling this method. You MUST NOT
-    /// do it after.
-    ///
-    pub fn start(&'static self, irq: impl InterruptNumber) -> SendSpawner {
-        if self
-            .started
-            .compare_exchange(false, true, Ordering::Acquire, Ordering::Relaxed)
-            .is_err()
-        {
-            panic!("InterruptExecutor::start() called multiple times on the same executor.");
-        }
-        unsafe {
-            (&mut *self.executor.get()).as_mut_ptr().write(raw::Executor::new(
-                |ctx| pend_by_number(ctx as u16),
-                irq.number() as *mut (),
-            ))
-        }
-        let executor = unsafe { (&*self.executor.get()).assume_init_ref() };
-        unsafe { NVIC::unmask(irq) }
-        executor.spawner().make_send()
-    }
-}
diff --git a/embassy-cortex-m/src/lib.rs b/embassy-cortex-m/src/lib.rs
index fba23367b..e4b713a06 100644
--- a/embassy-cortex-m/src/lib.rs
+++ b/embassy-cortex-m/src/lib.rs
@@ -5,6 +5,6 @@
 // This mod MUST go first, so that the others see its macros.
 pub(crate) mod fmt;
-pub mod executor;
+pub use embassy_executor as executor;
 pub mod interrupt;
 pub mod peripheral;
diff --git a/embassy-executor/Cargo.toml b/embassy-executor/Cargo.toml
index 8ad3fd698..bb8a46c82 100644
--- a/embassy-executor/Cargo.toml
+++ b/embassy-executor/Cargo.toml
@@ -31,9 +31,22 @@ flavors = [
 features = ["std", "nightly", "defmt"]
 [features]
-default = []
-std = ["critical-section/std"]
+# Architecture
-wasm = ["dep:wasm-bindgen", "dep:js-sys"]
+_arch = [] # some arch was picked
+arch-std = ["_arch", "critical-section/std"]
+arch-cortex-m = ["_arch", "dep:cortex-m"]
+arch-xtensa = ["_arch"]
+arch-riscv32 = ["_arch"]
+arch-wasm = ["_arch", "dep:wasm-bindgen", "dep:js-sys"]
+# Enable creating a `Pender` from an arbitrary function pointer callback.
+pender-callback = []
+# Enable the thread-mode executor (using WFE/SEV in Cortex-M, WFI in other embedded archs)
+executor-thread = []
+# Enable the interrupt-mode executor (available in Cortex-M only)
+executor-interrupt = []
 # Enable nightly-only features
 nightly = []
@@ -55,9 +68,11 @@ embassy-macros  = { version = "0.1.0", path = "../embassy-macros" }
 embassy-time  = { version = "0.1.0", path = "../embassy-time", optional = true}
 atomic-polyfill = "1.0.1"
 critical-section = "1.1"
-cfg-if = "1.0.0"
 static_cell = "1.0"
-# WASM dependencies
+# arch-cortex-m dependencies
+cortex-m = { version = "0.7.6", optional = true }
+# arch-wasm dependencies
 wasm-bindgen = { version = "0.2.82", optional = true }
 js-sys = { version = "0.3", optional = true }
diff --git a/embassy-executor/src/arch/cortex_m.rs b/embassy-executor/src/arch/cortex_m.rs
index 4b27a264e..d6a55c4c7 100644
--- a/embassy-executor/src/arch/cortex_m.rs
+++ b/embassy-executor/src/arch/cortex_m.rs
@@ -1,59 +1,209 @@
-use core::arch::asm;
+#[cfg(feature = "executor-thread")]
-use core::marker::PhantomData;
+pub use thread::*;
-use core::ptr;
+#[cfg(feature = "executor-thread")]
+mod thread {
-use super::{raw, Spawner};
+    use core::arch::asm;
+    use core::marker::PhantomData;
-/// Thread mode executor, using WFE/SEV.
-///
+    #[cfg(feature = "nightly")]
-/// This is the simplest and most common kind of executor. It runs on
+    pub use embassy_macros::main_cortex_m as main;
-/// thread mode (at the lowest priority level), and uses the `WFE` ARM instruction
-/// to sleep when it has no more work to do. When a task is woken, a `SEV` instruction
+    use crate::raw::{Pender, PenderInner};
-/// is executed, to make the `WFE` exit from sleep and poll the task.
+    use crate::{raw, Spawner};
-///
-/// This executor allows for ultra low power consumption for chips where `WFE`
+    #[derive(Copy, Clone)]
-/// triggers low-power sleep without extra steps. If your chip requires extra steps,
+    pub(crate) struct ThreadPender;
-/// you may use [`raw::Executor`] directly to program custom behavior.
-pub struct Executor {
+    impl ThreadPender {
-    inner: raw::Executor,
+        pub(crate) fn pend(self) {
-    not_send: PhantomData<*mut ()>,
+            unsafe { core::arch::asm!("sev") }
+        }
+    }
+    /// Thread mode executor, using WFE/SEV.
+    ///
+    /// This is the simplest and most common kind of executor. It runs on
+    /// thread mode (at the lowest priority level), and uses the `WFE` ARM instruction
+    /// to sleep when it has no more work to do. When a task is woken, a `SEV` instruction
+    /// is executed, to make the `WFE` exit from sleep and poll the task.
+    ///
+    /// This executor allows for ultra low power consumption for chips where `WFE`
+    /// triggers low-power sleep without extra steps. If your chip requires extra steps,
+    /// you may use [`raw::Executor`] directly to program custom behavior.
+    pub struct Executor {
+        inner: raw::Executor,
+        not_send: PhantomData<*mut ()>,
+    }
+    impl Executor {
+        /// Create a new Executor.
+        pub fn new() -> Self {
+            Self {
+                inner: raw::Executor::new(Pender(PenderInner::Thread(ThreadPender))),
+                not_send: PhantomData,
+            }
+        }
+        /// Run the executor.
+        ///
+        /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+        /// this executor. Use it to spawn the initial task(s). After `init` returns,
+        /// the executor starts running the tasks.
+        ///
+        /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        /// for example by passing it as an argument to the initial tasks.
+        ///
+        /// This function requires `&'static mut self`. This means you have to store the
+        /// Executor instance in a place where it'll live forever and grants you mutable
+        /// access. There's a few ways to do this:
+        ///
+        /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// - a `static mut` (unsafe)
+        /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        ///
+        /// This function never returns.
+        pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+            init(self.inner.spawner());
+            loop {
+                unsafe {
+                    self.inner.poll();
+                    asm!("wfe");
+                };
+            }
+        }
+    }
 }
-impl Executor {
+#[cfg(feature = "executor-interrupt")]
-    /// Create a new Executor.
+pub use interrupt::*;
-    pub fn new() -> Self {
+#[cfg(feature = "executor-interrupt")]
-        Self {
+mod interrupt {
-            inner: raw::Executor::new(|_| unsafe { asm!("sev") }, ptr::null_mut()),
+    use core::cell::UnsafeCell;
-            not_send: PhantomData,
+    use core::mem::MaybeUninit;
+    use atomic_polyfill::{AtomicBool, Ordering};
+    use cortex_m::interrupt::InterruptNumber;
+    use cortex_m::peripheral::NVIC;
+    use crate::raw::{self, Pender, PenderInner};
+    #[derive(Clone, Copy)]
+    pub(crate) struct InterruptPender(u16);
+    impl InterruptPender {
+        pub(crate) fn pend(self) {
+            // STIR is faster, but is only available in v7 and higher.
+            #[cfg(not(armv6m))]
+            {
+                let mut nvic: cortex_m::peripheral::NVIC = unsafe { core::mem::transmute(()) };
+                nvic.request(self);
+            }
+            #[cfg(armv6m)]
+            cortex_m::peripheral::NVIC::pend(self);
        }
    }
-    /// Run the executor.
+    unsafe impl cortex_m::interrupt::InterruptNumber for InterruptPender {
+        fn number(self) -> u16 {
+            self.0
+        }
+    }
+    /// Interrupt mode executor.
    ///
-    /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+    /// This executor runs tasks in interrupt mode. The interrupt handler is set up
-    /// this executor. Use it to spawn the initial task(s). After `init` returns,
+    /// to poll tasks, and when a task is woken the interrupt is pended from software.
-    /// the executor starts running the tasks.
    ///
-    /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+    /// This allows running async tasks at a priority higher than thread mode. One
-    /// for example by passing it as an argument to the initial tasks.
+    /// use case is to leave thread mode free for non-async tasks. Another use case is
+    /// to run multiple executors: one in thread mode for low priority tasks and another in
+    /// interrupt mode for higher priority tasks. Higher priority tasks will preempt lower
+    /// priority ones.
    ///
-    /// This function requires `&'static mut self`. This means you have to store the
+    /// It is even possible to run multiple interrupt mode executors at different priorities,
-    /// Executor instance in a place where it'll live forever and grants you mutable
+    /// by assigning different priorities to the interrupts. For an example on how to do this,
-    /// access. There's a few ways to do this:
+    /// See the 'multiprio' example for 'embassy-nrf'.
    ///
-    /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+    /// To use it, you have to pick an interrupt that won't be used by the hardware.
-    /// - a `static mut` (unsafe)
+    /// Some chips reserve some interrupts for this purpose, sometimes named "software interrupts" (SWI).
-    /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+    /// If this is not the case, you may use an interrupt from any unused peripheral.
    ///
-    /// This function never returns.
+    /// It is somewhat more complex to use, it's recommended to use the thread-mode
-    pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+    /// [`Executor`] instead, if it works for your use case.
-        init(self.inner.spawner());
+    pub struct InterruptExecutor {
+        started: AtomicBool,
+        executor: UnsafeCell<MaybeUninit<raw::Executor>>,
+    }
+    unsafe impl Send for InterruptExecutor {}
+    unsafe impl Sync for InterruptExecutor {}
+    impl InterruptExecutor {
+        /// Create a new, not started `InterruptExecutor`.
+        #[inline]
+        pub const fn new() -> Self {
+            Self {
+                started: AtomicBool::new(false),
+                executor: UnsafeCell::new(MaybeUninit::uninit()),
+            }
+        }
+        /// Executor interrupt callback.
+        ///
+        /// # Safety
+        ///
+        /// You MUST call this from the interrupt handler, and from nowhere else.
+        pub unsafe fn on_interrupt(&'static self) {
+            let executor = unsafe { (&*self.executor.get()).assume_init_ref() };
+            executor.poll();
+        }
+        /// Start the executor.
+        ///
+        /// This initializes the executor, enables the interrupt, and returns.
+        /// The executor keeps running in the background through the interrupt.
+        ///
+        /// This returns a [`SendSpawner`] you can use to spawn tasks on it. A [`SendSpawner`]
+        /// is returned instead of a [`Spawner`](embassy_executor::Spawner) because the executor effectively runs in a
+        /// different "thread" (the interrupt), so spawning tasks on it is effectively
+        /// sending them.
+        ///
+        /// To obtain a [`Spawner`](embassy_executor::Spawner) for this executor, use [`Spawner::for_current_executor()`](embassy_executor::Spawner::for_current_executor()) from
+        /// a task running in it.
+        ///
+        /// # Interrupt requirements
+        ///
+        /// You must write the interrupt handler yourself, and make it call [`on_interrupt()`](Self::on_interrupt).
+        ///
+        /// This method already enables (unmasks) the interrupt, you must NOT do it yourself.
+        ///
+        /// You must set the interrupt priority before calling this method. You MUST NOT
+        /// do it after.
+        ///
+        pub fn start(&'static self, irq: impl InterruptNumber) -> crate::SendSpawner {
+            if self
+                .started
+                .compare_exchange(false, true, Ordering::Acquire, Ordering::Relaxed)
+                .is_err()
+            {
+                panic!("InterruptExecutor::start() called multiple times on the same executor.");
+            }
-        loop {
            unsafe {
-                self.inner.poll();
+                (&mut *self.executor.get())
-                asm!("wfe");
+                    .as_mut_ptr()
-            };
+                    .write(raw::Executor::new(Pender(PenderInner::Interrupt(InterruptPender(
+                        irq.number(),
+                    )))))
+            }
+            let executor = unsafe { (&*self.executor.get()).assume_init_ref() };
+            unsafe { NVIC::unmask(irq) }
+            executor.spawner().make_send()
        }
    }
 }
diff --git a/embassy-executor/src/arch/riscv32.rs b/embassy-executor/src/arch/riscv32.rs
index e97a56cda..f66daeae4 100644
--- a/embassy-executor/src/arch/riscv32.rs
+++ b/embassy-executor/src/arch/riscv32.rs
@@ -1,72 +1,83 @@
-use core::marker::PhantomData;
+#[cfg(feature = "executor-interrupt")]
-use core::ptr;
+compile_error!("`executor-interrupt` is not supported with `arch-riscv32`.");
-use core::sync::atomic::{AtomicBool, Ordering};
-use super::{raw, Spawner};
+#[cfg(feature = "executor-thread")]
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
+    use core::marker::PhantomData;
+    use core::sync::atomic::{AtomicBool, Ordering};
-/// global atomic used to keep track of whether there is work to do since sev() is not available on RISCV
+    use crate::raw::{Pender, PenderInner};
-///
+    use crate::{raw, Spawner};
-static SIGNAL_WORK_THREAD_MODE: AtomicBool = AtomicBool::new(false);
-/// RISCV32 Executor
+    #[derive(Copy, Clone)]
-pub struct Executor {
+    pub(crate) struct ThreadPender;
-    inner: raw::Executor,
-    not_send: PhantomData<*mut ()>,
-}
-impl Executor {
+    impl ThreadPender {
-    /// Create a new Executor.
+        #[allow(unused)]
-    pub fn new() -> Self {
+        pub(crate) fn pend(self) {
-        Self {
+            SIGNAL_WORK_THREAD_MODE.store(true, core::sync::atomic::Ordering::SeqCst);
-            // use Signal_Work_Thread_Mode as substitute for local interrupt register
-            inner: raw::Executor::new(
-                |_| {
-                    SIGNAL_WORK_THREAD_MODE.store(true, Ordering::SeqCst);
-                },
-                ptr::null_mut(),
-            ),
-            not_send: PhantomData,
        }
    }
-    /// Run the executor.
+    /// global atomic used to keep track of whether there is work to do since sev() is not available on RISCV
-    ///
+    static SIGNAL_WORK_THREAD_MODE: AtomicBool = AtomicBool::new(false);
-    /// The `init` closure is called with a [`Spawner`] that spawns tasks on
-    /// this executor. Use it to spawn the initial task(s). After `init` returns,
+    /// RISCV32 Executor
-    /// the executor starts running the tasks.
+    pub struct Executor {
-    ///
+        inner: raw::Executor,
-    /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        not_send: PhantomData<*mut ()>,
-    /// for example by passing it as an argument to the initial tasks.
+    }
-    ///
-    /// This function requires `&'static mut self`. This means you have to store the
+    impl Executor {
-    /// Executor instance in a place where it'll live forever and grants you mutable
+        /// Create a new Executor.
-    /// access. There's a few ways to do this:
+        pub fn new() -> Self {
-    ///
+            Self {
-    /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+                inner: raw::Executor::new(Pender(PenderInner::Thread(ThreadPender))),
-    /// - a `static mut` (unsafe)
+                not_send: PhantomData,
-    /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+            }
-    ///
+        }
-    /// This function never returns.
-    pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+        /// Run the executor.
-        init(self.inner.spawner());
+        ///
+        /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+        /// this executor. Use it to spawn the initial task(s). After `init` returns,
+        /// the executor starts running the tasks.
+        ///
+        /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        /// for example by passing it as an argument to the initial tasks.
+        ///
+        /// This function requires `&'static mut self`. This means you have to store the
+        /// Executor instance in a place where it'll live forever and grants you mutable
+        /// access. There's a few ways to do this:
+        ///
+        /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// - a `static mut` (unsafe)
+        /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        ///
+        /// This function never returns.
+        pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+            init(self.inner.spawner());
-        loop {
+            loop {
-            unsafe {
+                unsafe {
-                self.inner.poll();
+                    self.inner.poll();
-                // we do not care about race conditions between the load and store operations, interrupts
+                    // we do not care about race conditions between the load and store operations, interrupts
-                //will only set this value to true.
+                    //will only set this value to true.
-                critical_section::with(|_| {
+                    critical_section::with(|_| {
-                    // if there is work to do, loop back to polling
+                        // if there is work to do, loop back to polling
-                    // TODO can we relax this?
+                        // TODO can we relax this?
-                    if SIGNAL_WORK_THREAD_MODE.load(Ordering::SeqCst) {
+                        if SIGNAL_WORK_THREAD_MODE.load(Ordering::SeqCst) {
-                        SIGNAL_WORK_THREAD_MODE.store(false, Ordering::SeqCst);
+                            SIGNAL_WORK_THREAD_MODE.store(false, Ordering::SeqCst);
-                    }
+                        }
-                    // if not, wait for interrupt
+                        // if not, wait for interrupt
-                    else {
+                        else {
-                        core::arch::asm!("wfi");
+                            core::arch::asm!("wfi");
-                    }
+                        }
-                });
+                    });
-                // if an interrupt occurred while waiting, it will be serviced here
+                    // if an interrupt occurred while waiting, it will be serviced here
+                }
            }
        }
    }
diff --git a/embassy-executor/src/arch/std.rs b/embassy-executor/src/arch/std.rs
index 701f0eb18..4e4a178f0 100644
--- a/embassy-executor/src/arch/std.rs
+++ b/embassy-executor/src/arch/std.rs
@@ -1,84 +1,100 @@
-use std::marker::PhantomData;
+#[cfg(feature = "executor-interrupt")]
-use std::sync::{Condvar, Mutex};
+compile_error!("`executor-interrupt` is not supported with `arch-std`.");
-use super::{raw, Spawner};
+#[cfg(feature = "executor-thread")]
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
+    use std::marker::PhantomData;
+    use std::sync::{Condvar, Mutex};
-/// Single-threaded std-based executor.
+    #[cfg(feature = "nightly")]
-pub struct Executor {
+    pub use embassy_macros::main_std as main;
-    inner: raw::Executor,
-    not_send: PhantomData<*mut ()>,
+    use crate::raw::{Pender, PenderInner};
-    signaler: &'static Signaler,
+    use crate::{raw, Spawner};
-}
-impl Executor {
+    #[derive(Copy, Clone)]
-    /// Create a new Executor.
+    pub(crate) struct ThreadPender(&'static Signaler);
-    pub fn new() -> Self {
-        let signaler = &*Box::leak(Box::new(Signaler::new()));
+    impl ThreadPender {
-        Self {
+        #[allow(unused)]
-            inner: raw::Executor::new(
+        pub(crate) fn pend(self) {
-                |p| unsafe {
+            self.0.signal()
-                    let s = &*(p as *const () as *const Signaler);
-                    s.signal()
-                },
-                signaler as *const _ as _,
-            ),
-            not_send: PhantomData,
-            signaler,
        }
    }
-    /// Run the executor.
+    /// Single-threaded std-based executor.
-    ///
+    pub struct Executor {
-    /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+        inner: raw::Executor,
-    /// this executor. Use it to spawn the initial task(s). After `init` returns,
+        not_send: PhantomData<*mut ()>,
-    /// the executor starts running the tasks.
+        signaler: &'static Signaler,
-    ///
+    }
-    /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
-    /// for example by passing it as an argument to the initial tasks.
-    ///
-    /// This function requires `&'static mut self`. This means you have to store the
-    /// Executor instance in a place where it'll live forever and grants you mutable
-    /// access. There's a few ways to do this:
-    ///
-    /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
-    /// - a `static mut` (unsafe)
-    /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
-    ///
-    /// This function never returns.
-    pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
-        init(self.inner.spawner());
-        loop {
+    impl Executor {
-            unsafe { self.inner.poll() };
+        /// Create a new Executor.
-            self.signaler.wait()
+        pub fn new() -> Self {
+            let signaler = &*Box::leak(Box::new(Signaler::new()));
+            Self {
+                inner: raw::Executor::new(Pender(PenderInner::Thread(ThreadPender(signaler)))),
+                not_send: PhantomData,
+                signaler,
+            }
        }
-    }
-}
-struct Signaler {
+        /// Run the executor.
-    mutex: Mutex<bool>,
+        ///
-    condvar: Condvar,
+        /// The `init` closure is called with a [`Spawner`] that spawns tasks on
-}
+        /// this executor. Use it to spawn the initial task(s). After `init` returns,
+        /// the executor starts running the tasks.
+        ///
+        /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        /// for example by passing it as an argument to the initial tasks.
+        ///
+        /// This function requires `&'static mut self`. This means you have to store the
+        /// Executor instance in a place where it'll live forever and grants you mutable
+        /// access. There's a few ways to do this:
+        ///
+        /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// - a `static mut` (unsafe)
+        /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        ///
+        /// This function never returns.
+        pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+            init(self.inner.spawner());
-impl Signaler {
+            loop {
-    fn new() -> Self {
+                unsafe { self.inner.poll() };
-        Self {
+                self.signaler.wait()
-            mutex: Mutex::new(false),
+            }
-            condvar: Condvar::new(),
        }
    }
-    fn wait(&self) {
+    struct Signaler {
-        let mut signaled = self.mutex.lock().unwrap();
+        mutex: Mutex<bool>,
-        while !*signaled {
+        condvar: Condvar,
-            signaled = self.condvar.wait(signaled).unwrap();
-        }
-        *signaled = false;
    }
-    fn signal(&self) {
+    impl Signaler {
-        let mut signaled = self.mutex.lock().unwrap();
+        fn new() -> Self {
-        *signaled = true;
+            Self {
-        self.condvar.notify_one();
+                mutex: Mutex::new(false),
+                condvar: Condvar::new(),
+            }
+        }
+        fn wait(&self) {
+            let mut signaled = self.mutex.lock().unwrap();
+            while !*signaled {
+                signaled = self.condvar.wait(signaled).unwrap();
+            }
+            *signaled = false;
+        }
+        fn signal(&self) {
+            let mut signaled = self.mutex.lock().unwrap();
+            *signaled = true;
+            self.condvar.notify_one();
+        }
    }
 }
diff --git a/embassy-executor/src/arch/wasm.rs b/embassy-executor/src/arch/wasm.rs
index 98091cfbb..08ab16b99 100644
--- a/embassy-executor/src/arch/wasm.rs
+++ b/embassy-executor/src/arch/wasm.rs
@@ -1,74 +1,88 @@
-use core::marker::PhantomData;
+#[cfg(feature = "executor-interrupt")]
+compile_error!("`executor-interrupt` is not supported with `arch-wasm`.");
-use js_sys::Promise;
+#[cfg(feature = "executor-thread")]
-use wasm_bindgen::prelude::*;
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
-use super::raw::util::UninitCell;
+    use core::marker::PhantomData;
-use super::raw::{self};
-use super::Spawner;
-/// WASM executor, wasm_bindgen to schedule tasks on the JS event loop.
+    #[cfg(feature = "nightly")]
-pub struct Executor {
+    pub use embassy_macros::main_wasm as main;
-    inner: raw::Executor,
+    use js_sys::Promise;
-    ctx: &'static WasmContext,
+    use wasm_bindgen::prelude::*;
-    not_send: PhantomData<*mut ()>,
-}
-pub(crate) struct WasmContext {
+    use crate::raw::util::UninitCell;
-    promise: Promise,
+    use crate::raw::{Pender, PenderInner};
-    closure: UninitCell<Closure<dyn FnMut(JsValue)>>,
+    use crate::{raw, Spawner};
-}
+    /// WASM executor, wasm_bindgen to schedule tasks on the JS event loop.
+    pub struct Executor {
+        inner: raw::Executor,
+        ctx: &'static WasmContext,
+        not_send: PhantomData<*mut ()>,
+    }
+    pub(crate) struct WasmContext {
+        promise: Promise,
+        closure: UninitCell<Closure<dyn FnMut(JsValue)>>,
+    }
+    #[derive(Copy, Clone)]
+    pub(crate) struct ThreadPender(&'static WasmContext);
-impl WasmContext {
+    impl ThreadPender {
-    pub fn new() -> Self {
+        #[allow(unused)]
-        Self {
+        pub(crate) fn pend(self) {
-            promise: Promise::resolve(&JsValue::undefined()),
+            let _ = self.0.promise.then(unsafe { self.0.closure.as_mut() });
-            closure: UninitCell::uninit(),
        }
    }
-}
-impl Executor {
+    impl WasmContext {
-    /// Create a new Executor.
+        pub fn new() -> Self {
-    pub fn new() -> Self {
+            Self {
-        let ctx = &*Box::leak(Box::new(WasmContext::new()));
+                promise: Promise::resolve(&JsValue::undefined()),
-        let inner = raw::Executor::new(
+                closure: UninitCell::uninit(),
-            |p| unsafe {
+            }
-                let ctx = &*(p as *const () as *const WasmContext);
-                let _ = ctx.promise.then(ctx.closure.as_mut());
-            },
-            ctx as *const _ as _,
-        );
-        Self {
-            inner,
-            not_send: PhantomData,
-            ctx,
        }
    }
-    /// Run the executor.
+    impl Executor {
-    ///
+        /// Create a new Executor.
-    /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+        pub fn new() -> Self {
-    /// this executor. Use it to spawn the initial task(s). After `init` returns,
+            let ctx = &*Box::leak(Box::new(WasmContext::new()));
-    /// the executor starts running the tasks.
+            Self {
-    ///
+                inner: raw::Executor::new(Pender(PenderInner::Thread(ThreadPender(ctx)))),
-    /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+                not_send: PhantomData,
-    /// for example by passing it as an argument to the initial tasks.
+                ctx,
-    ///
+            }
-    /// This function requires `&'static mut self`. This means you have to store the
+        }
-    /// Executor instance in a place where it'll live forever and grants you mutable
-    /// access. There's a few ways to do this:
+        /// Run the executor.
-    ///
+        ///
-    /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// The `init` closure is called with a [`Spawner`] that spawns tasks on
-    /// - a `static mut` (unsafe)
+        /// this executor. Use it to spawn the initial task(s). After `init` returns,
-    /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        /// the executor starts running the tasks.
-    pub fn start(&'static mut self, init: impl FnOnce(Spawner)) {
+        ///
-        unsafe {
+        /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
-            let executor = &self.inner;
+        /// for example by passing it as an argument to the initial tasks.
-            self.ctx.closure.write(Closure::new(move |_| {
+        ///
-                executor.poll();
+        /// This function requires `&'static mut self`. This means you have to store the
-            }));
+        /// Executor instance in a place where it'll live forever and grants you mutable
-            init(self.inner.spawner());
+        /// access. There's a few ways to do this:
+        ///
+        /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// - a `static mut` (unsafe)
+        /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        pub fn start(&'static mut self, init: impl FnOnce(Spawner)) {
+            unsafe {
+                let executor = &self.inner;
+                self.ctx.closure.write(Closure::new(move |_| {
+                    executor.poll();
+                }));
+                init(self.inner.spawner());
+            }
        }
    }
 }
diff --git a/embassy-executor/src/arch/xtensa.rs b/embassy-executor/src/arch/xtensa.rs
index 4ee0d9f78..61ea92c16 100644
--- a/embassy-executor/src/arch/xtensa.rs
+++ b/embassy-executor/src/arch/xtensa.rs
@@ -1,73 +1,84 @@
-use core::marker::PhantomData;
+#[cfg(feature = "executor-interrupt")]
-use core::ptr;
+compile_error!("`executor-interrupt` is not supported with `arch-xtensa`.");
-use core::sync::atomic::{AtomicBool, Ordering};
-use super::{raw, Spawner};
+#[cfg(feature = "executor-thread")]
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
+    use core::marker::PhantomData;
+    use core::sync::atomic::{AtomicBool, Ordering};
-/// global atomic used to keep track of whether there is work to do since sev() is not available on Xtensa
+    use crate::raw::{Pender, PenderInner};
-///
+    use crate::{raw, Spawner};
-static SIGNAL_WORK_THREAD_MODE: AtomicBool = AtomicBool::new(false);
-/// Xtensa Executor
+    #[derive(Copy, Clone)]
-pub struct Executor {
+    pub(crate) struct ThreadPender;
-    inner: raw::Executor,
-    not_send: PhantomData<*mut ()>,
-}
-impl Executor {
+    impl ThreadPender {
-    /// Create a new Executor.
+        #[allow(unused)]
-    pub fn new() -> Self {
+        pub(crate) fn pend(self) {
-        Self {
+            SIGNAL_WORK_THREAD_MODE.store(true, core::sync::atomic::Ordering::SeqCst);
-            // use Signal_Work_Thread_Mode as substitute for local interrupt register
-            inner: raw::Executor::new(
-                |_| {
-                    SIGNAL_WORK_THREAD_MODE.store(true, Ordering::SeqCst);
-                },
-                ptr::null_mut(),
-            ),
-            not_send: PhantomData,
        }
    }
-    /// Run the executor.
+    /// global atomic used to keep track of whether there is work to do since sev() is not available on Xtensa
-    ///
+    static SIGNAL_WORK_THREAD_MODE: AtomicBool = AtomicBool::new(false);
-    /// The `init` closure is called with a [`Spawner`] that spawns tasks on
-    /// this executor. Use it to spawn the initial task(s). After `init` returns,
+    /// Xtensa Executor
-    /// the executor starts running the tasks.
+    pub struct Executor {
-    ///
+        inner: raw::Executor,
-    /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        not_send: PhantomData<*mut ()>,
-    /// for example by passing it as an argument to the initial tasks.
+    }
-    ///
-    /// This function requires `&'static mut self`. This means you have to store the
+    impl Executor {
-    /// Executor instance in a place where it'll live forever and grants you mutable
+        /// Create a new Executor.
-    /// access. There's a few ways to do this:
+        pub fn new() -> Self {
-    ///
+            Self {
-    /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+                inner: raw::Executor::new(Pender(PenderInner::Thread(ThreadPender))),
-    /// - a `static mut` (unsafe)
+                not_send: PhantomData,
-    /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+            }
-    ///
+        }
-    /// This function never returns.
-    pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+        /// Run the executor.
-        init(self.inner.spawner());
+        ///
+        /// The `init` closure is called with a [`Spawner`] that spawns tasks on
+        /// this executor. Use it to spawn the initial task(s). After `init` returns,
+        /// the executor starts running the tasks.
+        ///
+        /// To spawn more tasks later, you may keep copies of the [`Spawner`] (it is `Copy`),
+        /// for example by passing it as an argument to the initial tasks.
+        ///
+        /// This function requires `&'static mut self`. This means you have to store the
+        /// Executor instance in a place where it'll live forever and grants you mutable
+        /// access. There's a few ways to do this:
+        ///
+        /// - a [StaticCell](https://docs.rs/static_cell/latest/static_cell/) (safe)
+        /// - a `static mut` (unsafe)
+        /// - a local variable in a function you know never returns (like `fn main() -> !`), upgrading its lifetime with `transmute`. (unsafe)
+        ///
+        /// This function never returns.
+        pub fn run(&'static mut self, init: impl FnOnce(Spawner)) -> ! {
+            init(self.inner.spawner());
-        loop {
+            loop {
-            unsafe {
+                unsafe {
-                self.inner.poll();
+                    self.inner.poll();
-                // we do not care about race conditions between the load and store operations, interrupts
+                    // we do not care about race conditions between the load and store operations, interrupts
-                // will only set this value to true.
+                    // will only set this value to true.
-                // if there is work to do, loop back to polling
+                    // if there is work to do, loop back to polling
-                // TODO can we relax this?
+                    // TODO can we relax this?
-                critical_section::with(|_| {
+                    critical_section::with(|_| {
-                    if SIGNAL_WORK_THREAD_MODE.load(Ordering::SeqCst) {
+                        if SIGNAL_WORK_THREAD_MODE.load(Ordering::SeqCst) {
-                        SIGNAL_WORK_THREAD_MODE.store(false, Ordering::SeqCst);
+                            SIGNAL_WORK_THREAD_MODE.store(false, Ordering::SeqCst);
-                    } else {
+                        } else {
-                        // waiti sets the PS.INTLEVEL when slipping into sleep
+                            // waiti sets the PS.INTLEVEL when slipping into sleep
-                        // because critical sections in Xtensa are implemented via increasing
+                            // because critical sections in Xtensa are implemented via increasing
-                        // PS.INTLEVEL the critical section ends here
+                            // PS.INTLEVEL the critical section ends here
-                        // take care not add code after `waiti` if it needs to be inside the CS
+                            // take care not add code after `waiti` if it needs to be inside the CS
-                        core::arch::asm!("waiti 0"); // critical section ends here
+                            core::arch::asm!("waiti 0"); // critical section ends here
-                    }
+                        }
-                });
+                    });
+                }
            }
        }
    }
diff --git a/embassy-executor/src/lib.rs b/embassy-executor/src/lib.rs
index 4c7e2f4cd..3ce687eb6 100644
--- a/embassy-executor/src/lib.rs
+++ b/embassy-executor/src/lib.rs
@@ -1,5 +1,5 @@
-#![cfg_attr(not(any(feature = "std", feature = "wasm")), no_std)]
+#![cfg_attr(not(any(feature = "arch-std", feature = "arch-wasm")), no_std)]
-#![cfg_attr(all(feature = "nightly", target_arch = "xtensa"), feature(asm_experimental_arch))]
+#![cfg_attr(all(feature = "nightly", feature = "arch-xtensa"), feature(asm_experimental_arch))]
 #![allow(clippy::new_without_default)]
 #![doc = include_str!("../README.md")]
 #![warn(missing_docs)]
@@ -10,47 +10,43 @@ pub(crate) mod fmt;
 #[cfg(feature = "nightly")]
 pub use embassy_macros::task;
-cfg_if::cfg_if! {
+macro_rules! check_at_most_one {
-    if #[cfg(cortex_m)] {
+    (@amo [$($feats:literal)*] [] [$($res:tt)*]) => {
-        #[path="arch/cortex_m.rs"]
+        #[cfg(any($($res)*))]
-        mod arch;
+        compile_error!(concat!("At most one of these features can be enabled at the same time:", $(" `", $feats, "`",)*));
-        pub use arch::*;
+    };
-        #[cfg(feature = "nightly")]
+    (@amo $feats:tt [$curr:literal $($rest:literal)*] [$($res:tt)*]) => {
-        pub use embassy_macros::main_cortex_m as main;
+        check_at_most_one!(@amo $feats [$($rest)*] [$($res)* $(all(feature=$curr, feature=$rest),)*]);
-    }
+    };
-    else if #[cfg(target_arch="riscv32")] {
+    ($($f:literal),*$(,)?) => {
-        #[path="arch/riscv32.rs"]
+        check_at_most_one!(@amo [$($f)*] [$($f)*] []);
-        mod arch;
+    };
-        pub use arch::*;
-        #[cfg(feature = "nightly")]
-        pub use embassy_macros::main_riscv as main;
-    }
-    else if #[cfg(all(target_arch="xtensa", feature = "nightly"))] {
-        #[path="arch/xtensa.rs"]
-        mod arch;
-        pub use arch::*;
-    }
-    else if #[cfg(feature="wasm")] {
-        #[path="arch/wasm.rs"]
-        mod arch;
-        pub use arch::*;
-        #[cfg(feature = "nightly")]
-        pub use embassy_macros::main_wasm as main;
-    }
-    else if #[cfg(feature="std")] {
-        #[path="arch/std.rs"]
-        mod arch;
-        pub use arch::*;
-        #[cfg(feature = "nightly")]
-        pub use embassy_macros::main_std as main;
-    }
 }
+check_at_most_one!("arch-cortex-m", "arch-riscv32", "arch-xtensa", "arch-std", "arch-wasm",);
+#[cfg(feature = "_arch")]
+#[cfg_attr(feature = "arch-cortex-m", path = "arch/cortex_m.rs")]
+#[cfg_attr(feature = "arch-riscv32", path = "arch/riscv32.rs")]
+#[cfg_attr(feature = "arch-xtensa", path = "arch/xtensa.rs")]
+#[cfg_attr(feature = "arch-std", path = "arch/std.rs")]
+#[cfg_attr(feature = "arch-wasm", path = "arch/wasm.rs")]
+mod arch;
+#[cfg(feature = "_arch")]
+pub use arch::*;
+pub mod raw;
+mod spawner;
+pub use spawner::*;
+/// Implementation details for embassy macros.
+/// Do not use. Used for macros and HALs only. Not covered by semver guarantees.
 #[doc(hidden)]
-/// Implementation details for embassy macros. DO NOT USE.
+pub mod _export {
-pub mod export {
    #[cfg(feature = "rtos-trace")]
    pub use rtos_trace::trace;
+    pub use static_cell::StaticCell;
    /// Expands the given block of code when `embassy-executor` is compiled with
    /// the `rtos-trace-interrupt` feature.
@@ -70,14 +66,3 @@ pub mod export {
        ($($tt:tt)*) => {};
    }
 }
-pub mod raw;
-mod spawner;
-pub use spawner::*;
-/// Do not use. Used for macros and HALs only. Not covered by semver guarantees.
-#[doc(hidden)]
-pub mod _export {
-    pub use static_cell::StaticCell;
-}
diff --git a/embassy-executor/src/raw/mod.rs b/embassy-executor/src/raw/mod.rs
index 72c367c33..f6c66da5a 100644
--- a/embassy-executor/src/raw/mod.rs
+++ b/embassy-executor/src/raw/mod.rs
@@ -19,7 +19,6 @@ use core::marker::PhantomData;
 use core::mem;
 use core::pin::Pin;
 use core::ptr::NonNull;
-use core::sync::atomic::AtomicPtr;
 use core::task::{Context, Poll};
 use atomic_polyfill::{AtomicU32, Ordering};
@@ -290,10 +289,60 @@ impl<F: Future + 'static, const N: usize> TaskPool<F, N> {
    }
 }
+#[derive(Clone, Copy)]
+pub(crate) enum PenderInner {
+    #[cfg(feature = "executor-thread")]
+    Thread(crate::arch::ThreadPender),
+    #[cfg(feature = "executor-interrupt")]
+    Interrupt(crate::arch::InterruptPender),
+    #[cfg(feature = "pender-callback")]
+    Callback { func: fn(*mut ()), context: *mut () },
+}
+unsafe impl Send for PenderInner {}
+unsafe impl Sync for PenderInner {}
+/// Platform/architecture-specific action executed when an executor has pending work.
+///
+/// When a task within an executor is woken, the `Pender` is called. This does a
+/// platform/architecture-specific action to signal there is pending work in the executor.
+/// When this happens, you must arrange for [`Executor::poll`] to be called.
+///
+/// You can think of it as a waker, but for the whole executor.
+pub struct Pender(pub(crate) PenderInner);
+impl Pender {
+    /// Create a `Pender` that will call an arbitrary function pointer.
+    ///
+    /// # Arguments
+    ///
+    /// - `func`: The function pointer to call.
+    /// - `context`: Opaque context pointer, that will be passed to the function pointer.
+    #[cfg(feature = "pender-callback")]
+    pub fn new_from_callback(func: fn(*mut ()), context: *mut ()) -> Self {
+        Self(PenderInner::Callback {
+            func,
+            context: context.into(),
+        })
+    }
+}
+impl Pender {
+    pub(crate) fn pend(&self) {
+        match self.0 {
+            #[cfg(feature = "executor-thread")]
+            PenderInner::Thread(x) => x.pend(),
+            #[cfg(feature = "executor-interrupt")]
+            PenderInner::Interrupt(x) => x.pend(),
+            #[cfg(feature = "pender-callback")]
+            PenderInner::Callback { func, context } => func(context),
+        }
+    }
+}
 pub(crate) struct SyncExecutor {
    run_queue: RunQueue,
-    signal_fn: fn(*mut ()),
+    pender: Pender,
-    signal_ctx: AtomicPtr<()>,
    #[cfg(feature = "integrated-timers")]
    pub(crate) timer_queue: timer_queue::TimerQueue,
@@ -302,16 +351,13 @@ pub(crate) struct SyncExecutor {
 }
 impl SyncExecutor {
-    pub(crate) fn new(signal_fn: fn(*mut ()), signal_ctx: *mut ()) -> Self {
+    pub(crate) fn new(pender: Pender) -> Self {
        #[cfg(feature = "integrated-timers")]
        let alarm = unsafe { unwrap!(driver::allocate_alarm()) };
-        #[cfg(feature = "integrated-timers")]
-        driver::set_alarm_callback(alarm, signal_fn, signal_ctx);
        Self {
            run_queue: RunQueue::new(),
-            signal_fn,
+            pender,
-            signal_ctx: AtomicPtr::new(signal_ctx),
            #[cfg(feature = "integrated-timers")]
            timer_queue: timer_queue::TimerQueue::new(),
@@ -332,10 +378,16 @@ impl SyncExecutor {
        trace::task_ready_begin(task.as_ptr() as u32);
        if self.run_queue.enqueue(cs, task) {
-            (self.signal_fn)(self.signal_ctx.load(Ordering::Relaxed))
+            self.pender.pend();
        }
    }
+    #[cfg(feature = "integrated-timers")]
+    fn alarm_callback(ctx: *mut ()) {
+        let this: &Self = unsafe { &*(ctx as *const Self) };
+        this.pender.pend();
+    }
    pub(super) unsafe fn spawn(&'static self, task: TaskRef) {
        task.header().executor.set(Some(self));
@@ -351,6 +403,9 @@ impl SyncExecutor {
    ///
    /// Same as [`Executor::poll`], plus you must only call this on the thread this executor was created.
    pub(crate) unsafe fn poll(&'static self) {
+        #[cfg(feature = "integrated-timers")]
+        driver::set_alarm_callback(self.alarm, Self::alarm_callback, self as *const _ as *mut ());
        #[allow(clippy::never_loop)]
        loop {
            #[cfg(feature = "integrated-timers")]
@@ -417,14 +472,14 @@ impl SyncExecutor {
 ///
 /// - To get the executor to do work, call `poll()`. This will poll all queued tasks (all tasks
 ///   that "want to run").
-/// - You must supply a `signal_fn`. The executor will call it to notify you it has work
+/// - You must supply a [`Pender`]. The executor will call it to notify you it has work
 ///   to do. You must arrange for `poll()` to be called as soon as possible.
 ///
-/// `signal_fn` can be called from *any* context: any thread, any interrupt priority
+/// The [`Pender`] can be called from *any* context: any thread, any interrupt priority
 /// level, etc. It may be called synchronously from any `Executor` method call as well.
 /// You must deal with this correctly.
 ///
-/// In particular, you must NOT call `poll` directly from `signal_fn`, as this violates
+/// In particular, you must NOT call `poll` directly from the pender callback, as this violates
 /// the requirement for `poll` to not be called reentrantly.
 #[repr(transparent)]
 pub struct Executor {
@@ -437,15 +492,15 @@ impl Executor {
    pub(crate) unsafe fn wrap(inner: &SyncExecutor) -> &Self {
        mem::transmute(inner)
    }
    /// Create a new executor.
    ///
-    /// When the executor has work to do, it will call `signal_fn` with
+    /// When the executor has work to do, it will call the [`Pender`].
-    /// `signal_ctx` as argument.
    ///
-    /// See [`Executor`] docs for details on `signal_fn`.
+    /// See [`Executor`] docs for details on `Pender`.
-    pub fn new(signal_fn: fn(*mut ()), signal_ctx: *mut ()) -> Self {
+    pub fn new(pender: Pender) -> Self {
        Self {
-            inner: SyncExecutor::new(signal_fn, signal_ctx),
+            inner: SyncExecutor::new(pender),
            _not_sync: PhantomData,
        }
    }
@@ -468,16 +523,16 @@ impl Executor {
    /// This loops over all tasks that are queued to be polled (i.e. they're
    /// freshly spawned or they've been woken). Other tasks are not polled.
    ///
-    /// You must call `poll` after receiving a call to `signal_fn`. It is OK
+    /// You must call `poll` after receiving a call to the [`Pender`]. It is OK
-    /// to call `poll` even when not requested by `signal_fn`, but it wastes
+    /// to call `poll` even when not requested by the `Pender`, but it wastes
    /// energy.
    ///
    /// # Safety
    ///
    /// You must NOT call `poll` reentrantly on the same executor.
    ///
-    /// In particular, note that `poll` may call `signal_fn` synchronously. Therefore, you
+    /// In particular, note that `poll` may call the `Pender` synchronously. Therefore, you
-    /// must NOT directly call `poll()` from your `signal_fn`. Instead, `signal_fn` has to
+    /// must NOT directly call `poll()` from the `Pender` callback. Instead, the callback has to
    /// somehow schedule for `poll()` to be called later, at a time you know for sure there's
    /// no `poll()` already running.
    pub unsafe fn poll(&'static self) {
diff --git a/embassy-macros/src/macros/cortex_m_interrupt_take.rs b/embassy-macros/src/macros/cortex_m_interrupt_take.rs
index e2ebf98c7..4806d1c12 100644
--- a/embassy-macros/src/macros/cortex_m_interrupt_take.rs
+++ b/embassy-macros/src/macros/cortex_m_interrupt_take.rs
@@ -10,12 +10,12 @@ pub fn run(name: syn::Ident) -> Result<TokenStream, TokenStream> {
    let (isr_enter, isr_exit) = (
        quote! {
            ::embassy_executor::rtos_trace_interrupt! {
-                ::embassy_executor::export::trace::isr_enter();
+                ::embassy_executor::_export::trace::isr_enter();
            }
        },
        quote! {
            ::embassy_executor::rtos_trace_interrupt! {
-                ::embassy_executor::export::trace::isr_exit();
+                ::embassy_executor::_export::trace::isr_exit();
            }
        },
    );
diff --git a/embassy-stm32/build.rs b/embassy-stm32/build.rs
index 3780c5a40..481fec677 100644
--- a/embassy-stm32/build.rs
+++ b/embassy-stm32/build.rs
@@ -50,10 +50,13 @@ fn main() {
                // We *shouldn't* have singletons for these, but the HAL currently requires
                // singletons, for using with RccPeripheral to enable/disable clocks to them.
                "rcc" => {
-                    if r.version.starts_with("h7") {
+                    if r.version.starts_with("h7") || r.version.starts_with("f4") {
                        singletons.push("MCO1".to_string());
                        singletons.push("MCO2".to_string());
                    }
+                    if r.version.starts_with("l4") {
+                        singletons.push("MCO".to_string());
+                    }
                    singletons.push(p.name.to_string());
                }
                //"dbgmcu" => {}
@@ -258,6 +261,7 @@ fn main() {
        (("i2c", "SCL"), quote!(crate::i2c::SclPin)),
        (("rcc", "MCO_1"), quote!(crate::rcc::McoPin)),
        (("rcc", "MCO_2"), quote!(crate::rcc::McoPin)),
+        (("rcc", "MCO"), quote!(crate::rcc::McoPin)),
        (("dcmi", "D0"), quote!(crate::dcmi::D0Pin)),
        (("dcmi", "D1"), quote!(crate::dcmi::D1Pin)),
        (("dcmi", "D2"), quote!(crate::dcmi::D2Pin)),
@@ -447,13 +451,22 @@ fn main() {
                    // MCO is special
                    if pin.signal.starts_with("MCO_") {
                        // Supported in H7 only for now
-                        if regs.version.starts_with("h7") {
+                        if regs.version.starts_with("h7") || regs.version.starts_with("f4") {
                            peri = format_ident!("{}", pin.signal.replace("_", ""));
                        } else {
                            continue;
                        }
                    }
+                    if pin.signal == "MCO" {
+                        // Supported in H7 only for now
+                        if regs.version.starts_with("l4") {
+                            peri = format_ident!("MCO");
+                        } else {
+                            continue;
+                        }
+                    }
                    g.extend(quote! {
                        pin_trait_impl!(#tr, #peri, #pin_name, #af);
                    })
diff --git a/embassy-stm32/src/rcc/f4.rs b/embassy-stm32/src/rcc/f4.rs
index 200bcce9c..2a17eb9b0 100644
--- a/embassy-stm32/src/rcc/f4.rs
+++ b/embassy-stm32/src/rcc/f4.rs
@@ -1,8 +1,16 @@
+use core::marker::PhantomData;
+use embassy_hal_common::into_ref;
+use stm32_metapac::rcc::vals::{Mco1, Mco2, Mcopre};
 use super::sealed::RccPeripheral;
+use crate::gpio::sealed::AFType;
+use crate::gpio::Speed;
 use crate::pac::rcc::vals::{Hpre, Ppre, Sw};
 use crate::pac::{FLASH, PWR, RCC};
 use crate::rcc::{set_freqs, Clocks};
 use crate::time::Hertz;
+use crate::{peripherals, Peripheral};
 /// HSI speed
 pub const HSI_FREQ: Hertz = Hertz(16_000_000);
@@ -96,6 +104,164 @@ unsafe fn setup_pll(pllsrcclk: u32, use_hse: bool, pllsysclk: Option<u32>, pll48
    }
 }
+pub enum McoClock {
+    DIV1,
+    DIV2,
+    DIV3,
+    DIV4,
+    DIV5,
+}
+impl McoClock {
+    fn into_raw(&self) -> Mcopre {
+        match self {
+            McoClock::DIV1 => Mcopre::DIV1,
+            McoClock::DIV2 => Mcopre::DIV2,
+            McoClock::DIV3 => Mcopre::DIV3,
+            McoClock::DIV4 => Mcopre::DIV4,
+            McoClock::DIV5 => Mcopre::DIV5,
+        }
+    }
+}
+#[derive(Copy, Clone)]
+pub enum Mco1Source {
+    Hsi,
+    Lse,
+    Hse,
+    Pll,
+}
+impl Default for Mco1Source {
+    fn default() -> Self {
+        Self::Hsi
+    }
+}
+pub trait McoSource {
+    type Raw;
+    fn into_raw(&self) -> Self::Raw;
+}
+impl McoSource for Mco1Source {
+    type Raw = Mco1;
+    fn into_raw(&self) -> Self::Raw {
+        match self {
+            Mco1Source::Hsi => Mco1::HSI,
+            Mco1Source::Lse => Mco1::LSE,
+            Mco1Source::Hse => Mco1::HSE,
+            Mco1Source::Pll => Mco1::PLL,
+        }
+    }
+}
+#[derive(Copy, Clone)]
+pub enum Mco2Source {
+    SysClk,
+    Plli2s,
+    Hse,
+    Pll,
+}
+impl Default for Mco2Source {
+    fn default() -> Self {
+        Self::SysClk
+    }
+}
+impl McoSource for Mco2Source {
+    type Raw = Mco2;
+    fn into_raw(&self) -> Self::Raw {
+        match self {
+            Mco2Source::SysClk => Mco2::SYSCLK,
+            Mco2Source::Plli2s => Mco2::PLLI2S,
+            Mco2Source::Hse => Mco2::HSE,
+            Mco2Source::Pll => Mco2::PLL,
+        }
+    }
+}
+pub(crate) mod sealed {
+    use stm32_metapac::rcc::vals::Mcopre;
+    pub trait McoInstance {
+        type Source;
+        unsafe fn apply_clock_settings(source: Self::Source, prescaler: Mcopre);
+    }
+}
+pub trait McoInstance: sealed::McoInstance + 'static {}
+pin_trait!(McoPin, McoInstance);
+impl sealed::McoInstance for peripherals::MCO1 {
+    type Source = Mco1;
+    unsafe fn apply_clock_settings(source: Self::Source, prescaler: Mcopre) {
+        RCC.cfgr().modify(|w| {
+            w.set_mco1(source);
+            w.set_mco1pre(prescaler);
+        });
+        match source {
+            Mco1::PLL => {
+                RCC.cr().modify(|w| w.set_pllon(true));
+                while !RCC.cr().read().pllrdy() {}
+            }
+            Mco1::HSI => {
+                RCC.cr().modify(|w| w.set_hsion(true));
+                while !RCC.cr().read().hsirdy() {}
+            }
+            _ => {}
+        }
+    }
+}
+impl McoInstance for peripherals::MCO1 {}
+impl sealed::McoInstance for peripherals::MCO2 {
+    type Source = Mco2;
+    unsafe fn apply_clock_settings(source: Self::Source, prescaler: Mcopre) {
+        RCC.cfgr().modify(|w| {
+            w.set_mco2(source);
+            w.set_mco2pre(prescaler);
+        });
+        match source {
+            Mco2::PLL => {
+                RCC.cr().modify(|w| w.set_pllon(true));
+                while !RCC.cr().read().pllrdy() {}
+            }
+            #[cfg(not(stm32f410))]
+            Mco2::PLLI2S => {
+                RCC.cr().modify(|w| w.set_plli2son(true));
+                while !RCC.cr().read().plli2srdy() {}
+            }
+            _ => {}
+        }
+    }
+}
+impl McoInstance for peripherals::MCO2 {}
+pub struct Mco<'d, T: McoInstance> {
+    phantom: PhantomData<&'d mut T>,
+}
+impl<'d, T: McoInstance> Mco<'d, T> {
+    pub fn new(
+        _peri: impl Peripheral<P = T> + 'd,
+        pin: impl Peripheral<P = impl McoPin<T>> + 'd,
+        source: impl McoSource<Raw = T::Source>,
+        prescaler: McoClock,
+    ) -> Self {
+        into_ref!(pin);
+        critical_section::with(|_| unsafe {
+            T::apply_clock_settings(source.into_raw(), prescaler.into_raw());
+            pin.set_as_af(pin.af_num(), AFType::OutputPushPull);
+            pin.set_speed(Speed::VeryHigh);
+        });
+        Self { phantom: PhantomData }
+    }
+}
 unsafe fn flash_setup(sysclk: u32) {
    use crate::pac::flash::vals::Latency;
diff --git a/embassy-stm32/src/rcc/l4.rs b/embassy-stm32/src/rcc/l4.rs
index e650490fe..c1bf7d0cd 100644
--- a/embassy-stm32/src/rcc/l4.rs
+++ b/embassy-stm32/src/rcc/l4.rs
@@ -1,7 +1,15 @@
+use core::marker::PhantomData;
+use embassy_hal_common::into_ref;
+use stm32_metapac::rcc::vals::{Mcopre, Mcosel};
+use crate::gpio::sealed::AFType;
+use crate::gpio::Speed;
 use crate::pac::rcc::vals::{Hpre, Msirange, Pllsrc, Ppre, Sw};
 use crate::pac::{FLASH, RCC};
 use crate::rcc::{set_freqs, Clocks};
 use crate::time::Hertz;
+use crate::{peripherals, Peripheral};
 /// HSI speed
 pub const HSI_FREQ: Hertz = Hertz(16_000_000);
@@ -298,6 +306,131 @@ impl Default for Config {
    }
 }
+pub enum McoClock {
+    DIV1,
+    DIV2,
+    DIV4,
+    DIV8,
+    DIV16,
+}
+impl McoClock {
+    fn into_raw(&self) -> Mcopre {
+        match self {
+            McoClock::DIV1 => Mcopre::DIV1,
+            McoClock::DIV2 => Mcopre::DIV2,
+            McoClock::DIV4 => Mcopre::DIV4,
+            McoClock::DIV8 => Mcopre::DIV8,
+            McoClock::DIV16 => Mcopre::DIV16,
+        }
+    }
+}
+#[derive(Copy, Clone)]
+pub enum Mco1Source {
+    Disabled,
+    Lse,
+    Lsi,
+    Hse,
+    Hsi16,
+    PllClk,
+    SysClk,
+    Msi,
+    #[cfg(not(any(stm32l471, stm32l475, stm32l476, stm32l486)))]
+    Hsi48,
+}
+impl Default for Mco1Source {
+    fn default() -> Self {
+        Self::Hsi16
+    }
+}
+pub trait McoSource {
+    type Raw;
+    fn into_raw(&self) -> Self::Raw;
+}
+impl McoSource for Mco1Source {
+    type Raw = Mcosel;
+    fn into_raw(&self) -> Self::Raw {
+        match self {
+            Mco1Source::Disabled => Mcosel::NOCLOCK,
+            Mco1Source::Lse => Mcosel::LSE,
+            Mco1Source::Lsi => Mcosel::LSI,
+            Mco1Source::Hse => Mcosel::HSE,
+            Mco1Source::Hsi16 => Mcosel::HSI16,
+            Mco1Source::PllClk => Mcosel::PLL,
+            Mco1Source::SysClk => Mcosel::SYSCLK,
+            Mco1Source::Msi => Mcosel::MSI,
+            #[cfg(not(any(stm32l471, stm32l475, stm32l476, stm32l486)))]
+            Mco1Source::Hsi48 => Mcosel::HSI48,
+        }
+    }
+}
+pub(crate) mod sealed {
+    use stm32_metapac::rcc::vals::Mcopre;
+    pub trait McoInstance {
+        type Source;
+        unsafe fn apply_clock_settings(source: Self::Source, prescaler: Mcopre);
+    }
+}
+pub trait McoInstance: sealed::McoInstance + 'static {}
+pin_trait!(McoPin, McoInstance);
+impl sealed::McoInstance for peripherals::MCO {
+    type Source = Mcosel;
+    unsafe fn apply_clock_settings(source: Self::Source, prescaler: Mcopre) {
+        RCC.cfgr().modify(|w| {
+            w.set_mcosel(source);
+            w.set_mcopre(prescaler);
+        });
+        match source {
+            Mcosel::HSI16 => {
+                RCC.cr().modify(|w| w.set_hsion(true));
+                while !RCC.cr().read().hsirdy() {}
+            }
+            #[cfg(not(any(stm32l471, stm32l475, stm32l476, stm32l486)))]
+            Mcosel::HSI48 => {
+                RCC.crrcr().modify(|w| w.set_hsi48on(true));
+                while !RCC.crrcr().read().hsi48rdy() {}
+            }
+            _ => {}
+        }
+    }
+}
+impl McoInstance for peripherals::MCO {}
+pub struct Mco<'d, T: McoInstance> {
+    phantom: PhantomData<&'d mut T>,
+}
+impl<'d, T: McoInstance> Mco<'d, T> {
+    pub fn new(
+        _peri: impl Peripheral<P = T> + 'd,
+        pin: impl Peripheral<P = impl McoPin<T>> + 'd,
+        source: impl McoSource<Raw = T::Source>,
+        prescaler: McoClock,
+    ) -> Self {
+        into_ref!(pin);
+        critical_section::with(|_| unsafe {
+            T::apply_clock_settings(source.into_raw(), prescaler.into_raw());
+            pin.set_as_af(pin.af_num(), AFType::OutputPushPull);
+            pin.set_speed(Speed::VeryHigh);
+        });
+        Self { phantom: PhantomData }
+    }
+}
 pub(crate) unsafe fn init(config: Config) {
    let (sys_clk, sw) = match config.mux {
        ClockSrc::MSI(range) => {
diff --git a/embassy-stm32/src/usart/buffered.rs b/embassy-stm32/src/usart/buffered.rs
index cd7d72f91..3e23e7ca1 100644
--- a/embassy-stm32/src/usart/buffered.rs
+++ b/embassy-stm32/src/usart/buffered.rs
@@ -1,55 +1,51 @@
-use core::cell::RefCell;
 use core::future::poll_fn;
-use core::sync::atomic::{compiler_fence, Ordering};
+use core::slice;
 use core::task::Poll;
-use embassy_cortex_m::peripheral::{PeripheralMutex, PeripheralState, StateStorage};
+use embassy_cortex_m::interrupt::Interrupt;
-use embassy_hal_common::ring_buffer::RingBuffer;
+use embassy_hal_common::atomic_ring_buffer::RingBuffer;
-use embassy_sync::waitqueue::WakerRegistration;
+use embassy_sync::waitqueue::AtomicWaker;
 use super::*;
-pub struct State<'d, T: BasicInstance>(StateStorage<StateInner<'d, T>>);
+pub struct State {
-impl<'d, T: BasicInstance> State<'d, T> {
+    rx_waker: AtomicWaker,
-    pub const fn new() -> Self {
+    rx_buf: RingBuffer,
-        Self(StateStorage::new())
-    }
-}
-struct StateInner<'d, T: BasicInstance> {
-    phantom: PhantomData<&'d mut T>,
-    rx_waker: WakerRegistration,
-    rx: RingBuffer<'d>,
-    tx_waker: WakerRegistration,
+    tx_waker: AtomicWaker,
-    tx: RingBuffer<'d>,
+    tx_buf: RingBuffer,
 }
-unsafe impl<'d, T: BasicInstance> Send for StateInner<'d, T> {}
+impl State {
-unsafe impl<'d, T: BasicInstance> Sync for StateInner<'d, T> {}
+    pub const fn new() -> Self {
+        Self {
+            rx_buf: RingBuffer::new(),
+            tx_buf: RingBuffer::new(),
+            rx_waker: AtomicWaker::new(),
+            tx_waker: AtomicWaker::new(),
+        }
+    }
+}
 pub struct BufferedUart<'d, T: BasicInstance> {
-    inner: RefCell<PeripheralMutex<'d, StateInner<'d, T>>>,
+    rx: BufferedUartRx<'d, T>,
+    tx: BufferedUartTx<'d, T>,
 }
-pub struct BufferedUartTx<'u, 'd, T: BasicInstance> {
+pub struct BufferedUartTx<'d, T: BasicInstance> {
-    inner: &'u BufferedUart<'d, T>,
+    phantom: PhantomData<&'d mut T>,
 }
-pub struct BufferedUartRx<'u, 'd, T: BasicInstance> {
+pub struct BufferedUartRx<'d, T: BasicInstance> {
-    inner: &'u BufferedUart<'d, T>,
+    phantom: PhantomData<&'d mut T>,
 }
-impl<'d, T: BasicInstance> Unpin for BufferedUart<'d, T> {}
 impl<'d, T: BasicInstance> BufferedUart<'d, T> {
    pub fn new(
-        state: &'d mut State<'d, T>,
        peri: impl Peripheral<P = T> + 'd,
+        irq: impl Peripheral<P = T::Interrupt> + 'd,
        rx: impl Peripheral<P = impl RxPin<T>> + 'd,
        tx: impl Peripheral<P = impl TxPin<T>> + 'd,
-        irq: impl Peripheral<P = T::Interrupt> + 'd,
        tx_buffer: &'d mut [u8],
        rx_buffer: &'d mut [u8],
        config: Config,
@@ -57,15 +53,14 @@ impl<'d, T: BasicInstance> BufferedUart<'d, T> {
        T::enable();
        T::reset();
-        Self::new_inner(state, peri, rx, tx, irq, tx_buffer, rx_buffer, config)
+        Self::new_inner(peri, irq, rx, tx, tx_buffer, rx_buffer, config)
    }
    pub fn new_with_rtscts(
-        state: &'d mut State<'d, T>,
        peri: impl Peripheral<P = T> + 'd,
+        irq: impl Peripheral<P = T::Interrupt> + 'd,
        rx: impl Peripheral<P = impl RxPin<T>> + 'd,
        tx: impl Peripheral<P = impl TxPin<T>> + 'd,
-        irq: impl Peripheral<P = T::Interrupt> + 'd,
        rts: impl Peripheral<P = impl RtsPin<T>> + 'd,
        cts: impl Peripheral<P = impl CtsPin<T>> + 'd,
        tx_buffer: &'d mut [u8],
@@ -86,16 +81,15 @@ impl<'d, T: BasicInstance> BufferedUart<'d, T> {
            });
        }
-        Self::new_inner(state, peri, rx, tx, irq, tx_buffer, rx_buffer, config)
+        Self::new_inner(peri, irq, rx, tx, tx_buffer, rx_buffer, config)
    }
    #[cfg(not(usart_v1))]
    pub fn new_with_de(
-        state: &'d mut State<'d, T>,
        peri: impl Peripheral<P = T> + 'd,
+        irq: impl Peripheral<P = T::Interrupt> + 'd,
        rx: impl Peripheral<P = impl RxPin<T>> + 'd,
        tx: impl Peripheral<P = impl TxPin<T>> + 'd,
-        irq: impl Peripheral<P = T::Interrupt> + 'd,
        de: impl Peripheral<P = impl DePin<T>> + 'd,
        tx_buffer: &'d mut [u8],
        rx_buffer: &'d mut [u8],
@@ -113,23 +107,27 @@ impl<'d, T: BasicInstance> BufferedUart<'d, T> {
            });
        }
-        Self::new_inner(state, peri, rx, tx, irq, tx_buffer, rx_buffer, config)
+        Self::new_inner(peri, irq, rx, tx, tx_buffer, rx_buffer, config)
    }
    fn new_inner(
-        state: &'d mut State<'d, T>,
        _peri: impl Peripheral<P = T> + 'd,
+        irq: impl Peripheral<P = T::Interrupt> + 'd,
        rx: impl Peripheral<P = impl RxPin<T>> + 'd,
        tx: impl Peripheral<P = impl TxPin<T>> + 'd,
-        irq: impl Peripheral<P = T::Interrupt> + 'd,
        tx_buffer: &'d mut [u8],
        rx_buffer: &'d mut [u8],
        config: Config,
    ) -> BufferedUart<'d, T> {
        into_ref!(_peri, rx, tx, irq);
-        let r = T::regs();
+        let state = T::buffered_state();
+        let len = tx_buffer.len();
+        unsafe { state.tx_buf.init(tx_buffer.as_mut_ptr(), len) };
+        let len = rx_buffer.len();
+        unsafe { state.rx_buf.init(rx_buffer.as_mut_ptr(), len) };
+        let r = T::regs();
        unsafe {
            rx.set_as_af(rx.af_num(), AFType::Input);
            tx.set_as_af(tx.af_num(), AFType::OutputPushPull);
@@ -147,273 +145,259 @@ impl<'d, T: BasicInstance> BufferedUart<'d, T> {
            });
        }
-        Self {
+        irq.set_handler(on_interrupt::<T>);
-            inner: RefCell::new(PeripheralMutex::new(irq, &mut state.0, move || StateInner {
+        irq.unpend();
-                phantom: PhantomData,
+        irq.enable();
-                tx: RingBuffer::new(tx_buffer),
-                tx_waker: WakerRegistration::new(),
-                rx: RingBuffer::new(rx_buffer),
+        Self {
-                rx_waker: WakerRegistration::new(),
+            rx: BufferedUartRx { phantom: PhantomData },
-            })),
+            tx: BufferedUartTx { phantom: PhantomData },
        }
    }
-    pub fn split<'u>(&'u mut self) -> (BufferedUartRx<'u, 'd, T>, BufferedUartTx<'u, 'd, T>) {
+    pub fn split(self) -> (BufferedUartTx<'d, T>, BufferedUartRx<'d, T>) {
-        (BufferedUartRx { inner: self }, BufferedUartTx { inner: self })
+        (self.tx, self.rx)
    }
+}
-    async fn inner_read<'a>(&'a self, buf: &'a mut [u8]) -> Result<usize, Error> {
+impl<'d, T: BasicInstance> BufferedUartRx<'d, T> {
+    async fn read(&self, buf: &mut [u8]) -> Result<usize, Error> {
        poll_fn(move |cx| {
-            let mut do_pend = false;
+            let state = T::buffered_state();
-            let mut inner = self.inner.borrow_mut();
+            let mut rx_reader = unsafe { state.rx_buf.reader() };
-            let res = inner.with(|state| {
+            let data = rx_reader.pop_slice();
-                compiler_fence(Ordering::SeqCst);
-                // We have data ready in buffer? Return it.
-                let data = state.rx.pop_buf();
-                if !data.is_empty() {
-                    let len = data.len().min(buf.len());
-                    buf[..len].copy_from_slice(&data[..len]);
-                    if state.rx.is_full() {
-                        do_pend = true;
-                    }
-                    state.rx.pop(len);
-                    return Poll::Ready(Ok(len));
-                }
-                state.rx_waker.register(cx.waker());
+            if !data.is_empty() {
-                Poll::Pending
+                let len = data.len().min(buf.len());
-            });
+                buf[..len].copy_from_slice(&data[..len]);
+                let do_pend = state.rx_buf.is_full();
+                rx_reader.pop_done(len);
-            if do_pend {
+                if do_pend {
-                inner.pend();
+                    unsafe { T::Interrupt::steal().pend() };
+                }
+                return Poll::Ready(Ok(len));
            }
-            res
+            state.rx_waker.register(cx.waker());
+            Poll::Pending
        })
        .await
    }
-    fn inner_blocking_read(&self, buf: &mut [u8]) -> Result<usize, Error> {
+    fn blocking_read(&self, buf: &mut [u8]) -> Result<usize, Error> {
        loop {
-            let mut do_pend = false;
+            let state = T::buffered_state();
-            let mut inner = self.inner.borrow_mut();
+            let mut rx_reader = unsafe { state.rx_buf.reader() };
-            let n = inner.with(|state| {
+            let data = rx_reader.pop_slice();
-                compiler_fence(Ordering::SeqCst);
-                // We have data ready in buffer? Return it.
-                let data = state.rx.pop_buf();
-                if !data.is_empty() {
-                    let len = data.len().min(buf.len());
-                    buf[..len].copy_from_slice(&data[..len]);
-                    if state.rx.is_full() {
-                        do_pend = true;
-                    }
-                    state.rx.pop(len);
-                    return len;
-                }
-                0
+            if !data.is_empty() {
-            });
+                let len = data.len().min(buf.len());
+                buf[..len].copy_from_slice(&data[..len]);
+                let do_pend = state.rx_buf.is_full();
+                rx_reader.pop_done(len);
+                if do_pend {
+                    unsafe { T::Interrupt::steal().pend() };
+                }
-            if do_pend {
+                return Ok(len);
-                inner.pend();
            }
+        }
+    }
-            if n > 0 {
+    async fn fill_buf(&self) -> Result<&[u8], Error> {
-                return Ok(n);
+        poll_fn(move |cx| {
+            let state = T::buffered_state();
+            let mut rx_reader = unsafe { state.rx_buf.reader() };
+            let (p, n) = rx_reader.pop_buf();
+            if n == 0 {
+                state.rx_waker.register(cx.waker());
+                return Poll::Pending;
            }
+            let buf = unsafe { slice::from_raw_parts(p, n) };
+            Poll::Ready(Ok(buf))
+        })
+        .await
+    }
+    fn consume(&self, amt: usize) {
+        let state = T::buffered_state();
+        let mut rx_reader = unsafe { state.rx_buf.reader() };
+        let full = state.rx_buf.is_full();
+        rx_reader.pop_done(amt);
+        if full {
+            unsafe { T::Interrupt::steal().pend() };
        }
    }
+}
-    async fn inner_write<'a>(&'a self, buf: &'a [u8]) -> Result<usize, Error> {
+impl<'d, T: BasicInstance> BufferedUartTx<'d, T> {
+    async fn write(&self, buf: &[u8]) -> Result<usize, Error> {
        poll_fn(move |cx| {
-            let mut inner = self.inner.borrow_mut();
+            let state = T::buffered_state();
-            let (poll, empty) = inner.with(|state| {
+            let empty = state.tx_buf.is_empty();
-                let empty = state.tx.is_empty();
-                let tx_buf = state.tx.push_buf();
+            let mut tx_writer = unsafe { state.tx_buf.writer() };
-                if tx_buf.is_empty() {
+            let data = tx_writer.push_slice();
-                    state.tx_waker.register(cx.waker());
+            if data.is_empty() {
-                    return (Poll::Pending, empty);
+                state.tx_waker.register(cx.waker());
-                }
+                return Poll::Pending;
+            }
-                let n = core::cmp::min(tx_buf.len(), buf.len());
+            let n = data.len().min(buf.len());
-                tx_buf[..n].copy_from_slice(&buf[..n]);
+            data[..n].copy_from_slice(&buf[..n]);
-                state.tx.push(n);
+            tx_writer.push_done(n);
-                (Poll::Ready(Ok(n)), empty)
-            });
            if empty {
-                inner.pend();
+                unsafe { T::Interrupt::steal() }.pend();
            }
-            poll
+            Poll::Ready(Ok(n))
        })
        .await
    }
-    async fn inner_flush<'a>(&'a self) -> Result<(), Error> {
+    async fn flush(&self) -> Result<(), Error> {
        poll_fn(move |cx| {
-            self.inner.borrow_mut().with(|state| {
+            let state = T::buffered_state();
-                if !state.tx.is_empty() {
+            if !state.tx_buf.is_empty() {
-                    state.tx_waker.register(cx.waker());
+                state.tx_waker.register(cx.waker());
-                    return Poll::Pending;
+                return Poll::Pending;
-                }
+            }
-                Poll::Ready(Ok(()))
+            Poll::Ready(Ok(()))
-            })
        })
        .await
    }
-    fn inner_blocking_write(&self, buf: &[u8]) -> Result<usize, Error> {
+    fn blocking_write(&self, buf: &[u8]) -> Result<usize, Error> {
        loop {
-            let mut inner = self.inner.borrow_mut();
+            let state = T::buffered_state();
-            let (n, empty) = inner.with(|state| {
+            let empty = state.tx_buf.is_empty();
-                let empty = state.tx.is_empty();
-                let tx_buf = state.tx.push_buf();
+            let mut tx_writer = unsafe { state.tx_buf.writer() };
-                if tx_buf.is_empty() {
+            let data = tx_writer.push_slice();
-                    return (0, empty);
+            if !data.is_empty() {
+                let n = data.len().min(buf.len());
+                data[..n].copy_from_slice(&buf[..n]);
+                tx_writer.push_done(n);
+                if empty {
+                    unsafe { T::Interrupt::steal() }.pend();
                }
-                let n = core::cmp::min(tx_buf.len(), buf.len());
-                tx_buf[..n].copy_from_slice(&buf[..n]);
-                state.tx.push(n);
-                (n, empty)
-            });
-            if empty {
-                inner.pend();
-            }
-            if n != 0 {
                return Ok(n);
            }
        }
    }
-    fn inner_blocking_flush(&self) -> Result<(), Error> {
+    fn blocking_flush(&self) -> Result<(), Error> {
        loop {
-            if !self.inner.borrow_mut().with(|state| state.tx.is_empty()) {
+            let state = T::buffered_state();
+            if state.tx_buf.is_empty() {
                return Ok(());
            }
        }
    }
+}
-    async fn inner_fill_buf<'a>(&'a self) -> Result<&'a [u8], Error> {
+impl<'d, T: BasicInstance> Drop for BufferedUartRx<'d, T> {
-        poll_fn(move |cx| {
+    fn drop(&mut self) {
-            self.inner.borrow_mut().with(|state| {
+        let state = T::buffered_state();
-                compiler_fence(Ordering::SeqCst);
+        unsafe {
+            state.rx_buf.deinit();
-                // We have data ready in buffer? Return it.
-                let buf = state.rx.pop_buf();
-                if !buf.is_empty() {
-                    let buf: &[u8] = buf;
-                    // Safety: buffer lives as long as uart
-                    let buf: &[u8] = unsafe { core::mem::transmute(buf) };
-                    return Poll::Ready(Ok(buf));
-                }
-                state.rx_waker.register(cx.waker());
-                Poll::<Result<&[u8], Error>>::Pending
-            })
-        })
-        .await
-    }
-    fn inner_consume(&self, amt: usize) {
+            // TX is inactive if the the buffer is not available.
-        let mut inner = self.inner.borrow_mut();
+            // We can now unregister the interrupt handler
-        let signal = inner.with(|state| {
+            if state.tx_buf.len() == 0 {
-            let full = state.rx.is_full();
+                T::Interrupt::steal().disable();
-            state.rx.pop(amt);
+            }
-            full
-        });
-        if signal {
-            inner.pend();
        }
    }
 }
-impl<'d, T: BasicInstance> StateInner<'d, T>
+impl<'d, T: BasicInstance> Drop for BufferedUartTx<'d, T> {
-where
+    fn drop(&mut self) {
-    Self: 'd,
+        let state = T::buffered_state();
-{
-    fn on_rx(&mut self) {
-        let r = T::regs();
        unsafe {
-            let sr = sr(r).read();
+            state.tx_buf.deinit();
-            clear_interrupt_flags(r, sr);
-            // This read also clears the error and idle interrupt flags on v1.
+            // RX is inactive if the the buffer is not available.
-            let b = rdr(r).read_volatile();
+            // We can now unregister the interrupt handler
+            if state.rx_buf.len() == 0 {
+                T::Interrupt::steal().disable();
+            }
+        }
+    }
+}
-            if sr.rxne() {
+unsafe fn on_interrupt<T: BasicInstance>(_: *mut ()) {
-                if sr.pe() {
+    let r = T::regs();
-                    warn!("Parity error");
+    let state = T::buffered_state();
-                }
-                if sr.fe() {
-                    warn!("Framing error");
-                }
-                if sr.ne() {
-                    warn!("Noise error");
-                }
-                if sr.ore() {
-                    warn!("Overrun error");
-                }
-                let buf = self.rx.push_buf();
+    // RX
-                if !buf.is_empty() {
+    unsafe {
-                    buf[0] = b;
+        let sr = sr(r).read();
-                    self.rx.push(1);
+        clear_interrupt_flags(r, sr);
-                } else {
-                    warn!("RX buffer full, discard received byte");
-                }
-                if self.rx.is_full() {
+        if sr.rxne() {
-                    self.rx_waker.wake();
+            if sr.pe() {
-                }
+                warn!("Parity error");
+            }
+            if sr.fe() {
+                warn!("Framing error");
+            }
+            if sr.ne() {
+                warn!("Noise error");
+            }
+            if sr.ore() {
+                warn!("Overrun error");
            }
-            if sr.idle() {
+            let mut rx_writer = state.rx_buf.writer();
-                self.rx_waker.wake();
+            let buf = rx_writer.push_slice();
-            };
+            if !buf.is_empty() {
-        }
+                // This read also clears the error and idle interrupt flags on v1.
-    }
+                buf[0] = rdr(r).read_volatile();
+                rx_writer.push_done(1);
+            } else {
+                // FIXME: Should we disable any further RX interrupts when the buffer becomes full.
+            }
-    fn on_tx(&mut self) {
+            if state.rx_buf.is_full() {
-        let r = T::regs();
+                state.rx_waker.wake();
-        unsafe {
-            if sr(r).read().txe() {
-                let buf = self.tx.pop_buf();
-                if !buf.is_empty() {
-                    r.cr1().modify(|w| {
-                        w.set_txeie(true);
-                    });
-                    tdr(r).write_volatile(buf[0].into());
-                    self.tx.pop(1);
-                    self.tx_waker.wake();
-                } else {
-                    // Disable interrupt until we have something to transmit again
-                    r.cr1().modify(|w| {
-                        w.set_txeie(false);
-                    });
-                }
            }
        }
-    }
-}
-impl<'d, T: BasicInstance> PeripheralState for StateInner<'d, T>
+        if sr.idle() {
-where
+            state.rx_waker.wake();
-    Self: 'd,
+        };
-{
+    }
-    type Interrupt = T::Interrupt;
-    fn on_interrupt(&mut self) {
+    // TX
-        self.on_rx();
+    unsafe {
-        self.on_tx();
+        if sr(r).read().txe() {
+            let mut tx_reader = state.tx_buf.reader();
+            let buf = tx_reader.pop_slice();
+            if !buf.is_empty() {
+                r.cr1().modify(|w| {
+                    w.set_txeie(true);
+                });
+                tdr(r).write_volatile(buf[0].into());
+                tx_reader.pop_done(1);
+                state.tx_waker.wake();
+            } else {
+                // Disable interrupt until we have something to transmit again
+                r.cr1().modify(|w| {
+                    w.set_txeie(false);
+                });
+            }
+        }
    }
 }
@@ -427,94 +411,284 @@ impl<'d, T: BasicInstance> embedded_io::Io for BufferedUart<'d, T> {
    type Error = Error;
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::Io for BufferedUartRx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::Io for BufferedUartRx<'d, T> {
    type Error = Error;
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::Io for BufferedUartTx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::Io for BufferedUartTx<'d, T> {
    type Error = Error;
 }
 impl<'d, T: BasicInstance> embedded_io::asynch::Read for BufferedUart<'d, T> {
    async fn read(&mut self, buf: &mut [u8]) -> Result<usize, Self::Error> {
-        self.inner_read(buf).await
+        self.rx.read(buf).await
    }
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::asynch::Read for BufferedUartRx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::asynch::Read for BufferedUartRx<'d, T> {
    async fn read(&mut self, buf: &mut [u8]) -> Result<usize, Self::Error> {
-        self.inner.inner_read(buf).await
+        Self::read(self, buf).await
    }
 }
 impl<'d, T: BasicInstance> embedded_io::asynch::BufRead for BufferedUart<'d, T> {
    async fn fill_buf(&mut self) -> Result<&[u8], Self::Error> {
-        self.inner_fill_buf().await
+        self.rx.fill_buf().await
    }
    fn consume(&mut self, amt: usize) {
-        self.inner_consume(amt)
+        self.rx.consume(amt)
    }
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::asynch::BufRead for BufferedUartRx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::asynch::BufRead for BufferedUartRx<'d, T> {
    async fn fill_buf(&mut self) -> Result<&[u8], Self::Error> {
-        self.inner.inner_fill_buf().await
+        Self::fill_buf(self).await
    }
    fn consume(&mut self, amt: usize) {
-        self.inner.inner_consume(amt)
+        Self::consume(self, amt)
    }
 }
 impl<'d, T: BasicInstance> embedded_io::asynch::Write for BufferedUart<'d, T> {
    async fn write(&mut self, buf: &[u8]) -> Result<usize, Self::Error> {
-        self.inner_write(buf).await
+        self.tx.write(buf).await
    }
    async fn flush(&mut self) -> Result<(), Self::Error> {
-        self.inner_flush().await
+        self.tx.flush().await
    }
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::asynch::Write for BufferedUartTx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::asynch::Write for BufferedUartTx<'d, T> {
    async fn write(&mut self, buf: &[u8]) -> Result<usize, Self::Error> {
-        self.inner.inner_write(buf).await
+        Self::write(self, buf).await
    }
    async fn flush(&mut self) -> Result<(), Self::Error> {
-        self.inner.inner_flush().await
+        Self::flush(self).await
    }
 }
 impl<'d, T: BasicInstance> embedded_io::blocking::Read for BufferedUart<'d, T> {
    fn read(&mut self, buf: &mut [u8]) -> Result<usize, Self::Error> {
-        self.inner_blocking_read(buf)
+        self.rx.blocking_read(buf)
    }
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::blocking::Read for BufferedUartRx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::blocking::Read for BufferedUartRx<'d, T> {
    fn read(&mut self, buf: &mut [u8]) -> Result<usize, Self::Error> {
-        self.inner.inner_blocking_read(buf)
+        self.blocking_read(buf)
    }
 }
 impl<'d, T: BasicInstance> embedded_io::blocking::Write for BufferedUart<'d, T> {
    fn write(&mut self, buf: &[u8]) -> Result<usize, Self::Error> {
-        self.inner_blocking_write(buf)
+        self.tx.blocking_write(buf)
    }
    fn flush(&mut self) -> Result<(), Self::Error> {
-        self.inner_blocking_flush()
+        self.tx.blocking_flush()
    }
 }
-impl<'u, 'd, T: BasicInstance> embedded_io::blocking::Write for BufferedUartTx<'u, 'd, T> {
+impl<'d, T: BasicInstance> embedded_io::blocking::Write for BufferedUartTx<'d, T> {
    fn write(&mut self, buf: &[u8]) -> Result<usize, Self::Error> {
-        self.inner.inner_blocking_write(buf)
+        Self::blocking_write(self, buf)
    }
    fn flush(&mut self) -> Result<(), Self::Error> {
-        self.inner.inner_blocking_flush()
+        Self::blocking_flush(self)
+    }
+}
+mod eh02 {
+    use super::*;
+    impl<'d, T: BasicInstance> embedded_hal_02::serial::Read<u8> for BufferedUartRx<'d, T> {
+        type Error = Error;
+        fn read(&mut self) -> Result<u8, nb::Error<Self::Error>> {
+            let r = T::regs();
+            unsafe {
+                let sr = sr(r).read();
+                if sr.pe() {
+                    rdr(r).read_volatile();
+                    Err(nb::Error::Other(Error::Parity))
+                } else if sr.fe() {
+                    rdr(r).read_volatile();
+                    Err(nb::Error::Other(Error::Framing))
+                } else if sr.ne() {
+                    rdr(r).read_volatile();
+                    Err(nb::Error::Other(Error::Noise))
+                } else if sr.ore() {
+                    rdr(r).read_volatile();
+                    Err(nb::Error::Other(Error::Overrun))
+                } else if sr.rxne() {
+                    Ok(rdr(r).read_volatile())
+                } else {
+                    Err(nb::Error::WouldBlock)
+                }
+            }
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_02::blocking::serial::Write<u8> for BufferedUartTx<'d, T> {
+        type Error = Error;
+        fn bwrite_all(&mut self, mut buffer: &[u8]) -> Result<(), Self::Error> {
+            while !buffer.is_empty() {
+                match self.blocking_write(buffer) {
+                    Ok(0) => panic!("zero-length write."),
+                    Ok(n) => buffer = &buffer[n..],
+                    Err(e) => return Err(e),
+                }
+            }
+            Ok(())
+        }
+        fn bflush(&mut self) -> Result<(), Self::Error> {
+            self.blocking_flush()
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_02::serial::Read<u8> for BufferedUart<'d, T> {
+        type Error = Error;
+        fn read(&mut self) -> Result<u8, nb::Error<Self::Error>> {
+            embedded_hal_02::serial::Read::read(&mut self.rx)
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_02::blocking::serial::Write<u8> for BufferedUart<'d, T> {
+        type Error = Error;
+        fn bwrite_all(&mut self, mut buffer: &[u8]) -> Result<(), Self::Error> {
+            while !buffer.is_empty() {
+                match self.tx.blocking_write(buffer) {
+                    Ok(0) => panic!("zero-length write."),
+                    Ok(n) => buffer = &buffer[n..],
+                    Err(e) => return Err(e),
+                }
+            }
+            Ok(())
+        }
+        fn bflush(&mut self) -> Result<(), Self::Error> {
+            self.tx.blocking_flush()
+        }
+    }
+}
+#[cfg(feature = "unstable-traits")]
+mod eh1 {
+    use super::*;
+    impl<'d, T: BasicInstance> embedded_hal_1::serial::ErrorType for BufferedUart<'d, T> {
+        type Error = Error;
+    }
+    impl<'d, T: BasicInstance> embedded_hal_1::serial::ErrorType for BufferedUartTx<'d, T> {
+        type Error = Error;
+    }
+    impl<'d, T: BasicInstance> embedded_hal_1::serial::ErrorType for BufferedUartRx<'d, T> {
+        type Error = Error;
+    }
+    impl<'d, T: BasicInstance> embedded_hal_nb::serial::Read for BufferedUartRx<'d, T> {
+        fn read(&mut self) -> nb::Result<u8, Self::Error> {
+            embedded_hal_02::serial::Read::read(self)
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_1::serial::Write for BufferedUartTx<'d, T> {
+        fn write(&mut self, buffer: &[u8]) -> Result<(), Self::Error> {
+            self.blocking_write(buffer).map(drop)
+        }
+        fn flush(&mut self) -> Result<(), Self::Error> {
+            self.blocking_flush()
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_nb::serial::Write for BufferedUartTx<'d, T> {
+        fn write(&mut self, char: u8) -> nb::Result<(), Self::Error> {
+            self.blocking_write(&[char]).map(drop).map_err(nb::Error::Other)
+        }
+        fn flush(&mut self) -> nb::Result<(), Self::Error> {
+            self.blocking_flush().map_err(nb::Error::Other)
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_nb::serial::Read for BufferedUart<'d, T> {
+        fn read(&mut self) -> Result<u8, nb::Error<Self::Error>> {
+            embedded_hal_02::serial::Read::read(&mut self.rx)
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_1::serial::Write for BufferedUart<'d, T> {
+        fn write(&mut self, buffer: &[u8]) -> Result<(), Self::Error> {
+            self.tx.blocking_write(buffer).map(drop)
+        }
+        fn flush(&mut self) -> Result<(), Self::Error> {
+            self.tx.blocking_flush()
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_nb::serial::Write for BufferedUart<'d, T> {
+        fn write(&mut self, char: u8) -> nb::Result<(), Self::Error> {
+            self.tx.blocking_write(&[char]).map(drop).map_err(nb::Error::Other)
+        }
+        fn flush(&mut self) -> nb::Result<(), Self::Error> {
+            self.tx.blocking_flush().map_err(nb::Error::Other)
+        }
+    }
+}
+#[cfg(all(
+    feature = "unstable-traits",
+    feature = "nightly",
+    feature = "_todo_embedded_hal_serial"
+))]
+mod eha {
+    use core::future::Future;
+    use super::*;
+    impl<'d, T: BasicInstance> embedded_hal_async::serial::Write for BufferedUartTx<'d, T> {
+        async fn write(&mut self, buf: &[u8]) -> Result<(), Self::Error> {
+            Self::write(buf)
+        }
+        async fn flush(&mut self) -> Result<(), Self::Error> {
+            Self::flush()
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_async::serial::Read for BufferedUartRx<'d, T> {
+        async fn read(&mut self, buf: &mut [u8]) -> Result<(), Self::Error> {
+            Self::read(buf)
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_async::serial::Write for BufferedUart<'d, T> {
+        async fn write(&mut self, buf: &[u8]) -> Result<(), Self::Error> {
+            self.tx.write(buf)
+        }
+        async fn flush(&mut self) -> Result<(), Self::Error> {
+            self.tx.flush()
+        }
+    }
+    impl<'d, T: BasicInstance> embedded_hal_async::serial::Read for BufferedUart<'d, T> {
+        async fn read(&mut self, buf: &mut [u8]) -> Result<(), Self::Error> {
+            self.rx.read(buf)
+        }
    }
 }
diff --git a/embassy-stm32/src/usart/mod.rs b/embassy-stm32/src/usart/mod.rs
index f80323e37..a42eede18 100644
--- a/embassy-stm32/src/usart/mod.rs
+++ b/embassy-stm32/src/usart/mod.rs
@@ -1112,6 +1112,9 @@ pub(crate) mod sealed {
        fn regs() -> Regs;
        fn state() -> &'static State;
+        #[cfg(feature = "nightly")]
+        fn buffered_state() -> &'static buffered::State;
    }
    pub trait FullInstance: BasicInstance {
@@ -1147,6 +1150,12 @@ macro_rules! impl_lpuart {
                static STATE: crate::usart::sealed::State = crate::usart::sealed::State::new();
                &STATE
            }
+            #[cfg(feature = "nightly")]
+            fn buffered_state() -> &'static buffered::State {
+                static STATE: buffered::State = buffered::State::new();
+                &STATE
+            }
        }
        impl BasicInstance for peripherals::$inst {}
diff --git a/examples/boot/application/nrf/Cargo.toml b/examples/boot/application/nrf/Cargo.toml
index 888993255..e75c73cbd 100644
--- a/examples/boot/application/nrf/Cargo.toml
+++ b/examples/boot/application/nrf/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync" }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers", "arch-cortex-m", "executor-thread"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly"] }
 embassy-nrf = { version = "0.1.0", path = "../../../../embassy-nrf", features = ["time-driver-rtc1", "gpiote", "nightly"] }
 embassy-boot = { version = "0.1.0", path = "../../../../embassy-boot/boot" }
diff --git a/examples/boot/application/rp/Cargo.toml b/examples/boot/application/rp/Cargo.toml
index 8d826790b..8de2d2ebd 100644
--- a/examples/boot/application/rp/Cargo.toml
+++ b/examples/boot/application/rp/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync" }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers", "arch-cortex-m", "executor-thread"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly"] }
 embassy-rp = { version = "0.1.0", path = "../../../../embassy-rp", features = ["time-driver", "unstable-traits", "nightly"] }
 embassy-boot-rp = { version = "0.1.0", path = "../../../../embassy-boot/rp" }
diff --git a/examples/boot/application/stm32f3/Cargo.toml b/examples/boot/application/stm32f3/Cargo.toml
index aa279fb76..083607de5 100644
--- a/examples/boot/application/stm32f3/Cargo.toml
+++ b/examples/boot/application/stm32f3/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32f303re", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32f7/Cargo.toml b/examples/boot/application/stm32f7/Cargo.toml
index 1ec0643a6..74f508515 100644
--- a/examples/boot/application/stm32f7/Cargo.toml
+++ b/examples/boot/application/stm32f7/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32f767zi", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32h7/Cargo.toml b/examples/boot/application/stm32h7/Cargo.toml
index a4eefe2a5..898b9a47e 100644
--- a/examples/boot/application/stm32h7/Cargo.toml
+++ b/examples/boot/application/stm32h7/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync" }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32h743zi", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32l0/Cargo.toml b/examples/boot/application/stm32l0/Cargo.toml
index 36eada29b..e142c8481 100644
--- a/examples/boot/application/stm32l0/Cargo.toml
+++ b/examples/boot/application/stm32l0/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32l072cz", "time-driver-any", "exti", "memory-x"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32l1/Cargo.toml b/examples/boot/application/stm32l1/Cargo.toml
index 67efda748..f0e92e1ac 100644
--- a/examples/boot/application/stm32l1/Cargo.toml
+++ b/examples/boot/application/stm32l1/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32l151cb-a", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32l4/Cargo.toml b/examples/boot/application/stm32l4/Cargo.toml
index 4b2e02dd2..87689e9a9 100644
--- a/examples/boot/application/stm32l4/Cargo.toml
+++ b/examples/boot/application/stm32l4/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32l475vg", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/boot/application/stm32wl/Cargo.toml b/examples/boot/application/stm32wl/Cargo.toml
index fecbfc51d..a6708bf51 100644
--- a/examples/boot/application/stm32wl/Cargo.toml
+++ b/examples/boot/application/stm32wl/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../../../embassy-time", features = ["nightly", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../../../embassy-stm32", features = ["unstable-traits", "nightly", "stm32wl55jc-cm4", "time-driver-any", "exti"]  }
 embassy-boot-stm32 = { version = "0.1.0", path = "../../../../embassy-boot/stm32" }
diff --git a/examples/nrf-rtos-trace/Cargo.toml b/examples/nrf-rtos-trace/Cargo.toml
index d8c24dfad..7910b372a 100644
--- a/examples/nrf-rtos-trace/Cargo.toml
+++ b/examples/nrf-rtos-trace/Cargo.toml
@@ -17,7 +17,7 @@ log = [
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync" }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features=["rtos-trace", "rtos-trace-interrupt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "rtos-trace", "rtos-trace-interrupt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time" }
 embassy-nrf = { version = "0.1.0", path = "../../embassy-nrf", features = ["nrf52840", "time-driver-rtc1", "gpiote", "unstable-pac"] }
diff --git a/examples/nrf52840/Cargo.toml b/examples/nrf52840/Cargo.toml
index cc88d92c7..3ece24066 100644
--- a/examples/nrf52840/Cargo.toml
+++ b/examples/nrf52840/Cargo.toml
@@ -12,7 +12,7 @@ nightly = ["embassy-executor/nightly", "embassy-nrf/nightly", "embassy-net/night
 [dependencies]
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime"] }
 embassy-nrf = { version = "0.1.0", path = "../../embassy-nrf", features = ["defmt", "nrf52840", "time-driver-rtc1", "gpiote", "unstable-pac", "time"] }
 embassy-net = { version = "0.1.0", path = "../../embassy-net", features = ["defmt", "tcp", "dhcpv4", "medium-ethernet"], optional = true }
diff --git a/examples/nrf5340/Cargo.toml b/examples/nrf5340/Cargo.toml
index e88ddf2f7..4134db46f 100644
--- a/examples/nrf5340/Cargo.toml
+++ b/examples/nrf5340/Cargo.toml
@@ -9,7 +9,7 @@ embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = [
    "defmt",
 ] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = [
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", 
    "nightly",
    "defmt",
    "integrated-timers",
diff --git a/examples/rp/Cargo.toml b/examples/rp/Cargo.toml
index 1e8870ed7..aea61eec5 100644
--- a/examples/rp/Cargo.toml
+++ b/examples/rp/Cargo.toml
@@ -7,7 +7,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime"] }
 embassy-rp = { version = "0.1.0", path = "../../embassy-rp", features = ["defmt", "unstable-traits", "nightly", "unstable-pac", "time-driver", "pio", "critical-section-impl"] }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/std/Cargo.toml b/examples/std/Cargo.toml
index 8087df09a..ff08e378c 100644
--- a/examples/std/Cargo.toml
+++ b/examples/std/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["log"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["log", "std", "nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-std", "executor-thread", "log", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["log", "std", "nightly"] }
 embassy-net = { version = "0.1.0", path = "../../embassy-net", features=[ "std", "nightly", "log", "medium-ethernet", "tcp", "udp", "dns", "dhcpv4", "unstable-traits", "proto-ipv6"] }
 embassy-net-driver = { version = "0.1.0", path = "../../embassy-net-driver" }
diff --git a/examples/stm32c0/Cargo.toml b/examples/stm32c0/Cargo.toml
index 0095a680c..3b1d888f6 100644
--- a/examples/stm32c0/Cargo.toml
+++ b/examples/stm32c0/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32c031c6", "memory-x", "unstable-pac", "exti"]  }
diff --git a/examples/stm32f0/Cargo.toml b/examples/stm32f0/Cargo.toml
index 89d99b6d3..5c82c5579 100644
--- a/examples/stm32f0/Cargo.toml
+++ b/examples/stm32f0/Cargo.toml
@@ -13,7 +13,7 @@ defmt = "0.3"
 defmt-rtt = "0.4"
 panic-probe = "0.3"
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "memory-x", "stm32f091rc", "time-driver-any", "exti", "unstable-pac"] }
 static_cell = "1.0"
diff --git a/examples/stm32f0/src/bin/multiprio.rs b/examples/stm32f0/src/bin/multiprio.rs
index e0dc8c989..430a805fc 100644
--- a/examples/stm32f0/src/bin/multiprio.rs
+++ b/examples/stm32f0/src/bin/multiprio.rs
@@ -62,7 +62,7 @@ use core::mem;
 use cortex_m::peripheral::NVIC;
 use cortex_m_rt::entry;
 use defmt::*;
-use embassy_stm32::executor::{Executor, InterruptExecutor};
+use embassy_executor::{Executor, InterruptExecutor};
 use embassy_stm32::interrupt;
 use embassy_stm32::pac::Interrupt;
 use embassy_time::{Duration, Instant, Timer};
diff --git a/examples/stm32f1/Cargo.toml b/examples/stm32f1/Cargo.toml
index 53f369b3a..387af783a 100644
--- a/examples/stm32f1/Cargo.toml
+++ b/examples/stm32f1/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f103c8", "unstable-pac", "memory-x", "time-driver-any"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32f2/Cargo.toml b/examples/stm32f2/Cargo.toml
index afaf9a0c9..ffb232310 100644
--- a/examples/stm32f2/Cargo.toml
+++ b/examples/stm32f2/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f207zg", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
diff --git a/examples/stm32f3/Cargo.toml b/examples/stm32f3/Cargo.toml
index 69ebef786..38f11201d 100644
--- a/examples/stm32f3/Cargo.toml
+++ b/examples/stm32f3/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f303ze", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32f3/src/bin/multiprio.rs b/examples/stm32f3/src/bin/multiprio.rs
index 77df51ac7..5d010f799 100644
--- a/examples/stm32f3/src/bin/multiprio.rs
+++ b/examples/stm32f3/src/bin/multiprio.rs
@@ -62,7 +62,7 @@ use core::mem;
 use cortex_m::peripheral::NVIC;
 use cortex_m_rt::entry;
 use defmt::*;
-use embassy_stm32::executor::{Executor, InterruptExecutor};
+use embassy_executor::{Executor, InterruptExecutor};
 use embassy_stm32::interrupt;
 use embassy_stm32::pac::Interrupt;
 use embassy_time::{Duration, Instant, Timer};
diff --git a/examples/stm32f4/Cargo.toml b/examples/stm32f4/Cargo.toml
index 7a7bab5bb..d967d8501 100644
--- a/examples/stm32f4/Cargo.toml
+++ b/examples/stm32f4/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers", "arch-cortex-m", "executor-thread", "executor-interrupt"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "unstable-traits", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "unstable-traits", "defmt", "stm32f429zi", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32f4/src/bin/mco.rs b/examples/stm32f4/src/bin/mco.rs
new file mode 100644
index 000000000..2b9ceebc3
--- /dev/null
+++ b/examples/stm32f4/src/bin/mco.rs
@@ -0,0 +1,30 @@
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+use defmt::*;
+use embassy_executor::Spawner;
+use embassy_stm32::gpio::{Level, Output, Speed};
+use embassy_stm32::rcc::{Mco, Mco1Source, Mco2Source, McoClock};
+use embassy_time::{Duration, Timer};
+use {defmt_rtt as _, panic_probe as _};
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    let p = embassy_stm32::init(Default::default());
+    info!("Hello World!");
+    let _mco1 = Mco::new(p.MCO1, p.PA8, Mco1Source::Hsi, McoClock::DIV1);
+    let _mco2 = Mco::new(p.MCO2, p.PC9, Mco2Source::Pll, McoClock::DIV4);
+    let mut led = Output::new(p.PB7, Level::High, Speed::Low);
+    loop {
+        info!("high");
+        led.set_high();
+        Timer::after(Duration::from_millis(300)).await;
+        info!("low");
+        led.set_low();
+        Timer::after(Duration::from_millis(300)).await;
+    }
+}
diff --git a/examples/stm32f4/src/bin/multiprio.rs b/examples/stm32f4/src/bin/multiprio.rs
index 77df51ac7..5d010f799 100644
--- a/examples/stm32f4/src/bin/multiprio.rs
+++ b/examples/stm32f4/src/bin/multiprio.rs
@@ -62,7 +62,7 @@ use core::mem;
 use cortex_m::peripheral::NVIC;
 use cortex_m_rt::entry;
 use defmt::*;
-use embassy_stm32::executor::{Executor, InterruptExecutor};
+use embassy_executor::{Executor, InterruptExecutor};
 use embassy_stm32::interrupt;
 use embassy_stm32::pac::Interrupt;
 use embassy_time::{Duration, Instant, Timer};
diff --git a/examples/stm32f4/src/bin/usart_buffered.rs b/examples/stm32f4/src/bin/usart_buffered.rs
index dd171fe13..a93f8baeb 100644
--- a/examples/stm32f4/src/bin/usart_buffered.rs
+++ b/examples/stm32f4/src/bin/usart_buffered.rs
@@ -5,7 +5,7 @@
 use defmt::*;
 use embassy_executor::Spawner;
 use embassy_stm32::interrupt;
-use embassy_stm32::usart::{BufferedUart, Config, State};
+use embassy_stm32::usart::{BufferedUart, Config};
 use embedded_io::asynch::BufRead;
 use {defmt_rtt as _, panic_probe as _};
@@ -16,20 +16,10 @@ async fn main(_spawner: Spawner) {
    let config = Config::default();
-    let mut state = State::new();
    let irq = interrupt::take!(USART3);
    let mut tx_buf = [0u8; 32];
    let mut rx_buf = [0u8; 32];
-    let mut buf_usart = BufferedUart::new(
+    let mut buf_usart = BufferedUart::new(p.USART3, irq, p.PD9, p.PD8, &mut tx_buf, &mut rx_buf, config);
-        &mut state,
-        p.USART3,
-        p.PD9,
-        p.PD8,
-        irq,
-        &mut tx_buf,
-        &mut rx_buf,
-        config,
-    );
    loop {
        let buf = buf_usart.fill_buf().await.unwrap();
diff --git a/examples/stm32f7/Cargo.toml b/examples/stm32f7/Cargo.toml
index ea4cbd808..74e7bf53d 100644
--- a/examples/stm32f7/Cargo.toml
+++ b/examples/stm32f7/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f767zi", "unstable-pac", "time-driver-any", "exti"]  }
 embassy-net = { path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet"] }
diff --git a/examples/stm32g0/Cargo.toml b/examples/stm32g0/Cargo.toml
index e7273c9fc..03bdbcea3 100644
--- a/examples/stm32g0/Cargo.toml
+++ b/examples/stm32g0/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g071rb", "memory-x", "unstable-pac", "exti"]  }
diff --git a/examples/stm32g4/Cargo.toml b/examples/stm32g4/Cargo.toml
index 8a57a8ef0..4e4150350 100644
--- a/examples/stm32g4/Cargo.toml
+++ b/examples/stm32g4/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g491re", "memory-x", "unstable-pac", "exti"]  }
 embassy-hal-common = {version = "0.1.0", path = "../../embassy-hal-common" }
diff --git a/examples/stm32h7/Cargo.toml b/examples/stm32h7/Cargo.toml
index a04134789..d0d6a9497 100644
--- a/examples/stm32h7/Cargo.toml
+++ b/examples/stm32h7/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "unstable-traits", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32h743bi", "time-driver-any", "exti", "unstable-pac", "unstable-traits"] }
 embassy-net = { path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet", "unstable-traits", "proto-ipv6"] }
diff --git a/examples/stm32l0/Cargo.toml b/examples/stm32l0/Cargo.toml
index 86933a629..413d5c18f 100644
--- a/examples/stm32l0/Cargo.toml
+++ b/examples/stm32l0/Cargo.toml
@@ -10,7 +10,7 @@ nightly = ["embassy-stm32/nightly", "embassy-lora", "lorawan-device", "lorawan",
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["defmt", "stm32l072cz", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-lora = { version = "0.1.0", path = "../../embassy-lora", features = ["sx127x", "time", "defmt"], optional = true}
diff --git a/examples/stm32l0/src/bin/usart_irq.rs b/examples/stm32l0/src/bin/usart_irq.rs
index 8e84cd092..465347004 100644
--- a/examples/stm32l0/src/bin/usart_irq.rs
+++ b/examples/stm32l0/src/bin/usart_irq.rs
@@ -5,7 +5,7 @@
 use defmt::*;
 use embassy_executor::Spawner;
 use embassy_stm32::interrupt;
-use embassy_stm32::usart::{BufferedUart, Config, State};
+use embassy_stm32::usart::{BufferedUart, Config};
 use embedded_io::asynch::{Read, Write};
 use {defmt_rtt as _, panic_probe as _};
@@ -20,20 +20,8 @@ async fn main(_spawner: Spawner) {
    let mut config = Config::default();
    config.baudrate = 9600;
-    let mut state = State::new();
    let irq = interrupt::take!(USART2);
-    let mut usart = unsafe {
+    let mut usart = unsafe { BufferedUart::new(p.USART2, irq, p.PA3, p.PA2, &mut TX_BUFFER, &mut RX_BUFFER, config) };
-        BufferedUart::new(
-            &mut state,
-            p.USART2,
-            p.PA3,
-            p.PA2,
-            irq,
-            &mut TX_BUFFER,
-            &mut RX_BUFFER,
-            config,
-        )
-    };
    usart.write_all(b"Hello Embassy World!\r\n").await.unwrap();
    info!("wrote Hello, starting echo");
diff --git a/examples/stm32l1/Cargo.toml b/examples/stm32l1/Cargo.toml
index 6e3b2103c..cd9508d57 100644
--- a/examples/stm32l1/Cargo.toml
+++ b/examples/stm32l1/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32l151cb-a", "time-driver-any", "memory-x"]  }
diff --git a/examples/stm32l4/Cargo.toml b/examples/stm32l4/Cargo.toml
index 644c90b1a..7c254eba3 100644
--- a/examples/stm32l4/Cargo.toml
+++ b/examples/stm32l4/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-embedded-hal = { version = "0.1.0", path = "../../embassy-embedded-hal" }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l4s5vi", "time-driver-any", "exti", "unstable-traits"]  }
diff --git a/examples/stm32l4/src/bin/mco.rs b/examples/stm32l4/src/bin/mco.rs
new file mode 100644
index 000000000..dea0c66e0
--- /dev/null
+++ b/examples/stm32l4/src/bin/mco.rs
@@ -0,0 +1,27 @@
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+use defmt::*;
+use embassy_executor::Spawner;
+use embassy_stm32::gpio::{Level, Output, Speed};
+use embassy_stm32::rcc::{Mco, Mco1Source, McoClock};
+use embassy_time::{Duration, Timer};
+use {defmt_rtt as _, panic_probe as _};
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    let p = embassy_stm32::init(Default::default());
+    info!("Hello World!");
+    let _mco = Mco::new(p.MCO, p.PA8, Mco1Source::Hsi16, McoClock::DIV1);
+    let mut led = Output::new(p.PB14, Level::High, Speed::Low);
+    loop {
+        led.set_high();
+        Timer::after(Duration::from_millis(300)).await;
+        led.set_low();
+        Timer::after(Duration::from_millis(300)).await;
+    }
+}
diff --git a/examples/stm32l5/Cargo.toml b/examples/stm32l5/Cargo.toml
index f880328dc..1c662b9da 100644
--- a/examples/stm32l5/Cargo.toml
+++ b/examples/stm32l5/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l552ze", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32u5/Cargo.toml b/examples/stm32u5/Cargo.toml
index 2b02eda92..ebef0a4f7 100644
--- a/examples/stm32u5/Cargo.toml
+++ b/examples/stm32u5/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32u585ai", "time-driver-any", "memory-x" ]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32wb/Cargo.toml b/examples/stm32wb/Cargo.toml
index e27b4527c..ddf9729e6 100644
--- a/examples/stm32wb/Cargo.toml
+++ b/examples/stm32wb/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32wb55cc", "time-driver-any", "exti"]  }
diff --git a/examples/stm32wl/Cargo.toml b/examples/stm32wl/Cargo.toml
index 690481bbf..9fc7e0f4a 100644
--- a/examples/stm32wl/Cargo.toml
+++ b/examples/stm32wl/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32wl55jc-cm4", "time-driver-any", "memory-x", "subghz", "unstable-pac", "exti"]  }
 embassy-lora = { version = "0.1.0", path = "../../embassy-lora", features = ["stm32wl", "time", "defmt"] }
diff --git a/examples/wasm/Cargo.toml b/examples/wasm/Cargo.toml
index e0e799a34..430d0b4c7 100644
--- a/examples/wasm/Cargo.toml
+++ b/examples/wasm/Cargo.toml
@@ -9,7 +9,7 @@ crate-type = ["cdylib"]
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["log"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["log", "wasm", "nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-wasm", "executor-thread", "log", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["log", "wasm", "nightly"] }
 wasm-logger = "0.2.0"
diff --git a/tests/nrf/Cargo.toml b/tests/nrf/Cargo.toml
index 2a4e8cf41..912749e5d 100644
--- a/tests/nrf/Cargo.toml
+++ b/tests/nrf/Cargo.toml
@@ -7,7 +7,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt", "nightly"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "nightly", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "nightly", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "nightly", "defmt-timestamp-uptime"] }
 embassy-nrf = { version = "0.1.0", path = "../../embassy-nrf", features = ["defmt", "nightly", "unstable-traits", "nrf52840", "time-driver-rtc1", "gpiote", "unstable-pac"] }
 embedded-io = { version = "0.4.0", features = ["async"] }
diff --git a/tests/rp/Cargo.toml b/tests/rp/Cargo.toml
index 572a9ce88..eb447be35 100644
--- a/tests/rp/Cargo.toml
+++ b/tests/rp/Cargo.toml
@@ -6,7 +6,7 @@ license = "MIT OR Apache-2.0"
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt"] }
 embassy-rp = { version = "0.1.0", path = "../../embassy-rp", features = ["nightly", "defmt", "unstable-pac", "unstable-traits", "time-driver", "critical-section-impl"]  }
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
diff --git a/tests/stm32/Cargo.toml b/tests/stm32/Cargo.toml
index 08a775eae..17b640797 100644
--- a/tests/stm32/Cargo.toml
+++ b/tests/stm32/Cargo.toml
@@ -15,7 +15,7 @@ stm32u585ai = ["embassy-stm32/stm32u585ai"]     # IoT board
 [dependencies]
 embassy-sync = { version = "0.1.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["defmt", "integrated-timers"] }
+embassy-executor = { version = "0.1.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.0", path = "../../embassy-time", features = ["defmt", "tick-hz-32_768"] }
 embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "memory-x", "time-driver-tim2"]  }