81 files changed, 2681 insertions, 966 deletions

diff --git a/embassy-executor/CHANGELOG.md b/embassy-executor/CHANGELOG.md
index 77c64fd8e..5fbb8cf13 100644
--- a/embassy-executor/CHANGELOG.md
+++ b/embassy-executor/CHANGELOG.md
@@ -5,7 +5,81 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
-## Unreleased
+<!-- next-header -->
+## Unreleased - ReleaseDate
+
+- Added new metadata API for tasks.
+- Main task automatically gets a name of `main` when the `metadata-name` feature is enabled.
+- Upgraded rtos-trace
+- Added optional "highest priority" scheduling
+- Added optional "earliest deadline first" EDF scheduling
+- Migrate `cortex-ar` to `aarch32-cpu`. The feature name `arch-cortex-ar` remains the same and
+  legacy ARM architectures are not supported.
+
+## 0.9.1 - 2025-08-31
+
+- Fixed performance regression on some ESP32 MCUs.
+
+## 0.9.0 - 2025-08-26
+
+- Added `extern "Rust" fn __embassy_time_queue_item_from_waker`
+- Removed `TaskRef::dangling`
+- Added `embassy-executor-timer-queue` as a dependency
+- Moved the `TimeQueueItem` struct and `timer-item-payload-size-*` features (as `timer-item-size-X-words`) into `embassy-executor-timer-queue`
+
+## 0.8.0 - 2025-07-31
+
+- Added `SpawnToken::id`
+- Task pools are now statically allocated on stable rust. All `task-arena-size-*` features have been removed and are no longer necessary.
+- New trace hooks: `_embassy_trace_poll_start` & `_embassy_trace_task_end`
+- Added task naming capability to tracing infrastructure
+- Added `Executor::id` & `Spawner::executor_id`
+- Disable `critical-section/std` for arch-std
+- Added possibility to select an executor in `#[embassy_executor::main]`
+- Fix AVR executor
+- executor: Make state implementations and their conditions match
+- Added support for Cortex-A and Cortex-R
+- Added support for `-> impl Future<Output = ()>` in `#[task]`
+- Fixed `Send` unsoundness with `-> impl Future` tasks
+- Marked `Spawner::for_current_executor` as `unsafe`
+- `#[task]` now properly marks the generated function as unsafe if the task is marked unsafe
+
+## 0.7.0 - 2025-01-02
+
+- Performance optimizations.
+- Remove feature `integrated-timers`. Starting with `embassy-time-driver` v0.2, `embassy-time` v0.4 the timer queue is now part of the time driver, so it's no longer the executor's responsibility. Therefore, `embassy-executor` no longer provides an `embassy-time-queue-driver` implementation.
+- Added the possibility for timer driver implementations to store arbitrary data in task headers. This can be used to make a timer queue intrusive list, similar to the previous `integrated-timers` feature. Payload size is controlled by the `timer-item-payload-size-X` features.
+- Added `TaskRef::executor` to obtain a reference to a task's executor
+
+## 0.6.3 - 2024-11-12
+
+- Building with the `nightly` feature now works with the Xtensa Rust compiler 1.82.
+- Compare vtable address instead of contents. Saves 44 bytes of flash on cortex-m.
+
+## 0.6.2 - 2024-11-06
+
+- The `nightly` feature no longer requires `nightly-2024-09-06` or newer.
+
+## 0.6.1 - 2024-10-21
+
+- Soundness fix: Deny using `impl Trait` in task arguments. This was previously accidentally allowed when not using the `nightly` feature,
+  and could cause out of bounds memory accesses if spawning the same task mulitple times with different underlying types
+  for the `impl Trait`. Affected versions are 0.4.x, 0.5.0 and 0.6.0, which have been yanked.
+- Add an architecture-agnostic executor that spins waiting for tasks to run, enabled with the `arch-spin` feature.
+- Update for breaking change in the nightly waker_getters API. The `nightly` feature now requires `nightly-2024-09-06` or newer.
+- Improve macro error messages.
+
+## 0.6.0 - 2024-08-05
+
+- Add collapse_debuginfo to fmt.rs macros.
+- initial support for AVR
+- use nightly waker_getters APIs
+
+## 0.5.1 - 2024-10-21
+
+- Soundness fix: Deny using `impl Trait` in task arguments. This was previously accidentally allowed when not using the `nightly` feature,
+  and could cause out of bounds memory accesses if spawning the same task mulitple times with different underlying types
+  for the `impl Trait`. Affected versions are 0.4.x, 0.5.0 and 0.6.0, which have been yanked.
 
 ## 0.5.0 - 2024-01-11
 
diff --git a/embassy-executor/Cargo.toml b/embassy-executor/Cargo.toml
index 431165cee..d3e5b241a 100644
--- a/embassy-executor/Cargo.toml
+++ b/embassy-executor/Cargo.toml
@@ -1,7 +1,7 @@
 [package]
 name = "embassy-executor"
-version = "0.5.0"
-edition = "2021"
+version = "0.9.1"
+edition = "2024"
 license = "MIT OR Apache-2.0"
 description = "async/await executor designed for embedded usage"
 repository = "https://github.com/embassy-rs/embassy"
@@ -12,10 +12,65 @@ categories = [
     "asynchronous",
 ]
 
+[package.metadata.embassy]
+build = [
+    {target = "thumbv7em-none-eabi", features = []},
+    {target = "thumbv7em-none-eabi", features = ["log"]},
+    {target = "thumbv7em-none-eabi", features = ["defmt"]},
+    {target = "thumbv6m-none-eabi", features = ["defmt"]},
+    {target = "thumbv6m-none-eabi", features = ["arch-cortex-m", "defmt", "executor-interrupt", "executor-thread"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "rtos-trace"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-thread"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "embassy-time-driver"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "embassy-time-driver", "scheduler-priority"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "embassy-time-driver", "scheduler-priority", "scheduler-deadline"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "embassy-time-driver", "scheduler-deadline"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "scheduler-priority", "scheduler-deadline"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "scheduler-deadline"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-cortex-m", "executor-interrupt", "executor-thread", "embassy-time-driver", "scheduler-priority", "scheduler-deadline", "trace"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-spin"]},
+    {target = "thumbv7em-none-eabi", features = ["arch-spin", "scheduler-deadline"]},
+    {target = "armv7a-none-eabi", features = ["arch-cortex-ar", "executor-thread"]},
+    {target = "armv7r-none-eabi", features = ["arch-cortex-ar", "executor-thread"]},
+    {target = "armv7r-none-eabihf", features = ["arch-cortex-ar", "executor-thread"]},
+    {target = "riscv32imac-unknown-none-elf", features = ["arch-riscv32"]},
+    {target = "riscv32imac-unknown-none-elf", features = ["arch-riscv32", "executor-thread"]},
+    {target = "riscv32imac-unknown-none-elf", features = ["arch-riscv32", "executor-thread", "trace"]},
+    # Nightly builds
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "log"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "defmt"]},
+    {group = "nightly", target = "thumbv6m-none-eabi", features = ["nightly", "defmt"]},
+    {group = "nightly", target = "thumbv6m-none-eabi", features = ["nightly", "defmt", "arch-cortex-m", "executor-thread", "executor-interrupt"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "arch-cortex-m"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "arch-cortex-m", "executor-thread"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "arch-cortex-m", "executor-interrupt"]},
+    {group = "nightly", target = "thumbv7em-none-eabi", features = ["nightly", "arch-cortex-m", "executor-thread", "executor-interrupt"]},
+    {group = "nightly", target = "riscv32imac-unknown-none-elf", features = ["nightly", "arch-riscv32"]},
+    {group = "nightly", target = "riscv32imac-unknown-none-elf", features = ["nightly", "arch-riscv32", "executor-thread"]},
+    {group = "nightly", target = "armv7a-none-eabi", features = ["nightly", "arch-cortex-ar", "executor-thread"]},
+    {group = "nightly", target = "avr-none", features = ["nightly", "arch-avr", "avr-device/atmega328p"], build-std = ["core", "alloc"], env = { RUSTFLAGS = "-Ctarget-cpu=atmega328p" }},
+    # Xtensa builds
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = []},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["log"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["defmt"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32s2-none-elf", features = ["defmt"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["defmt", "arch-spin", "executor-thread"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32s2-none-elf", features = ["defmt", "arch-spin", "executor-thread"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32s3-none-elf", features = ["defmt", "arch-spin", "executor-thread"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["arch-spin"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["arch-spin", "rtos-trace"]},
+    {group = "xtensa", build-std = ["core", "alloc"],  target = "xtensa-esp32-none-elf", features = ["arch-spin", "executor-thread"]},
+]
+
+
 [package.metadata.embassy_docs]
 src_base = "https://github.com/embassy-rs/embassy/blob/embassy-executor-v$VERSION/embassy-executor/src/"
 src_base_git = "https://github.com/embassy-rs/embassy/blob/$COMMIT/embassy-executor/src/"
-features = ["defmt"]
+features = ["defmt", "scheduler-deadline", "scheduler-priority"]
 flavors = [
     { name = "std",             target = "x86_64-unknown-linux-gnu",     features = ["arch-std", "executor-thread"] },
     { name = "wasm",            target = "wasm32-unknown-unknown",       features = ["arch-wasm", "executor-thread"] },
@@ -26,16 +81,16 @@ flavors = [
 [package.metadata.docs.rs]
 default-target = "thumbv7em-none-eabi"
 targets = ["thumbv7em-none-eabi"]
-features = ["defmt", "arch-cortex-m", "executor-thread", "executor-interrupt"]
+features = ["defmt", "arch-cortex-m", "executor-thread", "executor-interrupt", "scheduler-deadline", "scheduler-priority", "embassy-time-driver"]
 
 [dependencies]
-defmt = { version = "0.3", optional = true }
+defmt = { version = "1.0.1", optional = true }
 log = { version = "0.4.14", optional = true }
-rtos-trace = { version = "0.1.2", optional = true }
+rtos-trace = { version = "0.2", optional = true }
 
-embassy-executor-macros = { version = "0.4.0", path = "../embassy-executor-macros" }
-embassy-time-driver = { version = "0.1.0", path = "../embassy-time-driver", optional = true }
-embassy-time-queue-driver = { version = "0.1.0", path = "../embassy-time-queue-driver", optional = true }
+embassy-executor-macros = { version = "0.7.0", path = "../embassy-executor-macros" }
+embassy-time-driver = { version = "0.2.1", path = "../embassy-time-driver", optional = true }
+embassy-executor-timer-queue = { version = "0.1", path = "../embassy-executor-timer-queue" }
 critical-section = "1.1"
 
 document-features = "0.2.7"
@@ -46,42 +101,64 @@ portable-atomic = { version = "1.5", optional = true }
 # arch-cortex-m dependencies
 cortex-m = { version = "0.7.6", optional = true }
 
+# arch-cortex-ar dependencies
+aarch32-cpu = { version = "0.1", optional = true }
+
 # arch-wasm dependencies
 wasm-bindgen = { version = "0.2.82", optional = true }
 js-sys = { version = "0.3", optional = true }
 
 # arch-avr dependencies
-avr-device = { version = "0.5.3", optional = true }
+avr-device = { version = "0.7.0", optional = true }
+
+
+[dependencies.cordyceps]
+version = "0.3.4"
+features = ["no-cache-pad"]
 
 [dev-dependencies]
 critical-section = { version = "1.1", features = ["std"] }
-
+trybuild = "1.0"
+embassy-sync = { path = "../embassy-sync" }
+rustversion = "1.0.21"
 
 [features]
 
 ## Enable nightly-only features
 nightly = ["embassy-executor-macros/nightly"]
 
+## Enable defmt logging
+defmt = ["dep:defmt"]
+
+## Enable log logging
+log = ["dep:log"]
+
 # Enables turbo wakers, which requires patching core. Not surfaced in the docs by default due to
 # being an complicated advanced and undocumented feature.
 # See: https://github.com/embassy-rs/embassy/pull/1263
 turbowakers = []
 
-## Use the executor-integrated `embassy-time` timer queue.
-integrated-timers = ["dep:embassy-time-driver", "dep:embassy-time-queue-driver"]
-
 #! ### Architecture
 _arch = [] # some arch was picked
 ## std
-arch-std = ["_arch", "critical-section/std"]
+arch-std = ["_arch"]
 ## Cortex-M
 arch-cortex-m = ["_arch", "dep:cortex-m"]
+## Cortex-A/R
+arch-cortex-ar = ["_arch", "dep:aarch32-cpu", "dep:arm-targets"]
 ## RISC-V 32
 arch-riscv32 = ["_arch"]
 ## WASM
 arch-wasm = ["_arch", "dep:wasm-bindgen", "dep:js-sys"]
 ## AVR
 arch-avr = ["_arch", "dep:portable-atomic", "dep:avr-device"]
+## spin (architecture agnostic; never sleeps)
+arch-spin = ["_arch"]
+
+#! ### Metadata
+
+## Enable the `name` field in task metadata.
+metadata-name = ["embassy-executor-macros/metadata-name"]
 
 #! ### Executor
 
@@ -89,98 +166,22 @@ arch-avr = ["_arch", "dep:portable-atomic", "dep:avr-device"]
 executor-thread = []
 ## Enable the interrupt-mode executor (available in Cortex-M only)
 executor-interrupt = []
+## Enable tracing hooks
+trace = ["_any_trace"]
+## Enable support for rtos-trace framework
+rtos-trace = ["_any_trace", "metadata-name", "dep:rtos-trace", "embassy-time-driver"]
+_any_trace = []
+
+## Enable "Earliest Deadline First" Scheduler, using soft-realtime "deadlines" to prioritize
+## tasks based on the remaining time before their deadline. Adds some overhead.
+scheduler-deadline = []
+
+## Enable "Highest Priority First" Scheduler. Adds some overhead.
+scheduler-priority = []
+
+## Enable the embassy_time_driver dependency.
+## This can unlock extra APIs, for example for the `sheduler-deadline`
+embassy-time-driver = ["dep:embassy-time-driver"]
 
-#! ### Task Arena Size
-#! Sets the [task arena](#task-arena) size. Necessary if you’re not using `nightly`.
-#!
-#! <details>
-#!  <summary>Preconfigured Task Arena Sizes:</summary>
-#! <!-- rustdoc requires the following blank line for the feature list to render correctly! -->
-#!
-
-# BEGIN AUTOGENERATED CONFIG FEATURES
-# Generated by gen_config.py. DO NOT EDIT.
-## 64
-task-arena-size-64 = []
-## 128
-task-arena-size-128 = []
-## 192
-task-arena-size-192 = []
-## 256
-task-arena-size-256 = []
-## 320
-task-arena-size-320 = []
-## 384
-task-arena-size-384 = []
-## 512
-task-arena-size-512 = []
-## 640
-task-arena-size-640 = []
-## 768
-task-arena-size-768 = []
-## 1024
-task-arena-size-1024 = []
-## 1280
-task-arena-size-1280 = []
-## 1536
-task-arena-size-1536 = []
-## 2048
-task-arena-size-2048 = []
-## 2560
-task-arena-size-2560 = []
-## 3072
-task-arena-size-3072 = []
-## 4096 (default)
-task-arena-size-4096 = [] # Default
-## 5120
-task-arena-size-5120 = []
-## 6144
-task-arena-size-6144 = []
-## 8192
-task-arena-size-8192 = []
-## 10240
-task-arena-size-10240 = []
-## 12288
-task-arena-size-12288 = []
-## 16384
-task-arena-size-16384 = []
-## 20480
-task-arena-size-20480 = []
-## 24576
-task-arena-size-24576 = []
-## 32768
-task-arena-size-32768 = []
-## 40960
-task-arena-size-40960 = []
-## 49152
-task-arena-size-49152 = []
-## 65536
-task-arena-size-65536 = []
-## 81920
-task-arena-size-81920 = []
-## 98304
-task-arena-size-98304 = []
-## 131072
-task-arena-size-131072 = []
-## 163840
-task-arena-size-163840 = []
-## 196608
-task-arena-size-196608 = []
-## 262144
-task-arena-size-262144 = []
-## 327680
-task-arena-size-327680 = []
-## 393216
-task-arena-size-393216 = []
-## 524288
-task-arena-size-524288 = []
-## 655360
-task-arena-size-655360 = []
-## 786432
-task-arena-size-786432 = []
-## 1048576
-task-arena-size-1048576 = []
-
-# END AUTOGENERATED CONFIG FEATURES
-
-#! </details>
+[build-dependencies]
+arm-targets = { version = "0.4", optional = true }
diff --git a/embassy-executor/README.md b/embassy-executor/README.md
index aa9d59907..85f15edbb 100644
--- a/embassy-executor/README.md
+++ b/embassy-executor/README.md
@@ -3,35 +3,10 @@
 An async/await executor designed for embedded usage.
 
 - No `alloc`, no heap needed.
-- With nightly Rust, task futures can be fully statically allocated.
+- Tasks are statically allocated. Each task gets its own `static`, with the exact size to hold the task (or multiple instances of it, if using `pool_size`) calculated automatically at compile time. If tasks don't fit in RAM, this is detected at compile time by the linker. Runtime panics due to running out of memory are not possible.
 - No "fixed capacity" data structures, executor works with 1 or 1000 tasks without needing config/tuning.
 - Integrated timer queue: sleeping is easy, just do `Timer::after_secs(1).await;`.
 - No busy-loop polling: CPU sleeps when there's no work to do, using interrupts or `WFE/SEV`.
 - Efficient polling: a wake will only poll the woken task, not all of them.
 - Fair: a task can't monopolize CPU time even if it's constantly being woken. All other tasks get a chance to run before a given task gets polled for the second time.
 - Creating multiple executor instances is supported, to run tasks with multiple priority levels. This allows higher-priority tasks to preempt lower-priority tasks.
-
-## Task arena
-
-When the `nightly` Cargo feature is not enabled, `embassy-executor` allocates tasks out of an arena (a very simple bump allocator).
-
-If the task arena gets full, the program will panic at runtime. To guarantee this doesn't happen, you must set the size to the sum of sizes of all tasks.
-
-Tasks are allocated from the arena when spawned for the first time. If the task exists, the allocation is not released to the arena, but can be reused to spawn the task again. For multiple-instance tasks (like `#[embassy_executor::task(pool_size = 4)]`), the first spawn will allocate memory for all instances. This is done for performance and to increase predictability (for example, spawning at least 1 instance of every task at boot guarantees an immediate panic if the arena is too small, while allocating instances on-demand could delay the panic to only when the program is under load).
-
-The arena size can be configured in two ways:
-
-- Via Cargo features: enable a Cargo feature like `task-arena-size-8192`. Only a selection of values
-  is available, see [Task Area Sizes](#task-arena-size) for reference.
-- Via environment variables at build time: set the variable named `EMBASSY_EXECUTOR_TASK_ARENA_SIZE`. For example
-  `EMBASSY_EXECUTOR_TASK_ARENA_SIZE=4321 cargo build`. You can also set them in the `[env]` section of `.cargo/config.toml`.
-  Any value can be set, unlike with Cargo features.
-
-Environment variables take precedence over Cargo features. If two Cargo features are enabled for the same setting
-with different values, compilation fails.
-
-## Statically allocating tasks
-
-When using nightly Rust, enable the `nightly` Cargo feature. This will make `embassy-executor` use the `type_alias_impl_trait` feature to allocate all tasks in `static`s. Each task gets its own `static`, with the exact size to hold the task (or multiple instances of it, if using `pool_size`) calculated automatically at compile time. If tasks don't fit in RAM, this is detected at compile time by the linker. Runtime panics due to running out of memory are not possible.
-
-The configured arena size is ignored, no arena is used at all.
diff --git a/embassy-executor/build.rs b/embassy-executor/build.rs
index 8a41d7503..36e23a632 100644
--- a/embassy-executor/build.rs
+++ b/embassy-executor/build.rs
@@ -1,99 +1,12 @@
-use std::collections::HashMap;
-use std::fmt::Write;
-use std::path::PathBuf;
-use std::{env, fs};
-
 #[path = "./build_common.rs"]
 mod common;
 
-static CONFIGS: &[(&str, usize)] = &[
-    // BEGIN AUTOGENERATED CONFIG FEATURES
-    // Generated by gen_config.py. DO NOT EDIT.
-    ("TASK_ARENA_SIZE", 4096),
-    // END AUTOGENERATED CONFIG FEATURES
-];
-
-struct ConfigState {
-    value: usize,
-    seen_feature: bool,
-    seen_env: bool,
-}
-
 fn main() {
-    let crate_name = env::var("CARGO_PKG_NAME")
-        .unwrap()
-        .to_ascii_uppercase()
-        .replace('-', "_");
-
-    // only rebuild if build.rs changed. Otherwise Cargo will rebuild if any
-    // other file changed.
-    println!("cargo:rerun-if-changed=build.rs");
-
-    // Rebuild if config envvar changed.
-    for (name, _) in CONFIGS {
-        println!("cargo:rerun-if-env-changed={crate_name}_{name}");
-    }
-
-    let mut configs = HashMap::new();
-    for (name, default) in CONFIGS {
-        configs.insert(
-            *name,
-            ConfigState {
-                value: *default,
-                seen_env: false,
-                seen_feature: false,
-            },
-        );
-    }
-
-    let prefix = format!("{crate_name}_");
-    for (var, value) in env::vars() {
-        if let Some(name) = var.strip_prefix(&prefix) {
-            let Some(cfg) = configs.get_mut(name) else {
-                panic!("Unknown env var {name}")
-            };
-
-            let Ok(value) = value.parse::<usize>() else {
-                panic!("Invalid value for env var {name}: {value}")
-            };
-
-            cfg.value = value;
-            cfg.seen_env = true;
-        }
-
-        if let Some(feature) = var.strip_prefix("CARGO_FEATURE_") {
-            if let Some(i) = feature.rfind('_') {
-                let name = &feature[..i];
-                let value = &feature[i + 1..];
-                if let Some(cfg) = configs.get_mut(name) {
-                    let Ok(value) = value.parse::<usize>() else {
-                        panic!("Invalid value for feature {name}: {value}")
-                    };
-
-                    // envvars take priority.
-                    if !cfg.seen_env {
-                        if cfg.seen_feature {
-                            panic!("multiple values set for feature {}: {} and {}", name, cfg.value, value);
-                        }
-
-                        cfg.value = value;
-                        cfg.seen_feature = true;
-                    }
-                }
-            }
-        }
-    }
-
-    let mut data = String::new();
-
-    for (name, cfg) in &configs {
-        writeln!(&mut data, "pub const {}: usize = {};", name, cfg.value).unwrap();
-    }
-
-    let out_dir = PathBuf::from(env::var_os("OUT_DIR").unwrap());
-    let out_file = out_dir.join("config.rs").to_string_lossy().to_string();
-    fs::write(out_file, data).unwrap();
-
     let mut rustc_cfgs = common::CfgSet::new();
     common::set_target_cfgs(&mut rustc_cfgs);
+
+    // This is used to exclude legacy architecture support. The raw executor needs to be used for
+    // those architectures because SEV/WFE are not supported.
+    #[cfg(feature = "arch-cortex-ar")]
+    arm_targets::process();
 }
diff --git a/embassy-executor/build_common.rs b/embassy-executor/build_common.rs
index 0487eb3c5..4f24e6d37 100644
--- a/embassy-executor/build_common.rs
+++ b/embassy-executor/build_common.rs
@@ -8,8 +8,6 @@
 
 use std::collections::HashSet;
 use std::env;
-use std::ffi::OsString;
-use std::process::Command;
 
 /// Helper for emitting cargo instruction for enabling configs (`cargo:rustc-cfg=X`) and declaring
 /// them (`cargo:rust-check-cfg=cfg(X)`).
@@ -17,7 +15,6 @@ use std::process::Command;
 pub struct CfgSet {
     enabled: HashSet<String>,
     declared: HashSet<String>,
-    emit_declared: bool,
 }
 
 impl CfgSet {
@@ -25,7 +22,6 @@ impl CfgSet {
         Self {
             enabled: HashSet::new(),
             declared: HashSet::new(),
-            emit_declared: is_rustc_nightly(),
         }
     }
 
@@ -49,7 +45,7 @@ impl CfgSet {
     ///
     /// This enables rustc to check that the configs in `#[cfg(...)]` attributes are valid.
     pub fn declare(&mut self, cfg: impl AsRef<str>) {
-        if self.declared.insert(cfg.as_ref().to_owned()) && self.emit_declared {
+        if self.declared.insert(cfg.as_ref().to_owned()) {
             println!("cargo:rustc-check-cfg=cfg({})", cfg.as_ref());
         }
     }
@@ -69,21 +65,6 @@ impl CfgSet {
     }
 }
 
-fn is_rustc_nightly() -> bool {
-    if env::var_os("EMBASSY_FORCE_CHECK_CFG").is_some() {
-        return true;
-    }
-
-    let rustc = env::var_os("RUSTC").unwrap_or_else(|| OsString::from("rustc"));
-
-    let output = Command::new(rustc)
-        .arg("--version")
-        .output()
-        .expect("failed to run `rustc --version`");
-
-    String::from_utf8_lossy(&output.stdout).contains("nightly")
-}
-
 /// Sets configs that describe the target platform.
 pub fn set_target_cfgs(cfgs: &mut CfgSet) {
     let target = env::var("TARGET").unwrap();
diff --git a/embassy-executor/src/arch/avr.rs b/embassy-executor/src/arch/avr.rs
index 70085d04d..a841afe15 100644
--- a/embassy-executor/src/arch/avr.rs
+++ b/embassy-executor/src/arch/avr.rs
@@ -10,11 +10,11 @@ mod thread {
     pub use embassy_executor_macros::main_avr as main;
     use portable_atomic::{AtomicBool, Ordering};
 
-    use crate::{raw, Spawner};
+    use crate::{Spawner, raw};
 
     static SIGNAL_WORK_THREAD_MODE: AtomicBool = AtomicBool::new(false);
 
-    #[export_name = "__pender"]
+    #[unsafe(export_name = "__pender")]
     fn __pender(_context: *mut ()) {
         SIGNAL_WORK_THREAD_MODE.store(true, Ordering::SeqCst);
     }
diff --git a/embassy-executor/src/arch/cortex_ar.rs b/embassy-executor/src/arch/cortex_ar.rs
new file mode 100644
index 000000000..ce572738a
--- /dev/null
+++ b/embassy-executor/src/arch/cortex_ar.rs
@@ -0,0 +1,87 @@
+#[cfg(arm_profile = "legacy")]
+compile_error!("`arch-cortex-ar` does not support the legacy ARM profile, WFE/SEV are not available.");
+
+#[cfg(feature = "executor-interrupt")]
+compile_error!("`executor-interrupt` is not supported with `arch-cortex-ar`.");
+
+#[unsafe(export_name = "__pender")]
+#[cfg(any(feature = "executor-thread", feature = "executor-interrupt"))]
+fn __pender(context: *mut ()) {
+    // `context` is always `usize::MAX` created by `Executor::run`.
+    let context = context as usize;
+
+    #[cfg(feature = "executor-thread")]
+    // Try to make Rust optimize the branching away if we only use thread mode.
+    if !cfg!(feature = "executor-interrupt") || context == THREAD_PENDER {
+        aarch32_cpu::asm::sev();
+        return;
+    }
+}
+
+#[cfg(feature = "executor-thread")]
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
+    pub(super) const THREAD_PENDER: usize = usize::MAX;
+
+    use core::marker::PhantomData;
+
+    use aarch32_cpu::asm::wfe;
+    pub use embassy_executor_macros::main_cortex_ar as main;
+
+    use crate::{Spawner, raw};
+
+    /// 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(THREAD_PENDER as *mut ()),
+                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();
+                }
+                wfe();
+            }
+        }
+    }
+}
diff --git a/embassy-executor/src/arch/cortex_m.rs b/embassy-executor/src/arch/cortex_m.rs
index 5c517e0a2..1ce96d1d5 100644
--- a/embassy-executor/src/arch/cortex_m.rs
+++ b/embassy-executor/src/arch/cortex_m.rs
@@ -1,4 +1,4 @@
-#[export_name = "__pender"]
+#[unsafe(export_name = "__pender")]
 #[cfg(any(feature = "executor-thread", feature = "executor-interrupt"))]
 fn __pender(context: *mut ()) {
     unsafe {
@@ -53,7 +53,7 @@ mod thread {
 
     pub use embassy_executor_macros::main_cortex_m as main;
 
-    use crate::{raw, Spawner};
+    use crate::{Spawner, raw};
 
     /// Thread mode executor, using WFE/SEV.
     ///
@@ -143,7 +143,7 @@ mod interrupt {
     /// 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.
+    /// [`Executor`](crate::Executor) instead, if it works for your use case.
     pub struct InterruptExecutor {
         started: Mutex<Cell<bool>>,
         executor: UnsafeCell<MaybeUninit<raw::Executor>>,
@@ -179,11 +179,11 @@ mod interrupt {
         /// 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
+        /// is returned instead of a [`Spawner`](crate::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
+        /// To obtain a [`Spawner`](crate::Spawner) for this executor, use [`Spawner::for_current_executor()`](crate::Spawner::for_current_executor()) from
         /// a task running in it.
         ///
         /// # Interrupt requirements
@@ -195,6 +195,7 @@ mod interrupt {
         /// You must set the interrupt priority before calling this method. You MUST NOT
         /// do it after.
         ///
+        /// [`SendSpawner`]: crate::SendSpawner
         pub fn start(&'static self, irq: impl InterruptNumber) -> crate::SendSpawner {
             if critical_section::with(|cs| self.started.borrow(cs).replace(true)) {
                 panic!("InterruptExecutor::start() called multiple times on the same executor.");
@@ -215,7 +216,7 @@ mod interrupt {
 
         /// Get a SendSpawner for this executor
         ///
-        /// This returns a [`SendSpawner`] you can use to spawn tasks on this
+        /// This returns a [`SendSpawner`](crate::SendSpawner) you can use to spawn tasks on this
         /// executor.
         ///
         /// This MUST only be called on an executor that has already been started.
diff --git a/embassy-executor/src/arch/riscv32.rs b/embassy-executor/src/arch/riscv32.rs
index 01e63a9fd..c70c1344a 100644
--- a/embassy-executor/src/arch/riscv32.rs
+++ b/embassy-executor/src/arch/riscv32.rs
@@ -10,12 +10,12 @@ mod thread {
 
     pub use embassy_executor_macros::main_riscv as main;
 
-    use crate::{raw, Spawner};
+    use crate::{Spawner, raw};
 
     /// 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);
 
-    #[export_name = "__pender"]
+    #[unsafe(export_name = "__pender")]
     fn __pender(_context: *mut ()) {
         SIGNAL_WORK_THREAD_MODE.store(true, Ordering::SeqCst);
     }
diff --git a/embassy-executor/src/arch/spin.rs b/embassy-executor/src/arch/spin.rs
new file mode 100644
index 000000000..49f3356a6
--- /dev/null
+++ b/embassy-executor/src/arch/spin.rs
@@ -0,0 +1,58 @@
+#[cfg(feature = "executor-interrupt")]
+compile_error!("`executor-interrupt` is not supported with `arch-spin`.");
+
+#[cfg(feature = "executor-thread")]
+pub use thread::*;
+#[cfg(feature = "executor-thread")]
+mod thread {
+    use core::marker::PhantomData;
+
+    pub use embassy_executor_macros::main_spin as main;
+
+    use crate::{Spawner, raw};
+
+    #[unsafe(export_name = "__pender")]
+    fn __pender(_context: *mut ()) {}
+
+    /// Spin Executor
+    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(core::ptr::null_mut()),
+                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() };
+            }
+        }
+    }
+}
diff --git a/embassy-executor/src/arch/std.rs b/embassy-executor/src/arch/std.rs
index b02b15988..c62ab723b 100644
--- a/embassy-executor/src/arch/std.rs
+++ b/embassy-executor/src/arch/std.rs
@@ -10,9 +10,9 @@ mod thread {
 
     pub use embassy_executor_macros::main_std as main;
 
-    use crate::{raw, Spawner};
+    use crate::{Spawner, raw};
 
-    #[export_name = "__pender"]
+    #[unsafe(export_name = "__pender")]
     fn __pender(context: *mut ()) {
         let signaler: &'static Signaler = unsafe { std::mem::transmute(context) };
         signaler.signal()
diff --git a/embassy-executor/src/arch/wasm.rs b/embassy-executor/src/arch/wasm.rs
index f9d0f935c..d2ff2fe51 100644
--- a/embassy-executor/src/arch/wasm.rs
+++ b/embassy-executor/src/arch/wasm.rs
@@ -13,9 +13,9 @@ mod thread {
     use wasm_bindgen::prelude::*;
 
     use crate::raw::util::UninitCell;
-    use crate::{raw, Spawner};
+    use crate::{Spawner, raw};
 
-    #[export_name = "__pender"]
+    #[unsafe(export_name = "__pender")]
     fn __pender(context: *mut ()) {
         let signaler: &'static WasmContext = unsafe { std::mem::transmute(context) };
         let _ = signaler.promise.then(unsafe { signaler.closure.as_mut() });
diff --git a/embassy-executor/src/fmt.rs b/embassy-executor/src/fmt.rs
index 2ac42c557..8ca61bc39 100644
--- a/embassy-executor/src/fmt.rs
+++ b/embassy-executor/src/fmt.rs
@@ -6,6 +6,7 @@ use core::fmt::{Debug, Display, LowerHex};
 #[cfg(all(feature = "defmt", feature = "log"))]
 compile_error!("You may not enable both `defmt` and `log` features.");
 
+#[collapse_debuginfo(yes)]
 macro_rules! assert {
     ($($x:tt)*) => {
         {
@@ -17,6 +18,7 @@ macro_rules! assert {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! assert_eq {
     ($($x:tt)*) => {
         {
@@ -28,6 +30,7 @@ macro_rules! assert_eq {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! assert_ne {
     ($($x:tt)*) => {
         {
@@ -39,6 +42,7 @@ macro_rules! assert_ne {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! debug_assert {
     ($($x:tt)*) => {
         {
@@ -50,6 +54,7 @@ macro_rules! debug_assert {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! debug_assert_eq {
     ($($x:tt)*) => {
         {
@@ -61,6 +66,7 @@ macro_rules! debug_assert_eq {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! debug_assert_ne {
     ($($x:tt)*) => {
         {
@@ -72,6 +78,7 @@ macro_rules! debug_assert_ne {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! todo {
     ($($x:tt)*) => {
         {
@@ -83,20 +90,19 @@ macro_rules! todo {
     };
 }
 
-#[cfg(not(feature = "defmt"))]
+#[collapse_debuginfo(yes)]
 macro_rules! unreachable {
     ($($x:tt)*) => {
-        ::core::unreachable!($($x)*)
-    };
-}
-
-#[cfg(feature = "defmt")]
-macro_rules! unreachable {
-    ($($x:tt)*) => {
-        ::defmt::unreachable!($($x)*)
+        {
+            #[cfg(not(feature = "defmt"))]
+            ::core::unreachable!($($x)*);
+            #[cfg(feature = "defmt")]
+            ::defmt::unreachable!($($x)*);
+        }
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! panic {
     ($($x:tt)*) => {
         {
@@ -108,6 +114,7 @@ macro_rules! panic {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! trace {
     ($s:literal $(, $x:expr)* $(,)?) => {
         {
@@ -121,6 +128,7 @@ macro_rules! trace {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! debug {
     ($s:literal $(, $x:expr)* $(,)?) => {
         {
@@ -134,6 +142,7 @@ macro_rules! debug {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! info {
     ($s:literal $(, $x:expr)* $(,)?) => {
         {
@@ -147,6 +156,7 @@ macro_rules! info {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! warn {
     ($s:literal $(, $x:expr)* $(,)?) => {
         {
@@ -160,6 +170,7 @@ macro_rules! warn {
     };
 }
 
+#[collapse_debuginfo(yes)]
 macro_rules! error {
     ($s:literal $(, $x:expr)* $(,)?) => {
         {
@@ -174,6 +185,7 @@ macro_rules! error {
 }
 
 #[cfg(feature = "defmt")]
+#[collapse_debuginfo(yes)]
 macro_rules! unwrap {
     ($($x:tt)*) => {
         ::defmt::unwrap!($($x)*)
@@ -181,6 +193,7 @@ macro_rules! unwrap {
 }
 
 #[cfg(not(feature = "defmt"))]
+#[collapse_debuginfo(yes)]
 macro_rules! unwrap {
     ($arg:expr) => {
         match $crate::fmt::Try::into_result($arg) {
diff --git a/embassy-executor/src/lib.rs b/embassy-executor/src/lib.rs
index 553ed76d3..cffc76699 100644
--- a/embassy-executor/src/lib.rs
+++ b/embassy-executor/src/lib.rs
@@ -1,6 +1,6 @@
 #![cfg_attr(not(any(feature = "arch-std", feature = "arch-wasm")), no_std)]
-#![cfg_attr(feature = "nightly", feature(waker_getters))]
 #![allow(clippy::new_without_default)]
+#![allow(unsafe_op_in_unsafe_fn)]
 #![doc = include_str!("../README.md")]
 #![warn(missing_docs)]
 
@@ -24,120 +24,236 @@ macro_rules! check_at_most_one {
         check_at_most_one!(@amo [$($f)*] [$($f)*] []);
     };
 }
-check_at_most_one!("arch-avr", "arch-cortex-m", "arch-riscv32", "arch-std", "arch-wasm",);
+check_at_most_one!(
+    "arch-avr",
+    "arch-cortex-m",
+    "arch-cortex-ar",
+    "arch-riscv32",
+    "arch-std",
+    "arch-wasm",
+    "arch-spin",
+);
 
 #[cfg(feature = "_arch")]
 #[cfg_attr(feature = "arch-avr", path = "arch/avr.rs")]
 #[cfg_attr(feature = "arch-cortex-m", path = "arch/cortex_m.rs")]
+#[cfg_attr(feature = "arch-cortex-ar", path = "arch/cortex_ar.rs")]
 #[cfg_attr(feature = "arch-riscv32", path = "arch/riscv32.rs")]
 #[cfg_attr(feature = "arch-std", path = "arch/std.rs")]
 #[cfg_attr(feature = "arch-wasm", path = "arch/wasm.rs")]
+#[cfg_attr(feature = "arch-spin", path = "arch/spin.rs")]
 mod arch;
 
 #[cfg(feature = "_arch")]
 #[allow(unused_imports)] // don't warn if the module is empty.
 pub use arch::*;
+#[cfg(not(feature = "_arch"))]
+pub use embassy_executor_macros::main_unspecified as main;
 
 pub mod raw;
 
 mod spawner;
 pub use spawner::*;
 
-mod config {
-    #![allow(unused)]
-    include!(concat!(env!("OUT_DIR"), "/config.rs"));
-}
+mod metadata;
+pub use metadata::*;
 
 /// Implementation details for embassy macros.
 /// Do not use. Used for macros and HALs only. Not covered by semver guarantees.
 #[doc(hidden)]
 #[cfg(not(feature = "nightly"))]
 pub mod _export {
-    use core::alloc::Layout;
-    use core::cell::{Cell, UnsafeCell};
+    use core::cell::UnsafeCell;
     use core::future::Future;
     use core::mem::MaybeUninit;
-    use core::ptr::null_mut;
-
-    use critical_section::{CriticalSection, Mutex};
 
     use crate::raw::TaskPool;
 
-    struct Arena<const N: usize> {
-        buf: UnsafeCell<MaybeUninit<[u8; N]>>,
-        ptr: Mutex<Cell<*mut u8>>,
-    }
+    trait TaskReturnValue {}
+    impl TaskReturnValue for () {}
+    impl TaskReturnValue for Never {}
 
-    unsafe impl<const N: usize> Sync for Arena<N> {}
-    unsafe impl<const N: usize> Send for Arena<N> {}
+    #[diagnostic::on_unimplemented(
+        message = "task futures must resolve to `()` or `!`",
+        note = "use `async fn` or change the return type to `impl Future<Output = ()>`"
+    )]
+    #[allow(private_bounds)]
+    pub trait TaskFn<Args>: Copy {
+        type Fut: Future<Output: TaskReturnValue> + 'static;
+    }
 
-    impl<const N: usize> Arena<N> {
-        const fn new() -> Self {
-            Self {
-                buf: UnsafeCell::new(MaybeUninit::uninit()),
-                ptr: Mutex::new(Cell::new(null_mut())),
+    macro_rules! task_fn_impl {
+        ($($Tn:ident),*) => {
+            impl<F, Fut, $($Tn,)*> TaskFn<($($Tn,)*)> for F
+            where
+                F: Copy + FnOnce($($Tn,)*) -> Fut,
+                Fut: Future<Output: TaskReturnValue> + 'static,
+            {
+                type Fut = Fut;
             }
-        }
+        };
+    }
 
-        fn alloc<T>(&'static self, cs: CriticalSection) -> &'static mut MaybeUninit<T> {
-            let layout = Layout::new::<T>();
+    task_fn_impl!();
+    task_fn_impl!(T0);
+    task_fn_impl!(T0, T1);
+    task_fn_impl!(T0, T1, T2);
+    task_fn_impl!(T0, T1, T2, T3);
+    task_fn_impl!(T0, T1, T2, T3, T4);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14);
+    task_fn_impl!(T0, T1, T2, T3, T4, T5, T6, T7, T8, T9, T10, T11, T12, T13, T14, T15);
 
-            let start = self.buf.get().cast::<u8>();
-            let end = unsafe { start.add(N) };
+    #[allow(private_bounds)]
+    #[repr(C)]
+    pub struct TaskPoolHolder<const SIZE: usize, const ALIGN: usize>
+    where
+        Align<ALIGN>: Alignment,
+    {
+        data: UnsafeCell<[MaybeUninit<u8>; SIZE]>,
+        align: Align<ALIGN>,
+    }
 
-            let mut ptr = self.ptr.borrow(cs).get();
-            if ptr.is_null() {
-                ptr = self.buf.get().cast::<u8>();
-            }
+    unsafe impl<const SIZE: usize, const ALIGN: usize> Send for TaskPoolHolder<SIZE, ALIGN> where Align<ALIGN>: Alignment {}
+    unsafe impl<const SIZE: usize, const ALIGN: usize> Sync for TaskPoolHolder<SIZE, ALIGN> where Align<ALIGN>: Alignment {}
 
-            let bytes_left = (end as usize) - (ptr as usize);
-            let align_offset = (ptr as usize).next_multiple_of(layout.align()) - (ptr as usize);
+    #[allow(private_bounds)]
+    impl<const SIZE: usize, const ALIGN: usize> TaskPoolHolder<SIZE, ALIGN>
+    where
+        Align<ALIGN>: Alignment,
+    {
+        pub const fn get(&self) -> *const u8 {
+            self.data.get().cast()
+        }
+    }
 
-            if align_offset + layout.size() > bytes_left {
-                panic!("embassy-executor: task arena is full. You must increase the arena size, see the documentation for details: https://docs.embassy.dev/embassy-executor/");
-            }
+    pub const fn task_pool_size<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+    where
+        F: TaskFn<Args, Fut = Fut>,
+        Fut: Future + 'static,
+    {
+        size_of::<TaskPool<Fut, POOL_SIZE>>()
+    }
+
+    pub const fn task_pool_align<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+    where
+        F: TaskFn<Args, Fut = Fut>,
+        Fut: Future + 'static,
+    {
+        align_of::<TaskPool<Fut, POOL_SIZE>>()
+    }
 
-            let res = unsafe { ptr.add(align_offset) };
-            let ptr = unsafe { ptr.add(align_offset + layout.size()) };
+    pub const fn task_pool_new<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> TaskPool<Fut, POOL_SIZE>
+    where
+        F: TaskFn<Args, Fut = Fut>,
+        Fut: Future + 'static,
+    {
+        TaskPool::new()
+    }
 
-            self.ptr.borrow(cs).set(ptr);
+    #[allow(private_bounds)]
+    #[repr(transparent)]
+    pub struct Align<const N: usize>([<Self as Alignment>::Archetype; 0])
+    where
+        Self: Alignment;
 
-            unsafe { &mut *(res as *mut MaybeUninit<T>) }
-        }
+    trait Alignment {
+        /// A zero-sized type of particular alignment.
+        type Archetype: Copy + Eq + PartialEq + Send + Sync + Unpin;
     }
 
-    static ARENA: Arena<{ crate::config::TASK_ARENA_SIZE }> = Arena::new();
+    macro_rules! aligns {
+        ($($AlignX:ident: $n:literal,)*) => {
+            $(
+                #[derive(Copy, Clone, Eq, PartialEq)]
+                #[repr(align($n))]
+                struct $AlignX {}
+                impl Alignment for Align<$n> {
+                    type Archetype = $AlignX;
+                }
+            )*
+        };
+    }
+
+    aligns!(
+        Align1:         1,
+        Align2:         2,
+        Align4:         4,
+        Align8:         8,
+        Align16:        16,
+        Align32:        32,
+        Align64:        64,
+        Align128:       128,
+        Align256:       256,
+        Align512:       512,
+        Align1024:      1024,
+        Align2048:      2048,
+        Align4096:      4096,
+        Align8192:      8192,
+        Align16384:     16384,
+    );
+    #[cfg(any(target_pointer_width = "32", target_pointer_width = "64"))]
+    aligns!(
+        Align32768:     32768,
+        Align65536:     65536,
+        Align131072:    131072,
+        Align262144:    262144,
+        Align524288:    524288,
+        Align1048576:   1048576,
+        Align2097152:   2097152,
+        Align4194304:   4194304,
+        Align8388608:   8388608,
+        Align16777216:  16777216,
+        Align33554432:  33554432,
+        Align67108864:  67108864,
+        Align134217728: 134217728,
+        Align268435456: 268435456,
+        Align536870912: 536870912,
+    );
 
-    pub struct TaskPoolRef {
-        // type-erased `&'static mut TaskPool<F, N>`
-        // Needed because statics can't have generics.
-        ptr: Mutex<Cell<*mut ()>>,
+    #[allow(dead_code)]
+    pub trait HasOutput {
+        type Output;
     }
-    unsafe impl Sync for TaskPoolRef {}
-    unsafe impl Send for TaskPoolRef {}
 
-    impl TaskPoolRef {
-        pub const fn new() -> Self {
-            Self {
-                ptr: Mutex::new(Cell::new(null_mut())),
-            }
-        }
+    impl<O> HasOutput for fn() -> O {
+        type Output = O;
+    }
 
-        /// Get the pool for this ref, allocating it from the arena the first time.
-        ///
-        /// safety: for a given TaskPoolRef instance, must always call with the exact
-        /// same generic params.
-        pub unsafe fn get<F: Future, const N: usize>(&'static self) -> &'static TaskPool<F, N> {
-            critical_section::with(|cs| {
-                let ptr = self.ptr.borrow(cs);
-                if ptr.get().is_null() {
-                    let pool = ARENA.alloc::<TaskPool<F, N>>(cs);
-                    pool.write(TaskPool::new());
-                    ptr.set(pool as *mut _ as _);
-                }
+    #[allow(dead_code)]
+    pub type Never = <fn() -> ! as HasOutput>::Output;
+}
 
-                unsafe { &*(ptr.get() as *const _) }
-            })
-        }
+/// Implementation details for embassy macros.
+/// Do not use. Used for macros and HALs only. Not covered by semver guarantees.
+#[doc(hidden)]
+#[cfg(feature = "nightly")]
+pub mod _export {
+    #[diagnostic::on_unimplemented(
+        message = "task futures must resolve to `()` or `!`",
+        note = "use `async fn` or change the return type to `impl Future<Output = ()>`"
+    )]
+    pub trait TaskReturnValue {}
+    impl TaskReturnValue for () {}
+    impl TaskReturnValue for Never {}
+
+    #[allow(dead_code)]
+    pub trait HasOutput {
+        type Output;
+    }
+
+    impl<O> HasOutput for fn() -> O {
+        type Output = O;
     }
+
+    #[allow(dead_code)]
+    pub type Never = <fn() -> ! as HasOutput>::Output;
 }
diff --git a/embassy-executor/src/metadata.rs b/embassy-executor/src/metadata.rs
new file mode 100644
index 000000000..76504ab0b
--- /dev/null
+++ b/embassy-executor/src/metadata.rs
@@ -0,0 +1,148 @@
+#[cfg(feature = "metadata-name")]
+use core::cell::Cell;
+use core::future::{Future, poll_fn};
+#[cfg(feature = "scheduler-priority")]
+use core::sync::atomic::{AtomicU8, Ordering};
+use core::task::Poll;
+
+#[cfg(feature = "metadata-name")]
+use critical_section::Mutex;
+
+use crate::raw;
+#[cfg(feature = "scheduler-deadline")]
+use crate::raw::Deadline;
+
+/// Metadata associated with a task.
+pub struct Metadata {
+    #[cfg(feature = "metadata-name")]
+    name: Mutex<Cell<Option<&'static str>>>,
+    #[cfg(feature = "scheduler-priority")]
+    priority: AtomicU8,
+    #[cfg(feature = "scheduler-deadline")]
+    deadline: raw::Deadline,
+}
+
+impl Metadata {
+    pub(crate) const fn new() -> Self {
+        Self {
+            #[cfg(feature = "metadata-name")]
+            name: Mutex::new(Cell::new(None)),
+            #[cfg(feature = "scheduler-priority")]
+            priority: AtomicU8::new(0),
+            // NOTE: The deadline is set to zero to allow the initializer to reside in `.bss`. This
+            // will be lazily initalized in `initialize_impl`
+            #[cfg(feature = "scheduler-deadline")]
+            deadline: raw::Deadline::new_unset(),
+        }
+    }
+
+    pub(crate) fn reset(&self) {
+        #[cfg(feature = "metadata-name")]
+        critical_section::with(|cs| self.name.borrow(cs).set(None));
+
+        #[cfg(feature = "scheduler-priority")]
+        self.set_priority(0);
+
+        // By default, deadlines are set to the maximum value, so that any task WITH
+        // a set deadline will ALWAYS be scheduled BEFORE a task WITHOUT a set deadline
+        #[cfg(feature = "scheduler-deadline")]
+        self.unset_deadline();
+    }
+
+    /// Get the metadata for the current task.
+    ///
+    /// You can use this to read or modify the current task's metadata.
+    ///
+    /// This function is `async` just to get access to the current async
+    /// context. It returns instantly, it does not block/yield.
+    pub fn for_current_task() -> impl Future<Output = &'static Self> {
+        poll_fn(|cx| Poll::Ready(raw::task_from_waker(cx.waker()).metadata()))
+    }
+
+    /// Get this task's name
+    ///
+    /// NOTE: this takes a critical section.
+    #[cfg(feature = "metadata-name")]
+    pub fn name(&self) -> Option<&'static str> {
+        critical_section::with(|cs| self.name.borrow(cs).get())
+    }
+
+    /// Set this task's name
+    ///
+    /// NOTE: this takes a critical section.
+    #[cfg(feature = "metadata-name")]
+    pub fn set_name(&self, name: &'static str) {
+        critical_section::with(|cs| self.name.borrow(cs).set(Some(name)))
+    }
+
+    /// Get this task's priority.
+    #[cfg(feature = "scheduler-priority")]
+    pub fn priority(&self) -> u8 {
+        self.priority.load(Ordering::Relaxed)
+    }
+
+    /// Set this task's priority.
+    #[cfg(feature = "scheduler-priority")]
+    pub fn set_priority(&self, priority: u8) {
+        self.priority.store(priority, Ordering::Relaxed)
+    }
+
+    /// Get this task's deadline.
+    #[cfg(feature = "scheduler-deadline")]
+    pub fn deadline(&self) -> u64 {
+        self.deadline.instant_ticks()
+    }
+
+    /// Set this task's deadline.
+    ///
+    /// This method does NOT check whether the deadline has already passed.
+    #[cfg(feature = "scheduler-deadline")]
+    pub fn set_deadline(&self, instant_ticks: u64) {
+        self.deadline.set(instant_ticks);
+    }
+
+    /// Remove this task's deadline.
+    /// This brings it back to the defaul where it's not scheduled ahead of other tasks.
+    #[cfg(feature = "scheduler-deadline")]
+    pub fn unset_deadline(&self) {
+        self.deadline.set(Deadline::UNSET_TICKS);
+    }
+
+    /// Set this task's deadline `duration_ticks` in the future from when
+    /// this future is polled. This deadline is saturated to the max tick value.
+    ///
+    /// Analogous to `Timer::after`.
+    #[cfg(all(feature = "scheduler-deadline", feature = "embassy-time-driver"))]
+    pub fn set_deadline_after(&self, duration_ticks: u64) {
+        let now = embassy_time_driver::now();
+
+        // Since ticks is a u64, saturating add is PROBABLY overly cautious, leave
+        // it for now, we can probably make this wrapping_add for performance
+        // reasons later.
+        let deadline = now.saturating_add(duration_ticks);
+
+        self.set_deadline(deadline);
+    }
+
+    /// Set the this task's deadline `increment_ticks` from the previous deadline.
+    ///
+    /// This deadline is saturated to the max tick value.
+    ///
+    /// Note that by default (unless otherwise set), tasks start life with the deadline
+    /// not set, which means this method will have no effect.
+    ///
+    /// Analogous to one increment of `Ticker::every().next()`.
+    ///
+    /// Returns the deadline that was set.
+    #[cfg(feature = "scheduler-deadline")]
+    pub fn increment_deadline(&self, duration_ticks: u64) {
+        let last = self.deadline();
+
+        // Since ticks is a u64, saturating add is PROBABLY overly cautious, leave
+        // it for now, we can probably make this wrapping_add for performance
+        // reasons later.
+        let deadline = last.saturating_add(duration_ticks);
+
+        self.set_deadline(deadline);
+    }
+}
diff --git a/embassy-executor/src/raw/deadline.rs b/embassy-executor/src/raw/deadline.rs
new file mode 100644
index 000000000..cc89fadb0
--- /dev/null
+++ b/embassy-executor/src/raw/deadline.rs
@@ -0,0 +1,44 @@
+use core::sync::atomic::{AtomicU32, Ordering};
+
+/// A type for interacting with the deadline of the current task
+///
+/// Requires the `scheduler-deadline` feature.
+///
+/// Note: Interacting with the deadline should be done locally in a task.
+/// In theory you could try to set or read the deadline from another task,
+/// but that will result in weird (though not unsound) behavior.
+pub(crate) struct Deadline {
+    instant_ticks_hi: AtomicU32,
+    instant_ticks_lo: AtomicU32,
+}
+
+impl Deadline {
+    pub(crate) const fn new(instant_ticks: u64) -> Self {
+        Self {
+            instant_ticks_hi: AtomicU32::new((instant_ticks >> 32) as u32),
+            instant_ticks_lo: AtomicU32::new(instant_ticks as u32),
+        }
+    }
+
+    pub(crate) const fn new_unset() -> Self {
+        Self::new(Self::UNSET_TICKS)
+    }
+
+    pub(crate) fn set(&self, instant_ticks: u64) {
+        self.instant_ticks_hi
+            .store((instant_ticks >> 32) as u32, Ordering::Relaxed);
+        self.instant_ticks_lo.store(instant_ticks as u32, Ordering::Relaxed);
+    }
+
+    /// Deadline value in ticks, same time base and ticks as `embassy-time`
+    pub(crate) fn instant_ticks(&self) -> u64 {
+        let hi = self.instant_ticks_hi.load(Ordering::Relaxed) as u64;
+        let lo = self.instant_ticks_lo.load(Ordering::Relaxed) as u64;
+
+        (hi << 32) | lo
+    }
+
+    /// Sentinel value representing an "unset" deadline, which has lower priority
+    /// than any other set deadline value
+    pub(crate) const UNSET_TICKS: u64 = u64::MAX;
+}
diff --git a/embassy-executor/src/raw/mod.rs b/embassy-executor/src/raw/mod.rs
index d9ea5c005..ab845ed3b 100644
--- a/embassy-executor/src/raw/mod.rs
+++ b/embassy-executor/src/raw/mod.rs
@@ -7,54 +7,113 @@
 //! Using this module requires respecting subtle safety contracts. If you can, prefer using the safe
 //! [executor wrappers](crate::Executor) and the [`embassy_executor::task`](embassy_executor_macros::task) macro, which are fully safe.
 
-#[cfg_attr(target_has_atomic = "ptr", path = "run_queue_atomics.rs")]
-#[cfg_attr(not(target_has_atomic = "ptr"), path = "run_queue_critical_section.rs")]
 mod run_queue;
 
-#[cfg_attr(all(cortex_m, target_has_atomic = "8"), path = "state_atomics_arm.rs")]
-#[cfg_attr(all(not(cortex_m), target_has_atomic = "8"), path = "state_atomics.rs")]
-#[cfg_attr(not(target_has_atomic = "8"), path = "state_critical_section.rs")]
+#[cfg_attr(all(cortex_m, target_has_atomic = "32"), path = "state_atomics_arm.rs")]
+#[cfg_attr(
+    all(not(cortex_m), any(target_has_atomic = "8", target_has_atomic = "32")),
+    path = "state_atomics.rs"
+)]
+#[cfg_attr(
+    not(any(target_has_atomic = "8", target_has_atomic = "32")),
+    path = "state_critical_section.rs"
+)]
 mod state;
 
-#[cfg(feature = "integrated-timers")]
-mod timer_queue;
+#[cfg(feature = "_any_trace")]
+pub mod trace;
 pub(crate) mod util;
 #[cfg_attr(feature = "turbowakers", path = "waker_turbo.rs")]
 mod waker;
 
+#[cfg(feature = "scheduler-deadline")]
+mod deadline;
+
 use core::future::Future;
 use core::marker::PhantomData;
 use core::mem;
 use core::pin::Pin;
 use core::ptr::NonNull;
-use core::task::{Context, Poll};
+#[cfg(not(feature = "arch-avr"))]
+use core::sync::atomic::AtomicPtr;
+use core::sync::atomic::Ordering;
+use core::task::{Context, Poll, Waker};
 
-#[cfg(feature = "integrated-timers")]
-use embassy_time_driver::AlarmHandle;
-#[cfg(feature = "rtos-trace")]
-use rtos_trace::trace;
+#[cfg(feature = "scheduler-deadline")]
+pub(crate) use deadline::Deadline;
+use embassy_executor_timer_queue::TimerQueueItem;
+#[cfg(feature = "arch-avr")]
+use portable_atomic::AtomicPtr;
 
 use self::run_queue::{RunQueue, RunQueueItem};
 use self::state::State;
 use self::util::{SyncUnsafeCell, UninitCell};
 pub use self::waker::task_from_waker;
 use super::SpawnToken;
+use crate::{Metadata, SpawnError};
+
+#[unsafe(no_mangle)]
+extern "Rust" fn __embassy_time_queue_item_from_waker(waker: &Waker) -> &'static mut TimerQueueItem {
+    unsafe { task_from_waker(waker).timer_queue_item() }
+}
 
 /// Raw task header for use in task pointers.
+///
+/// A task can be in one of the following states:
+///
+/// - Not spawned: the task is ready to spawn.
+/// - `SPAWNED`: the task is currently spawned and may be running.
+/// - `RUN_ENQUEUED`: the task is enqueued to be polled. Note that the task may be `!SPAWNED`.
+///    In this case, the `RUN_ENQUEUED` state will be cleared when the task is next polled, without
+///    polling the task's future.
+///
+/// A task's complete life cycle is as follows:
+///
+/// ```text
+/// ┌────────────┐   ┌────────────────────────┐
+/// │Not spawned │◄─5┤Not spawned|Run enqueued│
+/// │            ├6─►│                        │
+/// └─────┬──────┘   └──────▲─────────────────┘
+///       1                 │
+///       │    ┌────────────┘
+///       │    4
+/// ┌─────▼────┴─────────┐
+/// │Spawned|Run enqueued│
+/// │                    │
+/// └─────┬▲─────────────┘
+///       2│
+///       │3
+/// ┌─────▼┴─────┐
+/// │  Spawned   │
+/// │            │
+/// └────────────┘
+/// ```
+///
+/// Transitions:
+/// - 1: Task is spawned - `AvailableTask::claim -> Executor::spawn`
+/// - 2: During poll - `RunQueue::dequeue_all -> State::run_dequeue`
+/// - 3: Task wakes itself, waker wakes task, or task exits - `Waker::wake -> wake_task -> State::run_enqueue`
+/// - 4: A run-queued task exits - `TaskStorage::poll -> Poll::Ready`
+/// - 5: Task is dequeued. The task's future is not polled, because exiting the task replaces its `poll_fn`.
+/// - 6: A task is waken when it is not spawned - `wake_task -> State::run_enqueue`
 pub(crate) struct TaskHeader {
     pub(crate) state: State,
     pub(crate) run_queue_item: RunQueueItem,
-    pub(crate) executor: SyncUnsafeCell<Option<&'static SyncExecutor>>,
+
+    pub(crate) executor: AtomicPtr<SyncExecutor>,
     poll_fn: SyncUnsafeCell<Option<unsafe fn(TaskRef)>>,
 
-    #[cfg(feature = "integrated-timers")]
-    pub(crate) expires_at: SyncUnsafeCell<u64>,
-    #[cfg(feature = "integrated-timers")]
-    pub(crate) timer_queue_item: timer_queue::TimerQueueItem,
+    /// Integrated timer queue storage. This field should not be accessed outside of the timer queue.
+    pub(crate) timer_queue_item: TimerQueueItem,
+
+    pub(crate) metadata: Metadata,
+
+    #[cfg(feature = "rtos-trace")]
+    all_tasks_next: AtomicPtr<TaskHeader>,
 }
 
 /// This is essentially a `&'static TaskStorage<F>` where the type of the future has been erased.
-#[derive(Clone, Copy)]
+#[derive(Debug, Clone, Copy, PartialEq)]
 pub struct TaskRef {
     ptr: NonNull<TaskHeader>,
 }
@@ -80,10 +139,35 @@ impl TaskRef {
         unsafe { self.ptr.as_ref() }
     }
 
+    pub(crate) fn metadata(self) -> &'static Metadata {
+        unsafe { &self.ptr.as_ref().metadata }
+    }
+
+    /// Returns a reference to the executor that the task is currently running on.
+    pub unsafe fn executor(self) -> Option<&'static Executor> {
+        let executor = self.header().executor.load(Ordering::Relaxed);
+        executor.as_ref().map(|e| Executor::wrap(e))
+    }
+
+    /// Returns a mutable reference to the timer queue item.
+    ///
+    /// Safety
+    ///
+    /// This function must only be called in the context of the integrated timer queue.
+    pub unsafe fn timer_queue_item(mut self) -> &'static mut TimerQueueItem {
+        unsafe { &mut self.ptr.as_mut().timer_queue_item }
+    }
+
     /// The returned pointer is valid for the entire TaskStorage.
     pub(crate) fn as_ptr(self) -> *const TaskHeader {
         self.ptr.as_ptr()
     }
+
+    /// Returns the task ID.
+    /// This can be used in combination with rtos-trace to match task names with IDs
+    pub fn id(&self) -> u32 {
+        self.as_ptr() as u32
+    }
 }
 
 /// Raw storage in which a task can be spawned.
@@ -107,6 +191,10 @@ pub struct TaskStorage<F: Future + 'static> {
     future: UninitCell<F>, // Valid if STATE_SPAWNED
 }
 
+unsafe fn poll_exited(_p: TaskRef) {
+    // Nothing to do, the task is already !SPAWNED and dequeued.
+}
+
 impl<F: Future + 'static> TaskStorage<F> {
     const NEW: Self = Self::new();
 
@@ -116,14 +204,14 @@ impl<F: Future + 'static> TaskStorage<F> {
             raw: TaskHeader {
                 state: State::new(),
                 run_queue_item: RunQueueItem::new(),
-                executor: SyncUnsafeCell::new(None),
+                executor: AtomicPtr::new(core::ptr::null_mut()),
                 // Note: this is lazily initialized so that a static `TaskStorage` will go in `.bss`
                 poll_fn: SyncUnsafeCell::new(None),
 
-                #[cfg(feature = "integrated-timers")]
-                expires_at: SyncUnsafeCell::new(0),
-                #[cfg(feature = "integrated-timers")]
-                timer_queue_item: timer_queue::TimerQueueItem::new(),
+                timer_queue_item: TimerQueueItem::new(),
+                metadata: Metadata::new(),
+                #[cfg(feature = "rtos-trace")]
+                all_tasks_next: AtomicPtr::new(core::ptr::null_mut()),
             },
             future: UninitCell::uninit(),
         }
@@ -142,27 +230,39 @@ impl<F: Future + 'static> TaskStorage<F> {
     ///
     /// Once the task has finished running, you may spawn it again. It is allowed to spawn it
     /// on a different executor.
-    pub fn spawn(&'static self, future: impl FnOnce() -> F) -> SpawnToken<impl Sized> {
+    pub fn spawn(&'static self, future: impl FnOnce() -> F) -> Result<SpawnToken<impl Sized>, SpawnError> {
         let task = AvailableTask::claim(self);
         match task {
-            Some(task) => task.initialize(future),
-            None => SpawnToken::new_failed(),
+            Some(task) => Ok(task.initialize(future)),
+            None => Err(SpawnError::Busy),
         }
     }
 
     unsafe fn poll(p: TaskRef) {
-        let this = &*(p.as_ptr() as *const TaskStorage<F>);
+        let this = &*p.as_ptr().cast::<TaskStorage<F>>();
 
         let future = Pin::new_unchecked(this.future.as_mut());
         let waker = waker::from_task(p);
         let mut cx = Context::from_waker(&waker);
         match future.poll(&mut cx) {
             Poll::Ready(_) => {
+                #[cfg(feature = "_any_trace")]
+                let exec_ptr: *const SyncExecutor = this.raw.executor.load(Ordering::Relaxed);
+
+                // As the future has finished and this function will not be called
+                // again, we can safely drop the future here.
                 this.future.drop_in_place();
+
+                // We replace the poll_fn with a despawn function, so that the task is cleaned up
+                // when the executor polls it next.
+                this.raw.poll_fn.set(Some(poll_exited));
+
+                // Make sure we despawn last, so that other threads can only spawn the task
+                // after we're done with it.
                 this.raw.state.despawn();
 
-                #[cfg(feature = "integrated-timers")]
-                this.raw.expires_at.set(u64::MAX);
+                #[cfg(feature = "_any_trace")]
+                trace::task_end(exec_ptr, &p);
             }
             Poll::Pending => {}
         }
@@ -196,6 +296,7 @@ impl<F: Future + 'static> AvailableTask<F> {
 
     fn initialize_impl<S>(self, future: impl FnOnce() -> F) -> SpawnToken<S> {
         unsafe {
+            self.task.raw.metadata.reset();
             self.task.raw.poll_fn.set(Some(TaskStorage::<F>::poll));
             self.task.future.write_in_place(future);
 
@@ -262,10 +363,10 @@ impl<F: Future + 'static, const N: usize> TaskPool<F, N> {
         }
     }
 
-    fn spawn_impl<T>(&'static self, future: impl FnOnce() -> F) -> SpawnToken<T> {
+    fn spawn_impl<T>(&'static self, future: impl FnOnce() -> F) -> Result<SpawnToken<T>, SpawnError> {
         match self.pool.iter().find_map(AvailableTask::claim) {
-            Some(task) => task.initialize_impl::<T>(future),
-            None => SpawnToken::new_failed(),
+            Some(task) => Ok(task.initialize_impl::<T>(future)),
+            None => Err(SpawnError::Busy),
         }
     }
 
@@ -276,7 +377,7 @@ impl<F: Future + 'static, const N: usize> TaskPool<F, N> {
     /// This will loop over the pool and spawn the task in the first storage that
     /// is currently free. If none is free, a "poisoned" SpawnToken is returned,
     /// which will cause [`Spawner::spawn()`](super::Spawner::spawn) to return the error.
-    pub fn spawn(&'static self, future: impl FnOnce() -> F) -> SpawnToken<impl Sized> {
+    pub fn spawn(&'static self, future: impl FnOnce() -> F) -> Result<SpawnToken<impl Sized>, SpawnError> {
         self.spawn_impl::<F>(future)
     }
 
@@ -289,7 +390,7 @@ impl<F: Future + 'static, const N: usize> TaskPool<F, N> {
     /// SAFETY: `future` must be a closure of the form `move || my_async_fn(args)`, where `my_async_fn`
     /// is an `async fn`, NOT a hand-written `Future`.
     #[doc(hidden)]
-    pub unsafe fn _spawn_async_fn<FutFn>(&'static self, future: FutFn) -> SpawnToken<impl Sized>
+    pub unsafe fn _spawn_async_fn<FutFn>(&'static self, future: FutFn) -> Result<SpawnToken<impl Sized>, SpawnError>
     where
         FutFn: FnOnce() -> F,
     {
@@ -306,7 +407,7 @@ unsafe impl Sync for Pender {}
 
 impl Pender {
     pub(crate) fn pend(self) {
-        extern "Rust" {
+        unsafe extern "Rust" {
             fn __pender(context: *mut ());
         }
         unsafe { __pender(self.0) };
@@ -316,26 +417,13 @@ impl Pender {
 pub(crate) struct SyncExecutor {
     run_queue: RunQueue,
     pender: Pender,
-
-    #[cfg(feature = "integrated-timers")]
-    pub(crate) timer_queue: timer_queue::TimerQueue,
-    #[cfg(feature = "integrated-timers")]
-    alarm: AlarmHandle,
 }
 
 impl SyncExecutor {
     pub(crate) fn new(pender: Pender) -> Self {
-        #[cfg(feature = "integrated-timers")]
-        let alarm = unsafe { unwrap!(embassy_time_driver::allocate_alarm()) };
-
         Self {
             run_queue: RunQueue::new(),
             pender,
-
-            #[cfg(feature = "integrated-timers")]
-            timer_queue: timer_queue::TimerQueue::new(),
-            #[cfg(feature = "integrated-timers")]
-            alarm,
         }
     }
 
@@ -346,90 +434,50 @@ impl SyncExecutor {
     /// - `task` must be set up to run in this executor.
     /// - `task` must NOT be already enqueued (in this executor or another one).
     #[inline(always)]
-    unsafe fn enqueue(&self, task: TaskRef) {
-        #[cfg(feature = "rtos-trace")]
-        trace::task_ready_begin(task.as_ptr() as u32);
+    unsafe fn enqueue(&self, task: TaskRef, l: state::Token) {
+        #[cfg(feature = "_any_trace")]
+        trace::task_ready_begin(self, &task);
 
-        if self.run_queue.enqueue(task) {
+        if self.run_queue.enqueue(task, l) {
             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));
+        task.header()
+            .executor
+            .store((self as *const Self).cast_mut(), Ordering::Relaxed);
 
-        #[cfg(feature = "rtos-trace")]
-        trace::task_new(task.as_ptr() as u32);
+        #[cfg(feature = "_any_trace")]
+        trace::task_new(self, &task);
 
-        self.enqueue(task);
+        state::locked(|l| {
+            self.enqueue(task, l);
+        })
     }
 
     /// # Safety
     ///
     /// 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")]
-        embassy_time_driver::set_alarm_callback(self.alarm, Self::alarm_callback, self as *const _ as *mut ());
-
-        #[allow(clippy::never_loop)]
-        loop {
-            #[cfg(feature = "integrated-timers")]
-            self.timer_queue
-                .dequeue_expired(embassy_time_driver::now(), wake_task_no_pend);
-
-            self.run_queue.dequeue_all(|p| {
-                let task = p.header();
-
-                #[cfg(feature = "integrated-timers")]
-                task.expires_at.set(u64::MAX);
-
-                if !task.state.run_dequeue() {
-                    // If task is not running, ignore it. This can happen in the following scenario:
-                    //   - Task gets dequeued, poll starts
-                    //   - While task is being polled, it gets woken. It gets placed in the queue.
-                    //   - Task poll finishes, returning done=true
-                    //   - RUNNING bit is cleared, but the task is already in the queue.
-                    return;
-                }
+        #[cfg(feature = "_any_trace")]
+        trace::poll_start(self);
 
-                #[cfg(feature = "rtos-trace")]
-                trace::task_exec_begin(p.as_ptr() as u32);
+        self.run_queue.dequeue_all(|p| {
+            let task = p.header();
 
-                // Run the task
-                task.poll_fn.get().unwrap_unchecked()(p);
+            #[cfg(feature = "_any_trace")]
+            trace::task_exec_begin(self, &p);
 
-                #[cfg(feature = "rtos-trace")]
-                trace::task_exec_end();
-
-                // Enqueue or update into timer_queue
-                #[cfg(feature = "integrated-timers")]
-                self.timer_queue.update(p);
-            });
-
-            #[cfg(feature = "integrated-timers")]
-            {
-                // If this is already in the past, set_alarm might return false
-                // In that case do another poll loop iteration.
-                let next_expiration = self.timer_queue.next_expiration();
-                if embassy_time_driver::set_alarm(self.alarm, next_expiration) {
-                    break;
-                }
-            }
+            // Run the task
+            task.poll_fn.get().unwrap_unchecked()(p);
 
-            #[cfg(not(feature = "integrated-timers"))]
-            {
-                break;
-            }
-        }
+            #[cfg(feature = "_any_trace")]
+            trace::task_exec_end(self, &p);
+        });
 
-        #[cfg(feature = "rtos-trace")]
-        trace::system_idle();
+        #[cfg(feature = "_any_trace")]
+        trace::executor_idle(self)
     }
 }
 
@@ -459,7 +507,7 @@ impl SyncExecutor {
 /// The pender function must be exported with the name `__pender` and have the following signature:
 ///
 /// ```rust
-/// #[export_name = "__pender"]
+/// #[unsafe(export_name = "__pender")]
 /// fn pender(context: *mut ()) {
 ///    // schedule `poll()` to be called
 /// }
@@ -533,6 +581,11 @@ impl Executor {
     pub fn spawner(&'static self) -> super::Spawner {
         super::Spawner::new(self)
     }
+
+    /// Get a unique ID for this Executor.
+    pub fn id(&'static self) -> usize {
+        &self.inner as *const SyncExecutor as usize
+    }
 }
 
 /// Wake a task by `TaskRef`.
@@ -540,13 +593,13 @@ impl Executor {
 /// You can obtain a `TaskRef` from a `Waker` using [`task_from_waker`].
 pub fn wake_task(task: TaskRef) {
     let header = task.header();
-    if header.state.run_enqueue() {
+    header.state.run_enqueue(|l| {
         // We have just marked the task as scheduled, so enqueue it.
         unsafe {
-            let executor = header.executor.get().unwrap_unchecked();
-            executor.enqueue(task);
+            let executor = header.executor.load(Ordering::Relaxed).as_ref().unwrap_unchecked();
+            executor.enqueue(task, l);
         }
-    }
+    });
 }
 
 /// Wake a task by `TaskRef` without calling pend.
@@ -554,57 +607,11 @@ pub fn wake_task(task: TaskRef) {
 /// You can obtain a `TaskRef` from a `Waker` using [`task_from_waker`].
 pub fn wake_task_no_pend(task: TaskRef) {
     let header = task.header();
-    if header.state.run_enqueue() {
+    header.state.run_enqueue(|l| {
         // We have just marked the task as scheduled, so enqueue it.
         unsafe {
-            let executor = header.executor.get().unwrap_unchecked();
-            executor.run_queue.enqueue(task);
+            let executor = header.executor.load(Ordering::Relaxed).as_ref().unwrap_unchecked();
+            executor.run_queue.enqueue(task, l);
         }
-    }
+    });
 }
-
-#[cfg(feature = "integrated-timers")]
-struct TimerQueue;
-
-#[cfg(feature = "integrated-timers")]
-impl embassy_time_queue_driver::TimerQueue for TimerQueue {
-    fn schedule_wake(&'static self, at: u64, waker: &core::task::Waker) {
-        let task = waker::task_from_waker(waker);
-        let task = task.header();
-        unsafe {
-            let expires_at = task.expires_at.get();
-            task.expires_at.set(expires_at.min(at));
-        }
-    }
-}
-
-#[cfg(feature = "integrated-timers")]
-embassy_time_queue_driver::timer_queue_impl!(static TIMER_QUEUE: TimerQueue = TimerQueue);
-
-#[cfg(all(feature = "rtos-trace", feature = "integrated-timers"))]
-const fn gcd(a: u64, b: u64) -> u64 {
-    if b == 0 {
-        a
-    } else {
-        gcd(b, a % b)
-    }
-}
-
-#[cfg(feature = "rtos-trace")]
-impl rtos_trace::RtosTraceOSCallbacks for Executor {
-    fn task_list() {
-        // We don't know what tasks exist, so we can't send them.
-    }
-    #[cfg(feature = "integrated-timers")]
-    fn time() -> u64 {
-        const GCD_1M: u64 = gcd(embassy_time_driver::TICK_HZ, 1_000_000);
-        embassy_time_driver::now() * (1_000_000 / GCD_1M) / (embassy_time_driver::TICK_HZ / GCD_1M)
-    }
-    #[cfg(not(feature = "integrated-timers"))]
-    fn time() -> u64 {
-        0
-    }
-}
-
-#[cfg(feature = "rtos-trace")]
-rtos_trace::global_os_callbacks! {Executor}
diff --git a/embassy-executor/src/raw/run_queue.rs b/embassy-executor/src/raw/run_queue.rs
new file mode 100644
index 000000000..6f2abdbd0
--- /dev/null
+++ b/embassy-executor/src/raw/run_queue.rs
@@ -0,0 +1,213 @@
+use core::ptr::{NonNull, addr_of_mut};
+
+use cordyceps::Linked;
+#[cfg(any(feature = "scheduler-priority", feature = "scheduler-deadline"))]
+use cordyceps::SortedList;
+use cordyceps::sorted_list::Links;
+
+#[cfg(target_has_atomic = "ptr")]
+type TransferStack<T> = cordyceps::TransferStack<T>;
+
+#[cfg(not(target_has_atomic = "ptr"))]
+type TransferStack<T> = MutexTransferStack<T>;
+
+use super::{TaskHeader, TaskRef};
+
+/// Use `cordyceps::sorted_list::Links` as the singly linked list
+/// for RunQueueItems.
+pub(crate) type RunQueueItem = Links<TaskHeader>;
+
+/// Implements the `Linked` trait, allowing for singly linked list usage
+/// of any of cordyceps' `TransferStack` (used for the atomic runqueue),
+/// `SortedList` (used with the DRS scheduler), or `Stack`, which is
+/// popped atomically from the `TransferStack`.
+unsafe impl Linked<Links<TaskHeader>> for TaskHeader {
+    type Handle = TaskRef;
+
+    // Convert a TaskRef into a TaskHeader ptr
+    fn into_ptr(r: TaskRef) -> NonNull<TaskHeader> {
+        r.ptr
+    }
+
+    // Convert a TaskHeader into a TaskRef
+    unsafe fn from_ptr(ptr: NonNull<TaskHeader>) -> TaskRef {
+        TaskRef { ptr }
+    }
+
+    // Given a pointer to a TaskHeader, obtain a pointer to the Links structure,
+    // which can be used to traverse to other TaskHeader nodes in the linked list
+    unsafe fn links(ptr: NonNull<TaskHeader>) -> NonNull<Links<TaskHeader>> {
+        let ptr: *mut TaskHeader = ptr.as_ptr();
+        NonNull::new_unchecked(addr_of_mut!((*ptr).run_queue_item))
+    }
+}
+
+/// Atomic task queue using a very, very simple lock-free linked-list queue:
+///
+/// To enqueue a task, task.next is set to the old head, and head is atomically set to task.
+///
+/// Dequeuing is done in batches: the queue is emptied by atomically replacing head with
+/// null. Then the batch is iterated following the next pointers until null is reached.
+///
+/// Note that batches will be iterated in the reverse order as they were enqueued. This is OK
+/// for our purposes: it can't create fairness problems since the next batch won't run until the
+/// current batch is completely processed, so even if a task enqueues itself instantly (for example
+/// by waking its own waker) can't prevent other tasks from running.
+pub(crate) struct RunQueue {
+    stack: TransferStack<TaskHeader>,
+}
+
+impl RunQueue {
+    pub const fn new() -> Self {
+        Self {
+            stack: TransferStack::new(),
+        }
+    }
+
+    /// Enqueues an item. Returns true if the queue was empty.
+    ///
+    /// # Safety
+    ///
+    /// `item` must NOT be already enqueued in any queue.
+    #[inline(always)]
+    pub(crate) unsafe fn enqueue(&self, task: TaskRef, _tok: super::state::Token) -> bool {
+        self.stack.push_was_empty(
+            task,
+            #[cfg(not(target_has_atomic = "ptr"))]
+            _tok,
+        )
+    }
+
+    /// # Standard atomic runqueue
+    ///
+    /// Empty the queue, then call `on_task` for each task that was in the queue.
+    /// NOTE: It is OK for `on_task` to enqueue more tasks. In this case they're left in the queue
+    /// and will be processed by the *next* call to `dequeue_all`, *not* the current one.
+    #[cfg(not(any(feature = "scheduler-priority", feature = "scheduler-deadline")))]
+    pub(crate) fn dequeue_all(&self, on_task: impl Fn(TaskRef)) {
+        let taken = self.stack.take_all();
+        for taskref in taken {
+            run_dequeue(&taskref);
+            on_task(taskref);
+        }
+    }
+
+    /// # Earliest Deadline First Scheduler
+    ///
+    /// This algorithm will loop until all enqueued tasks are processed.
+    ///
+    /// Before polling a task, all currently enqueued tasks will be popped from the
+    /// runqueue, and will be added to the working `sorted` list, a linked-list that
+    /// sorts tasks by their deadline, with nearest deadline items in the front, and
+    /// furthest deadline items in the back.
+    ///
+    /// After popping and sorting all pending tasks, the SOONEST task will be popped
+    /// from the front of the queue, and polled by calling `on_task` on it.
+    ///
+    /// This process will repeat until the local `sorted` queue AND the global
+    /// runqueue are both empty, at which point this function will return.
+    #[cfg(any(feature = "scheduler-priority", feature = "scheduler-deadline"))]
+    pub(crate) fn dequeue_all(&self, on_task: impl Fn(TaskRef)) {
+        let mut sorted = SortedList::<TaskHeader>::new_with_cmp(|lhs, rhs| {
+            // compare by priority first
+            #[cfg(feature = "scheduler-priority")]
+            {
+                let lp = lhs.metadata.priority();
+                let rp = rhs.metadata.priority();
+                if lp != rp {
+                    return lp.cmp(&rp).reverse();
+                }
+            }
+            // compare deadlines in case of tie.
+            #[cfg(feature = "scheduler-deadline")]
+            {
+                let ld = lhs.metadata.deadline();
+                let rd = rhs.metadata.deadline();
+                if ld != rd {
+                    return ld.cmp(&rd);
+                }
+            }
+            core::cmp::Ordering::Equal
+        });
+
+        loop {
+            // For each loop, grab any newly pended items
+            let taken = self.stack.take_all();
+
+            // Sort these into the list - this is potentially expensive! We do an
+            // insertion sort of new items, which iterates the linked list.
+            //
+            // Something on the order of `O(n * m)`, where `n` is the number
+            // of new tasks, and `m` is the number of already pending tasks.
+            sorted.extend(taken);
+
+            // Pop the task with the SOONEST deadline. If there are no tasks
+            // pending, then we are done.
+            let Some(taskref) = sorted.pop_front() else {
+                return;
+            };
+
+            // We got one task, mark it as dequeued, and process the task.
+            run_dequeue(&taskref);
+            on_task(taskref);
+        }
+    }
+}
+
+/// atomic state does not require a cs...
+#[cfg(target_has_atomic = "ptr")]
+#[inline(always)]
+fn run_dequeue(taskref: &TaskRef) {
+    taskref.header().state.run_dequeue();
+}
+
+/// ...while non-atomic state does
+#[cfg(not(target_has_atomic = "ptr"))]
+#[inline(always)]
+fn run_dequeue(taskref: &TaskRef) {
+    critical_section::with(|cs| {
+        taskref.header().state.run_dequeue(cs);
+    })
+}
+
+/// A wrapper type that acts like TransferStack by wrapping a normal Stack in a CS mutex
+#[cfg(not(target_has_atomic = "ptr"))]
+struct MutexTransferStack<T: Linked<cordyceps::stack::Links<T>>> {
+    inner: critical_section::Mutex<core::cell::UnsafeCell<cordyceps::Stack<T>>>,
+}
+
+#[cfg(not(target_has_atomic = "ptr"))]
+impl<T: Linked<cordyceps::stack::Links<T>>> MutexTransferStack<T> {
+    const fn new() -> Self {
+        Self {
+            inner: critical_section::Mutex::new(core::cell::UnsafeCell::new(cordyceps::Stack::new())),
+        }
+    }
+
+    /// Push an item to the transfer stack, returning whether the stack was previously empty
+    fn push_was_empty(&self, item: T::Handle, token: super::state::Token) -> bool {
+        // SAFETY: The critical-section mutex guarantees that there is no *concurrent* access
+        // for the lifetime of the token, but does NOT protect against re-entrant access.
+        // However, we never *return* the reference, nor do we recurse (or call another method
+        // like `take_all`) that could ever allow for re-entrant aliasing. Therefore, the
+        // presence of the critical section is sufficient to guarantee exclusive access to
+        // the `inner` field for the purposes of this function.
+        let inner = unsafe { &mut *self.inner.borrow(token).get() };
+        let is_empty = inner.is_empty();
+        inner.push(item);
+        is_empty
+    }
+
+    fn take_all(&self) -> cordyceps::Stack<T> {
+        critical_section::with(|cs| {
+            // SAFETY: The critical-section mutex guarantees that there is no *concurrent* access
+            // for the lifetime of the token, but does NOT protect against re-entrant access.
+            // However, we never *return* the reference, nor do we recurse (or call another method
+            // like `push_was_empty`) that could ever allow for re-entrant aliasing. Therefore, the
+            // presence of the critical section is sufficient to guarantee exclusive access to
+            // the `inner` field for the purposes of this function.
+            let inner = unsafe { &mut *self.inner.borrow(cs).get() };
+            inner.take_all()
+        })
+    }
+}
diff --git a/embassy-executor/src/raw/run_queue_atomics.rs b/embassy-executor/src/raw/run_queue_atomics.rs
deleted file mode 100644
index 90907cfda..000000000
--- a/embassy-executor/src/raw/run_queue_atomics.rs
+++ /dev/null
@@ -1,87 +0,0 @@
-use core::ptr;
-use core::ptr::NonNull;
-use core::sync::atomic::{AtomicPtr, Ordering};
-
-use super::{TaskHeader, TaskRef};
-use crate::raw::util::SyncUnsafeCell;
-
-pub(crate) struct RunQueueItem {
-    next: SyncUnsafeCell<Option<TaskRef>>,
-}
-
-impl RunQueueItem {
-    pub const fn new() -> Self {
-        Self {
-            next: SyncUnsafeCell::new(None),
-        }
-    }
-}
-
-/// Atomic task queue using a very, very simple lock-free linked-list queue:
-///
-/// To enqueue a task, task.next is set to the old head, and head is atomically set to task.
-///
-/// Dequeuing is done in batches: the queue is emptied by atomically replacing head with
-/// null. Then the batch is iterated following the next pointers until null is reached.
-///
-/// Note that batches will be iterated in the reverse order as they were enqueued. This is OK
-/// for our purposes: it can't create fairness problems since the next batch won't run until the
-/// current batch is completely processed, so even if a task enqueues itself instantly (for example
-/// by waking its own waker) can't prevent other tasks from running.
-pub(crate) struct RunQueue {
-    head: AtomicPtr<TaskHeader>,
-}
-
-impl RunQueue {
-    pub const fn new() -> Self {
-        Self {
-            head: AtomicPtr::new(ptr::null_mut()),
-        }
-    }
-
-    /// Enqueues an item. Returns true if the queue was empty.
-    ///
-    /// # Safety
-    ///
-    /// `item` must NOT be already enqueued in any queue.
-    #[inline(always)]
-    pub(crate) unsafe fn enqueue(&self, task: TaskRef) -> bool {
-        let mut was_empty = false;
-
-        self.head
-            .fetch_update(Ordering::SeqCst, Ordering::SeqCst, |prev| {
-                was_empty = prev.is_null();
-                unsafe {
-                    // safety: the pointer is either null or valid
-                    let prev = NonNull::new(prev).map(|ptr| TaskRef::from_ptr(ptr.as_ptr()));
-                    // safety: there are no concurrent accesses to `next`
-                    task.header().run_queue_item.next.set(prev);
-                }
-                Some(task.as_ptr() as *mut _)
-            })
-            .ok();
-
-        was_empty
-    }
-
-    /// Empty the queue, then call `on_task` for each task that was in the queue.
-    /// NOTE: It is OK for `on_task` to enqueue more tasks. In this case they're left in the queue
-    /// and will be processed by the *next* call to `dequeue_all`, *not* the current one.
-    pub(crate) fn dequeue_all(&self, on_task: impl Fn(TaskRef)) {
-        // Atomically empty the queue.
-        let ptr = self.head.swap(ptr::null_mut(), Ordering::AcqRel);
-
-        // safety: the pointer is either null or valid
-        let mut next = unsafe { NonNull::new(ptr).map(|ptr| TaskRef::from_ptr(ptr.as_ptr())) };
-
-        // Iterate the linked list of tasks that were previously in the queue.
-        while let Some(task) = next {
-            // If the task re-enqueues itself, the `next` pointer will get overwritten.
-            // Therefore, first read the next pointer, and only then process the task.
-            // safety: there are no concurrent accesses to `next`
-            next = unsafe { task.header().run_queue_item.next.get() };
-
-            on_task(task);
-        }
-    }
-}
diff --git a/embassy-executor/src/raw/run_queue_critical_section.rs b/embassy-executor/src/raw/run_queue_critical_section.rs
deleted file mode 100644
index ba59c8f29..000000000
--- a/embassy-executor/src/raw/run_queue_critical_section.rs
+++ /dev/null
@@ -1,75 +0,0 @@
-use core::cell::Cell;
-
-use critical_section::{CriticalSection, Mutex};
-
-use super::TaskRef;
-
-pub(crate) struct RunQueueItem {
-    next: Mutex<Cell<Option<TaskRef>>>,
-}
-
-impl RunQueueItem {
-    pub const fn new() -> Self {
-        Self {
-            next: Mutex::new(Cell::new(None)),
-        }
-    }
-}
-
-/// Atomic task queue using a very, very simple lock-free linked-list queue:
-///
-/// To enqueue a task, task.next is set to the old head, and head is atomically set to task.
-///
-/// Dequeuing is done in batches: the queue is emptied by atomically replacing head with
-/// null. Then the batch is iterated following the next pointers until null is reached.
-///
-/// Note that batches will be iterated in the reverse order as they were enqueued. This is OK
-/// for our purposes: it can't create fairness problems since the next batch won't run until the
-/// current batch is completely processed, so even if a task enqueues itself instantly (for example
-/// by waking its own waker) can't prevent other tasks from running.
-pub(crate) struct RunQueue {
-    head: Mutex<Cell<Option<TaskRef>>>,
-}
-
-impl RunQueue {
-    pub const fn new() -> Self {
-        Self {
-            head: Mutex::new(Cell::new(None)),
-        }
-    }
-
-    /// Enqueues an item. Returns true if the queue was empty.
-    ///
-    /// # Safety
-    ///
-    /// `item` must NOT be already enqueued in any queue.
-    #[inline(always)]
-    pub(crate) unsafe fn enqueue(&self, task: TaskRef) -> bool {
-        critical_section::with(|cs| {
-            let prev = self.head.borrow(cs).replace(Some(task));
-            task.header().run_queue_item.next.borrow(cs).set(prev);
-
-            prev.is_none()
-        })
-    }
-
-    /// Empty the queue, then call `on_task` for each task that was in the queue.
-    /// NOTE: It is OK for `on_task` to enqueue more tasks. In this case they're left in the queue
-    /// and will be processed by the *next* call to `dequeue_all`, *not* the current one.
-    pub(crate) fn dequeue_all(&self, on_task: impl Fn(TaskRef)) {
-        // Atomically empty the queue.
-        let mut next = critical_section::with(|cs| self.head.borrow(cs).take());
-
-        // Iterate the linked list of tasks that were previously in the queue.
-        while let Some(task) = next {
-            // If the task re-enqueues itself, the `next` pointer will get overwritten.
-            // Therefore, first read the next pointer, and only then process the task.
-
-            // safety: we know if the task is enqueued, no one else will touch the `next` pointer.
-            let cs = unsafe { CriticalSection::new() };
-            next = task.header().run_queue_item.next.borrow(cs).get();
-
-            on_task(task);
-        }
-    }
-}
diff --git a/embassy-executor/src/raw/state_atomics.rs b/embassy-executor/src/raw/state_atomics.rs
index e1279ac0b..6675875be 100644
--- a/embassy-executor/src/raw/state_atomics.rs
+++ b/embassy-executor/src/raw/state_atomics.rs
@@ -1,21 +1,39 @@
-use core::sync::atomic::{AtomicU32, Ordering};
+// Prefer pointer-width atomic operations, as narrower ones may be slower.
+#[cfg(all(target_pointer_width = "32", target_has_atomic = "32"))]
+type AtomicState = core::sync::atomic::AtomicU32;
+#[cfg(not(all(target_pointer_width = "32", target_has_atomic = "32")))]
+type AtomicState = core::sync::atomic::AtomicU8;
+
+#[cfg(all(target_pointer_width = "32", target_has_atomic = "32"))]
+type StateBits = u32;
+#[cfg(not(all(target_pointer_width = "32", target_has_atomic = "32")))]
+type StateBits = u8;
+
+use core::sync::atomic::Ordering;
+
+#[derive(Clone, Copy)]
+pub(crate) struct Token(());
+
+/// Creates a token and passes it to the closure.
+///
+/// This is a no-op replacement for `CriticalSection::with` because we don't need any locking.
+pub(crate) fn locked<R>(f: impl FnOnce(Token) -> R) -> R {
+    f(Token(()))
+}
 
 /// Task is spawned (has a future)
-pub(crate) const STATE_SPAWNED: u32 = 1 << 0;
+pub(crate) const STATE_SPAWNED: StateBits = 1 << 0;
 /// Task is in the executor run queue
-pub(crate) const STATE_RUN_QUEUED: u32 = 1 << 1;
-/// Task is in the executor timer queue
-#[cfg(feature = "integrated-timers")]
-pub(crate) const STATE_TIMER_QUEUED: u32 = 1 << 2;
+pub(crate) const STATE_RUN_QUEUED: StateBits = 1 << 1;
 
 pub(crate) struct State {
-    state: AtomicU32,
+    state: AtomicState,
 }
 
 impl State {
     pub const fn new() -> State {
         Self {
-            state: AtomicU32::new(0),
+            state: AtomicState::new(0),
         }
     }
 
@@ -33,41 +51,19 @@ impl State {
         self.state.fetch_and(!STATE_SPAWNED, Ordering::AcqRel);
     }
 
-    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Return true on success.
+    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Run the given
+    /// function if the task was successfully marked.
     #[inline(always)]
-    pub fn run_enqueue(&self) -> bool {
-        self.state
-            .fetch_update(Ordering::SeqCst, Ordering::SeqCst, |state| {
-                // If already scheduled, or if not started,
-                if (state & STATE_RUN_QUEUED != 0) || (state & STATE_SPAWNED == 0) {
-                    None
-                } else {
-                    // Mark it as scheduled
-                    Some(state | STATE_RUN_QUEUED)
-                }
-            })
-            .is_ok()
+    pub fn run_enqueue(&self, f: impl FnOnce(Token)) {
+        let prev = self.state.fetch_or(STATE_RUN_QUEUED, Ordering::AcqRel);
+        if prev & STATE_RUN_QUEUED == 0 {
+            locked(f);
+        }
     }
 
     /// Unmark the task as run-queued. Return whether the task is spawned.
     #[inline(always)]
-    pub fn run_dequeue(&self) -> bool {
-        let state = self.state.fetch_and(!STATE_RUN_QUEUED, Ordering::AcqRel);
-        state & STATE_SPAWNED != 0
-    }
-
-    /// Mark the task as timer-queued. Return whether it was newly queued (i.e. not queued before)
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_enqueue(&self) -> bool {
-        let old_state = self.state.fetch_or(STATE_TIMER_QUEUED, Ordering::AcqRel);
-        old_state & STATE_TIMER_QUEUED == 0
-    }
-
-    /// Unmark the task as timer-queued.
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_dequeue(&self) {
-        self.state.fetch_and(!STATE_TIMER_QUEUED, Ordering::AcqRel);
+    pub fn run_dequeue(&self) {
+        self.state.fetch_and(!STATE_RUN_QUEUED, Ordering::AcqRel);
     }
 }
diff --git a/embassy-executor/src/raw/state_atomics_arm.rs b/embassy-executor/src/raw/state_atomics_arm.rs
index e4dfe5093..f68de955f 100644
--- a/embassy-executor/src/raw/state_atomics_arm.rs
+++ b/embassy-executor/src/raw/state_atomics_arm.rs
@@ -1,5 +1,14 @@
-use core::arch::asm;
-use core::sync::atomic::{compiler_fence, AtomicBool, AtomicU32, Ordering};
+use core::sync::atomic::{AtomicBool, AtomicU32, Ordering, compiler_fence};
+
+#[derive(Clone, Copy)]
+pub(crate) struct Token(());
+
+/// Creates a token and passes it to the closure.
+///
+/// This is a no-op replacement for `CriticalSection::with` because we don't need any locking.
+pub(crate) fn locked<R>(f: impl FnOnce(Token) -> R) -> R {
+    f(Token(()))
+}
 
 // Must be kept in sync with the layout of `State`!
 pub(crate) const STATE_SPAWNED: u32 = 1 << 0;
@@ -11,9 +20,8 @@ pub(crate) struct State {
     spawned: AtomicBool,
     /// Task is in the executor run queue
     run_queued: AtomicBool,
-    /// Task is in the executor timer queue
-    timer_queued: AtomicBool,
     pad: AtomicBool,
+    pad2: AtomicBool,
 }
 
 impl State {
@@ -21,8 +29,8 @@ impl State {
         Self {
             spawned: AtomicBool::new(false),
             run_queued: AtomicBool::new(false),
-            timer_queued: AtomicBool::new(false),
             pad: AtomicBool::new(false),
+            pad2: AtomicBool::new(false),
         }
     }
 
@@ -54,50 +62,22 @@ impl State {
         self.spawned.store(false, Ordering::Relaxed);
     }
 
-    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Return true on success.
+    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Run the given
+    /// function if the task was successfully marked.
     #[inline(always)]
-    pub fn run_enqueue(&self) -> bool {
-        unsafe {
-            loop {
-                let state: u32;
-                asm!("ldrex {}, [{}]", out(reg) state, in(reg) self, options(nostack));
+    pub fn run_enqueue(&self, f: impl FnOnce(Token)) {
+        let old = self.run_queued.swap(true, Ordering::AcqRel);
 
-                if (state & STATE_RUN_QUEUED != 0) || (state & STATE_SPAWNED == 0) {
-                    asm!("clrex", options(nomem, nostack));
-                    return false;
-                }
-
-                let outcome: usize;
-                let new_state = state | STATE_RUN_QUEUED;
-                asm!("strex {}, {}, [{}]", out(reg) outcome, in(reg) new_state, in(reg) self, options(nostack));
-                if outcome == 0 {
-                    return true;
-                }
-            }
+        if !old {
+            locked(f);
         }
     }
 
     /// Unmark the task as run-queued. Return whether the task is spawned.
     #[inline(always)]
-    pub fn run_dequeue(&self) -> bool {
+    pub fn run_dequeue(&self) {
         compiler_fence(Ordering::Release);
 
-        let r = self.spawned.load(Ordering::Relaxed);
         self.run_queued.store(false, Ordering::Relaxed);
-        r
-    }
-
-    /// Mark the task as timer-queued. Return whether it was newly queued (i.e. not queued before)
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_enqueue(&self) -> bool {
-        !self.timer_queued.swap(true, Ordering::Relaxed)
-    }
-
-    /// Unmark the task as timer-queued.
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_dequeue(&self) {
-        self.timer_queued.store(false, Ordering::Relaxed);
     }
 }
diff --git a/embassy-executor/src/raw/state_critical_section.rs b/embassy-executor/src/raw/state_critical_section.rs
index c3cc1b0b7..8d7ef2892 100644
--- a/embassy-executor/src/raw/state_critical_section.rs
+++ b/embassy-executor/src/raw/state_critical_section.rs
@@ -1,17 +1,20 @@
 use core::cell::Cell;
 
-use critical_section::Mutex;
+use critical_section::{CriticalSection, Mutex};
+pub(crate) use critical_section::{CriticalSection as Token, with as locked};
+
+#[cfg(target_arch = "avr")]
+type StateBits = u8;
+#[cfg(not(target_arch = "avr"))]
+type StateBits = usize;
 
 /// Task is spawned (has a future)
-pub(crate) const STATE_SPAWNED: u32 = 1 << 0;
+pub(crate) const STATE_SPAWNED: StateBits = 1 << 0;
 /// Task is in the executor run queue
-pub(crate) const STATE_RUN_QUEUED: u32 = 1 << 1;
-/// Task is in the executor timer queue
-#[cfg(feature = "integrated-timers")]
-pub(crate) const STATE_TIMER_QUEUED: u32 = 1 << 2;
+pub(crate) const STATE_RUN_QUEUED: StateBits = 1 << 1;
 
 pub(crate) struct State {
-    state: Mutex<Cell<u32>>,
+    state: Mutex<Cell<StateBits>>,
 }
 
 impl State {
@@ -21,14 +24,16 @@ impl State {
         }
     }
 
-    fn update<R>(&self, f: impl FnOnce(&mut u32) -> R) -> R {
-        critical_section::with(|cs| {
-            let s = self.state.borrow(cs);
-            let mut val = s.get();
-            let r = f(&mut val);
-            s.set(val);
-            r
-        })
+    fn update<R>(&self, f: impl FnOnce(&mut StateBits) -> R) -> R {
+        critical_section::with(|cs| self.update_with_cs(cs, f))
+    }
+
+    fn update_with_cs<R>(&self, cs: CriticalSection<'_>, f: impl FnOnce(&mut StateBits) -> R) -> R {
+        let s = self.state.borrow(cs);
+        let mut val = s.get();
+        let r = f(&mut val);
+        s.set(val);
+        r
     }
 
     /// If task is idle, mark it as spawned + run_queued and return true.
@@ -50,44 +55,24 @@ impl State {
         self.update(|s| *s &= !STATE_SPAWNED);
     }
 
-    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Return true on success.
+    /// Mark the task as run-queued if it's spawned and isn't already run-queued. Run the given
+    /// function if the task was successfully marked.
     #[inline(always)]
-    pub fn run_enqueue(&self) -> bool {
-        self.update(|s| {
-            if (*s & STATE_RUN_QUEUED != 0) || (*s & STATE_SPAWNED == 0) {
-                false
-            } else {
+    pub fn run_enqueue(&self, f: impl FnOnce(Token)) {
+        critical_section::with(|cs| {
+            if self.update_with_cs(cs, |s| {
+                let ok = *s & STATE_RUN_QUEUED == 0;
                 *s |= STATE_RUN_QUEUED;
-                true
+                ok
+            }) {
+                f(cs);
             }
-        })
+        });
     }
 
     /// Unmark the task as run-queued. Return whether the task is spawned.
     #[inline(always)]
-    pub fn run_dequeue(&self) -> bool {
-        self.update(|s| {
-            let ok = *s & STATE_SPAWNED != 0;
-            *s &= !STATE_RUN_QUEUED;
-            ok
-        })
-    }
-
-    /// Mark the task as timer-queued. Return whether it was newly queued (i.e. not queued before)
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_enqueue(&self) -> bool {
-        self.update(|s| {
-            let ok = *s & STATE_TIMER_QUEUED == 0;
-            *s |= STATE_TIMER_QUEUED;
-            ok
-        })
-    }
-
-    /// Unmark the task as timer-queued.
-    #[cfg(feature = "integrated-timers")]
-    #[inline(always)]
-    pub fn timer_dequeue(&self) {
-        self.update(|s| *s &= !STATE_TIMER_QUEUED);
+    pub fn run_dequeue(&self, cs: CriticalSection<'_>) {
+        self.update_with_cs(cs, |s| *s &= !STATE_RUN_QUEUED)
     }
 }
diff --git a/embassy-executor/src/raw/timer_queue.rs b/embassy-executor/src/raw/timer_queue.rs
deleted file mode 100644
index 94a5f340b..000000000
--- a/embassy-executor/src/raw/timer_queue.rs
+++ /dev/null
@@ -1,76 +0,0 @@
-use core::cmp::min;
-
-use super::TaskRef;
-use crate::raw::util::SyncUnsafeCell;
-
-pub(crate) struct TimerQueueItem {
-    next: SyncUnsafeCell<Option<TaskRef>>,
-}
-
-impl TimerQueueItem {
-    pub const fn new() -> Self {
-        Self {
-            next: SyncUnsafeCell::new(None),
-        }
-    }
-}
-
-pub(crate) struct TimerQueue {
-    head: SyncUnsafeCell<Option<TaskRef>>,
-}
-
-impl TimerQueue {
-    pub const fn new() -> Self {
-        Self {
-            head: SyncUnsafeCell::new(None),
-        }
-    }
-
-    pub(crate) unsafe fn update(&self, p: TaskRef) {
-        let task = p.header();
-        if task.expires_at.get() != u64::MAX {
-            if task.state.timer_enqueue() {
-                task.timer_queue_item.next.set(self.head.get());
-                self.head.set(Some(p));
-            }
-        }
-    }
-
-    pub(crate) unsafe fn next_expiration(&self) -> u64 {
-        let mut res = u64::MAX;
-        self.retain(|p| {
-            let task = p.header();
-            let expires = task.expires_at.get();
-            res = min(res, expires);
-            expires != u64::MAX
-        });
-        res
-    }
-
-    pub(crate) unsafe fn dequeue_expired(&self, now: u64, on_task: impl Fn(TaskRef)) {
-        self.retain(|p| {
-            let task = p.header();
-            if task.expires_at.get() <= now {
-                on_task(p);
-                false
-            } else {
-                true
-            }
-        });
-    }
-
-    pub(crate) unsafe fn retain(&self, mut f: impl FnMut(TaskRef) -> bool) {
-        let mut prev = &self.head;
-        while let Some(p) = prev.get() {
-            let task = p.header();
-            if f(p) {
-                // Skip to next
-                prev = &task.timer_queue_item.next;
-            } else {
-                // Remove it
-                prev.set(task.timer_queue_item.next.get());
-                task.state.timer_dequeue();
-            }
-        }
-    }
-}
diff --git a/embassy-executor/src/raw/trace.rs b/embassy-executor/src/raw/trace.rs
new file mode 100644
index 000000000..830162039
--- /dev/null
+++ b/embassy-executor/src/raw/trace.rs
@@ -0,0 +1,380 @@
+//! # Tracing
+//!
+//! The `trace` feature enables a number of callbacks that can be used to track the
+//! lifecycle of tasks and/or executors.
+//!
+//! Callbacks will have one or both of the following IDs passed to them:
+//!
+//! 1. A `task_id`, a `u32` value unique to a task for the duration of the time it is valid
+//! 2. An `executor_id`, a `u32` value unique to an executor for the duration of the time it is
+//!    valid
+//!
+//! Today, both `task_id` and `executor_id` are u32s containing the least significant 32 bits of
+//! the address of the task or executor, however this is NOT a stable guarantee, and MAY change
+//! at any time.
+//!
+//! IDs are only guaranteed to be unique for the duration of time the item is valid. If a task
+//! ends, and is re-spawned, it MAY or MAY NOT have the same ID. For tasks, this valid time is defined
+//! as the time between `_embassy_trace_task_new` and `_embassy_trace_task_end` for a given task.
+//! For executors, this time is not defined, but is often "forever" for practical embedded
+//! programs.
+//!
+//! Callbacks can be used by enabling the `trace` feature, and providing implementations of the
+//! `extern "Rust"` functions below. All callbacks must be implemented.
+//!
+//! ## Task Tracing lifecycle
+//!
+//! ```text
+//! ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
+//!        │(1)                                            │
+//! │      │
+//!   ╔════▼════╗ (2) ┌─────────┐ (3) ┌─────────┐          │
+//! │ ║ SPAWNED ║────▶│ WAITING │────▶│ RUNNING │
+//!   ╚═════════╝     └─────────┘     └─────────┘          │
+//! │                 ▲         ▲     │    │    │
+//!                   │           (4)      │    │(6)       │
+//! │                 │(7)      └ ─ ─ ┘    │    │
+//!                   │                    │    │          │
+//! │             ┌──────┐             (5) │    │  ┌─────┐
+//!               │ IDLE │◀────────────────┘    └─▶│ END │ │
+//! │             └──────┘                         └─────┘
+//!   ┌──────────────────────┐                             │
+//! └ ┤ Task Trace Lifecycle │─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
+//!   └──────────────────────┘
+//! ```
+//!
+//! 1. A task is spawned, `_embassy_trace_task_new` is called
+//! 2. A task is enqueued for the first time, `_embassy_trace_task_ready_begin` is called
+//! 3. A task is polled, `_embassy_trace_task_exec_begin` is called
+//! 4. WHILE a task is polled, the task is re-awoken, and `_embassy_trace_task_ready_begin` is
+//!      called. The task does not IMMEDIATELY move state, until polling is complete and the
+//!      RUNNING state is existed. `_embassy_trace_task_exec_end` is called when polling is
+//!      complete, marking the transition to WAITING
+//! 5. Polling is complete, `_embassy_trace_task_exec_end` is called
+//! 6. The task has completed, and `_embassy_trace_task_end` is called
+//! 7. A task is awoken, `_embassy_trace_task_ready_begin` is called
+//!
+//! ## Executor Tracing lifecycle
+//!
+//! ```text
+//! ┌ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
+//!       │(1)                                             │
+//! │     │
+//!   ╔═══▼══╗   (2)     ┌────────────┐  (3)  ┌─────────┐  │
+//! │ ║ IDLE ║──────────▶│ SCHEDULING │──────▶│ POLLING │
+//!   ╚══════╝           └────────────┘       └─────────┘  │
+//! │     ▲              │            ▲            │
+//!       │      (5)     │            │  (4)       │       │
+//! │     └──────────────┘            └────────────┘
+//!   ┌──────────────────────────┐                         │
+//! └ ┤ Executor Trace Lifecycle │─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─ ─
+//!   └──────────────────────────┘
+//! ```
+//!
+//! 1. The executor is started (no associated trace)
+//! 2. A task on this executor is awoken. `_embassy_trace_task_ready_begin` is called
+//!      when this occurs, and `_embassy_trace_poll_start` is called when the executor
+//!      actually begins running
+//! 3. The executor has decided a task to poll. `_embassy_trace_task_exec_begin` is called
+//! 4. The executor finishes polling the task. `_embassy_trace_task_exec_end` is called
+//! 5. The executor has finished polling tasks. `_embassy_trace_executor_idle` is called
+
+#![allow(unused)]
+
+use core::cell::UnsafeCell;
+use core::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
+
+#[cfg(feature = "rtos-trace")]
+use rtos_trace::TaskInfo;
+
+use crate::raw::{SyncExecutor, TaskHeader, TaskRef};
+use crate::spawner::{SpawnError, SpawnToken, Spawner};
+
+/// Global task tracker instance
+///
+/// This static provides access to the global task tracker which maintains
+/// a list of all tasks in the system. It's automatically updated by the
+/// task lifecycle hooks in the trace module.
+#[cfg(feature = "rtos-trace")]
+pub(crate) static TASK_TRACKER: TaskTracker = TaskTracker::new();
+
+/// A thread-safe tracker for all tasks in the system
+///
+/// This struct uses an intrusive linked list approach to track all tasks
+/// without additional memory allocations. It maintains a global list of
+/// tasks that can be traversed to find all currently existing tasks.
+#[cfg(feature = "rtos-trace")]
+pub(crate) struct TaskTracker {
+    head: AtomicPtr<TaskHeader>,
+}
+
+#[cfg(feature = "rtos-trace")]
+impl TaskTracker {
+    /// Creates a new empty task tracker
+    ///
+    /// Initializes a tracker with no tasks in its list.
+    pub const fn new() -> Self {
+        Self {
+            head: AtomicPtr::new(core::ptr::null_mut()),
+        }
+    }
+
+    /// Adds a task to the tracker
+    ///
+    /// This method inserts a task at the head of the intrusive linked list.
+    /// The operation is thread-safe and lock-free, using atomic operations
+    /// to ensure consistency even when called from different contexts.
+    ///
+    /// # Arguments
+    /// * `task` - The task reference to add to the tracker
+    pub fn add(&self, task: TaskRef) {
+        let task_ptr = task.as_ptr();
+
+        loop {
+            let current_head = self.head.load(Ordering::Acquire);
+            unsafe {
+                (*task_ptr).all_tasks_next.store(current_head, Ordering::Relaxed);
+            }
+
+            if self
+                .head
+                .compare_exchange(current_head, task_ptr.cast_mut(), Ordering::Release, Ordering::Relaxed)
+                .is_ok()
+            {
+                break;
+            }
+        }
+    }
+
+    /// Performs an operation on each task in the tracker
+    ///
+    /// This method traverses the entire list of tasks and calls the provided
+    /// function for each task. This allows inspecting or processing all tasks
+    /// in the system without modifying the tracker's structure.
+    ///
+    /// # Arguments
+    /// * `f` - A function to call for each task in the tracker
+    pub fn for_each<F>(&self, mut f: F)
+    where
+        F: FnMut(TaskRef),
+    {
+        let mut current = self.head.load(Ordering::Acquire);
+        while !current.is_null() {
+            let task = unsafe { TaskRef::from_ptr(current) };
+            f(task);
+
+            current = unsafe { (*current).all_tasks_next.load(Ordering::Acquire) };
+        }
+    }
+}
+
+#[cfg(feature = "trace")]
+unsafe extern "Rust" {
+    /// This callback is called when the executor begins polling. This will always
+    /// be paired with a later call to `_embassy_trace_executor_idle`.
+    ///
+    /// This marks the EXECUTOR state transition from IDLE -> SCHEDULING.
+    fn _embassy_trace_poll_start(executor_id: u32);
+
+    /// This callback is called AFTER a task is initialized/allocated, and BEFORE
+    /// it is enqueued to run for the first time. If the task ends (and does not
+    /// loop "forever"), there will be a matching call to `_embassy_trace_task_end`.
+    ///
+    /// Tasks start life in the SPAWNED state.
+    fn _embassy_trace_task_new(executor_id: u32, task_id: u32);
+
+    /// This callback is called AFTER a task is destructed/freed. This will always
+    /// have a prior matching call to `_embassy_trace_task_new`.
+    fn _embassy_trace_task_end(executor_id: u32, task_id: u32);
+
+    /// This callback is called AFTER a task has been dequeued from the runqueue,
+    /// and BEFORE the task is polled. There will always be a matching call to
+    /// `_embassy_trace_task_exec_end`.
+    ///
+    /// This marks the TASK state transition from WAITING -> RUNNING
+    /// This marks the EXECUTOR state transition from SCHEDULING -> POLLING
+    fn _embassy_trace_task_exec_begin(executor_id: u32, task_id: u32);
+
+    /// This callback is called AFTER a task has completed polling. There will
+    /// always be a matching call to `_embassy_trace_task_exec_begin`.
+    ///
+    /// This marks the TASK state transition from either:
+    /// * RUNNING -> IDLE - if there were no `_embassy_trace_task_ready_begin` events
+    ///     for this task since the last `_embassy_trace_task_exec_begin` for THIS task
+    /// * RUNNING -> WAITING - if there WAS a `_embassy_trace_task_ready_begin` event
+    ///     for this task since the last `_embassy_trace_task_exec_begin` for THIS task
+    ///
+    /// This marks the EXECUTOR state transition from POLLING -> SCHEDULING
+    fn _embassy_trace_task_exec_end(excutor_id: u32, task_id: u32);
+
+    /// This callback is called AFTER the waker for a task is awoken, and BEFORE it
+    /// is added to the run queue.
+    ///
+    /// If the given task is currently RUNNING, this marks no state change, BUT the
+    /// RUNNING task will then move to the WAITING stage when polling is complete.
+    ///
+    /// If the given task is currently IDLE, this marks the TASK state transition
+    /// from IDLE -> WAITING.
+    ///
+    /// NOTE: This may be called from an interrupt, outside the context of the current
+    /// task or executor.
+    fn _embassy_trace_task_ready_begin(executor_id: u32, task_id: u32);
+
+    /// This callback is called AFTER all dequeued tasks in a single call to poll
+    /// have been processed. This will always be paired with a call to
+    /// `_embassy_trace_executor_idle`.
+    ///
+    /// This marks the EXECUTOR state transition from SCHEDULING -> IDLE
+    fn _embassy_trace_executor_idle(executor_id: u32);
+}
+
+#[inline]
+pub(crate) fn poll_start(executor: &SyncExecutor) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_poll_start(executor as *const _ as u32)
+    }
+}
+
+#[inline]
+pub(crate) fn task_new(executor: &SyncExecutor, task: &TaskRef) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_task_new(executor as *const _ as u32, task.as_ptr() as u32)
+    }
+
+    #[cfg(feature = "rtos-trace")]
+    {
+        rtos_trace::trace::task_new(task.as_ptr() as u32);
+        let name = task.metadata().name().unwrap_or("unnamed task\0");
+        let info = rtos_trace::TaskInfo {
+            name,
+            priority: 0,
+            stack_base: 0,
+            stack_size: 0,
+        };
+        rtos_trace::trace::task_send_info(task.id(), info);
+    }
+
+    #[cfg(feature = "rtos-trace")]
+    TASK_TRACKER.add(*task);
+}
+
+#[inline]
+pub(crate) fn task_end(executor: *const SyncExecutor, task: &TaskRef) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_task_end(executor as u32, task.as_ptr() as u32)
+    }
+}
+
+#[inline]
+pub(crate) fn task_ready_begin(executor: &SyncExecutor, task: &TaskRef) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_task_ready_begin(executor as *const _ as u32, task.as_ptr() as u32)
+    }
+    #[cfg(feature = "rtos-trace")]
+    rtos_trace::trace::task_ready_begin(task.as_ptr() as u32);
+}
+
+#[inline]
+pub(crate) fn task_exec_begin(executor: &SyncExecutor, task: &TaskRef) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_task_exec_begin(executor as *const _ as u32, task.as_ptr() as u32)
+    }
+    #[cfg(feature = "rtos-trace")]
+    rtos_trace::trace::task_exec_begin(task.as_ptr() as u32);
+}
+
+#[inline]
+pub(crate) fn task_exec_end(executor: &SyncExecutor, task: &TaskRef) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_task_exec_end(executor as *const _ as u32, task.as_ptr() as u32)
+    }
+    #[cfg(feature = "rtos-trace")]
+    rtos_trace::trace::task_exec_end();
+}
+
+#[inline]
+pub(crate) fn executor_idle(executor: &SyncExecutor) {
+    #[cfg(feature = "trace")]
+    unsafe {
+        _embassy_trace_executor_idle(executor as *const _ as u32)
+    }
+    #[cfg(feature = "rtos-trace")]
+    rtos_trace::trace::system_idle();
+}
+
+/// Returns an iterator over all active tasks in the system
+///
+/// This function provides a convenient way to iterate over all tasks
+/// that are currently tracked in the system. The returned iterator
+/// yields each task in the global task tracker.
+///
+/// # Returns
+/// An iterator that yields `TaskRef` items for each task
+#[cfg(feature = "rtos-trace")]
+fn get_all_active_tasks() -> impl Iterator<Item = TaskRef> + 'static {
+    struct TaskIterator<'a> {
+        tracker: &'a TaskTracker,
+        current: *mut TaskHeader,
+    }
+
+    impl<'a> Iterator for TaskIterator<'a> {
+        type Item = TaskRef;
+
+        fn next(&mut self) -> Option<Self::Item> {
+            if self.current.is_null() {
+                return None;
+            }
+
+            let task = unsafe { TaskRef::from_ptr(self.current) };
+            self.current = unsafe { (*self.current).all_tasks_next.load(Ordering::Acquire) };
+
+            Some(task)
+        }
+    }
+
+    TaskIterator {
+        tracker: &TASK_TRACKER,
+        current: TASK_TRACKER.head.load(Ordering::Acquire),
+    }
+}
+
+/// Perform an action on each active task
+#[cfg(feature = "rtos-trace")]
+fn with_all_active_tasks<F>(f: F)
+where
+    F: FnMut(TaskRef),
+{
+    TASK_TRACKER.for_each(f);
+}
+
+#[cfg(feature = "rtos-trace")]
+impl rtos_trace::RtosTraceOSCallbacks for crate::raw::SyncExecutor {
+    fn task_list() {
+        with_all_active_tasks(|task| {
+            let info = rtos_trace::TaskInfo {
+                name: task.metadata().name().unwrap_or("unnamed task\0"),
+                priority: 0,
+                stack_base: 0,
+                stack_size: 0,
+            };
+            rtos_trace::trace::task_send_info(task.id(), info);
+        });
+    }
+    fn time() -> u64 {
+        const fn gcd(a: u64, b: u64) -> u64 {
+            if b == 0 { a } else { gcd(b, a % b) }
+        }
+
+        const GCD_1M: u64 = gcd(embassy_time_driver::TICK_HZ, 1_000_000);
+        embassy_time_driver::now() * (1_000_000 / GCD_1M) / (embassy_time_driver::TICK_HZ / GCD_1M)
+    }
+}
+
+#[cfg(feature = "rtos-trace")]
+rtos_trace::global_os_callbacks! {SyncExecutor}
diff --git a/embassy-executor/src/raw/waker.rs b/embassy-executor/src/raw/waker.rs
index 8d3910a25..2706f0fdf 100644
--- a/embassy-executor/src/raw/waker.rs
+++ b/embassy-executor/src/raw/waker.rs
@@ -1,6 +1,6 @@
 use core::task::{RawWaker, RawWakerVTable, Waker};
 
-use super::{wake_task, TaskHeader, TaskRef};
+use super::{TaskHeader, TaskRef, wake_task};
 
 static VTABLE: RawWakerVTable = RawWakerVTable::new(clone, wake, wake, drop);
 
@@ -26,38 +26,19 @@ pub(crate) unsafe fn from_task(p: TaskRef) -> Waker {
 /// (1 word) instead of full Wakers (2 words). This saves a bit of RAM and helps
 /// avoid dynamic dispatch.
 ///
-/// You can use the returned task pointer to wake the task with [`wake_task`](super::wake_task).
+/// You can use the returned task pointer to wake the task with [`wake_task`].
 ///
 /// # Panics
 ///
 /// Panics if the waker is not created by the Embassy executor.
 pub fn task_from_waker(waker: &Waker) -> TaskRef {
-    let (vtable, data) = {
-        #[cfg(not(feature = "nightly"))]
-        {
-            struct WakerHack {
-                data: *const (),
-                vtable: &'static RawWakerVTable,
-            }
-
-            // safety: OK because WakerHack has the same layout as Waker.
-            // This is not really guaranteed because the structs are `repr(Rust)`, it is
-            // indeed the case in the current implementation.
-            // TODO use waker_getters when stable. https://github.com/rust-lang/rust/issues/96992
-            let hack: &WakerHack = unsafe { core::mem::transmute(waker) };
-            (hack.vtable, hack.data)
-        }
-
-        #[cfg(feature = "nightly")]
-        {
-            let raw_waker = waker.as_raw();
-            (raw_waker.vtable(), raw_waker.data())
-        }
-    };
-
-    if vtable != &VTABLE {
-        panic!("Found waker not created by the Embassy executor. `embassy_time::Timer` only works with the Embassy executor.")
+    // make sure to compare vtable addresses. Doing `==` on the references
+    // will compare the contents, which is slower.
+    if waker.vtable() as *const _ != &VTABLE as *const _ {
+        panic!(
+            "Found waker not created by the Embassy executor. `embassy_time::Timer` only works with the Embassy executor."
+        )
     }
     // safety: our wakers are always created with `TaskRef::as_ptr`
-    unsafe { TaskRef::from_ptr(data as *const TaskHeader) }
+    unsafe { TaskRef::from_ptr(waker.data() as *const TaskHeader) }
 }
diff --git a/embassy-executor/src/raw/waker_turbo.rs b/embassy-executor/src/raw/waker_turbo.rs
index 435a0ff7e..919bcc61a 100644
--- a/embassy-executor/src/raw/waker_turbo.rs
+++ b/embassy-executor/src/raw/waker_turbo.rs
@@ -1,7 +1,7 @@
 use core::ptr::NonNull;
 use core::task::Waker;
 
-use super::{wake_task, TaskHeader, TaskRef};
+use super::{TaskHeader, TaskRef, wake_task};
 
 pub(crate) unsafe fn from_task(p: TaskRef) -> Waker {
     Waker::from_turbo_ptr(NonNull::new_unchecked(p.as_ptr() as _))
@@ -26,7 +26,7 @@ pub fn task_from_waker(waker: &Waker) -> TaskRef {
 }
 
 #[inline(never)]
-#[no_mangle]
+#[unsafe(no_mangle)]
 fn _turbo_wake(ptr: NonNull<()>) {
     // safety: our wakers are always created with `TaskRef::as_ptr`
     let task = unsafe { TaskRef::from_ptr(ptr.as_ptr() as *const TaskHeader) };
diff --git a/embassy-executor/src/spawner.rs b/embassy-executor/src/spawner.rs
index 271606244..b73a1e7c6 100644
--- a/embassy-executor/src/spawner.rs
+++ b/embassy-executor/src/spawner.rs
@@ -1,9 +1,11 @@
-use core::future::poll_fn;
+use core::future::{Future, poll_fn};
 use core::marker::PhantomData;
 use core::mem;
+use core::sync::atomic::Ordering;
 use core::task::Poll;
 
 use super::raw;
+use crate::Metadata;
 
 /// Token to spawn a newly-created task in an executor.
 ///
@@ -21,24 +23,28 @@ use super::raw;
 /// Once you've invoked a task function and obtained a SpawnToken, you *must* spawn it.
 #[must_use = "Calling a task function does nothing on its own. You must spawn the returned SpawnToken, typically with Spawner::spawn()"]
 pub struct SpawnToken<S> {
-    raw_task: Option<raw::TaskRef>,
+    pub(crate) raw_task: raw::TaskRef,
     phantom: PhantomData<*mut S>,
 }
 
 impl<S> SpawnToken<S> {
     pub(crate) unsafe fn new(raw_task: raw::TaskRef) -> Self {
         Self {
-            raw_task: Some(raw_task),
+            raw_task,
             phantom: PhantomData,
         }
     }
 
-    /// Return a SpawnToken that represents a failed spawn.
-    pub fn new_failed() -> Self {
-        Self {
-            raw_task: None,
-            phantom: PhantomData,
-        }
+    /// Returns the task ID.
+    /// This can be used in combination with rtos-trace to match task names with IDs
+    pub fn id(&self) -> u32 {
+        self.raw_task.id()
+    }
+
+    /// Get the metadata for this task. You can use this to set metadata fields
+    /// prior to spawning it.
+    pub fn metadata(&self) -> &Metadata {
+        self.raw_task.metadata()
     }
 }
 
@@ -50,8 +56,7 @@ impl<S> Drop for SpawnToken<S> {
 }
 
 /// Error returned when spawning a task.
-#[derive(Copy, Clone, Debug)]
-#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+#[derive(Copy, Clone)]
 pub enum SpawnError {
     /// Too many instances of this task are already running.
     ///
@@ -61,6 +66,37 @@ pub enum SpawnError {
     Busy,
 }
 
+impl core::fmt::Debug for SpawnError {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        core::fmt::Display::fmt(self, f)
+    }
+}
+
+impl core::fmt::Display for SpawnError {
+    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
+        match self {
+            SpawnError::Busy => write!(
+                f,
+                "Busy - Too many instances of this task are already running. Check the `pool_size` attribute of the task."
+            ),
+        }
+    }
+}
+
+#[cfg(feature = "defmt")]
+impl defmt::Format for SpawnError {
+    fn format(&self, f: defmt::Formatter) {
+        match self {
+            SpawnError::Busy => defmt::write!(
+                f,
+                "Busy - Too many instances of this task are already running. Check the `pool_size` attribute of the task."
+            ),
+        }
+    }
+}
+
+impl core::error::Error for SpawnError {}
+
 /// Handle to spawn tasks into an executor.
 ///
 /// This Spawner can spawn any task (Send and non-Send ones), but it can
@@ -69,7 +105,7 @@ pub enum SpawnError {
 /// If you want to spawn tasks from another thread, use [SendSpawner].
 #[derive(Copy, Clone)]
 pub struct Spawner {
-    executor: &'static raw::Executor,
+    pub(crate) executor: &'static raw::Executor,
     not_send: PhantomData<*mut ()>,
 }
 
@@ -86,46 +122,47 @@ impl Spawner {
     /// This function is `async` just to get access to the current async
     /// context. It returns instantly, it does not block/yield.
     ///
+    /// Using this method is discouraged due to it being unsafe. Consider the following
+    /// alternatives instead:
+    ///
+    /// - Pass the initial `Spawner` as an argument to tasks. Note that it's `Copy`, so you can
+    ///   make as many copies of it as you want.
+    /// - Use `SendSpawner::for_current_executor()` instead, which is safe but can only be used
+    ///   if task arguments are `Send`.
+    ///
+    /// The only case where using this method is absolutely required is obtaining the `Spawner`
+    /// for an `InterruptExecutor`.
+    ///
+    /// # Safety
+    ///
+    /// You must only execute this with an async `Context` created by the Embassy executor.
+    /// You must not execute it with manually-created `Context`s.
+    ///
     /// # Panics
     ///
     /// Panics if the current executor is not an Embassy executor.
-    pub async fn for_current_executor() -> Self {
+    pub unsafe fn for_current_executor() -> impl Future<Output = Self> {
         poll_fn(|cx| {
             let task = raw::task_from_waker(cx.waker());
-            let executor = unsafe { task.header().executor.get().unwrap_unchecked() };
+            let executor = unsafe {
+                task.header()
+                    .executor
+                    .load(Ordering::Relaxed)
+                    .as_ref()
+                    .unwrap_unchecked()
+            };
             let executor = unsafe { raw::Executor::wrap(executor) };
             Poll::Ready(Self::new(executor))
         })
-        .await
     }
 
     /// Spawn a task into an executor.
     ///
     /// You obtain the `token` by calling a task function (i.e. one marked with `#[embassy_executor::task]`).
-    pub fn spawn<S>(&self, token: SpawnToken<S>) -> Result<(), SpawnError> {
+    pub fn spawn<S>(&self, token: SpawnToken<S>) {
         let task = token.raw_task;
         mem::forget(token);
-
-        match task {
-            Some(task) => {
-                unsafe { self.executor.spawn(task) };
-                Ok(())
-            }
-            None => Err(SpawnError::Busy),
-        }
-    }
-
-    // Used by the `embassy_executor_macros::main!` macro to throw an error when spawn
-    // fails. This is here to allow conditional use of `defmt::unwrap!`
-    // without introducing a `defmt` feature in the `embassy_executor_macros` package,
-    // which would require use of `-Z namespaced-features`.
-    /// Spawn a task into an executor, panicking on failure.
-    ///
-    /// # Panics
-    ///
-    /// Panics if the spawning fails.
-    pub fn must_spawn<S>(&self, token: SpawnToken<S>) {
-        unwrap!(self.spawn(token));
+        unsafe { self.executor.spawn(task) }
     }
 
     /// Convert this Spawner to a SendSpawner. This allows you to send the
@@ -134,6 +171,11 @@ impl Spawner {
     pub fn make_send(&self) -> SendSpawner {
         SendSpawner::new(&self.executor.inner)
     }
+
+    /// Return the unique ID of this Spawner's Executor.
+    pub fn executor_id(&self) -> usize {
+        self.executor.id()
+    }
 }
 
 /// Handle to spawn tasks into an executor from any thread.
@@ -161,37 +203,26 @@ impl SendSpawner {
     /// # Panics
     ///
     /// Panics if the current executor is not an Embassy executor.
-    pub async fn for_current_executor() -> Self {
+    pub fn for_current_executor() -> impl Future<Output = Self> {
         poll_fn(|cx| {
             let task = raw::task_from_waker(cx.waker());
-            let executor = unsafe { task.header().executor.get().unwrap_unchecked() };
+            let executor = unsafe {
+                task.header()
+                    .executor
+                    .load(Ordering::Relaxed)
+                    .as_ref()
+                    .unwrap_unchecked()
+            };
             Poll::Ready(Self::new(executor))
         })
-        .await
     }
 
     /// Spawn a task into an executor.
     ///
     /// You obtain the `token` by calling a task function (i.e. one marked with `#[embassy_executor::task]`).
-    pub fn spawn<S: Send>(&self, token: SpawnToken<S>) -> Result<(), SpawnError> {
+    pub fn spawn<S: Send>(&self, token: SpawnToken<S>) {
         let header = token.raw_task;
         mem::forget(token);
-
-        match header {
-            Some(header) => {
-                unsafe { self.executor.spawn(header) };
-                Ok(())
-            }
-            None => Err(SpawnError::Busy),
-        }
-    }
-
-    /// Spawn a task into an executor, panicking on failure.
-    ///
-    /// # Panics
-    ///
-    /// Panics if the spawning fails.
-    pub fn must_spawn<S: Send>(&self, token: SpawnToken<S>) {
-        unwrap!(self.spawn(token));
+        unsafe { self.executor.spawn(header) }
     }
 }
diff --git a/embassy-executor/tests/test.rs b/embassy-executor/tests/test.rs
index 348cc7dc4..a99976168 100644
--- a/embassy-executor/tests/test.rs
+++ b/embassy-executor/tests/test.rs
@@ -1,14 +1,15 @@
 #![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+#![cfg_attr(feature = "nightly", feature(never_type))]
 
 use std::boxed::Box;
-use std::future::poll_fn;
+use std::future::{Future, poll_fn};
 use std::sync::{Arc, Mutex};
 use std::task::Poll;
 
 use embassy_executor::raw::Executor;
-use embassy_executor::task;
+use embassy_executor::{Spawner, task};
 
-#[export_name = "__pender"]
+#[unsafe(export_name = "__pender")]
 fn __pender(context: *mut ()) {
     unsafe {
         let trace = &*(context as *const Trace);
@@ -40,6 +41,7 @@ fn setup() -> (&'static Executor, Trace) {
     let trace = Trace::new();
     let context = Box::leak(Box::new(trace.clone())) as *mut _ as *mut ();
     let executor = &*Box::leak(Box::new(Executor::new(context)));
+
     (executor, trace)
 }
 
@@ -57,8 +59,41 @@ fn executor_task() {
         trace.push("poll task1")
     }
 
+    #[task]
+    async fn task2() -> ! {
+        panic!()
+    }
+
     let (executor, trace) = setup();
-    executor.spawner().spawn(task1(trace.clone())).unwrap();
+    executor.spawner().spawn(task1(trace.clone()).unwrap());
+
+    unsafe { executor.poll() };
+    unsafe { executor.poll() };
+
+    assert_eq!(
+        trace.get(),
+        &[
+            "pend",       // spawning a task pends the executor
+            "poll task1", // poll only once.
+        ]
+    )
+}
+
+#[test]
+fn executor_task_rpit() {
+    #[task]
+    fn task1(trace: Trace) -> impl Future<Output = ()> {
+        async move { trace.push("poll task1") }
+    }
+
+    #[cfg(feature = "nightly")]
+    #[task]
+    fn task2() -> impl Future<Output = !> {
+        async { panic!() }
+    }
+
+    let (executor, trace) = setup();
+    executor.spawner().spawn(task1(trace.clone()).unwrap());
 
     unsafe { executor.poll() };
     unsafe { executor.poll() };
@@ -85,7 +120,7 @@ fn executor_task_self_wake() {
     }
 
     let (executor, trace) = setup();
-    executor.spawner().spawn(task1(trace.clone())).unwrap();
+    executor.spawner().spawn(task1(trace.clone()).unwrap());
 
     unsafe { executor.poll() };
     unsafe { executor.poll() };
@@ -117,7 +152,7 @@ fn executor_task_self_wake_twice() {
     }
 
     let (executor, trace) = setup();
-    executor.spawner().spawn(task1(trace.clone())).unwrap();
+    executor.spawner().spawn(task1(trace.clone()).unwrap());
 
     unsafe { executor.poll() };
     unsafe { executor.poll() };
@@ -137,6 +172,138 @@ fn executor_task_self_wake_twice() {
 }
 
 #[test]
+fn waking_after_completion_does_not_poll() {
+    use embassy_sync::waitqueue::AtomicWaker;
+
+    #[task]
+    async fn task1(trace: Trace, waker: &'static AtomicWaker) {
+        poll_fn(|cx| {
+            trace.push("poll task1");
+            waker.register(cx.waker());
+            Poll::Ready(())
+        })
+        .await
+    }
+
+    let waker = Box::leak(Box::new(AtomicWaker::new()));
+
+    let (executor, trace) = setup();
+    executor.spawner().spawn(task1(trace.clone(), waker).unwrap());
+
+    unsafe { executor.poll() };
+    waker.wake();
+    unsafe { executor.poll() };
+
+    // Exited task may be waken but is not polled
+    waker.wake();
+    waker.wake();
+    unsafe { executor.poll() }; // Clears running status
+
+    // Can respawn waken-but-dead task
+    executor.spawner().spawn(task1(trace.clone(), waker).unwrap());
+
+    unsafe { executor.poll() };
+
+    assert_eq!(
+        trace.get(),
+        &[
+            "pend",       // spawning a task pends the executor
+            "poll task1", //
+            "pend",       // manual wake, gets cleared by poll
+            "pend",       // manual wake, single pend for two wakes
+            "pend",       // respawning a task pends the executor
+            "poll task1", //
+        ]
+    )
+}
+
+#[test]
+fn waking_with_old_waker_after_respawn() {
+    use embassy_sync::waitqueue::AtomicWaker;
+
+    async fn yield_now(trace: Trace) {
+        let mut yielded = false;
+        poll_fn(|cx| {
+            if yielded {
+                Poll::Ready(())
+            } else {
+                trace.push("yield_now");
+                yielded = true;
+                cx.waker().wake_by_ref();
+                Poll::Pending
+            }
+        })
+        .await
+    }
+
+    #[task]
+    async fn task1(trace: Trace, waker: &'static AtomicWaker) {
+        yield_now(trace.clone()).await;
+        poll_fn(|cx| {
+            trace.push("poll task1");
+            waker.register(cx.waker());
+            Poll::Ready(())
+        })
+        .await;
+    }
+
+    let waker = Box::leak(Box::new(AtomicWaker::new()));
+
+    let (executor, trace) = setup();
+    executor.spawner().spawn(task1(trace.clone(), waker).unwrap());
+
+    unsafe { executor.poll() };
+    unsafe { executor.poll() }; // progress to registering the waker
+    waker.wake();
+    unsafe { executor.poll() };
+    // Task has exited
+
+    assert_eq!(
+        trace.get(),
+        &[
+            "pend",       // spawning a task pends the executor
+            "yield_now",  //
+            "pend",       // yield_now wakes the task
+            "poll task1", //
+            "pend",       // task self-wakes
+        ]
+    );
+
+    // Can respawn task on another executor
+    let (other_executor, other_trace) = setup();
+    other_executor
+        .spawner()
+        .spawn(task1(other_trace.clone(), waker).unwrap());
+
+    unsafe { other_executor.poll() }; // just run to the yield_now
+    waker.wake(); // trigger old waker registration
+    unsafe { executor.poll() };
+    unsafe { other_executor.poll() };
+
+    // First executor's trace has not changed
+    assert_eq!(
+        trace.get(),
+        &[
+            "pend",       // spawning a task pends the executor
+            "yield_now",  //
+            "pend",       // yield_now wakes the task
+            "poll task1", //
+            "pend",       // task self-wakes
+        ]
+    );
+
+    assert_eq!(
+        other_trace.get(),
+        &[
+            "pend",       // spawning a task pends the executor
+            "yield_now",  //
+            "pend",       // manual wake, gets cleared by poll
+            "poll task1", //
+        ]
+    );
+}
+
+#[test]
 fn executor_task_cfg_args() {
     // simulate cfg'ing away argument c
     #[task]
@@ -149,3 +316,43 @@ fn executor_task_cfg_args() {
         let (_, _, _) = (a, b, c);
     }
 }
+
+#[test]
+fn recursive_task() {
+    #[embassy_executor::task(pool_size = 2)]
+    async fn task1() {
+        let spawner = unsafe { Spawner::for_current_executor().await };
+        spawner.spawn(task1().unwrap());
+    }
+}
+
+#[cfg(feature = "metadata-name")]
+#[test]
+fn task_metadata() {
+    #[task]
+    async fn task1(expected_name: Option<&'static str>) {
+        use embassy_executor::Metadata;
+        assert_eq!(Metadata::for_current_task().await.name(), expected_name);
+    }
+
+    // check no task name
+    let (executor, _) = setup();
+    executor.spawner().spawn(task1(None).unwrap());
+    unsafe { executor.poll() };
+
+    // check setting task name
+    let token = task1(Some("foo")).unwrap();
+    token.metadata().set_name("foo");
+    executor.spawner().spawn(token);
+    unsafe { executor.poll() };
+
+    let token = task1(Some("bar")).unwrap();
+    token.metadata().set_name("bar");
+    executor.spawner().spawn(token);
+    unsafe { executor.poll() };
+
+    // check name is cleared if the task pool slot is recycled.
+    let (executor, _) = setup();
+    executor.spawner().spawn(task1(None).unwrap());
+    unsafe { executor.poll() };
+}
diff --git a/embassy-executor/tests/ui.rs b/embassy-executor/tests/ui.rs
new file mode 100644
index 000000000..5486a0624
--- /dev/null
+++ b/embassy-executor/tests/ui.rs
@@ -0,0 +1,38 @@
+#[cfg(not(miri))]
+#[test]
+fn ui() {
+    let t = trybuild::TestCases::new();
+    t.compile_fail("tests/ui/abi.rs");
+    t.compile_fail("tests/ui/bad_return.rs");
+    t.compile_fail("tests/ui/generics.rs");
+    t.compile_fail("tests/ui/impl_trait_nested.rs");
+    t.compile_fail("tests/ui/impl_trait.rs");
+    t.compile_fail("tests/ui/impl_trait_static.rs");
+    t.compile_fail("tests/ui/nonstatic_ref_anon_nested.rs");
+    t.compile_fail("tests/ui/nonstatic_ref_anon.rs");
+    t.compile_fail("tests/ui/nonstatic_ref_elided.rs");
+    t.compile_fail("tests/ui/nonstatic_ref_generic.rs");
+    t.compile_fail("tests/ui/nonstatic_struct_anon.rs");
+    #[cfg(not(feature = "nightly"))] // we can't catch this case with the macro, so the output changes on nightly.
+    t.compile_fail("tests/ui/nonstatic_struct_elided.rs");
+    t.compile_fail("tests/ui/nonstatic_struct_generic.rs");
+    t.compile_fail("tests/ui/not_async.rs");
+    t.compile_fail("tests/ui/spawn_nonsend.rs");
+    t.compile_fail("tests/ui/return_impl_future_nonsend.rs");
+    if rustversion::cfg!(stable) {
+        // output is slightly different on nightly
+        t.compile_fail("tests/ui/bad_return_impl_future.rs");
+        t.compile_fail("tests/ui/return_impl_send.rs");
+    }
+    if cfg!(feature = "nightly") {
+        t.compile_fail("tests/ui/bad_return_impl_future_nightly.rs");
+        t.compile_fail("tests/ui/return_impl_send_nightly.rs");
+    }
+    t.compile_fail("tests/ui/self_ref.rs");
+    t.compile_fail("tests/ui/self.rs");
+    t.compile_fail("tests/ui/type_error.rs");
+    t.compile_fail("tests/ui/where_clause.rs");
+    t.compile_fail("tests/ui/unsafe_op_in_unsafe_task.rs");
+
+    t.pass("tests/ui/task_safety_attribute.rs");
+}
diff --git a/embassy-executor/tests/ui/abi.rs b/embassy-executor/tests/ui/abi.rs
new file mode 100644
index 000000000..fd52f7e41
--- /dev/null
+++ b/embassy-executor/tests/ui/abi.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async extern "C" fn task() {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/abi.stderr b/embassy-executor/tests/ui/abi.stderr
new file mode 100644
index 000000000..e264e371a
--- /dev/null
+++ b/embassy-executor/tests/ui/abi.stderr
@@ -0,0 +1,5 @@
+error: task functions must not have an ABI qualifier
+ --> tests/ui/abi.rs:6:1
+  |
+6 | async extern "C" fn task() {}
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/bad_return.rs b/embassy-executor/tests/ui/bad_return.rs
new file mode 100644
index 000000000..f09a5205b
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return.rs
@@ -0,0 +1,10 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task() -> u32 {
+    5
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/bad_return.stderr b/embassy-executor/tests/ui/bad_return.stderr
new file mode 100644
index 000000000..e9d94dff8
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return.stderr
@@ -0,0 +1,5 @@
+error: task functions must either not return a value, return `()` or return `!`
+ --> tests/ui/bad_return.rs:6:1
+  |
+6 | async fn task() -> u32 {
+  | ^^^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/bad_return_impl_future.rs b/embassy-executor/tests/ui/bad_return_impl_future.rs
new file mode 100644
index 000000000..baaa7dc5a
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return_impl_future.rs
@@ -0,0 +1,9 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+use core::future::Future;
+
+#[embassy_executor::task]
+fn task() -> impl Future<Output = u32> {
+    async { 5 }
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/bad_return_impl_future.stderr b/embassy-executor/tests/ui/bad_return_impl_future.stderr
new file mode 100644
index 000000000..57f147714
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return_impl_future.stderr
@@ -0,0 +1,120 @@
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:5:4
+  |
+4 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+5 | fn task() -> impl Future<Output = u32> {
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_size`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_size<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  -------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_size`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:4:1
+  |
+4 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_size`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_size<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  -------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_size`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:5:4
+  |
+4 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+5 | fn task() -> impl Future<Output = u32> {
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_align`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_align<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  --------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_align`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:4:1
+  |
+4 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_align`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_align<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  --------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_align`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:5:4
+  |
+4 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+5 | fn task() -> impl Future<Output = u32> {
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `__task_pool_get`
+ --> tests/ui/bad_return_impl_future.rs:4:1
+  |
+4 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `__task_pool_get`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:5:4
+  |
+4 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+5 | fn task() -> impl Future<Output = u32> {
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_new`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_new<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> TaskPool<Fut, POOL_SIZE>
+  |                  ------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_new`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future.rs:4:1
+  |
+4 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Future<Output = u32> {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_new`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_new<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> TaskPool<Fut, POOL_SIZE>
+  |                  ------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_new`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/embassy-executor/tests/ui/bad_return_impl_future_nightly.rs b/embassy-executor/tests/ui/bad_return_impl_future_nightly.rs
new file mode 100644
index 000000000..baaa7dc5a
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return_impl_future_nightly.rs
@@ -0,0 +1,9 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+use core::future::Future;
+
+#[embassy_executor::task]
+fn task() -> impl Future<Output = u32> {
+    async { 5 }
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/bad_return_impl_future_nightly.stderr b/embassy-executor/tests/ui/bad_return_impl_future_nightly.stderr
new file mode 100644
index 000000000..3c3c9503b
--- /dev/null
+++ b/embassy-executor/tests/ui/bad_return_impl_future_nightly.stderr
@@ -0,0 +1,10 @@
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/bad_return_impl_future_nightly.rs:4:1
+  |
+4 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskReturnValue` is not implemented for `u32`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+  = help: the following other types implement trait `TaskReturnValue`:
+            ()
+            <fn() -> ! as HasOutput>::Output
diff --git a/embassy-executor/tests/ui/generics.rs b/embassy-executor/tests/ui/generics.rs
new file mode 100644
index 000000000..b83123bb1
--- /dev/null
+++ b/embassy-executor/tests/ui/generics.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task<T: Sized>(_t: T) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/generics.stderr b/embassy-executor/tests/ui/generics.stderr
new file mode 100644
index 000000000..197719a7b
--- /dev/null
+++ b/embassy-executor/tests/ui/generics.stderr
@@ -0,0 +1,5 @@
+error: task functions must not be generic
+ --> tests/ui/generics.rs:6:1
+  |
+6 | async fn task<T: Sized>(_t: T) {}
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/impl_trait.rs b/embassy-executor/tests/ui/impl_trait.rs
new file mode 100644
index 000000000..a21402aa0
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo(_x: impl Sized) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/impl_trait.stderr b/embassy-executor/tests/ui/impl_trait.stderr
new file mode 100644
index 000000000..099b1828f
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait.stderr
@@ -0,0 +1,5 @@
+error: `impl Trait` is not allowed in task arguments. It is syntax sugar for generics, and tasks can't be generic.
+ --> tests/ui/impl_trait.rs:4:18
+  |
+4 | async fn foo(_x: impl Sized) {}
+  |                  ^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/impl_trait_nested.rs b/embassy-executor/tests/ui/impl_trait_nested.rs
new file mode 100644
index 000000000..07442b8fa
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait_nested.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<T>(T);
+
+#[embassy_executor::task]
+async fn foo(_x: Foo<impl Sized + 'static>) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/impl_trait_nested.stderr b/embassy-executor/tests/ui/impl_trait_nested.stderr
new file mode 100644
index 000000000..39592f958
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait_nested.stderr
@@ -0,0 +1,5 @@
+error: `impl Trait` is not allowed in task arguments. It is syntax sugar for generics, and tasks can't be generic.
+ --> tests/ui/impl_trait_nested.rs:6:22
+  |
+6 | async fn foo(_x: Foo<impl Sized + 'static>) {}
+  |                      ^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/impl_trait_static.rs b/embassy-executor/tests/ui/impl_trait_static.rs
new file mode 100644
index 000000000..272470f98
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait_static.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo(_x: impl Sized + 'static) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/impl_trait_static.stderr b/embassy-executor/tests/ui/impl_trait_static.stderr
new file mode 100644
index 000000000..0032a20c9
--- /dev/null
+++ b/embassy-executor/tests/ui/impl_trait_static.stderr
@@ -0,0 +1,5 @@
+error: `impl Trait` is not allowed in task arguments. It is syntax sugar for generics, and tasks can't be generic.
+ --> tests/ui/impl_trait_static.rs:4:18
+  |
+4 | async fn foo(_x: impl Sized + 'static) {}
+  |                  ^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/nonstatic_ref_anon.rs b/embassy-executor/tests/ui/nonstatic_ref_anon.rs
new file mode 100644
index 000000000..417c360a1
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_anon.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo(_x: &'_ u32) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_ref_anon.stderr b/embassy-executor/tests/ui/nonstatic_ref_anon.stderr
new file mode 100644
index 000000000..0544de843
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_anon.stderr
@@ -0,0 +1,5 @@
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_ref_anon.rs:4:19
+  |
+4 | async fn foo(_x: &'_ u32) {}
+  |                   ^^
diff --git a/embassy-executor/tests/ui/nonstatic_ref_anon_nested.rs b/embassy-executor/tests/ui/nonstatic_ref_anon_nested.rs
new file mode 100644
index 000000000..175ebccc1
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_anon_nested.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo(_x: &'static &'_ u32) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_ref_anon_nested.stderr b/embassy-executor/tests/ui/nonstatic_ref_anon_nested.stderr
new file mode 100644
index 000000000..79f262e6b
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_anon_nested.stderr
@@ -0,0 +1,5 @@
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_ref_anon_nested.rs:4:28
+  |
+4 | async fn foo(_x: &'static &'_ u32) {}
+  |                            ^^
diff --git a/embassy-executor/tests/ui/nonstatic_ref_elided.rs b/embassy-executor/tests/ui/nonstatic_ref_elided.rs
new file mode 100644
index 000000000..cf49ad709
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_elided.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo(_x: &u32) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_ref_elided.stderr b/embassy-executor/tests/ui/nonstatic_ref_elided.stderr
new file mode 100644
index 000000000..7e2b9eb7c
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_elided.stderr
@@ -0,0 +1,5 @@
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_ref_elided.rs:4:18
+  |
+4 | async fn foo(_x: &u32) {}
+  |                  ^
diff --git a/embassy-executor/tests/ui/nonstatic_ref_generic.rs b/embassy-executor/tests/ui/nonstatic_ref_generic.rs
new file mode 100644
index 000000000..3f8a26cf8
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_generic.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn foo<'a>(_x: &'a u32) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_ref_generic.stderr b/embassy-executor/tests/ui/nonstatic_ref_generic.stderr
new file mode 100644
index 000000000..af8491ad7
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_ref_generic.stderr
@@ -0,0 +1,11 @@
+error: task functions must not be generic
+ --> tests/ui/nonstatic_ref_generic.rs:4:1
+  |
+4 | async fn foo<'a>(_x: &'a u32) {}
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_ref_generic.rs:4:23
+  |
+4 | async fn foo<'a>(_x: &'a u32) {}
+  |                       ^^
diff --git a/embassy-executor/tests/ui/nonstatic_struct_anon.rs b/embassy-executor/tests/ui/nonstatic_struct_anon.rs
new file mode 100644
index 000000000..ba95d1459
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_anon.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task(_x: Foo<'_>) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_struct_anon.stderr b/embassy-executor/tests/ui/nonstatic_struct_anon.stderr
new file mode 100644
index 000000000..5df2a6e06
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_anon.stderr
@@ -0,0 +1,5 @@
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_struct_anon.rs:6:23
+  |
+6 | async fn task(_x: Foo<'_>) {}
+  |                       ^^
diff --git a/embassy-executor/tests/ui/nonstatic_struct_elided.rs b/embassy-executor/tests/ui/nonstatic_struct_elided.rs
new file mode 100644
index 000000000..4cfe2966a
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_elided.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task(_x: Foo) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_struct_elided.stderr b/embassy-executor/tests/ui/nonstatic_struct_elided.stderr
new file mode 100644
index 000000000..e6829bf5d
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_elided.stderr
@@ -0,0 +1,24 @@
+error[E0726]: implicit elided lifetime not allowed here
+ --> tests/ui/nonstatic_struct_elided.rs:6:19
+  |
+6 | async fn task(_x: Foo) {}
+  |                   ^^^ expected lifetime parameter
+  |
+help: indicate the anonymous lifetime
+  |
+6 | async fn task(_x: Foo<'_>) {}
+  |                      ++++
+
+error: lifetime may not live long enough
+ --> tests/ui/nonstatic_struct_elided.rs:5:1
+  |
+5 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ returning this value requires that `'1` must outlive `'static`
+6 | async fn task(_x: Foo) {}
+  |               -- has type `Foo<'1>`
+  |
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+help: to declare that `impl Sized` captures data from argument `_x`, you can add an explicit `'_` lifetime bound
+  |
+5 | #[embassy_executor::task] + '_
+  |                           ++++
diff --git a/embassy-executor/tests/ui/nonstatic_struct_generic.rs b/embassy-executor/tests/ui/nonstatic_struct_generic.rs
new file mode 100644
index 000000000..ec3d908f6
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_generic.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task<'a>(_x: Foo<'a>) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/nonstatic_struct_generic.stderr b/embassy-executor/tests/ui/nonstatic_struct_generic.stderr
new file mode 100644
index 000000000..61d5231bc
--- /dev/null
+++ b/embassy-executor/tests/ui/nonstatic_struct_generic.stderr
@@ -0,0 +1,11 @@
+error: task functions must not be generic
+ --> tests/ui/nonstatic_struct_generic.rs:6:1
+  |
+6 | async fn task<'a>(_x: Foo<'a>) {}
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error: Arguments for tasks must live forever. Try using the `'static` lifetime.
+ --> tests/ui/nonstatic_struct_generic.rs:6:27
+  |
+6 | async fn task<'a>(_x: Foo<'a>) {}
+  |                           ^^
diff --git a/embassy-executor/tests/ui/not_async.rs b/embassy-executor/tests/ui/not_async.rs
new file mode 100644
index 000000000..f3f7e9bd2
--- /dev/null
+++ b/embassy-executor/tests/ui/not_async.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+fn task() {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/not_async.stderr b/embassy-executor/tests/ui/not_async.stderr
new file mode 100644
index 000000000..27f040d9c
--- /dev/null
+++ b/embassy-executor/tests/ui/not_async.stderr
@@ -0,0 +1,5 @@
+error: task functions must be async
+ --> tests/ui/not_async.rs:6:1
+  |
+6 | fn task() {}
+  | ^^^^^^^^^
diff --git a/embassy-executor/tests/ui/return_impl_future_nonsend.rs b/embassy-executor/tests/ui/return_impl_future_nonsend.rs
new file mode 100644
index 000000000..77b3119d6
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_future_nonsend.rs
@@ -0,0 +1,21 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+use core::future::Future;
+
+use embassy_executor::SendSpawner;
+
+#[embassy_executor::task]
+fn task() -> impl Future<Output = ()> {
+    // runs in spawning thread
+    let non_send: *mut () = core::ptr::null_mut();
+    async move {
+        // runs in executor thread
+        println!("{}", non_send as usize);
+    }
+}
+
+fn send_spawn(s: SendSpawner) {
+    s.spawn(task().unwrap());
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/return_impl_future_nonsend.stderr b/embassy-executor/tests/ui/return_impl_future_nonsend.stderr
new file mode 100644
index 000000000..51944ad65
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_future_nonsend.stderr
@@ -0,0 +1,17 @@
+error: future cannot be sent between threads safely
+  --> tests/ui/return_impl_future_nonsend.rs:18:13
+   |
+18 |     s.spawn(task().unwrap());
+   |             ^^^^^^^^^^^^^^^ future created by async block is not `Send`
+   |
+   = help: within `impl Sized`, the trait `Send` is not implemented for `*mut ()`
+note: captured value is not `Send`
+  --> tests/ui/return_impl_future_nonsend.rs:13:24
+   |
+13 |         println!("{}", non_send as usize);
+   |                        ^^^^^^^^ has type `*mut ()` which is not `Send`
+note: required by a bound in `SendSpawner::spawn`
+  --> src/spawner.rs
+   |
+   |     pub fn spawn<S: Send>(&self, token: SpawnToken<S>) {
+   |                     ^^^^ required by this bound in `SendSpawner::spawn`
diff --git a/embassy-executor/tests/ui/return_impl_send.rs b/embassy-executor/tests/ui/return_impl_send.rs
new file mode 100644
index 000000000..6ddb0e722
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_send.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+fn task() -> impl Send {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/return_impl_send.stderr b/embassy-executor/tests/ui/return_impl_send.stderr
new file mode 100644
index 000000000..5d19465ec
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_send.stderr
@@ -0,0 +1,137 @@
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:4:4
+  |
+3 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+4 | fn task() -> impl Send {}
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_size`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_size<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  -------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_size`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_size`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_size<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  -------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_size`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:4:4
+  |
+3 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+4 | fn task() -> impl Send {}
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_align`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_align<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  --------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_align`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_align`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_align<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> usize
+  |                  --------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_align`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:4:4
+  |
+3 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+4 | fn task() -> impl Send {}
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `__task_pool_get`
+ --> tests/ui/return_impl_send.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `__task_pool_get`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: `impl Send` is not a future
+ --> tests/ui/return_impl_send.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ `impl Send` is not a future
+  |
+  = help: the trait `Future` is not implemented for `impl Send`
+note: required by a bound in `TaskPool::<F, N>::spawn`
+ --> src/raw/mod.rs
+  |
+  | impl<F: Future + 'static, const N: usize> TaskPool<F, N> {
+  |         ^^^^^^ required by this bound in `TaskPool::<F, N>::spawn`
+...
+  |     pub fn spawn(&'static self, future: impl FnOnce() -> F) -> Result<SpawnToken<impl Sized>, SpawnError> {
+  |            ----- required by a bound in this associated function
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:4:4
+  |
+3 | #[embassy_executor::task]
+  | ------------------------- required by a bound introduced by this call
+4 | fn task() -> impl Send {}
+  |    ^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_new`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_new<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> TaskPool<Fut, POOL_SIZE>
+  |                  ------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |                         ^^^^^^^^^ required by this bound in `task_pool_new`
+
+error[E0277]: task futures must resolve to `()` or `!`
+ --> tests/ui/return_impl_send.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `TaskFn<_>` is not implemented for fn item `fn() -> impl Send {__task_task}`
+  |
+  = note: use `async fn` or change the return type to `impl Future<Output = ()>`
+note: required by a bound in `task_pool_new`
+ --> src/lib.rs
+  |
+  |     pub const fn task_pool_new<F, Args, Fut, const POOL_SIZE: usize>(_: F) -> TaskPool<Fut, POOL_SIZE>
+  |                  ------------- required by a bound in this function
+  |     where
+  |         F: TaskFn<Args, Fut = Fut>,
+  |            ^^^^^^^^^^^^^^^^^^^^^^^ required by this bound in `task_pool_new`
+  = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/embassy-executor/tests/ui/return_impl_send_nightly.rs b/embassy-executor/tests/ui/return_impl_send_nightly.rs
new file mode 100644
index 000000000..6ddb0e722
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_send_nightly.rs
@@ -0,0 +1,6 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+fn task() -> impl Send {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/return_impl_send_nightly.stderr b/embassy-executor/tests/ui/return_impl_send_nightly.stderr
new file mode 100644
index 000000000..de9ba6243
--- /dev/null
+++ b/embassy-executor/tests/ui/return_impl_send_nightly.stderr
@@ -0,0 +1,10 @@
+error[E0277]: `impl Send` is not a future
+ --> tests/ui/return_impl_send_nightly.rs:3:1
+  |
+3 | #[embassy_executor::task]
+  | ^^^^^^^^^^^^^^^^^^^^^^^^^
+  | |
+  | `impl Send` is not a future
+  | return type was inferred to be `impl Send` here
+  |
+  = help: the trait `Future` is not implemented for `impl Send`
diff --git a/embassy-executor/tests/ui/self.rs b/embassy-executor/tests/ui/self.rs
new file mode 100644
index 000000000..f83a962d1
--- /dev/null
+++ b/embassy-executor/tests/ui/self.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task(self) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/self.stderr b/embassy-executor/tests/ui/self.stderr
new file mode 100644
index 000000000..aaf031573
--- /dev/null
+++ b/embassy-executor/tests/ui/self.stderr
@@ -0,0 +1,13 @@
+error: task functions must not have `self` arguments
+ --> tests/ui/self.rs:6:15
+  |
+6 | async fn task(self) {}
+  |               ^^^^
+
+error: `self` parameter is only allowed in associated functions
+ --> tests/ui/self.rs:6:15
+  |
+6 | async fn task(self) {}
+  |               ^^^^ not semantically valid as function parameter
+  |
+  = note: associated functions are those in `impl` or `trait` definitions
diff --git a/embassy-executor/tests/ui/self_ref.rs b/embassy-executor/tests/ui/self_ref.rs
new file mode 100644
index 000000000..5e49bba5e
--- /dev/null
+++ b/embassy-executor/tests/ui/self_ref.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task(&mut self) {}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/self_ref.stderr b/embassy-executor/tests/ui/self_ref.stderr
new file mode 100644
index 000000000..dd2052977
--- /dev/null
+++ b/embassy-executor/tests/ui/self_ref.stderr
@@ -0,0 +1,13 @@
+error: task functions must not have `self` arguments
+ --> tests/ui/self_ref.rs:6:15
+  |
+6 | async fn task(&mut self) {}
+  |               ^^^^^^^^^
+
+error: `self` parameter is only allowed in associated functions
+ --> tests/ui/self_ref.rs:6:15
+  |
+6 | async fn task(&mut self) {}
+  |               ^^^^^^^^^ not semantically valid as function parameter
+  |
+  = note: associated functions are those in `impl` or `trait` definitions
diff --git a/embassy-executor/tests/ui/spawn_nonsend.rs b/embassy-executor/tests/ui/spawn_nonsend.rs
new file mode 100644
index 000000000..a06c0b37a
--- /dev/null
+++ b/embassy-executor/tests/ui/spawn_nonsend.rs
@@ -0,0 +1,14 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+use embassy_executor::SendSpawner;
+
+#[embassy_executor::task]
+async fn task(non_send: *mut ()) {
+    println!("{}", non_send as usize);
+}
+
+fn send_spawn(s: SendSpawner) {
+    s.spawn(task(core::ptr::null_mut()).unwrap());
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/spawn_nonsend.stderr b/embassy-executor/tests/ui/spawn_nonsend.stderr
new file mode 100644
index 000000000..31efadd49
--- /dev/null
+++ b/embassy-executor/tests/ui/spawn_nonsend.stderr
@@ -0,0 +1,33 @@
+error[E0277]: `*mut ()` cannot be sent between threads safely
+  --> tests/ui/spawn_nonsend.rs:11:13
+   |
+ 5 | #[embassy_executor::task]
+   | ------------------------- within this `impl Sized`
+...
+11 |     s.spawn(task(core::ptr::null_mut()).unwrap());
+   |       ----- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ `*mut ()` cannot be sent between threads safely
+   |       |
+   |       required by a bound introduced by this call
+   |
+   = help: within `impl Sized`, the trait `Send` is not implemented for `*mut ()`
+note: required because it's used within this closure
+  --> tests/ui/spawn_nonsend.rs:5:1
+   |
+ 5 | #[embassy_executor::task]
+   | ^^^^^^^^^^^^^^^^^^^^^^^^^
+note: required because it appears within the type `impl Sized`
+  --> src/raw/mod.rs
+   |
+   |     pub unsafe fn _spawn_async_fn<FutFn>(&'static self, future: FutFn) -> Result<SpawnToken<impl Sized>, SpawnError>
+   |                                                                                             ^^^^^^^^^^
+note: required because it appears within the type `impl Sized`
+  --> tests/ui/spawn_nonsend.rs:5:1
+   |
+ 5 | #[embassy_executor::task]
+   | ^^^^^^^^^^^^^^^^^^^^^^^^^
+note: required by a bound in `SendSpawner::spawn`
+  --> src/spawner.rs
+   |
+   |     pub fn spawn<S: Send>(&self, token: SpawnToken<S>) {
+   |                     ^^^^ required by this bound in `SendSpawner::spawn`
+   = note: this error originates in the attribute macro `embassy_executor::task` (in Nightly builds, run with -Z macro-backtrace for more info)
diff --git a/embassy-executor/tests/ui/task_safety_attribute.rs b/embassy-executor/tests/ui/task_safety_attribute.rs
new file mode 100644
index 000000000..46a5c665f
--- /dev/null
+++ b/embassy-executor/tests/ui/task_safety_attribute.rs
@@ -0,0 +1,25 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+#![deny(unused_unsafe)]
+
+use std::mem;
+
+#[embassy_executor::task]
+async fn safe() {}
+
+#[embassy_executor::task]
+async unsafe fn not_safe() {}
+
+#[unsafe(export_name = "__pender")]
+fn pender(_: *mut ()) {
+    // The test doesn't link if we don't include this.
+    // We never call this anyway.
+}
+
+fn main() {
+    let _forget_me = safe();
+    // SAFETY: not_safe has not safety preconditions
+    let _forget_me2 = unsafe { not_safe() };
+
+    mem::forget(_forget_me);
+    mem::forget(_forget_me2);
+}
diff --git a/embassy-executor/tests/ui/type_error.rs b/embassy-executor/tests/ui/type_error.rs
new file mode 100644
index 000000000..1734bc6c4
--- /dev/null
+++ b/embassy-executor/tests/ui/type_error.rs
@@ -0,0 +1,8 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+#[embassy_executor::task]
+async fn task() {
+    5
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/type_error.stderr b/embassy-executor/tests/ui/type_error.stderr
new file mode 100644
index 000000000..bce315811
--- /dev/null
+++ b/embassy-executor/tests/ui/type_error.stderr
@@ -0,0 +1,7 @@
+error[E0308]: mismatched types
+ --> tests/ui/type_error.rs:5:5
+  |
+4 | async fn task() {
+  |                - help: try adding a return type: `-> i32`
+5 |     5
+  |     ^ expected `()`, found integer
diff --git a/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.rs b/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.rs
new file mode 100644
index 000000000..ee7924838
--- /dev/null
+++ b/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.rs
@@ -0,0 +1,10 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+#![deny(unsafe_op_in_unsafe_fn)]
+
+#[embassy_executor::task]
+async unsafe fn task() {
+    let x = 5;
+    (&x as *const i32).read();
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.stderr b/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.stderr
new file mode 100644
index 000000000..033395584
--- /dev/null
+++ b/embassy-executor/tests/ui/unsafe_op_in_unsafe_task.stderr
@@ -0,0 +1,18 @@
+error[E0133]: call to unsafe function `std::ptr::const_ptr::<impl *const T>::read` is unsafe and requires unsafe block
+ --> tests/ui/unsafe_op_in_unsafe_task.rs:7:5
+  |
+7 |     (&x as *const i32).read();
+  |     ^^^^^^^^^^^^^^^^^^^^^^^^^ call to unsafe function
+  |
+  = note: for more information, see <https://doc.rust-lang.org/edition-guide/rust-2024/unsafe-op-in-unsafe-fn.html>
+  = note: consult the function's documentation for information on how to avoid undefined behavior
+note: an unsafe function restricts its caller, but its body is safe by default
+ --> tests/ui/unsafe_op_in_unsafe_task.rs:5:1
+  |
+5 | async unsafe fn task() {
+  | ^^^^^^^^^^^^^^^^^^^^^^
+note: the lint level is defined here
+ --> tests/ui/unsafe_op_in_unsafe_task.rs:2:9
+  |
+2 | #![deny(unsafe_op_in_unsafe_fn)]
+  |         ^^^^^^^^^^^^^^^^^^^^^^
diff --git a/embassy-executor/tests/ui/where_clause.rs b/embassy-executor/tests/ui/where_clause.rs
new file mode 100644
index 000000000..848d78149
--- /dev/null
+++ b/embassy-executor/tests/ui/where_clause.rs
@@ -0,0 +1,12 @@
+#![cfg_attr(feature = "nightly", feature(impl_trait_in_assoc_type))]
+
+struct Foo<'a>(&'a ());
+
+#[embassy_executor::task]
+async fn task()
+where
+    (): Sized,
+{
+}
+
+fn main() {}
diff --git a/embassy-executor/tests/ui/where_clause.stderr b/embassy-executor/tests/ui/where_clause.stderr
new file mode 100644
index 000000000..eba45af40
--- /dev/null
+++ b/embassy-executor/tests/ui/where_clause.stderr
@@ -0,0 +1,7 @@
+error: task functions must not have `where` clauses
+ --> tests/ui/where_clause.rs:6:1
+  |
+6 | / async fn task()
+7 | | where
+8 | |     (): Sized,
+  | |______________^