181 files changed, 2094 insertions, 878 deletions

diff --git a/.github/ci/test.sh b/.github/ci/test.sh
index d014e4bd7..2892bcf8d 100755
--- a/.github/ci/test.sh
+++ b/.github/ci/test.sh
@@ -13,7 +13,7 @@ hashtime save /ci/cache/filetime.json
 
 cargo test --manifest-path ./embassy-sync/Cargo.toml 
 cargo test --manifest-path ./embassy-embedded-hal/Cargo.toml 
-cargo test --manifest-path ./embassy-hal-common/Cargo.toml 
+cargo test --manifest-path ./embassy-hal-internal/Cargo.toml 
 cargo test --manifest-path ./embassy-time/Cargo.toml --features generic-queue
 
 cargo test --manifest-path ./embassy-boot/boot/Cargo.toml
diff --git a/README.md b/README.md
index b05e55aa5..c4c01dfbc 100644
--- a/README.md
+++ b/README.md
@@ -33,6 +33,7 @@ The <a href="https://docs.embassy.dev/embassy-net/">embassy-net</a> network stac
 
 - **Bluetooth** - 
 The <a href="https://github.com/embassy-rs/nrf-softdevice">nrf-softdevice</a> crate provides Bluetooth Low Energy 4.x and 5.x support for nRF52 microcontrollers.
+The <a href="https://github.com/embassy-rs/embassy/tree/main/embassy-stm32-wpan">embassy-stm32-wpan</a> crate provides Bluetooth Low Energy 5.x support for stm32wb microcontrollers.
 
 - **LoRa** - 
 <a href="https://docs.embassy.dev/embassy-lora/">embassy-lora</a> supports LoRa networking.
@@ -111,6 +112,12 @@ cargo install probe-rs --features cli
 cd examples/nrf52840
 ```
 
+- Ensure `Cargo.toml` sets the right feature for the name of the chip you are programming.
+  If this name is incorrect, the example may fail to run or immediately crash
+  after being programmed.
+
+- Ensure `.cargo/config.toml` contains the name of the chip you are programming.
+
 - Run the example
 
 For example:
@@ -119,6 +126,8 @@ For example:
 cargo run --release --bin blinky
 ```
 
+For more help getting started, see [Getting Started][1] and [Running the Examples][2].
+
 ## Developing Embassy with Rust Analyzer based editors
 
 The [Rust Analyzer](https://rust-analyzer.github.io/) is used by [Visual Studio Code](https://code.visualstudio.com/)
@@ -151,3 +160,5 @@ This work is licensed under either of
 
 at your option.
 
+[1]: https://github.com/embassy-rs/embassy/wiki/Getting-Started
+[2]: https://github.com/embassy-rs/embassy/wiki/Running-the-Examples
diff --git a/ci.sh b/ci.sh
index 6c10a0630..32ae7855e 100755
--- a/ci.sh
+++ b/ci.sh
@@ -27,6 +27,8 @@ cargo batch  \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,dhcpv4,medium-ethernet,nightly \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,dhcpv4,medium-ethernet,unstable-traits,nightly \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ethernet \
+    --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ieee802154 \
+    --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ethernet,medium-ieee802154 \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ethernet,unstable-traits \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ethernet,nightly \
     --- build --release --manifest-path embassy-net/Cargo.toml --target thumbv7em-none-eabi --features defmt,tcp,udp,dns,proto-ipv6,medium-ethernet,unstable-traits,nightly \
diff --git a/cyw43-firmware/43439A0.bin b/cyw43-firmware/43439A0.bin
index b46b3beff..017375277 100755
--- a/cyw43-firmware/43439A0.bin
+++ b/cyw43-firmware/43439A0.bin
diff --git a/cyw43-firmware/43439A0_clm.bin b/cyw43-firmware/43439A0_clm.bin
index 6e3ba786b..1fedd753a 100755
--- a/cyw43-firmware/43439A0_clm.bin
+++ b/cyw43-firmware/43439A0_clm.bin
diff --git a/cyw43-firmware/README.md b/cyw43-firmware/README.md
index 7381fdc56..db3d9c9cf 100644
--- a/cyw43-firmware/README.md
+++ b/cyw43-firmware/README.md
@@ -2,4 +2,8 @@
 
 Firmware obtained from https://github.com/Infineon/wifi-host-driver/tree/master/WiFi_Host_Driver/resources/firmware/COMPONENT_43439
 
-Licensed under the [Infineon Permissive Binary License](./LICENSE-permissive-binary-license-1.0.txt)
\ No newline at end of file
+Licensed under the [Infineon Permissive Binary License](./LICENSE-permissive-binary-license-1.0.txt)
+
+## Changelog
+
+* 2023-07-28: synced with `ad3bad0` - Update 43439 fw from 7.95.55 ot 7.95.62
diff --git a/cyw43-pio/src/lib.rs b/cyw43-pio/src/lib.rs
index dca30c74d..830a5b44a 100644
--- a/cyw43-pio/src/lib.rs
+++ b/cyw43-pio/src/lib.rs
@@ -8,7 +8,6 @@ use cyw43::SpiBusCyw43;
 use embassy_rp::dma::Channel;
 use embassy_rp::gpio::{Drive, Level, Output, Pin, Pull, SlewRate};
 use embassy_rp::pio::{Common, Config, Direction, Instance, Irq, PioPin, ShiftDirection, StateMachine};
-use embassy_rp::relocate::RelocatedProgram;
 use embassy_rp::{pio_instr_util, Peripheral, PeripheralRef};
 use fixed::FixedU32;
 use pio_proc::pio_asm;
@@ -88,8 +87,6 @@ where
             ".wrap"
         );
 
-        let relocated = RelocatedProgram::new(&program.program);
-
         let mut pin_io: embassy_rp::pio::Pin<PIO> = common.make_pio_pin(dio);
         pin_io.set_pull(Pull::None);
         pin_io.set_schmitt(true);
@@ -102,7 +99,8 @@ where
         pin_clk.set_slew_rate(SlewRate::Fast);
 
         let mut cfg = Config::default();
-        cfg.use_program(&common.load_program(&relocated), &[&pin_clk]);
+        let loaded_program = common.load_program(&program.program);
+        cfg.use_program(&loaded_program, &[&pin_clk]);
         cfg.set_out_pins(&[&pin_io]);
         cfg.set_in_pins(&[&pin_io]);
         cfg.set_set_pins(&[&pin_io]);
@@ -142,7 +140,7 @@ where
             sm,
             irq,
             dma: dma.into_ref(),
-            wrap_target: relocated.wrap().target,
+            wrap_target: loaded_program.wrap.target,
         }
     }
 
diff --git a/embassy-hal-common/Cargo.toml b/embassy-hal-internal/Cargo.toml
index 18c758d7b..42e03199c 100644
--- a/embassy-hal-common/Cargo.toml
+++ b/embassy-hal-internal/Cargo.toml
@@ -1,5 +1,5 @@
 [package]
-name = "embassy-hal-common"
+name = "embassy-hal-internal"
 version = "0.1.0"
 edition = "2021"
 license = "MIT OR Apache-2.0"
diff --git a/embassy-hal-internal/README.md b/embassy-hal-internal/README.md
new file mode 100644
index 000000000..d6539701b
--- /dev/null
+++ b/embassy-hal-internal/README.md
@@ -0,0 +1,16 @@
+# embassy-macros
+
+An [Embassy](https://embassy.dev) project.
+
+Internal implementation details for Embassy HALs. DO NOT USE DIRECTLY. Embassy HALs (`embassy-nrf`, `embassy-stm32`, `embassy-rp`) already reexport
+everything you need to use them effectively.
+
+## License
+
+This work is licensed under either of
+
+- Apache License, Version 2.0 ([LICENSE-APACHE](LICENSE-APACHE) or
+  <http://www.apache.org/licenses/LICENSE-2.0>)
+- MIT license ([LICENSE-MIT](LICENSE-MIT) or <http://opensource.org/licenses/MIT>)
+
+at your option.
diff --git a/embassy-hal-common/build.rs b/embassy-hal-internal/build.rs
index 6fe82b44f..6fe82b44f 100644
--- a/embassy-hal-common/build.rs
+++ b/embassy-hal-internal/build.rs
diff --git a/embassy-hal-common/src/atomic_ring_buffer.rs b/embassy-hal-internal/src/atomic_ring_buffer.rs
index ea84925c4..ea84925c4 100644
--- a/embassy-hal-common/src/atomic_ring_buffer.rs
+++ b/embassy-hal-internal/src/atomic_ring_buffer.rs
diff --git a/embassy-hal-common/src/drop.rs b/embassy-hal-internal/src/drop.rs
index 7cd16aaec..7cd16aaec 100644
--- a/embassy-hal-common/src/drop.rs
+++ b/embassy-hal-internal/src/drop.rs
diff --git a/embassy-hal-common/src/fmt.rs b/embassy-hal-internal/src/fmt.rs
index 066970813..066970813 100644
--- a/embassy-hal-common/src/fmt.rs
+++ b/embassy-hal-internal/src/fmt.rs
diff --git a/embassy-hal-common/src/interrupt.rs b/embassy-hal-internal/src/interrupt.rs
index b970aa2cd..b970aa2cd 100644
--- a/embassy-hal-common/src/interrupt.rs
+++ b/embassy-hal-internal/src/interrupt.rs
diff --git a/embassy-hal-common/src/lib.rs b/embassy-hal-internal/src/lib.rs
index 235964aa4..3640ea184 100644
--- a/embassy-hal-common/src/lib.rs
+++ b/embassy-hal-internal/src/lib.rs
@@ -1,5 +1,6 @@
 #![no_std]
 #![allow(clippy::new_without_default)]
+#![doc = include_str!("../README.md")]
 
 // This mod MUST go first, so that the others see its macros.
 pub(crate) mod fmt;
diff --git a/embassy-hal-common/src/macros.rs b/embassy-hal-internal/src/macros.rs
index f06b46002..f06b46002 100644
--- a/embassy-hal-common/src/macros.rs
+++ b/embassy-hal-internal/src/macros.rs
diff --git a/embassy-hal-common/src/peripheral.rs b/embassy-hal-internal/src/peripheral.rs
index 38b4c452e..38b4c452e 100644
--- a/embassy-hal-common/src/peripheral.rs
+++ b/embassy-hal-internal/src/peripheral.rs
diff --git a/embassy-hal-common/src/ratio.rs b/embassy-hal-internal/src/ratio.rs
index 9a8808a33..9a8808a33 100644
--- a/embassy-hal-common/src/ratio.rs
+++ b/embassy-hal-internal/src/ratio.rs
diff --git a/embassy-hal-common/src/ring_buffer.rs b/embassy-hal-internal/src/ring_buffer.rs
index fcad68bb1..fcad68bb1 100644
--- a/embassy-hal-common/src/ring_buffer.rs
+++ b/embassy-hal-internal/src/ring_buffer.rs
diff --git a/embassy-lora/Cargo.toml b/embassy-lora/Cargo.toml
index e4524af5b..402ad2d70 100644
--- a/embassy-lora/Cargo.toml
+++ b/embassy-lora/Cargo.toml
@@ -12,7 +12,7 @@ target = "thumbv7em-none-eabi"
 
 [features]
 stm32wl = ["dep:embassy-stm32"]
-time = []
+time = ["embassy-time", "lorawan-device"]
 defmt = ["dep:defmt", "lorawan-device/defmt"]
 
 [dependencies]
@@ -20,18 +20,12 @@ defmt = ["dep:defmt", "lorawan-device/defmt"]
 defmt = { version = "0.3", optional = true }
 log = { version = "0.4.14", optional = true }
 
-embassy-time = { version = "0.1.2", path = "../embassy-time" }
+embassy-time = { version = "0.1.2", path = "../embassy-time", optional = true }
 embassy-sync = { version = "0.2.0", path = "../embassy-sync" }
 embassy-stm32 = { version = "0.1.0", path = "../embassy-stm32", default-features = false, optional = true }
-embedded-hal-1 = { package = "embedded-hal", version = "=1.0.0-alpha.11" }
 embedded-hal-async = { version = "=0.2.0-alpha.2" }
-embassy-hal-common = { version = "0.1.0", path = "../embassy-hal-common", default-features = false }
-futures = { version = "0.3.17", default-features = false, features = [ "async-await" ] }
 embedded-hal = { version = "0.2", features = ["unproven"] }
-bit_field = { version = "0.10" }
 
+futures = { version = "0.3.17", default-features = false, features = [ "async-await" ] }
 lora-phy = { version = "1" }
-lorawan-device = { version = "0.10.0", default-features = false, features = ["async"] }
-
-[patch.crates-io]
-lora-phy = { git = "https://github.com/embassy-rs/lora-phy", rev = "ad289428fd44b02788e2fa2116445cc8f640a265" }
+lorawan-device = { version = "0.10.0", default-features = false, features = ["async"], optional = true }
diff --git a/embassy-net-esp-hosted/src/control.rs b/embassy-net-esp-hosted/src/control.rs
index 79f8cde7b..37f220da0 100644
--- a/embassy-net-esp-hosted/src/control.rs
+++ b/embassy-net-esp-hosted/src/control.rs
@@ -1,5 +1,4 @@
 use ch::driver::LinkState;
-use defmt::Debug2Format;
 use embassy_net_driver_channel as ch;
 use heapless::String;
 
@@ -57,7 +56,6 @@ impl<'a> Control<'a> {
         let proto::CtrlMsgPayload::RespConnectAp(resp) = resp.payload.unwrap() else {
             panic!("unexpected resp")
         };
-        debug!("======= {:?}", Debug2Format(&resp));
         assert_eq!(resp.resp, 0);
         self.state_ch.set_link_state(LinkState::Up);
     }
diff --git a/embassy-net/src/lib.rs b/embassy-net/src/lib.rs
index 81c751a2c..3f9150168 100644
--- a/embassy-net/src/lib.rs
+++ b/embassy-net/src/lib.rs
@@ -32,12 +32,14 @@ pub use smoltcp::iface::MulticastError;
 use smoltcp::iface::{Interface, SocketHandle, SocketSet, SocketStorage};
 #[cfg(feature = "dhcpv4")]
 use smoltcp::socket::dhcpv4::{self, RetryConfig};
+#[cfg(feature = "medium-ethernet")]
+pub use smoltcp::wire::EthernetAddress;
+#[cfg(any(feature = "medium-ethernet", feature = "medium-ieee802154"))]
+pub use smoltcp::wire::HardwareAddress;
 #[cfg(feature = "udp")]
 pub use smoltcp::wire::IpListenEndpoint;
-#[cfg(feature = "medium-ethernet")]
-pub use smoltcp::wire::{EthernetAddress, HardwareAddress};
 #[cfg(feature = "medium-ieee802154")]
-pub use smoltcp::wire::{HardwareAddress, Ieee802154Address};
+pub use smoltcp::wire::{Ieee802154Address, Ieee802154Frame};
 pub use smoltcp::wire::{IpAddress, IpCidr, IpEndpoint};
 #[cfg(feature = "proto-ipv4")]
 pub use smoltcp::wire::{Ipv4Address, Ipv4Cidr};
@@ -583,7 +585,7 @@ impl SocketStack {
 impl<D: Driver + 'static> Inner<D> {
     #[cfg(feature = "proto-ipv4")]
     fn apply_config_v4(&mut self, s: &mut SocketStack, config: StaticConfigV4) {
-        #[cfg(feature = "medium-ethernet")]
+        #[cfg(any(feature = "medium-ethernet", feature = "medium-ieee802154"))]
         let medium = self.device.capabilities().medium;
 
         debug!("Acquired IP configuration:");
diff --git a/embassy-nrf/Cargo.toml b/embassy-nrf/Cargo.toml
index 57dd22f1c..d10cd2c34 100644
--- a/embassy-nrf/Cargo.toml
+++ b/embassy-nrf/Cargo.toml
@@ -93,7 +93,7 @@ _gpio-p1 = []
 [dependencies]
 embassy-time = { version = "0.1.2", path = "../embassy-time", optional = true }
 embassy-sync = { version = "0.2.0", path = "../embassy-sync" }
-embassy-hal-common = {version = "0.1.0", path = "../embassy-hal-common", features = ["cortex-m", "prio-bits-3"] }
+embassy-hal-internal = {version = "0.1.0", path = "../embassy-hal-internal", features = ["cortex-m", "prio-bits-3"] }
 embassy-embedded-hal = {version = "0.1.0", path = "../embassy-embedded-hal" }
 embassy-usb-driver = {version = "0.1.0", path = "../embassy-usb-driver", optional=true }
 
diff --git a/embassy-nrf/src/buffered_uarte.rs b/embassy-nrf/src/buffered_uarte.rs
index 9bc1c1e7a..5a0a3c7c0 100644
--- a/embassy-nrf/src/buffered_uarte.rs
+++ b/embassy-nrf/src/buffered_uarte.rs
@@ -15,8 +15,8 @@ use core::slice;
 use core::sync::atomic::{compiler_fence, AtomicU8, AtomicUsize, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::atomic_ring_buffer::RingBuffer;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::atomic_ring_buffer::RingBuffer;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 // Re-export SVD variants to allow user to directly set values
 pub use pac::uarte0::{baudrate::BAUDRATE_A as Baudrate, config::PARITY_A as Parity};
diff --git a/embassy-nrf/src/chips/nrf52805.rs b/embassy-nrf/src/chips/nrf52805.rs
index 8776000c8..70e4b4863 100644
--- a/embassy-nrf/src/chips/nrf52805.rs
+++ b/embassy-nrf/src/chips/nrf52805.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 192 * 1024;
 
 pub const RESET_PIN: u32 = 21;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -208,7 +208,7 @@ impl_ppi_channel!(PPI_CH31, 31 => static);
 impl_saadc_input!(P0_04, ANALOG_INPUT2);
 impl_saadc_input!(P0_05, ANALOG_INPUT3);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52810.rs b/embassy-nrf/src/chips/nrf52810.rs
index 5519e8953..7416d3912 100644
--- a/embassy-nrf/src/chips/nrf52810.rs
+++ b/embassy-nrf/src/chips/nrf52810.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 192 * 1024;
 
 pub const RESET_PIN: u32 = 21;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -234,7 +234,7 @@ impl_saadc_input!(P0_29, ANALOG_INPUT5);
 impl_saadc_input!(P0_30, ANALOG_INPUT6);
 impl_saadc_input!(P0_31, ANALOG_INPUT7);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52811.rs b/embassy-nrf/src/chips/nrf52811.rs
index d5367c59a..588010685 100644
--- a/embassy-nrf/src/chips/nrf52811.rs
+++ b/embassy-nrf/src/chips/nrf52811.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 192 * 1024;
 
 pub const RESET_PIN: u32 = 21;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -236,7 +236,7 @@ impl_saadc_input!(P0_29, ANALOG_INPUT5);
 impl_saadc_input!(P0_30, ANALOG_INPUT6);
 impl_saadc_input!(P0_31, ANALOG_INPUT7);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52820.rs b/embassy-nrf/src/chips/nrf52820.rs
index 785170447..0ecddaf31 100644
--- a/embassy-nrf/src/chips/nrf52820.rs
+++ b/embassy-nrf/src/chips/nrf52820.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 256 * 1024;
 
 pub const RESET_PIN: u32 = 18;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // USB
     USBD,
 
@@ -224,7 +224,7 @@ impl_ppi_channel!(PPI_CH29, 29 => static);
 impl_ppi_channel!(PPI_CH30, 30 => static);
 impl_ppi_channel!(PPI_CH31, 31 => static);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52832.rs b/embassy-nrf/src/chips/nrf52832.rs
index b77564a5c..ae39628d2 100644
--- a/embassy-nrf/src/chips/nrf52832.rs
+++ b/embassy-nrf/src/chips/nrf52832.rs
@@ -12,7 +12,7 @@ pub const FLASH_SIZE: usize = 512 * 1024;
 
 pub const RESET_PIN: u32 = 21;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -263,7 +263,7 @@ impl_saadc_input!(P0_31, ANALOG_INPUT7);
 
 impl_i2s!(I2S, I2S, I2S);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52833.rs b/embassy-nrf/src/chips/nrf52833.rs
index bff7f4ebb..b8830b338 100644
--- a/embassy-nrf/src/chips/nrf52833.rs
+++ b/embassy-nrf/src/chips/nrf52833.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 512 * 1024;
 
 pub const RESET_PIN: u32 = 18;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // USB
     USBD,
 
@@ -306,7 +306,7 @@ impl_saadc_input!(P0_31, ANALOG_INPUT7);
 
 impl_i2s!(I2S, I2S, I2S);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf52840.rs b/embassy-nrf/src/chips/nrf52840.rs
index 9b0050823..a490cb079 100644
--- a/embassy-nrf/src/chips/nrf52840.rs
+++ b/embassy-nrf/src/chips/nrf52840.rs
@@ -8,7 +8,7 @@ pub const FLASH_SIZE: usize = 1024 * 1024;
 
 pub const RESET_PIN: u32 = 18;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // USB
     USBD,
 
@@ -311,7 +311,7 @@ impl_saadc_input!(P0_31, ANALOG_INPUT7);
 
 impl_i2s!(I2S, I2S, I2S);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     POWER_CLOCK,
     RADIO,
     UARTE0_UART0,
diff --git a/embassy-nrf/src/chips/nrf5340_app.rs b/embassy-nrf/src/chips/nrf5340_app.rs
index 410ae921c..afc2c4a7e 100644
--- a/embassy-nrf/src/chips/nrf5340_app.rs
+++ b/embassy-nrf/src/chips/nrf5340_app.rs
@@ -218,7 +218,7 @@ pub const FORCE_COPY_BUFFER_SIZE: usize = 1024;
 
 pub const FLASH_SIZE: usize = 1024 * 1024;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // USB
     USBD,
 
@@ -506,7 +506,7 @@ impl_saadc_input!(P0_18, ANALOG_INPUT5);
 impl_saadc_input!(P0_19, ANALOG_INPUT6);
 impl_saadc_input!(P0_20, ANALOG_INPUT7);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     FPU,
     CACHE,
     SPU,
diff --git a/embassy-nrf/src/chips/nrf5340_net.rs b/embassy-nrf/src/chips/nrf5340_net.rs
index 6ac783085..dee666a61 100644
--- a/embassy-nrf/src/chips/nrf5340_net.rs
+++ b/embassy-nrf/src/chips/nrf5340_net.rs
@@ -109,7 +109,7 @@ pub const FORCE_COPY_BUFFER_SIZE: usize = 1024;
 
 pub const FLASH_SIZE: usize = 256 * 1024;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -342,7 +342,7 @@ impl_ppi_channel!(PPI_CH29, 29 => configurable);
 impl_ppi_channel!(PPI_CH30, 30 => configurable);
 impl_ppi_channel!(PPI_CH31, 31 => configurable);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     CLOCK_POWER,
     RADIO,
     RNG,
diff --git a/embassy-nrf/src/chips/nrf9160.rs b/embassy-nrf/src/chips/nrf9160.rs
index 67ea032ff..495285ba3 100644
--- a/embassy-nrf/src/chips/nrf9160.rs
+++ b/embassy-nrf/src/chips/nrf9160.rs
@@ -169,7 +169,7 @@ pub const FORCE_COPY_BUFFER_SIZE: usize = 1024;
 
 pub const FLASH_SIZE: usize = 1024 * 1024;
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     // RTC
     RTC0,
     RTC1,
@@ -368,7 +368,7 @@ impl_saadc_input!(P0_18, ANALOG_INPUT5);
 impl_saadc_input!(P0_19, ANALOG_INPUT6);
 impl_saadc_input!(P0_20, ANALOG_INPUT7);
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     SPU,
     CLOCK_POWER,
     UARTE0_SPIM0_SPIS0_TWIM0_TWIS0,
diff --git a/embassy-nrf/src/gpio.rs b/embassy-nrf/src/gpio.rs
index 895ab9340..ea2b76096 100644
--- a/embassy-nrf/src/gpio.rs
+++ b/embassy-nrf/src/gpio.rs
@@ -5,7 +5,7 @@ use core::convert::Infallible;
 use core::hint::unreachable_unchecked;
 
 use cfg_if::cfg_if;
-use embassy_hal_common::{impl_peripheral, into_ref, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, PeripheralRef};
 
 use self::sealed::Pin as _;
 use crate::pac::p0 as gpio;
diff --git a/embassy-nrf/src/gpiote.rs b/embassy-nrf/src/gpiote.rs
index 6550f2abd..7488bc085 100644
--- a/embassy-nrf/src/gpiote.rs
+++ b/embassy-nrf/src/gpiote.rs
@@ -4,7 +4,7 @@ use core::convert::Infallible;
 use core::future::{poll_fn, Future};
 use core::task::{Context, Poll};
 
-use embassy_hal_common::{impl_peripheral, into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::gpio::sealed::Pin as _;
diff --git a/embassy-nrf/src/i2s.rs b/embassy-nrf/src/i2s.rs
index fea38c4c0..907acdf4c 100644
--- a/embassy-nrf/src/i2s.rs
+++ b/embassy-nrf/src/i2s.rs
@@ -9,8 +9,8 @@ use core::ops::{Deref, DerefMut};
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::gpio::{AnyPin, Pin as GpioPin};
 use crate::interrupt::typelevel::Interrupt;
diff --git a/embassy-nrf/src/lib.rs b/embassy-nrf/src/lib.rs
index d23759f9d..355a00497 100644
--- a/embassy-nrf/src/lib.rs
+++ b/embassy-nrf/src/lib.rs
@@ -98,7 +98,7 @@ mod chip;
 /// This defines the right interrupt handlers, and creates a unit struct (like `struct Irqs;`)
 /// and implements the right [`Binding`]s for it. You can pass this struct to drivers to
 /// prove at compile-time that the right interrupts have been bound.
-// developer note: this macro can't be in `embassy-hal-common` due to the use of `$crate`.
+// developer note: this macro can't be in `embassy-hal-internal` due to the use of `$crate`.
 #[macro_export]
 macro_rules! bind_interrupts {
         ($vis:vis struct $name:ident { $($irq:ident => $($handler:ty),*;)* }) => {
@@ -127,7 +127,7 @@ pub use chip::pac;
 #[cfg(not(feature = "unstable-pac"))]
 pub(crate) use chip::pac;
 pub use chip::{peripherals, Peripherals, EASY_DMA_SIZE};
-pub use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+pub use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 
 pub use crate::chip::interrupt;
 pub use crate::pac::NVIC_PRIO_BITS;
diff --git a/embassy-nrf/src/nvmc.rs b/embassy-nrf/src/nvmc.rs
index 91a5a272f..de840b886 100644
--- a/embassy-nrf/src/nvmc.rs
+++ b/embassy-nrf/src/nvmc.rs
@@ -2,7 +2,7 @@
 
 use core::{ptr, slice};
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embedded_storage::nor_flash::{
     ErrorType, MultiwriteNorFlash, NorFlash, NorFlashError, NorFlashErrorKind, ReadNorFlash,
 };
diff --git a/embassy-nrf/src/pdm.rs b/embassy-nrf/src/pdm.rs
index 217884d1c..01f41e9f9 100644
--- a/embassy-nrf/src/pdm.rs
+++ b/embassy-nrf/src/pdm.rs
@@ -6,8 +6,8 @@ use core::marker::PhantomData;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use fixed::types::I7F1;
 use futures::future::poll_fn;
 
diff --git a/embassy-nrf/src/ppi/dppi.rs b/embassy-nrf/src/ppi/dppi.rs
index 40ccb2f09..0bc7f821e 100644
--- a/embassy-nrf/src/ppi/dppi.rs
+++ b/embassy-nrf/src/ppi/dppi.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 
 use super::{Channel, ConfigurableChannel, Event, Ppi, Task};
 use crate::{pac, Peripheral};
diff --git a/embassy-nrf/src/ppi/mod.rs b/embassy-nrf/src/ppi/mod.rs
index ff6593bd5..5b4a64388 100644
--- a/embassy-nrf/src/ppi/mod.rs
+++ b/embassy-nrf/src/ppi/mod.rs
@@ -18,7 +18,7 @@
 use core::marker::PhantomData;
 use core::ptr::NonNull;
 
-use embassy_hal_common::{impl_peripheral, into_ref, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, PeripheralRef};
 
 use crate::{peripherals, Peripheral};
 
diff --git a/embassy-nrf/src/ppi/ppi.rs b/embassy-nrf/src/ppi/ppi.rs
index 1fe898625..3e9e9fc81 100644
--- a/embassy-nrf/src/ppi/ppi.rs
+++ b/embassy-nrf/src/ppi/ppi.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 
 use super::{Channel, ConfigurableChannel, Event, Ppi, StaticChannel, Task};
 use crate::{pac, Peripheral};
diff --git a/embassy-nrf/src/pwm.rs b/embassy-nrf/src/pwm.rs
index c8c81fa01..2f0397632 100644
--- a/embassy-nrf/src/pwm.rs
+++ b/embassy-nrf/src/pwm.rs
@@ -4,7 +4,7 @@
 
 use core::sync::atomic::{compiler_fence, Ordering};
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::gpio::sealed::Pin as _;
 use crate::gpio::{AnyPin, Pin as GpioPin, PselBits};
diff --git a/embassy-nrf/src/qdec.rs b/embassy-nrf/src/qdec.rs
index 8bac87d37..2aa50a2ba 100644
--- a/embassy-nrf/src/qdec.rs
+++ b/embassy-nrf/src/qdec.rs
@@ -6,7 +6,7 @@ use core::future::poll_fn;
 use core::marker::PhantomData;
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::gpio::sealed::Pin as _;
 use crate::gpio::{AnyPin, Pin as GpioPin};
diff --git a/embassy-nrf/src/qspi.rs b/embassy-nrf/src/qspi.rs
index baefc7967..36ee33f6d 100644
--- a/embassy-nrf/src/qspi.rs
+++ b/embassy-nrf/src/qspi.rs
@@ -7,8 +7,8 @@ use core::marker::PhantomData;
 use core::ptr;
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embedded_storage::nor_flash::{ErrorType, NorFlash, NorFlashError, NorFlashErrorKind, ReadNorFlash};
 
 use crate::gpio::{self, Pin as GpioPin};
diff --git a/embassy-nrf/src/rng.rs b/embassy-nrf/src/rng.rs
index 923b8b467..e2803f0d3 100644
--- a/embassy-nrf/src/rng.rs
+++ b/embassy-nrf/src/rng.rs
@@ -8,8 +8,8 @@ use core::ptr;
 use core::sync::atomic::{AtomicPtr, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::interrupt::typelevel::Interrupt;
diff --git a/embassy-nrf/src/saadc.rs b/embassy-nrf/src/saadc.rs
index 23292924c..662b05614 100644
--- a/embassy-nrf/src/saadc.rs
+++ b/embassy-nrf/src/saadc.rs
@@ -6,8 +6,8 @@ use core::future::poll_fn;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{impl_peripheral, into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{impl_peripheral, into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use pac::{saadc, SAADC};
 use saadc::ch::config::{GAIN_A, REFSEL_A, RESP_A, TACQ_A};
diff --git a/embassy-nrf/src/spim.rs b/embassy-nrf/src/spim.rs
index b7dc332e9..4673a0175 100644
--- a/embassy-nrf/src/spim.rs
+++ b/embassy-nrf/src/spim.rs
@@ -8,7 +8,7 @@ use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 pub use embedded_hal_02::spi::{Mode, Phase, Polarity, MODE_0, MODE_1, MODE_2, MODE_3};
 pub use pac::spim0::frequency::FREQUENCY_A as Frequency;
 
diff --git a/embassy-nrf/src/spis.rs b/embassy-nrf/src/spis.rs
index aa438415a..212825121 100644
--- a/embassy-nrf/src/spis.rs
+++ b/embassy-nrf/src/spis.rs
@@ -7,7 +7,7 @@ use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 pub use embedded_hal_02::spi::{Mode, Phase, Polarity, MODE_0, MODE_1, MODE_2, MODE_3};
 
 use crate::chip::FORCE_COPY_BUFFER_SIZE;
diff --git a/embassy-nrf/src/temp.rs b/embassy-nrf/src/temp.rs
index 491e92c04..cec46d8d0 100644
--- a/embassy-nrf/src/temp.rs
+++ b/embassy-nrf/src/temp.rs
@@ -3,8 +3,8 @@
 use core::future::poll_fn;
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use fixed::types::I30F2;
 
diff --git a/embassy-nrf/src/timer.rs b/embassy-nrf/src/timer.rs
index 04748238d..3dbfdac42 100644
--- a/embassy-nrf/src/timer.rs
+++ b/embassy-nrf/src/timer.rs
@@ -6,7 +6,7 @@
 
 #![macro_use]
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::ppi::{Event, Task};
 use crate::{pac, Peripheral};
diff --git a/embassy-nrf/src/twim.rs b/embassy-nrf/src/twim.rs
index 2ad0d19b1..fdea480e3 100644
--- a/embassy-nrf/src/twim.rs
+++ b/embassy-nrf/src/twim.rs
@@ -9,7 +9,7 @@ use core::sync::atomic::Ordering::SeqCst;
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 #[cfg(feature = "time")]
 use embassy_time::{Duration, Instant};
diff --git a/embassy-nrf/src/twis.rs b/embassy-nrf/src/twis.rs
index a115d5616..c6c020557 100644
--- a/embassy-nrf/src/twis.rs
+++ b/embassy-nrf/src/twis.rs
@@ -8,7 +8,7 @@ use core::sync::atomic::compiler_fence;
 use core::sync::atomic::Ordering::SeqCst;
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 #[cfg(feature = "time")]
 use embassy_time::{Duration, Instant};
diff --git a/embassy-nrf/src/uarte.rs b/embassy-nrf/src/uarte.rs
index 48d57fea4..e79df3561 100644
--- a/embassy-nrf/src/uarte.rs
+++ b/embassy-nrf/src/uarte.rs
@@ -18,8 +18,8 @@ use core::marker::PhantomData;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use pac::uarte0::RegisterBlock;
 // Re-export SVD variants to allow user to directly set values.
 pub use pac::uarte0::{baudrate::BAUDRATE_A as Baudrate, config::PARITY_A as Parity};
diff --git a/embassy-nrf/src/usb/mod.rs b/embassy-nrf/src/usb/mod.rs
index 76cf40ac7..e26b49db3 100644
--- a/embassy-nrf/src/usb/mod.rs
+++ b/embassy-nrf/src/usb/mod.rs
@@ -11,7 +11,7 @@ use core::sync::atomic::{compiler_fence, AtomicU32, Ordering};
 use core::task::Poll;
 
 use cortex_m::peripheral::NVIC;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use embassy_usb_driver as driver;
 use embassy_usb_driver::{Direction, EndpointAddress, EndpointError, EndpointInfo, EndpointType, Event, Unsupported};
diff --git a/embassy-rp/Cargo.toml b/embassy-rp/Cargo.toml
index 8f3ed885d..b53c7a01a 100644
--- a/embassy-rp/Cargo.toml
+++ b/embassy-rp/Cargo.toml
@@ -16,7 +16,7 @@ flavors = [
 default = [ "rt" ]
 rt = [ "rp-pac/rt" ]
 
-defmt = ["dep:defmt", "embassy-usb-driver?/defmt", "embassy-hal-common/defmt"]
+defmt = ["dep:defmt", "embassy-usb-driver?/defmt", "embassy-hal-internal/defmt"]
 
 # critical section that is safe for multicore use
 critical-section-impl = ["critical-section/restore-state-u8"]
@@ -58,7 +58,7 @@ unstable-traits = ["embedded-hal-1", "embedded-hal-nb"]
 embassy-sync = { version = "0.2.0", path = "../embassy-sync" }
 embassy-time = { version = "0.1.2", path = "../embassy-time", features = [ "tick-hz-1_000_000" ] }
 embassy-futures = { version = "0.1.0", path = "../embassy-futures" }
-embassy-hal-common = {version = "0.1.0", path = "../embassy-hal-common", features = ["cortex-m", "prio-bits-2"] }
+embassy-hal-internal = {version = "0.1.0", path = "../embassy-hal-internal", features = ["cortex-m", "prio-bits-2"] }
 embassy-embedded-hal = {version = "0.1.0", path = "../embassy-embedded-hal" }
 embassy-usb-driver = {version = "0.1.0", path = "../embassy-usb-driver", optional = true }
 atomic-polyfill = "1.0.1"
diff --git a/embassy-rp/src/adc.rs b/embassy-rp/src/adc.rs
index 95780c068..4fba31169 100644
--- a/embassy-rp/src/adc.rs
+++ b/embassy-rp/src/adc.rs
@@ -3,7 +3,7 @@ use core::marker::PhantomData;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::gpio::sealed::Pin as GpioPin;
diff --git a/embassy-rp/src/clocks.rs b/embassy-rp/src/clocks.rs
index acb21dce5..976d06de7 100644
--- a/embassy-rp/src/clocks.rs
+++ b/embassy-rp/src/clocks.rs
@@ -1,7 +1,7 @@
 use core::marker::PhantomData;
 use core::sync::atomic::{AtomicU16, AtomicU32, Ordering};
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use pac::clocks::vals::*;
 
 use crate::gpio::sealed::Pin;
diff --git a/embassy-rp/src/dma.rs b/embassy-rp/src/dma.rs
index 1a458778c..c8f741804 100644
--- a/embassy-rp/src/dma.rs
+++ b/embassy-rp/src/dma.rs
@@ -4,7 +4,7 @@ use core::pin::Pin;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::{Context, Poll};
 
-use embassy_hal_common::{impl_peripheral, into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use pac::dma::vals::DataSize;
 
diff --git a/embassy-rp/src/flash.rs b/embassy-rp/src/flash.rs
index 96d2d4541..0ed6808eb 100644
--- a/embassy-rp/src/flash.rs
+++ b/embassy-rp/src/flash.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::Peripheral;
+use embassy_hal_internal::Peripheral;
 use embedded_storage::nor_flash::{
     check_erase, check_read, check_write, ErrorType, MultiwriteNorFlash, NorFlash, NorFlashError, NorFlashErrorKind,
     ReadNorFlash,
diff --git a/embassy-rp/src/gpio.rs b/embassy-rp/src/gpio.rs
index a3d330cdc..2807eb678 100644
--- a/embassy-rp/src/gpio.rs
+++ b/embassy-rp/src/gpio.rs
@@ -3,7 +3,7 @@ use core::future::Future;
 use core::pin::Pin as FuturePin;
 use core::task::{Context, Poll};
 
-use embassy_hal_common::{impl_peripheral, into_ref, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::interrupt::InterruptExt;
diff --git a/embassy-rp/src/i2c.rs b/embassy-rp/src/i2c.rs
index 9b85b2345..536ad747d 100644
--- a/embassy-rp/src/i2c.rs
+++ b/embassy-rp/src/i2c.rs
@@ -2,7 +2,7 @@ use core::future;
 use core::marker::PhantomData;
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use pac::i2c;
 
diff --git a/embassy-rp/src/lib.rs b/embassy-rp/src/lib.rs
index 4f205a16e..45156458d 100644
--- a/embassy-rp/src/lib.rs
+++ b/embassy-rp/src/lib.rs
@@ -33,10 +33,10 @@ pub mod watchdog;
 // TODO: move `pio_instr_util` and `relocate` to inside `pio`
 pub mod pio;
 pub mod pio_instr_util;
-pub mod relocate;
+pub(crate) mod relocate;
 
 // Reexports
-pub use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+pub use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 #[cfg(feature = "unstable-pac")]
 pub use rp_pac as pac;
 #[cfg(not(feature = "unstable-pac"))]
@@ -45,7 +45,7 @@ pub(crate) use rp_pac as pac;
 #[cfg(feature = "rt")]
 pub use crate::pac::NVIC_PRIO_BITS;
 
-embassy_hal_common::interrupt_mod!(
+embassy_hal_internal::interrupt_mod!(
     TIMER_IRQ_0,
     TIMER_IRQ_1,
     TIMER_IRQ_2,
@@ -85,7 +85,7 @@ embassy_hal_common::interrupt_mod!(
 /// This defines the right interrupt handlers, and creates a unit struct (like `struct Irqs;`)
 /// and implements the right [`Binding`]s for it. You can pass this struct to drivers to
 /// prove at compile-time that the right interrupts have been bound.
-// developer note: this macro can't be in `embassy-hal-common` due to the use of `$crate`.
+// developer note: this macro can't be in `embassy-hal-internal` due to the use of `$crate`.
 #[macro_export]
 macro_rules! bind_interrupts {
     ($vis:vis struct $name:ident { $($irq:ident => $($handler:ty),*;)* }) => {
@@ -107,7 +107,7 @@ macro_rules! bind_interrupts {
     };
 }
 
-embassy_hal_common::peripherals! {
+embassy_hal_internal::peripherals! {
     PIN_0,
     PIN_1,
     PIN_2,
@@ -219,6 +219,74 @@ select_bootloader! {
     default => BOOT_LOADER_W25Q080
 }
 
+/// Installs a stack guard for the CORE0 stack in MPU region 0.
+/// Will fail if the MPU is already confgigured. This function requires
+/// a `_stack_end` symbol to be defined by the linker script, and expexcts
+/// `_stack_end` to be located at the lowest address (largest depth) of
+/// the stack.
+///
+/// This method can *only* set up stack guards on the currently
+/// executing core. Stack guards for CORE1 are set up automatically,
+/// only CORE0 should ever use this.
+///
+/// # Usage
+///
+/// ```no_run
+/// #![feature(type_alias_impl_trait)]
+/// use embassy_rp::install_core0_stack_guard;
+/// use embassy_executor::{Executor, Spawner};
+///
+/// #[embassy_executor::main]
+/// async fn main(_spawner: Spawner) {
+///     // set up by the linker as follows:
+///     //
+///     //     MEMORY {
+///     //       STACK0: ORIGIN = 0x20040000, LENGTH = 4K
+///     //     }
+///     //
+///     //     _stack_end = ORIGIN(STACK0);
+///     //     _stack_start = _stack_end + LENGTH(STACK0);
+///     //
+///     install_core0_stack_guard().expect("MPU already configured");
+///     let p = embassy_rp::init(Default::default());
+///
+///     // ...
+/// }
+/// ```
+pub fn install_core0_stack_guard() -> Result<(), ()> {
+    extern "C" {
+        static mut _stack_end: usize;
+    }
+    unsafe { install_stack_guard(&mut _stack_end as *mut usize) }
+}
+
+#[inline(always)]
+fn install_stack_guard(stack_bottom: *mut usize) -> Result<(), ()> {
+    let core = unsafe { cortex_m::Peripherals::steal() };
+
+    // Fail if MPU is already configured
+    if core.MPU.ctrl.read() != 0 {
+        return Err(());
+    }
+
+    // The minimum we can protect is 32 bytes on a 32 byte boundary, so round up which will
+    // just shorten the valid stack range a tad.
+    let addr = (stack_bottom as u32 + 31) & !31;
+    // Mask is 1 bit per 32 bytes of the 256 byte range... clear the bit for the segment we want
+    let subregion_select = 0xff ^ (1 << ((addr >> 5) & 7));
+    unsafe {
+        core.MPU.ctrl.write(5); // enable mpu with background default map
+        core.MPU.rbar.write((addr & !0xff) | (1 << 4)); // set address and update RNR
+        core.MPU.rasr.write(
+            1 // enable region
+               | (0x7 << 1) // size 2^(7 + 1) = 256
+               | (subregion_select << 8)
+               | 0x10000000, // XN = disable instruction fetch; no other bits means no permissions
+        );
+    }
+    Ok(())
+}
+
 pub mod config {
     use crate::clocks::ClockConfig;
 
diff --git a/embassy-rp/src/multicore.rs b/embassy-rp/src/multicore.rs
index 468e8470a..915761801 100644
--- a/embassy-rp/src/multicore.rs
+++ b/embassy-rp/src/multicore.rs
@@ -52,41 +52,20 @@ use core::sync::atomic::{compiler_fence, AtomicBool, Ordering};
 
 use crate::interrupt::InterruptExt;
 use crate::peripherals::CORE1;
-use crate::{gpio, interrupt, pac};
+use crate::{gpio, install_stack_guard, interrupt, pac};
 
 const PAUSE_TOKEN: u32 = 0xDEADBEEF;
 const RESUME_TOKEN: u32 = !0xDEADBEEF;
 static IS_CORE1_INIT: AtomicBool = AtomicBool::new(false);
 
 #[inline(always)]
-fn install_stack_guard(stack_bottom: *mut usize) {
-    let core = unsafe { cortex_m::Peripherals::steal() };
-
-    // Trap if MPU is already configured
-    if core.MPU.ctrl.read() != 0 {
+fn core1_setup(stack_bottom: *mut usize) {
+    if let Err(_) = install_stack_guard(stack_bottom) {
+        // currently only happens if the MPU was already set up, which
+        // would indicate that the core is already in use from outside
+        // embassy, somehow. trap if so since we can't deal with that.
         cortex_m::asm::udf();
     }
-
-    // The minimum we can protect is 32 bytes on a 32 byte boundary, so round up which will
-    // just shorten the valid stack range a tad.
-    let addr = (stack_bottom as u32 + 31) & !31;
-    // Mask is 1 bit per 32 bytes of the 256 byte range... clear the bit for the segment we want
-    let subregion_select = 0xff ^ (1 << ((addr >> 5) & 7));
-    unsafe {
-        core.MPU.ctrl.write(5); // enable mpu with background default map
-        core.MPU.rbar.write((addr & !0xff) | 0x8);
-        core.MPU.rasr.write(
-            1 // enable region
-               | (0x7 << 1) // size 2^(7 + 1) = 256
-               | (subregion_select << 8)
-               | 0x10000000, // XN = disable instruction fetch; no other bits means no permissions
-        );
-    }
-}
-
-#[inline(always)]
-fn core1_setup(stack_bottom: *mut usize) {
-    install_stack_guard(stack_bottom);
     unsafe {
         gpio::init();
     }
diff --git a/embassy-rp/src/pio.rs b/embassy-rp/src/pio.rs
index 72a2f44ed..3de398af7 100644
--- a/embassy-rp/src/pio.rs
+++ b/embassy-rp/src/pio.rs
@@ -5,13 +5,13 @@ use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::{Context, Poll};
 
 use atomic_polyfill::{AtomicU32, AtomicU8};
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use fixed::types::extra::U8;
 use fixed::FixedU32;
 use pac::io::vals::Gpio0ctrlFuncsel;
 use pac::pio::vals::SmExecctrlStatusSel;
-use pio::{SideSet, Wrap};
+use pio::{Program, SideSet, Wrap};
 
 use crate::dma::{Channel, Transfer, Word};
 use crate::gpio::sealed::Pin as SealedPin;
@@ -734,23 +734,67 @@ pub struct InstanceMemory<'d, PIO: Instance> {
 
 pub struct LoadedProgram<'d, PIO: Instance> {
     pub used_memory: InstanceMemory<'d, PIO>,
-    origin: u8,
-    wrap: Wrap,
-    side_set: SideSet,
+    pub origin: u8,
+    pub wrap: Wrap,
+    pub side_set: SideSet,
+}
+
+#[derive(Clone, Copy, PartialEq, Eq, Debug)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum LoadError {
+    /// Insufficient consecutive free instruction space to load program.
+    InsufficientSpace,
+    /// Loading the program would overwrite an instruction address already
+    /// used by another program.
+    AddressInUse(usize),
 }
 
 impl<'d, PIO: Instance> Common<'d, PIO> {
-    pub fn load_program<const SIZE: usize>(&mut self, prog: &RelocatedProgram<SIZE>) -> LoadedProgram<'d, PIO> {
+    /// Load a PIO program. This will automatically relocate the program to
+    /// an available chunk of free instruction memory if the program origin
+    /// was not explicitly specified, otherwise it will attempt to load the
+    /// program only at its origin.
+    pub fn load_program<const SIZE: usize>(&mut self, prog: &Program<SIZE>) -> LoadedProgram<'d, PIO> {
         match self.try_load_program(prog) {
             Ok(r) => r,
-            Err(at) => panic!("Trying to write already used PIO instruction memory at {}", at),
+            Err(e) => panic!("Failed to load PIO program: {:?}", e),
         }
     }
 
+    /// Load a PIO program. This will automatically relocate the program to
+    /// an available chunk of free instruction memory if the program origin
+    /// was not explicitly specified, otherwise it will attempt to load the
+    /// program only at its origin.
     pub fn try_load_program<const SIZE: usize>(
         &mut self,
-        prog: &RelocatedProgram<SIZE>,
+        prog: &Program<SIZE>,
+    ) -> Result<LoadedProgram<'d, PIO>, LoadError> {
+        match prog.origin {
+            Some(origin) => self
+                .try_load_program_at(prog, origin)
+                .map_err(|a| LoadError::AddressInUse(a)),
+            None => {
+                // naively search for free space, allowing wraparound since
+                // PIO does support that. with only 32 instruction slots it
+                // doesn't make much sense to do anything more fancy.
+                let mut origin = 0;
+                while origin < 32 {
+                    match self.try_load_program_at(prog, origin as _) {
+                        Ok(r) => return Ok(r),
+                        Err(a) => origin = a + 1,
+                    }
+                }
+                Err(LoadError::InsufficientSpace)
+            }
+        }
+    }
+
+    fn try_load_program_at<const SIZE: usize>(
+        &mut self,
+        prog: &Program<SIZE>,
+        origin: u8,
     ) -> Result<LoadedProgram<'d, PIO>, usize> {
+        let prog = RelocatedProgram::new_with_origin(prog, origin);
         let used_memory = self.try_write_instr(prog.origin() as _, prog.code())?;
         Ok(LoadedProgram {
             used_memory,
@@ -760,7 +804,7 @@ impl<'d, PIO: Instance> Common<'d, PIO> {
         })
     }
 
-    pub fn try_write_instr<I>(&mut self, start: usize, instrs: I) -> Result<InstanceMemory<'d, PIO>, usize>
+    fn try_write_instr<I>(&mut self, start: usize, instrs: I) -> Result<InstanceMemory<'d, PIO>, usize>
     where
         I: Iterator<Item = u16>,
     {
diff --git a/embassy-rp/src/pwm.rs b/embassy-rp/src/pwm.rs
index 20bb88446..c0ddb2a90 100644
--- a/embassy-rp/src/pwm.rs
+++ b/embassy-rp/src/pwm.rs
@@ -1,6 +1,6 @@
 //! Pulse Width Modulation (PWM)
 
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use fixed::traits::ToFixed;
 use fixed::FixedU16;
 use pac::pwm::regs::{ChDiv, Intr};
diff --git a/embassy-rp/src/relocate.rs b/embassy-rp/src/relocate.rs
index 9cb279ccd..b35b4ed72 100644
--- a/embassy-rp/src/relocate.rs
+++ b/embassy-rp/src/relocate.rs
@@ -41,11 +41,6 @@ pub struct RelocatedProgram<'a, const PROGRAM_SIZE: usize> {
 }
 
 impl<'a, const PROGRAM_SIZE: usize> RelocatedProgram<'a, PROGRAM_SIZE> {
-    pub fn new(program: &Program<PROGRAM_SIZE>) -> RelocatedProgram<PROGRAM_SIZE> {
-        let origin = program.origin.unwrap_or(0);
-        RelocatedProgram { program, origin }
-    }
-
     pub fn new_with_origin(program: &Program<PROGRAM_SIZE>, origin: u8) -> RelocatedProgram<PROGRAM_SIZE> {
         RelocatedProgram { program, origin }
     }
diff --git a/embassy-rp/src/rtc/mod.rs b/embassy-rp/src/rtc/mod.rs
index 1b33fdf8d..60ca8627b 100644
--- a/embassy-rp/src/rtc/mod.rs
+++ b/embassy-rp/src/rtc/mod.rs
@@ -1,6 +1,6 @@
 mod filter;
 
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 
 pub use self::filter::DateTimeFilter;
 
diff --git a/embassy-rp/src/spi.rs b/embassy-rp/src/spi.rs
index af101cf4a..544b542e8 100644
--- a/embassy-rp/src/spi.rs
+++ b/embassy-rp/src/spi.rs
@@ -3,7 +3,7 @@ use core::marker::PhantomData;
 
 use embassy_embedded_hal::SetConfig;
 use embassy_futures::join::join;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 pub use embedded_hal_02::spi::{Phase, Polarity};
 
 use crate::dma::{AnyChannel, Channel};
diff --git a/embassy-rp/src/uart/buffered.rs b/embassy-rp/src/uart/buffered.rs
index 30eeb5476..9d96db12c 100644
--- a/embassy-rp/src/uart/buffered.rs
+++ b/embassy-rp/src/uart/buffered.rs
@@ -3,7 +3,7 @@ use core::slice;
 use core::task::Poll;
 
 use atomic_polyfill::{AtomicU8, Ordering};
-use embassy_hal_common::atomic_ring_buffer::RingBuffer;
+use embassy_hal_internal::atomic_ring_buffer::RingBuffer;
 use embassy_sync::waitqueue::AtomicWaker;
 use embassy_time::{Duration, Timer};
 
diff --git a/embassy-rp/src/uart/mod.rs b/embassy-rp/src/uart/mod.rs
index 7b94bce5e..69c6ac2f1 100644
--- a/embassy-rp/src/uart/mod.rs
+++ b/embassy-rp/src/uart/mod.rs
@@ -4,7 +4,7 @@ use core::task::Poll;
 
 use atomic_polyfill::{AtomicU16, Ordering};
 use embassy_futures::select::{select, Either};
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use embassy_time::{Duration, Timer};
 use pac::uart::regs::Uartris;
diff --git a/embassy-stm32-wpan/Cargo.toml b/embassy-stm32-wpan/Cargo.toml
index 6cd122200..96c474845 100644
--- a/embassy-stm32-wpan/Cargo.toml
+++ b/embassy-stm32-wpan/Cargo.toml
@@ -15,7 +15,7 @@ embassy-stm32 = { version = "0.1.0", path = "../embassy-stm32" }
 embassy-sync = { version = "0.2.0", path = "../embassy-sync" }
 embassy-time = { version = "0.1.2", path = "../embassy-time", optional = true }
 embassy-futures = { version = "0.1.0", path = "../embassy-futures" }
-embassy-hal-common = { version = "0.1.0", path = "../embassy-hal-common" }
+embassy-hal-internal = { version = "0.1.0", path = "../embassy-hal-internal" }
 embassy-embedded-hal = { version = "0.1.0", path = "../embassy-embedded-hal" }
 embassy-net-driver = { version = "0.1.0", path = "../embassy-net-driver", optional=true }
 
@@ -26,12 +26,12 @@ aligned = "0.4.1"
 
 bit_field = "0.10.2"
 stm32-device-signature = { version = "0.3.3", features = ["stm32wb5x"] }
-stm32wb-hci = { version = "0.1.3", optional = true }
+stm32wb-hci = { version = "0.1.4", optional = true }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 bitflags = { version = "2.3.3", optional = true }
 
 [features]
-defmt = ["dep:defmt", "embassy-sync/defmt", "embassy-embedded-hal/defmt", "embassy-hal-common/defmt", "stm32wb-hci?/defmt"]
+defmt = ["dep:defmt", "embassy-sync/defmt", "embassy-embedded-hal/defmt", "embassy-hal-internal/defmt", "stm32wb-hci?/defmt"]
 
 ble = ["dep:stm32wb-hci"]
 mac = ["dep:bitflags", "dep:embassy-net-driver" ]
diff --git a/embassy-stm32-wpan/src/lib.rs b/embassy-stm32-wpan/src/lib.rs
index 6836d7a8b..5ecce2cc2 100644
--- a/embassy-stm32-wpan/src/lib.rs
+++ b/embassy-stm32-wpan/src/lib.rs
@@ -8,7 +8,7 @@ pub mod fmt;
 use core::mem::MaybeUninit;
 use core::sync::atomic::{compiler_fence, Ordering};
 
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use embassy_stm32::interrupt;
 use embassy_stm32::ipcc::{Config, Ipcc, ReceiveInterruptHandler, TransmitInterruptHandler};
 use embassy_stm32::peripherals::IPCC;
diff --git a/embassy-stm32/Cargo.toml b/embassy-stm32/Cargo.toml
index 0fb6fdb56..ba279f795 100644
--- a/embassy-stm32/Cargo.toml
+++ b/embassy-stm32/Cargo.toml
@@ -34,7 +34,7 @@ flavors = [
 embassy-sync = { version = "0.2.0", path = "../embassy-sync" }
 embassy-time = { version = "0.1.2", path = "../embassy-time", optional = true }
 embassy-futures = { version = "0.1.0", path = "../embassy-futures" }
-embassy-hal-common = {version = "0.1.0", path = "../embassy-hal-common", features = ["cortex-m", "prio-bits-4"] }
+embassy-hal-internal = {version = "0.1.0", path = "../embassy-hal-internal", features = ["cortex-m", "prio-bits-4"] }
 embassy-embedded-hal = {version = "0.1.0", path = "../embassy-embedded-hal" }
 embassy-net-driver = { version = "0.1.0", path = "../embassy-net-driver" }
 embassy-usb-driver = {version = "0.1.0", path = "../embassy-usb-driver", optional = true }
@@ -67,6 +67,7 @@ cfg-if = "1.0.0"
 embedded-io = { version = "0.4.0", features = ["async"], optional = true }
 chrono = { version = "^0.4", default-features = false, optional = true}
 bit_field = "0.10.2"
+document-features = "0.2.7"
 
 [dev-dependencies]
 critical-section = { version = "1.1", features = ["std"] }
@@ -78,40 +79,63 @@ stm32-metapac = { version = "13", default-features = false, features = ["metadat
 
 [features]
 default = ["rt"]
+
+## Enable `stm32-metapac`'s `rt` feature
 rt = ["stm32-metapac/rt"]
 
-defmt = ["dep:defmt", "bxcan/unstable-defmt", "embassy-sync/defmt", "embassy-embedded-hal/defmt", "embassy-hal-common/defmt", "embedded-io?/defmt", "embassy-usb-driver?/defmt", "embassy-net-driver/defmt"]
-memory-x = ["stm32-metapac/memory-x"]
+## Use [`defmt`](https://docs.rs/defmt/latest/defmt/) for logging
+defmt = ["dep:defmt", "bxcan/unstable-defmt", "embassy-sync/defmt", "embassy-embedded-hal/defmt", "embassy-hal-internal/defmt", "embedded-io?/defmt", "embassy-usb-driver?/defmt", "embassy-net-driver/defmt"]
+
 exti = []
 
-# Enables additional driver features that depend on embassy-time
+## Automatically generate `memory.x` file using [`stm32-metapac`](https://docs.rs/stm32-metapac/)
+memory-x = ["stm32-metapac/memory-x"]
+
+## Enable nightly-only features
+nightly = ["embedded-hal-1", "embedded-hal-async", "embedded-storage-async", "dep:embedded-io", "dep:embassy-usb-driver", "embassy-embedded-hal/nightly"]
+
+## Re-export stm32-metapac at `embassy_stm32::pac`.
+## This is unstable because semver-minor (non-breaking) releases of embassy-stm32 may major-bump (breaking) the stm32-metapac version.
+## If this is an issue for you, you're encouraged to directly depend on a fixed version of the PAC.
+## There are no plans to make this stable.
+unstable-pac = []
+
+## Implement embedded-hal 1.0 alpha traits.
+## Implement embedded-hal-async traits if `nightly` is set as well.
+unstable-traits = ["embedded-hal-1", "dep:embedded-hal-nb"]
+
+#! ## Time
+
+## Enables additional driver features that depend on embassy-time
 time = ["dep:embassy-time"]
 
 # Features starting with `_` are for internal use only. They're not intended
 # to be enabled by other crates, and are not covered by semver guarantees.
 _time-driver = ["time"]
+
+## Use any time driver
 time-driver-any = ["_time-driver"]
+## Use TIM2 as time driver
 time-driver-tim2 = ["_time-driver"]
+## Use TIM3 as time driver
 time-driver-tim3 = ["_time-driver"]
+## Use TIM4 as time driver
 time-driver-tim4 = ["_time-driver"]
+## Use TIM5 as time driver
 time-driver-tim5 = ["_time-driver"]
+## Use TIM12 as time driver
 time-driver-tim12 = ["_time-driver"]
+## Use TIM15 as time driver
 time-driver-tim15 = ["_time-driver"]
 
-# Enable nightly-only features
-nightly = ["embedded-hal-1", "embedded-hal-async", "embedded-storage-async", "dep:embedded-io", "dep:embassy-usb-driver", "embassy-embedded-hal/nightly"]
-
-# Reexport stm32-metapac at `embassy_stm32::pac`.
-# This is unstable because semver-minor (non-breaking) releases of embassy-stm32 may major-bump (breaking) the stm32-metapac version.
-# If this is an issue for you, you're encouraged to directly depend on a fixed version of the PAC.
-# There are no plans to make this stable.
-unstable-pac = []
 
-# Implement embedded-hal 1.0 alpha traits.
-# Implement embedded-hal-async traits if `nightly` is set as well.
-unstable-traits = ["embedded-hal-1", "dep:embedded-hal-nb"]
+#! ## Chip-selection features
+#! Select your chip by specifying the model as a feature, e.g. `stm32c011d6`.
+#! Check the `Cargo.toml` for the latest list of supported chips.
+#!
+#! **Important:** Do not forget to adapt the target chip in your toolchain,
+#! e.g. in `.cargo/config.toml`.
 
-# Chip-selection features
 stm32c011d6 = [ "stm32-metapac/stm32c011d6" ]
 stm32c011f4 = [ "stm32-metapac/stm32c011f4" ]
 stm32c011f6 = [ "stm32-metapac/stm32c011f6" ]
@@ -1445,4 +1469,4 @@ stm32wle5cb = [ "stm32-metapac/stm32wle5cb" ]
 stm32wle5cc = [ "stm32-metapac/stm32wle5cc" ]
 stm32wle5j8 = [ "stm32-metapac/stm32wle5j8" ]
 stm32wle5jb = [ "stm32-metapac/stm32wle5jb" ]
-stm32wle5jc = [ "stm32-metapac/stm32wle5jc" ]
\ No newline at end of file
+stm32wle5jc = [ "stm32-metapac/stm32wle5jc" ]
diff --git a/embassy-stm32/build.rs b/embassy-stm32/build.rs
index c2b84797d..9b3caefd5 100644
--- a/embassy-stm32/build.rs
+++ b/embassy-stm32/build.rs
@@ -138,7 +138,7 @@ fn main() {
     let singleton_tokens: Vec<_> = singletons.iter().map(|s| format_ident!("{}", s)).collect();
 
     g.extend(quote! {
-        embassy_hal_common::peripherals_definition!(#(#singleton_tokens),*);
+        embassy_hal_internal::peripherals_definition!(#(#singleton_tokens),*);
     });
 
     let singleton_tokens: Vec<_> = singletons
@@ -148,7 +148,7 @@ fn main() {
         .collect();
 
     g.extend(quote! {
-        embassy_hal_common::peripherals_struct!(#(#singleton_tokens),*);
+        embassy_hal_internal::peripherals_struct!(#(#singleton_tokens),*);
     });
 
     // ========
@@ -160,7 +160,7 @@ fn main() {
     }
 
     g.extend(quote! {
-        embassy_hal_common::interrupt_mod!(
+        embassy_hal_internal::interrupt_mod!(
             #(
                 #irqs,
             )*
@@ -211,7 +211,7 @@ fn main() {
         let region_type = format_ident!("{}", get_flash_region_type_name(region.name));
         flash_regions.extend(quote! {
             #[cfg(flash)]
-            pub struct #region_type<'d, MODE = crate::flash::Async>(pub &'static crate::flash::FlashRegion, pub(crate) embassy_hal_common::PeripheralRef<'d, crate::peripherals::FLASH>, pub(crate) core::marker::PhantomData<MODE>);
+            pub struct #region_type<'d, MODE = crate::flash::Async>(pub &'static crate::flash::FlashRegion, pub(crate) embassy_hal_internal::PeripheralRef<'d, crate::peripherals::FLASH>, pub(crate) core::marker::PhantomData<MODE>);
         });
     }
 
@@ -243,7 +243,7 @@ fn main() {
 
         #[cfg(flash)]
         impl<'d, MODE> FlashLayout<'d, MODE> {
-            pub(crate) fn new(p: embassy_hal_common::PeripheralRef<'d, crate::peripherals::FLASH>) -> Self {
+            pub(crate) fn new(p: embassy_hal_internal::PeripheralRef<'d, crate::peripherals::FLASH>) -> Self {
                 Self {
                     #(#inits),*,
                     _mode: core::marker::PhantomData,
@@ -572,21 +572,21 @@ fn main() {
         (("fmc", "Clk"), quote!(crate::fmc::ClkPin)),
         (("fmc", "BA0"), quote!(crate::fmc::BA0Pin)),
         (("fmc", "BA1"), quote!(crate::fmc::BA1Pin)),
-        (("timer", "CH1"), quote!(crate::pwm::Channel1Pin)),
-        (("timer", "CH1N"), quote!(crate::pwm::Channel1ComplementaryPin)),
-        (("timer", "CH2"), quote!(crate::pwm::Channel2Pin)),
-        (("timer", "CH2N"), quote!(crate::pwm::Channel2ComplementaryPin)),
-        (("timer", "CH3"), quote!(crate::pwm::Channel3Pin)),
-        (("timer", "CH3N"), quote!(crate::pwm::Channel3ComplementaryPin)),
-        (("timer", "CH4"), quote!(crate::pwm::Channel4Pin)),
-        (("timer", "CH4N"), quote!(crate::pwm::Channel4ComplementaryPin)),
-        (("timer", "ETR"), quote!(crate::pwm::ExternalTriggerPin)),
-        (("timer", "BKIN"), quote!(crate::pwm::BreakInputPin)),
-        (("timer", "BKIN_COMP1"), quote!(crate::pwm::BreakInputComparator1Pin)),
-        (("timer", "BKIN_COMP2"), quote!(crate::pwm::BreakInputComparator2Pin)),
-        (("timer", "BKIN2"), quote!(crate::pwm::BreakInput2Pin)),
-        (("timer", "BKIN2_COMP1"), quote!(crate::pwm::BreakInput2Comparator1Pin)),
-        (("timer", "BKIN2_COMP2"), quote!(crate::pwm::BreakInput2Comparator2Pin)),
+        (("timer", "CH1"), quote!(crate::timer::Channel1Pin)),
+        (("timer", "CH1N"), quote!(crate::timer::Channel1ComplementaryPin)),
+        (("timer", "CH2"), quote!(crate::timer::Channel2Pin)),
+        (("timer", "CH2N"), quote!(crate::timer::Channel2ComplementaryPin)),
+        (("timer", "CH3"), quote!(crate::timer::Channel3Pin)),
+        (("timer", "CH3N"), quote!(crate::timer::Channel3ComplementaryPin)),
+        (("timer", "CH4"), quote!(crate::timer::Channel4Pin)),
+        (("timer", "CH4N"), quote!(crate::timer::Channel4ComplementaryPin)),
+        (("timer", "ETR"), quote!(crate::timer::ExternalTriggerPin)),
+        (("timer", "BKIN"), quote!(crate::timer::BreakInputPin)),
+        (("timer", "BKIN_COMP1"), quote!(crate::timer::BreakInputComparator1Pin)),
+        (("timer", "BKIN_COMP2"), quote!(crate::timer::BreakInputComparator2Pin)),
+        (("timer", "BKIN2"), quote!(crate::timer::BreakInput2Pin)),
+        (("timer", "BKIN2_COMP1"), quote!(crate::timer::BreakInput2Comparator1Pin)),
+        (("timer", "BKIN2_COMP2"), quote!(crate::timer::BreakInput2Comparator2Pin)),
         (("hrtim", "CHA1"), quote!(crate::hrtim::ChannelAPin)),
         (("hrtim", "CHA2"), quote!(crate::hrtim::ChannelAComplementaryPin)),
         (("hrtim", "CHB1"), quote!(crate::hrtim::ChannelBPin)),
diff --git a/embassy-stm32/src/adc/f1.rs b/embassy-stm32/src/adc/f1.rs
index 2322204d5..e577ec289 100644
--- a/embassy-stm32/src/adc/f1.rs
+++ b/embassy-stm32/src/adc/f1.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use embedded_hal_02::blocking::delay::DelayUs;
 
 use crate::adc::{Adc, AdcPin, Instance, SampleTime};
diff --git a/embassy-stm32/src/adc/v1.rs b/embassy-stm32/src/adc/v1.rs
index d9af0c55e..e8245884e 100644
--- a/embassy-stm32/src/adc/v1.rs
+++ b/embassy-stm32/src/adc/v1.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use embedded_hal_02::blocking::delay::DelayUs;
 
 use crate::adc::{Adc, AdcPin, Instance, InternalChannel, Resolution, SampleTime};
diff --git a/embassy-stm32/src/adc/v2.rs b/embassy-stm32/src/adc/v2.rs
index 091c1d447..9a7acea53 100644
--- a/embassy-stm32/src/adc/v2.rs
+++ b/embassy-stm32/src/adc/v2.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use embedded_hal_02::blocking::delay::DelayUs;
 
 use super::InternalChannel;
diff --git a/embassy-stm32/src/adc/v3.rs b/embassy-stm32/src/adc/v3.rs
index 3a6e58cf6..821cc7f6a 100644
--- a/embassy-stm32/src/adc/v3.rs
+++ b/embassy-stm32/src/adc/v3.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use embedded_hal_02::blocking::delay::DelayUs;
 
 use crate::adc::{Adc, AdcPin, Instance, Resolution, SampleTime};
diff --git a/embassy-stm32/src/adc/v4.rs b/embassy-stm32/src/adc/v4.rs
index c51c6840f..64d0f0c75 100644
--- a/embassy-stm32/src/adc/v4.rs
+++ b/embassy-stm32/src/adc/v4.rs
@@ -226,7 +226,7 @@ impl Prescaler {
 
 impl<'d, T: Instance> Adc<'d, T> {
     pub fn new(adc: impl Peripheral<P = T> + 'd, delay: &mut impl DelayUs<u16>) -> Self {
-        embassy_hal_common::into_ref!(adc);
+        embassy_hal_internal::into_ref!(adc);
         T::enable();
         T::reset();
 
diff --git a/embassy-stm32/src/can/bxcan.rs b/embassy-stm32/src/can/bxcan.rs
index 5a0153464..1d0fdd532 100644
--- a/embassy-stm32/src/can/bxcan.rs
+++ b/embassy-stm32/src/can/bxcan.rs
@@ -6,7 +6,7 @@ use core::task::Poll;
 
 pub use bxcan;
 use bxcan::{Data, ExtendedId, Frame, Id, StandardId};
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use futures::FutureExt;
 
 use crate::gpio::sealed::AFType;
@@ -77,6 +77,7 @@ pub struct Can<'d, T: Instance> {
 }
 
 #[derive(Debug)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
 pub enum BusError {
     Stuff,
     Form,
@@ -90,6 +91,22 @@ pub enum BusError {
     BusWarning,
 }
 
+#[derive(Debug)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum TryReadError {
+    /// Bus error
+    BusError(BusError),
+    /// Receive buffer is empty
+    Empty,
+}
+
+#[derive(Debug)]
+#[cfg_attr(feature = "defmt", derive(defmt::Format))]
+pub enum TryWriteError {
+    /// All transmit mailboxes are full
+    Full,
+}
+
 impl<'d, T: Instance> Can<'d, T> {
     /// Creates a new Bxcan instance, keeping the peripheral in sleep mode.
     /// You must call [Can::enable_non_blocking] to use the peripheral.
@@ -161,58 +178,60 @@ impl<'d, T: Instance> Can<'d, T> {
             .leave_disabled();
     }
 
+    /// Enables the peripheral and synchronizes with the bus.
+    ///
+    /// This will wait for 11 consecutive recessive bits (bus idle state).
+    /// Contrary to enable method from bxcan library, this will not freeze the executor while waiting.
+    pub async fn enable(&mut self) {
+        while self.borrow_mut().enable_non_blocking().is_err() {
+            // SCE interrupt is only generated for entering sleep mode, but not leaving.
+            // Yield to allow other tasks to execute while can bus is initializing.
+            embassy_futures::yield_now().await;
+        }
+    }
+
     /// Queues the message to be sent but exerts backpressure
     pub async fn write(&mut self, frame: &Frame) -> bxcan::TransmitStatus {
-        poll_fn(|cx| {
-            T::state().tx_waker.register(cx.waker());
-            if let Ok(status) = self.can.borrow_mut().transmit(frame) {
-                return Poll::Ready(status);
-            }
+        CanTx { can: &self.can }.write(frame).await
+    }
 
-            Poll::Pending
-        })
-        .await
+    /// Attempts to transmit a frame without blocking.
+    ///
+    /// Returns [Err(TryWriteError::Full)] if all transmit mailboxes are full.
+    pub fn try_write(&mut self, frame: &Frame) -> Result<bxcan::TransmitStatus, TryWriteError> {
+        CanTx { can: &self.can }.try_write(frame)
     }
 
+    /// Waits for a specific transmit mailbox to become empty
     pub async fn flush(&self, mb: bxcan::Mailbox) {
-        poll_fn(|cx| {
-            T::state().tx_waker.register(cx.waker());
-            if T::regs().tsr().read().tme(mb.index()) {
-                return Poll::Ready(());
-            }
+        CanTx { can: &self.can }.flush(mb).await
+    }
 
-            Poll::Pending
-        })
-        .await;
+    /// Waits until any of the transmit mailboxes become empty
+    pub async fn flush_any(&self) {
+        CanTx { can: &self.can }.flush_any().await
+    }
+
+    /// Waits until all of the transmit mailboxes become empty
+    pub async fn flush_all(&self) {
+        CanTx { can: &self.can }.flush_all().await
     }
 
     /// Returns a tuple of the time the message was received and the message frame
     pub async fn read(&mut self) -> Result<(u16, bxcan::Frame), BusError> {
-        poll_fn(|cx| {
-            T::state().err_waker.register(cx.waker());
-            if let Poll::Ready((time, frame)) = T::state().rx_queue.recv().poll_unpin(cx) {
-                return Poll::Ready(Ok((time, frame)));
-            } else if let Some(err) = self.curr_error() {
-                return Poll::Ready(Err(err));
-            }
+        CanRx { can: &self.can }.read().await
+    }
 
-            Poll::Pending
-        })
-        .await
+    /// Attempts to read a can frame without blocking.
+    ///
+    /// Returns [Err(TryReadError::Empty)] if there are no frames in the rx queue.
+    pub fn try_read(&mut self) -> Result<(u16, bxcan::Frame), TryReadError> {
+        CanRx { can: &self.can }.try_read()
     }
 
-    fn curr_error(&self) -> Option<BusError> {
-        let err = { T::regs().esr().read() };
-        if err.boff() {
-            return Some(BusError::BusOff);
-        } else if err.epvf() {
-            return Some(BusError::BusPassive);
-        } else if err.ewgf() {
-            return Some(BusError::BusWarning);
-        } else if let Some(err) = err.lec().into_bus_err() {
-            return Some(err);
-        }
-        None
+    /// Waits while receive queue is empty.
+    pub async fn wait_not_empty(&mut self) {
+        CanRx { can: &self.can }.wait_not_empty().await
     }
 
     unsafe fn receive_fifo(fifo: RxFifo) {
@@ -374,6 +393,14 @@ impl<'c, 'd, T: Instance> CanTx<'c, 'd, T> {
         .await
     }
 
+    /// Attempts to transmit a frame without blocking.
+    ///
+    /// Returns [Err(TryWriteError::Full)] if all transmit mailboxes are full.
+    pub fn try_write(&mut self, frame: &Frame) -> Result<bxcan::TransmitStatus, TryWriteError> {
+        self.can.borrow_mut().transmit(frame).map_err(|_| TryWriteError::Full)
+    }
+
+    /// Waits for a specific transmit mailbox to become empty
     pub async fn flush(&self, mb: bxcan::Mailbox) {
         poll_fn(|cx| {
             T::state().tx_waker.register(cx.waker());
@@ -385,6 +412,42 @@ impl<'c, 'd, T: Instance> CanTx<'c, 'd, T> {
         })
         .await;
     }
+
+    /// Waits until any of the transmit mailboxes become empty
+    pub async fn flush_any(&self) {
+        poll_fn(|cx| {
+            T::state().tx_waker.register(cx.waker());
+
+            let tsr = T::regs().tsr().read();
+            if tsr.tme(bxcan::Mailbox::Mailbox0.index())
+                || tsr.tme(bxcan::Mailbox::Mailbox1.index())
+                || tsr.tme(bxcan::Mailbox::Mailbox2.index())
+            {
+                return Poll::Ready(());
+            }
+
+            Poll::Pending
+        })
+        .await;
+    }
+
+    /// Waits until all of the transmit mailboxes become empty
+    pub async fn flush_all(&self) {
+        poll_fn(|cx| {
+            T::state().tx_waker.register(cx.waker());
+
+            let tsr = T::regs().tsr().read();
+            if tsr.tme(bxcan::Mailbox::Mailbox0.index())
+                && tsr.tme(bxcan::Mailbox::Mailbox1.index())
+                && tsr.tme(bxcan::Mailbox::Mailbox2.index())
+            {
+                return Poll::Ready(());
+            }
+
+            Poll::Pending
+        })
+        .await;
+    }
 }
 
 #[allow(dead_code)]
@@ -407,6 +470,33 @@ impl<'c, 'd, T: Instance> CanRx<'c, 'd, T> {
         .await
     }
 
+    /// Attempts to read a CAN frame without blocking.
+    ///
+    /// Returns [Err(TryReadError::Empty)] if there are no frames in the rx queue.
+    pub fn try_read(&mut self) -> Result<(u16, bxcan::Frame), TryReadError> {
+        if let Ok(envelope) = T::state().rx_queue.try_recv() {
+            return Ok(envelope);
+        }
+
+        if let Some(err) = self.curr_error() {
+            return Err(TryReadError::BusError(err));
+        }
+
+        Err(TryReadError::Empty)
+    }
+
+    /// Waits while receive queue is empty.
+    pub async fn wait_not_empty(&mut self) {
+        poll_fn(|cx| {
+            if T::state().rx_queue.poll_ready_to_receive(cx) {
+                Poll::Ready(())
+            } else {
+                Poll::Pending
+            }
+        })
+        .await
+    }
+
     fn curr_error(&self) -> Option<BusError> {
         let err = { T::regs().esr().read() };
         if err.boff() {
diff --git a/embassy-stm32/src/can/fdcan.rs b/embassy-stm32/src/can/fdcan.rs
index c31a7fc63..f77788db3 100644
--- a/embassy-stm32/src/can/fdcan.rs
+++ b/embassy-stm32/src/can/fdcan.rs
@@ -1,5 +1,5 @@
 pub use bxcan;
-use embassy_hal_common::PeripheralRef;
+use embassy_hal_internal::PeripheralRef;
 
 use crate::peripherals;
 
diff --git a/embassy-stm32/src/crc/v1.rs b/embassy-stm32/src/crc/v1.rs
index 3946a2d47..154f2eb91 100644
--- a/embassy-stm32/src/crc/v1.rs
+++ b/embassy-stm32/src/crc/v1.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::pac::CRC as PAC_CRC;
 use crate::peripherals::CRC;
diff --git a/embassy-stm32/src/crc/v2v3.rs b/embassy-stm32/src/crc/v2v3.rs
index f337055a7..de0c08755 100644
--- a/embassy-stm32/src/crc/v2v3.rs
+++ b/embassy-stm32/src/crc/v2v3.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::pac::crc::vals;
 use crate::pac::CRC as PAC_CRC;
diff --git a/embassy-stm32/src/dac/mod.rs b/embassy-stm32/src/dac/mod.rs
index 3d58914b3..a2040b857 100644
--- a/embassy-stm32/src/dac/mod.rs
+++ b/embassy-stm32/src/dac/mod.rs
@@ -3,7 +3,7 @@
 //! Provide access to the STM32 digital-to-analog converter (DAC).
 use core::marker::PhantomData;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::pac::dac;
 use crate::rcc::RccPeripheral;
@@ -38,11 +38,30 @@ impl Channel {
 #[cfg_attr(feature = "defmt", derive(defmt::Format))]
 /// Trigger sources for CH1
 pub enum Ch1Trigger {
-    Tim6,
+    #[cfg(dac_v3)]
+    Tim1,
+    Tim2,
+    #[cfg(not(dac_v3))]
     Tim3,
+    #[cfg(dac_v3)]
+    Tim4,
+    #[cfg(dac_v3)]
+    Tim5,
+    Tim6,
     Tim7,
+    #[cfg(dac_v3)]
+    Tim8,
     Tim15,
-    Tim2,
+    #[cfg(dac_v3)]
+    Hrtim1Dactrg1,
+    #[cfg(dac_v3)]
+    Hrtim1Dactrg2,
+    #[cfg(dac_v3)]
+    Lptim1,
+    #[cfg(dac_v3)]
+    Lptim2,
+    #[cfg(dac_v3)]
+    Lptim3,
     Exti9,
     Software,
 }
@@ -50,14 +69,30 @@ pub enum Ch1Trigger {
 impl Ch1Trigger {
     fn tsel(&self) -> dac::vals::Tsel1 {
         match self {
-            Ch1Trigger::Tim6 => dac::vals::Tsel1::TIM6_TRGO,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Tim1 => dac::vals::Tsel1::TIM1_TRGO,
+            Ch1Trigger::Tim2 => dac::vals::Tsel1::TIM2_TRGO,
             #[cfg(not(dac_v3))]
             Ch1Trigger::Tim3 => dac::vals::Tsel1::TIM3_TRGO,
             #[cfg(dac_v3)]
-            Ch1Trigger::Tim3 => dac::vals::Tsel1::TIM1_TRGO,
+            Ch1Trigger::Tim4 => dac::vals::Tsel1::TIM4_TRGO,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Tim5 => dac::vals::Tsel1::TIM5_TRGO,
+            Ch1Trigger::Tim6 => dac::vals::Tsel1::TIM6_TRGO,
             Ch1Trigger::Tim7 => dac::vals::Tsel1::TIM7_TRGO,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Tim8 => dac::vals::Tsel1::TIM8_TRGO,
             Ch1Trigger::Tim15 => dac::vals::Tsel1::TIM15_TRGO,
-            Ch1Trigger::Tim2 => dac::vals::Tsel1::TIM2_TRGO,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Hrtim1Dactrg1 => dac::vals::Tsel1::HRTIM1_DACTRG1,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Hrtim1Dactrg2 => dac::vals::Tsel1::HRTIM1_DACTRG2,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Lptim1 => dac::vals::Tsel1::LPTIM1_OUT,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Lptim2 => dac::vals::Tsel1::LPTIM2_OUT,
+            #[cfg(dac_v3)]
+            Ch1Trigger::Lptim3 => dac::vals::Tsel1::LPTIM3_OUT,
             Ch1Trigger::Exti9 => dac::vals::Tsel1::EXTI9,
             Ch1Trigger::Software => dac::vals::Tsel1::SOFTWARE,
         }
@@ -129,7 +164,7 @@ pub trait DacChannel<T: Instance, Tx> {
     }
 
     /// Set mode register of the given channel
-    #[cfg(dac_v2)]
+    #[cfg(any(dac_v2, dac_v3))]
     fn set_channel_mode(&mut self, val: u8) -> Result<(), Error> {
         T::regs().mcr().modify(|reg| {
             reg.set_mode(Self::CHANNEL.index(), val);
@@ -216,8 +251,9 @@ impl<'d, T: Instance, Tx> DacCh1<'d, T, Tx> {
     pub fn new(
         peri: impl Peripheral<P = T> + 'd,
         dma: impl Peripheral<P = Tx> + 'd,
-        _pin: impl Peripheral<P = impl DacPin<T, 1>> + 'd,
+        pin: impl Peripheral<P = impl DacPin<T, 1>> + crate::gpio::sealed::Pin + 'd,
     ) -> Self {
+        pin.set_as_analog();
         into_ref!(peri, dma);
         T::enable();
         T::reset();
@@ -226,7 +262,7 @@ impl<'d, T: Instance, Tx> DacCh1<'d, T, Tx> {
 
         // Configure each activated channel. All results can be `unwrap`ed since they
         // will only error if the channel is not configured (i.e. ch1, ch2 are false)
-        #[cfg(dac_v2)]
+        #[cfg(any(dac_v2, dac_v3))]
         dac.set_channel_mode(0).unwrap();
         dac.enable_channel().unwrap();
         dac.set_trigger_enable(true).unwrap();
@@ -252,7 +288,6 @@ impl<'d, T: Instance, Tx> DacCh1<'d, T, Tx> {
     /// Note that for performance reasons in circular mode the transfer complete interrupt is disabled.
     ///
     /// **Important:** Channel 1 has to be configured for the DAC instance!
-    #[cfg(all(bdma, not(dma)))] // It currently only works with BDMA-only chips (DMA should theoretically work though)
     pub async fn write(&mut self, data: ValueArray<'_>, circular: bool) -> Result<(), Error>
     where
         Tx: DmaCh1<T>,
@@ -327,8 +362,9 @@ impl<'d, T: Instance, Tx> DacCh2<'d, T, Tx> {
     pub fn new(
         _peri: impl Peripheral<P = T> + 'd,
         dma: impl Peripheral<P = Tx> + 'd,
-        _pin: impl Peripheral<P = impl DacPin<T, 2>> + 'd,
+        pin: impl Peripheral<P = impl DacPin<T, 2>> + crate::gpio::sealed::Pin + 'd,
     ) -> Self {
+        pin.set_as_analog();
         into_ref!(_peri, dma);
         T::enable();
         T::reset();
@@ -340,7 +376,7 @@ impl<'d, T: Instance, Tx> DacCh2<'d, T, Tx> {
 
         // Configure each activated channel. All results can be `unwrap`ed since they
         // will only error if the channel is not configured (i.e. ch1, ch2 are false)
-        #[cfg(dac_v2)]
+        #[cfg(any(dac_v2, dac_v3))]
         dac.set_channel_mode(0).unwrap();
         dac.enable_channel().unwrap();
         dac.set_trigger_enable(true).unwrap();
@@ -364,7 +400,6 @@ impl<'d, T: Instance, Tx> DacCh2<'d, T, Tx> {
     /// Note that for performance reasons in circular mode the transfer complete interrupt is disabled.
     ///
     /// **Important:** Channel 2 has to be configured for the DAC instance!
-    #[cfg(all(bdma, not(dma)))] // It currently only works with BDMA-only chips (DMA should theoretically work though)
     pub async fn write(&mut self, data: ValueArray<'_>, circular: bool) -> Result<(), Error>
     where
         Tx: DmaCh2<T>,
@@ -442,9 +477,11 @@ impl<'d, T: Instance, TxCh1, TxCh2> Dac<'d, T, TxCh1, TxCh2> {
         peri: impl Peripheral<P = T> + 'd,
         dma_ch1: impl Peripheral<P = TxCh1> + 'd,
         dma_ch2: impl Peripheral<P = TxCh2> + 'd,
-        _pin_ch1: impl Peripheral<P = impl DacPin<T, 1>> + 'd,
-        _pin_ch2: impl Peripheral<P = impl DacPin<T, 2>> + 'd,
+        pin_ch1: impl Peripheral<P = impl DacPin<T, 1>> + crate::gpio::sealed::Pin + 'd,
+        pin_ch2: impl Peripheral<P = impl DacPin<T, 2>> + crate::gpio::sealed::Pin + 'd,
     ) -> Self {
+        pin_ch1.set_as_analog();
+        pin_ch2.set_as_analog();
         into_ref!(peri, dma_ch1, dma_ch2);
         T::enable();
         T::reset();
@@ -461,12 +498,12 @@ impl<'d, T: Instance, TxCh1, TxCh2> Dac<'d, T, TxCh1, TxCh2> {
 
         // Configure each activated channel. All results can be `unwrap`ed since they
         // will only error if the channel is not configured (i.e. ch1, ch2 are false)
-        #[cfg(dac_v2)]
+        #[cfg(any(dac_v2, dac_v3))]
         dac_ch1.set_channel_mode(0).unwrap();
         dac_ch1.enable_channel().unwrap();
         dac_ch1.set_trigger_enable(true).unwrap();
 
-        #[cfg(dac_v2)]
+        #[cfg(any(dac_v2, dac_v3))]
         dac_ch2.set_channel_mode(0).unwrap();
         dac_ch2.enable_channel().unwrap();
         dac_ch2.set_trigger_enable(true).unwrap();
diff --git a/embassy-stm32/src/dcmi.rs b/embassy-stm32/src/dcmi.rs
index 78b026cb6..7497f4aaa 100644
--- a/embassy-stm32/src/dcmi.rs
+++ b/embassy-stm32/src/dcmi.rs
@@ -2,7 +2,7 @@ use core::future::poll_fn;
 use core::marker::PhantomData;
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::dma::Transfer;
diff --git a/embassy-stm32/src/dma/bdma.rs b/embassy-stm32/src/dma/bdma.rs
index 5a87888b7..d956047d5 100644
--- a/embassy-stm32/src/dma/bdma.rs
+++ b/embassy-stm32/src/dma/bdma.rs
@@ -6,7 +6,7 @@ use core::sync::atomic::{fence, Ordering};
 use core::task::{Context, Poll, Waker};
 
 use atomic_polyfill::AtomicUsize;
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use super::ringbuffer::{DmaCtrl, DmaRingBuffer, OverrunError};
@@ -466,15 +466,53 @@ impl<'a, C: Channel, W: Word> RingBuffer<'a, C, W> {
         self.ringbuf.clear(DmaCtrlImpl(self.channel.reborrow()));
     }
 
-    /// Read bytes from the ring buffer
+    /// Read elements from the ring buffer
     /// Return a tuple of the length read and the length remaining in the buffer
-    /// If not all of the bytes were read, then there will be some bytes in the buffer remaining
-    /// The length remaining is the capacity, ring_buf.len(), less the bytes remaining after the read
+    /// If not all of the elements were read, then there will be some elements in the buffer remaining
+    /// The length remaining is the capacity, ring_buf.len(), less the elements remaining after the read
     /// OverrunError is returned if the portion to be read was overwritten by the DMA controller.
     pub fn read(&mut self, buf: &mut [W]) -> Result<(usize, usize), OverrunError> {
         self.ringbuf.read(DmaCtrlImpl(self.channel.reborrow()), buf)
     }
 
+    /// Read an exact number of elements from the ringbuffer.
+    ///
+    /// Returns the remaining number of elements available for immediate reading.
+    /// OverrunError is returned if the portion to be read was overwritten by the DMA controller.
+    ///
+    /// Async/Wake Behavior:
+    /// The underlying DMA peripheral only can wake us when its buffer pointer has reached the halfway point,
+    /// and when it wraps around. This means that when called with a buffer of length 'M', when this
+    /// ring buffer was created with a buffer of size 'N':
+    /// - If M equals N/2 or N/2 divides evenly into M, this function will return every N/2 elements read on the DMA source.
+    /// - Otherwise, this function may need up to N/2 extra elements to arrive before returning.
+    pub async fn read_exact(&mut self, buffer: &mut [W]) -> Result<usize, OverrunError> {
+        use core::future::poll_fn;
+        use core::sync::atomic::compiler_fence;
+
+        let mut read_data = 0;
+        let buffer_len = buffer.len();
+
+        poll_fn(|cx| {
+            self.set_waker(cx.waker());
+
+            compiler_fence(Ordering::SeqCst);
+
+            match self.read(&mut buffer[read_data..buffer_len]) {
+                Ok((len, remaining)) => {
+                    read_data += len;
+                    if read_data == buffer_len {
+                        Poll::Ready(Ok(remaining))
+                    } else {
+                        Poll::Pending
+                    }
+                }
+                Err(e) => Poll::Ready(Err(e)),
+            }
+        })
+        .await
+    }
+
     /// The capacity of the ringbuffer
     pub fn cap(&self) -> usize {
         self.ringbuf.cap()
diff --git a/embassy-stm32/src/dma/dma.rs b/embassy-stm32/src/dma/dma.rs
index 58d438af8..219ef2eb0 100644
--- a/embassy-stm32/src/dma/dma.rs
+++ b/embassy-stm32/src/dma/dma.rs
@@ -4,7 +4,7 @@ use core::pin::Pin;
 use core::sync::atomic::{fence, AtomicUsize, Ordering};
 use core::task::{Context, Poll, Waker};
 
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use super::ringbuffer::{DmaCtrl, DmaRingBuffer, OverrunError};
@@ -28,6 +28,12 @@ pub struct TransferOptions {
     pub flow_ctrl: FlowControl,
     /// FIFO threshold for DMA FIFO mode. If none, direct mode is used.
     pub fifo_threshold: Option<FifoThreshold>,
+    /// Enable circular DMA
+    pub circular: bool,
+    /// Enable half transfer interrupt
+    pub half_transfer_ir: bool,
+    /// Enable transfer complete interrupt
+    pub complete_transfer_ir: bool,
 }
 
 impl Default for TransferOptions {
@@ -37,6 +43,9 @@ impl Default for TransferOptions {
             mburst: Burst::Single,
             flow_ctrl: FlowControl::Dma,
             fifo_threshold: None,
+            circular: false,
+            half_transfer_ir: false,
+            complete_transfer_ir: true,
         }
     }
 }
@@ -365,7 +374,13 @@ impl<'a, C: Channel> Transfer<'a, C> {
             });
             w.set_pinc(vals::Inc::FIXED);
             w.set_teie(true);
-            w.set_tcie(true);
+            w.set_tcie(options.complete_transfer_ir);
+            if options.circular {
+                w.set_circ(vals::Circ::ENABLED);
+                debug!("Setting circular mode");
+            } else {
+                w.set_circ(vals::Circ::DISABLED);
+            }
             #[cfg(dma_v1)]
             w.set_trbuff(true);
 
@@ -646,7 +661,7 @@ impl<'a, C: Channel, W: Word> RingBuffer<'a, C, W> {
         w.set_minc(vals::Inc::INCREMENTED);
         w.set_pinc(vals::Inc::FIXED);
         w.set_teie(true);
-        w.set_htie(true);
+        w.set_htie(options.half_transfer_ir);
         w.set_tcie(true);
         w.set_circ(vals::Circ::ENABLED);
         #[cfg(dma_v1)]
@@ -696,15 +711,53 @@ impl<'a, C: Channel, W: Word> RingBuffer<'a, C, W> {
         self.ringbuf.clear(DmaCtrlImpl(self.channel.reborrow()));
     }
 
-    /// Read bytes from the ring buffer
+    /// Read elements from the ring buffer
     /// Return a tuple of the length read and the length remaining in the buffer
-    /// If not all of the bytes were read, then there will be some bytes in the buffer remaining
-    /// The length remaining is the capacity, ring_buf.len(), less the bytes remaining after the read
+    /// If not all of the elements were read, then there will be some elements in the buffer remaining
+    /// The length remaining is the capacity, ring_buf.len(), less the elements remaining after the read
     /// OverrunError is returned if the portion to be read was overwritten by the DMA controller.
     pub fn read(&mut self, buf: &mut [W]) -> Result<(usize, usize), OverrunError> {
         self.ringbuf.read(DmaCtrlImpl(self.channel.reborrow()), buf)
     }
 
+    /// Read an exact number of elements from the ringbuffer.
+    ///
+    /// Returns the remaining number of elements available for immediate reading.
+    /// OverrunError is returned if the portion to be read was overwritten by the DMA controller.
+    ///
+    /// Async/Wake Behavior:
+    /// The underlying DMA peripheral only can wake us when its buffer pointer has reached the halfway point,
+    /// and when it wraps around. This means that when called with a buffer of length 'M', when this
+    /// ring buffer was created with a buffer of size 'N':
+    /// - If M equals N/2 or N/2 divides evenly into M, this function will return every N/2 elements read on the DMA source.
+    /// - Otherwise, this function may need up to N/2 extra elements to arrive before returning.
+    pub async fn read_exact(&mut self, buffer: &mut [W]) -> Result<usize, OverrunError> {
+        use core::future::poll_fn;
+        use core::sync::atomic::compiler_fence;
+
+        let mut read_data = 0;
+        let buffer_len = buffer.len();
+
+        poll_fn(|cx| {
+            self.set_waker(cx.waker());
+
+            compiler_fence(Ordering::SeqCst);
+
+            match self.read(&mut buffer[read_data..buffer_len]) {
+                Ok((len, remaining)) => {
+                    read_data += len;
+                    if read_data == buffer_len {
+                        Poll::Ready(Ok(remaining))
+                    } else {
+                        Poll::Pending
+                    }
+                }
+                Err(e) => Poll::Ready(Err(e)),
+            }
+        })
+        .await
+    }
+
     // The capacity of the ringbuffer
     pub fn cap(&self) -> usize {
         self.ringbuf.cap()
diff --git a/embassy-stm32/src/dma/gpdma.rs b/embassy-stm32/src/dma/gpdma.rs
index b7bcf7795..97cc200d7 100644
--- a/embassy-stm32/src/dma/gpdma.rs
+++ b/embassy-stm32/src/dma/gpdma.rs
@@ -5,7 +5,7 @@ use core::pin::Pin;
 use core::sync::atomic::{fence, Ordering};
 use core::task::{Context, Poll};
 
-use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use super::word::{Word, WordSize};
diff --git a/embassy-stm32/src/dma/mod.rs b/embassy-stm32/src/dma/mod.rs
index 0858587bd..4f1a58ae2 100644
--- a/embassy-stm32/src/dma/mod.rs
+++ b/embassy-stm32/src/dma/mod.rs
@@ -26,7 +26,7 @@ pub mod word;
 
 use core::mem;
 
-use embassy_hal_common::impl_peripheral;
+use embassy_hal_internal::impl_peripheral;
 
 #[cfg(dmamux)]
 pub use self::dmamux::*;
diff --git a/embassy-stm32/src/dma/ringbuffer.rs b/embassy-stm32/src/dma/ringbuffer.rs
index a2bde986f..190793974 100644
--- a/embassy-stm32/src/dma/ringbuffer.rs
+++ b/embassy-stm32/src/dma/ringbuffer.rs
@@ -72,10 +72,10 @@ impl<'a, W: Word> DmaRingBuffer<'a, W> {
         self.cap() - remaining_transfers
     }
 
-    /// Read bytes from the ring buffer
+    /// Read elements from the ring buffer
     /// Return a tuple of the length read and the length remaining in the buffer
-    /// If not all of the bytes were read, then there will be some bytes in the buffer remaining
-    /// The length remaining is the capacity, ring_buf.len(), less the bytes remaining after the read
+    /// If not all of the elements were read, then there will be some elements in the buffer remaining
+    /// The length remaining is the capacity, ring_buf.len(), less the elements remaining after the read
     /// OverrunError is returned if the portion to be read was overwritten by the DMA controller.
     pub fn read(&mut self, mut dma: impl DmaCtrl, buf: &mut [W]) -> Result<(usize, usize), OverrunError> {
         /*
@@ -95,11 +95,11 @@ impl<'a, W: Word> DmaRingBuffer<'a, W> {
         */
         let end = self.pos(dma.get_remaining_transfers());
         if self.start == end && dma.get_complete_count() == 0 {
-            // No bytes are available in the buffer
+            // No elements are available in the buffer
             Ok((0, self.cap()))
         } else if self.start < end {
             // The available, unread portion in the ring buffer DOES NOT wrap
-            // Copy out the bytes from the dma buffer
+            // Copy out the elements from the dma buffer
             let len = self.copy_to(buf, self.start..end);
 
             compiler_fence(Ordering::SeqCst);
@@ -128,7 +128,7 @@ impl<'a, W: Word> DmaRingBuffer<'a, W> {
             // The DMA writer has wrapped since we last read and is currently
             // writing (or the next byte added will be) in the beginning of the ring buffer.
 
-            // The provided read buffer is not large enough to include all bytes from the tail of the dma buffer.
+            // The provided read buffer is not large enough to include all elements from the tail of the dma buffer.
 
             // Copy out from the dma buffer
             let len = self.copy_to(buf, self.start..self.cap());
@@ -154,8 +154,8 @@ impl<'a, W: Word> DmaRingBuffer<'a, W> {
             // The DMA writer has wrapped since we last read and is currently
             // writing (or the next byte added will be) in the beginning of the ring buffer.
 
-            // The provided read buffer is large enough to include all bytes from the tail of the dma buffer,
-            // so the next read will not have any unread tail bytes in the ring buffer.
+            // The provided read buffer is large enough to include all elements from the tail of the dma buffer,
+            // so the next read will not have any unread tail elements in the ring buffer.
 
             // Copy out from the dma buffer
             let tail = self.copy_to(buf, self.start..self.cap());
@@ -180,7 +180,7 @@ impl<'a, W: Word> DmaRingBuffer<'a, W> {
     }
     /// Copy from the dma buffer at `data_range` into `buf`
     fn copy_to(&mut self, buf: &mut [W], data_range: Range<usize>) -> usize {
-        // Limit the number of bytes that can be copied
+        // Limit the number of elements that can be copied
         let length = usize::min(data_range.len(), buf.len());
 
         // Copy from dma buffer into read buffer
diff --git a/embassy-stm32/src/eth/v1/mod.rs b/embassy-stm32/src/eth/v1/mod.rs
index 2a6ea35ff..a1e0240c8 100644
--- a/embassy-stm32/src/eth/v1/mod.rs
+++ b/embassy-stm32/src/eth/v1/mod.rs
@@ -6,7 +6,7 @@ mod tx_desc;
 use core::marker::PhantomData;
 use core::sync::atomic::{fence, Ordering};
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use stm32_metapac::eth::vals::{Apcs, Cr, Dm, DmaomrSr, Fes, Ftf, Ifg, MbProgress, Mw, Pbl, Rsf, St, Tsf};
 
 pub(crate) use self::rx_desc::{RDes, RDesRing};
diff --git a/embassy-stm32/src/eth/v2/mod.rs b/embassy-stm32/src/eth/v2/mod.rs
index bb681c42b..ada495fdb 100644
--- a/embassy-stm32/src/eth/v2/mod.rs
+++ b/embassy-stm32/src/eth/v2/mod.rs
@@ -3,7 +3,7 @@ mod descriptors;
 use core::marker::PhantomData;
 use core::sync::atomic::{fence, Ordering};
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 pub(crate) use self::descriptors::{RDes, RDesRing, TDes, TDesRing};
 use super::*;
diff --git a/embassy-stm32/src/exti.rs b/embassy-stm32/src/exti.rs
index 3ff92c9e6..925cf39be 100644
--- a/embassy-stm32/src/exti.rs
+++ b/embassy-stm32/src/exti.rs
@@ -3,7 +3,7 @@ use core::marker::PhantomData;
 use core::pin::Pin;
 use core::task::{Context, Poll};
 
-use embassy_hal_common::impl_peripheral;
+use embassy_hal_internal::impl_peripheral;
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::gpio::{AnyPin, Input, Pin as GpioPin};
diff --git a/embassy-stm32/src/flash/asynch.rs b/embassy-stm32/src/flash/asynch.rs
index f175349cd..e966e2a77 100644
--- a/embassy-stm32/src/flash/asynch.rs
+++ b/embassy-stm32/src/flash/asynch.rs
@@ -1,8 +1,8 @@
 use core::marker::PhantomData;
 use core::sync::atomic::{fence, Ordering};
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::into_ref;
 use embassy_sync::blocking_mutex::raw::CriticalSectionRawMutex;
 use embassy_sync::mutex::Mutex;
 
diff --git a/embassy-stm32/src/flash/common.rs b/embassy-stm32/src/flash/common.rs
index 2a374733d..16c511295 100644
--- a/embassy-stm32/src/flash/common.rs
+++ b/embassy-stm32/src/flash/common.rs
@@ -1,8 +1,8 @@
 use core::marker::PhantomData;
 use core::sync::atomic::{fence, Ordering};
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use stm32_metapac::FLASH_BASE;
 
 use super::{
diff --git a/embassy-stm32/src/flash/f4.rs b/embassy-stm32/src/flash/f4.rs
index 4cb39e033..728f6d604 100644
--- a/embassy-stm32/src/flash/f4.rs
+++ b/embassy-stm32/src/flash/f4.rs
@@ -14,7 +14,7 @@ use crate::pac;
 mod alt_regions {
     use core::marker::PhantomData;
 
-    use embassy_hal_common::PeripheralRef;
+    use embassy_hal_internal::PeripheralRef;
     use stm32_metapac::FLASH_SIZE;
 
     use crate::_generated::flash_regions::{OTPRegion, BANK1_REGION1, BANK1_REGION2, BANK1_REGION3, OTP_REGION};
diff --git a/embassy-stm32/src/fmc.rs b/embassy-stm32/src/fmc.rs
index 60d7a00ee..177e66a91 100644
--- a/embassy-stm32/src/fmc.rs
+++ b/embassy-stm32/src/fmc.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 
 use crate::gpio::sealed::AFType;
 use crate::gpio::{Pull, Speed};
diff --git a/embassy-stm32/src/gpio.rs b/embassy-stm32/src/gpio.rs
index af3a8eaca..cda597145 100644
--- a/embassy-stm32/src/gpio.rs
+++ b/embassy-stm32/src/gpio.rs
@@ -1,7 +1,7 @@
 #![macro_use]
 use core::convert::Infallible;
 
-use embassy_hal_common::{impl_peripheral, into_ref, PeripheralRef};
+use embassy_hal_internal::{impl_peripheral, into_ref, PeripheralRef};
 
 use crate::pac::gpio::{self, vals};
 use crate::{pac, peripherals, Peripheral};
diff --git a/embassy-stm32/src/i2c/v1.rs b/embassy-stm32/src/i2c/v1.rs
index aa485cd86..e5254a8cd 100644
--- a/embassy-stm32/src/i2c/v1.rs
+++ b/embassy-stm32/src/i2c/v1.rs
@@ -1,7 +1,7 @@
 use core::marker::PhantomData;
 
 use embassy_embedded_hal::SetConfig;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use crate::dma::NoDma;
 use crate::gpio::sealed::AFType;
diff --git a/embassy-stm32/src/i2c/v2.rs b/embassy-stm32/src/i2c/v2.rs
index 208d1527d..eaf980a4d 100644
--- a/embassy-stm32/src/i2c/v2.rs
+++ b/embassy-stm32/src/i2c/v2.rs
@@ -4,8 +4,8 @@ use core::marker::PhantomData;
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 
 use crate::dma::{NoDma, Transfer};
diff --git a/embassy-stm32/src/i2s.rs b/embassy-stm32/src/i2s.rs
index 62dda69b4..1ccad7328 100644
--- a/embassy-stm32/src/i2s.rs
+++ b/embassy-stm32/src/i2s.rs
@@ -1,4 +1,4 @@
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 
 use crate::gpio::sealed::{AFType, Pin as _};
 use crate::gpio::AnyPin;
diff --git a/embassy-stm32/src/ipcc.rs b/embassy-stm32/src/ipcc.rs
index a24cba9f0..e100ca5cc 100644
--- a/embassy-stm32/src/ipcc.rs
+++ b/embassy-stm32/src/ipcc.rs
@@ -265,63 +265,9 @@ pub(crate) mod sealed {
 }
 
 fn _configure_pwr() {
-    // TODO: move this to RCC
-
-    let pwr = crate::pac::PWR;
+    // TODO: move the rest of this to rcc
     let rcc = crate::pac::RCC;
 
-    rcc.cfgr().modify(|w| w.set_stopwuck(true));
-
-    pwr.cr1().modify(|w| w.set_dbp(true));
-    pwr.cr1().modify(|w| w.set_dbp(true));
-
-    // configure LSE
-    rcc.bdcr().modify(|w| w.set_lseon(true));
-
-    // select system clock source = PLL
-    // set PLL coefficients
-    // m: 2,
-    // n: 12,
-    // r: 3,
-    // q: 4,
-    // p: 3,
-    let src_bits = 0b11;
-    let pllp = (3 - 1) & 0b11111;
-    let pllq = (4 - 1) & 0b111;
-    let pllr = (3 - 1) & 0b111;
-    let plln = 12 & 0b1111111;
-    let pllm = (2 - 1) & 0b111;
-    rcc.pllcfgr().modify(|w| {
-        w.set_pllsrc(src_bits);
-        w.set_pllm(pllm);
-        w.set_plln(plln);
-        w.set_pllr(pllr);
-        w.set_pllp(pllp);
-        w.set_pllpen(true);
-        w.set_pllq(pllq);
-        w.set_pllqen(true);
-    });
-    // enable PLL
-    rcc.cr().modify(|w| w.set_pllon(true));
-    rcc.cr().write(|w| w.set_hsion(false));
-    // while !rcc.cr().read().pllrdy() {}
-
-    // configure SYSCLK mux to use PLL clocl
-    rcc.cfgr().modify(|w| w.set_sw(0b11));
-
-    // configure CPU1 & CPU2 dividers
-    rcc.cfgr().modify(|w| w.set_hpre(0)); // not divided
-    rcc.extcfgr().modify(|w| {
-        w.set_c2hpre(0b1000); // div2
-        w.set_shdhpre(0); // not divided
-    });
-
-    // apply APB1 / APB2 values
-    rcc.cfgr().modify(|w| {
-        w.set_ppre1(0b000); // not divided
-        w.set_ppre2(0b000); // not divided
-    });
-
     // TODO: required
     // set RF wake-up clock = LSE
     rcc.csr().modify(|w| w.set_rfwkpsel(0b01));
diff --git a/embassy-stm32/src/lib.rs b/embassy-stm32/src/lib.rs
index 8c005bfed..34220fbf5 100644
--- a/embassy-stm32/src/lib.rs
+++ b/embassy-stm32/src/lib.rs
@@ -1,6 +1,9 @@
 #![cfg_attr(not(test), no_std)]
 #![cfg_attr(feature = "nightly", feature(async_fn_in_trait, impl_trait_projections))]
 
+//! ## Feature flags
+#![doc = document_features::document_features!(feature_label = r#"<span class="stab portability"><code>{feature}</code></span>"#)]
+
 // This must go FIRST so that all the other modules see its macros.
 pub mod fmt;
 include!(concat!(env!("OUT_DIR"), "/_macros.rs"));
@@ -44,7 +47,6 @@ pub mod i2c;
 pub mod i2s;
 #[cfg(stm32wb)]
 pub mod ipcc;
-pub mod pwm;
 #[cfg(quadspi)]
 pub mod qspi;
 #[cfg(rng)]
@@ -80,7 +82,7 @@ pub use crate::_generated::interrupt;
 /// This defines the right interrupt handlers, and creates a unit struct (like `struct Irqs;`)
 /// and implements the right [`Binding`]s for it. You can pass this struct to drivers to
 /// prove at compile-time that the right interrupts have been bound.
-// developer note: this macro can't be in `embassy-hal-common` due to the use of `$crate`.
+// developer note: this macro can't be in `embassy-hal-internal` due to the use of `$crate`.
 #[macro_export]
 macro_rules! bind_interrupts {
     ($vis:vis struct $name:ident { $($irq:ident => $($handler:ty),*;)* }) => {
@@ -104,7 +106,7 @@ macro_rules! bind_interrupts {
 
 // Reexports
 pub use _generated::{peripherals, Peripherals};
-pub use embassy_hal_common::{into_ref, Peripheral, PeripheralRef};
+pub use embassy_hal_internal::{into_ref, Peripheral, PeripheralRef};
 #[cfg(feature = "unstable-pac")]
 pub use stm32_metapac as pac;
 #[cfg(not(feature = "unstable-pac"))]
diff --git a/embassy-stm32/src/pwm/mod.rs b/embassy-stm32/src/pwm/mod.rs
deleted file mode 100644
index 5aba2663e..000000000
--- a/embassy-stm32/src/pwm/mod.rs
+++ /dev/null
@@ -1,269 +0,0 @@
-pub mod complementary_pwm;
-pub mod simple_pwm;
-
-use stm32_metapac::timer::vals::Ckd;
-
-#[cfg(feature = "unstable-pac")]
-pub mod low_level {
-    pub use super::sealed::*;
-}
-
-#[derive(Clone, Copy)]
-pub enum Channel {
-    Ch1,
-    Ch2,
-    Ch3,
-    Ch4,
-}
-
-impl Channel {
-    pub fn raw(&self) -> usize {
-        match self {
-            Channel::Ch1 => 0,
-            Channel::Ch2 => 1,
-            Channel::Ch3 => 2,
-            Channel::Ch4 => 3,
-        }
-    }
-}
-
-#[derive(Clone, Copy)]
-pub enum OutputCompareMode {
-    Frozen,
-    ActiveOnMatch,
-    InactiveOnMatch,
-    Toggle,
-    ForceInactive,
-    ForceActive,
-    PwmMode1,
-    PwmMode2,
-}
-
-impl From<OutputCompareMode> for stm32_metapac::timer::vals::Ocm {
-    fn from(mode: OutputCompareMode) -> Self {
-        match mode {
-            OutputCompareMode::Frozen => stm32_metapac::timer::vals::Ocm::FROZEN,
-            OutputCompareMode::ActiveOnMatch => stm32_metapac::timer::vals::Ocm::ACTIVEONMATCH,
-            OutputCompareMode::InactiveOnMatch => stm32_metapac::timer::vals::Ocm::INACTIVEONMATCH,
-            OutputCompareMode::Toggle => stm32_metapac::timer::vals::Ocm::TOGGLE,
-            OutputCompareMode::ForceInactive => stm32_metapac::timer::vals::Ocm::FORCEINACTIVE,
-            OutputCompareMode::ForceActive => stm32_metapac::timer::vals::Ocm::FORCEACTIVE,
-            OutputCompareMode::PwmMode1 => stm32_metapac::timer::vals::Ocm::PWMMODE1,
-            OutputCompareMode::PwmMode2 => stm32_metapac::timer::vals::Ocm::PWMMODE2,
-        }
-    }
-}
-
-pub(crate) mod sealed {
-    use super::*;
-
-    pub trait CaptureCompare16bitInstance: crate::timer::sealed::GeneralPurpose16bitInstance {
-        /// Global output enable. Does not do anything on non-advanced timers.
-        fn enable_outputs(&mut self, enable: bool);
-
-        fn set_output_compare_mode(&mut self, channel: Channel, mode: OutputCompareMode);
-
-        fn enable_channel(&mut self, channel: Channel, enable: bool);
-
-        fn set_compare_value(&mut self, channel: Channel, value: u16);
-
-        fn get_max_compare_value(&self) -> u16;
-    }
-
-    pub trait ComplementaryCaptureCompare16bitInstance: CaptureCompare16bitInstance {
-        fn set_dead_time_clock_division(&mut self, value: Ckd);
-
-        fn set_dead_time_value(&mut self, value: u8);
-
-        fn enable_complementary_channel(&mut self, channel: Channel, enable: bool);
-    }
-
-    pub trait CaptureCompare32bitInstance: crate::timer::sealed::GeneralPurpose32bitInstance {
-        fn set_output_compare_mode(&mut self, channel: Channel, mode: OutputCompareMode);
-
-        fn enable_channel(&mut self, channel: Channel, enable: bool);
-
-        fn set_compare_value(&mut self, channel: Channel, value: u32);
-
-        fn get_max_compare_value(&self) -> u32;
-    }
-}
-
-pub trait CaptureCompare16bitInstance:
-    sealed::CaptureCompare16bitInstance + crate::timer::GeneralPurpose16bitInstance + 'static
-{
-}
-
-pub trait ComplementaryCaptureCompare16bitInstance:
-    sealed::ComplementaryCaptureCompare16bitInstance + crate::timer::AdvancedControlInstance + 'static
-{
-}
-
-pub trait CaptureCompare32bitInstance:
-    sealed::CaptureCompare32bitInstance + CaptureCompare16bitInstance + crate::timer::GeneralPurpose32bitInstance + 'static
-{
-}
-
-#[allow(unused)]
-macro_rules! impl_compare_capable_16bit {
-    ($inst:ident) => {
-        impl crate::pwm::sealed::CaptureCompare16bitInstance for crate::peripherals::$inst {
-            fn enable_outputs(&mut self, _enable: bool) {}
-
-            fn set_output_compare_mode(&mut self, channel: crate::pwm::Channel, mode: OutputCompareMode) {
-                use crate::timer::sealed::GeneralPurpose16bitInstance;
-                let r = Self::regs_gp16();
-                let raw_channel: usize = channel.raw();
-                r.ccmr_output(raw_channel / 2)
-                    .modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
-            }
-
-            fn enable_channel(&mut self, channel: Channel, enable: bool) {
-                use crate::timer::sealed::GeneralPurpose16bitInstance;
-                Self::regs_gp16()
-                    .ccer()
-                    .modify(|w| w.set_cce(channel.raw(), enable));
-            }
-
-            fn set_compare_value(&mut self, channel: Channel, value: u16) {
-                use crate::timer::sealed::GeneralPurpose16bitInstance;
-                Self::regs_gp16().ccr(channel.raw()).modify(|w| w.set_ccr(value));
-            }
-
-            fn get_max_compare_value(&self) -> u16 {
-                use crate::timer::sealed::GeneralPurpose16bitInstance;
-                Self::regs_gp16().arr().read().arr()
-            }
-        }
-    };
-}
-
-foreach_interrupt! {
-    ($inst:ident, timer, TIM_GP16, UP, $irq:ident) => {
-        impl_compare_capable_16bit!($inst);
-
-        impl CaptureCompare16bitInstance for crate::peripherals::$inst {
-
-        }
-    };
-
-    ($inst:ident, timer, TIM_GP32, UP, $irq:ident) => {
-        impl_compare_capable_16bit!($inst);
-        impl crate::pwm::sealed::CaptureCompare32bitInstance for crate::peripherals::$inst {
-            fn set_output_compare_mode(
-                &mut self,
-                channel: crate::pwm::Channel,
-                mode: OutputCompareMode,
-            ) {
-                use crate::timer::sealed::GeneralPurpose32bitInstance;
-                let raw_channel = channel.raw();
-                Self::regs_gp32().ccmr_output(raw_channel / 2).modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
-            }
-
-            fn enable_channel(&mut self, channel: Channel, enable: bool) {
-                use crate::timer::sealed::GeneralPurpose32bitInstance;
-                Self::regs_gp32().ccer().modify(|w| w.set_cce(channel.raw(), enable));
-            }
-
-            fn set_compare_value(&mut self, channel: Channel, value: u32) {
-                use crate::timer::sealed::GeneralPurpose32bitInstance;
-                Self::regs_gp32().ccr(channel.raw()).modify(|w| w.set_ccr(value));
-            }
-
-            fn get_max_compare_value(&self) -> u32 {
-                use crate::timer::sealed::GeneralPurpose32bitInstance;
-                Self::regs_gp32().arr().read().arr() as u32
-            }
-        }
-        impl CaptureCompare16bitInstance for crate::peripherals::$inst {
-
-        }
-        impl CaptureCompare32bitInstance for crate::peripherals::$inst {
-
-        }
-    };
-
-    ($inst:ident, timer, TIM_ADV, UP, $irq:ident) => {
-        impl crate::pwm::sealed::CaptureCompare16bitInstance for crate::peripherals::$inst {
-            fn enable_outputs(&mut self, enable: bool) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                let r = Self::regs_advanced();
-                r.bdtr().modify(|w| w.set_moe(enable));
-            }
-
-            fn set_output_compare_mode(
-                &mut self,
-                channel: crate::pwm::Channel,
-                mode: OutputCompareMode,
-            ) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                let r = Self::regs_advanced();
-                let raw_channel: usize = channel.raw();
-                r.ccmr_output(raw_channel / 2)
-                    .modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
-            }
-
-            fn enable_channel(&mut self, channel: Channel, enable: bool) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced()
-                    .ccer()
-                    .modify(|w| w.set_cce(channel.raw(), enable));
-            }
-
-            fn set_compare_value(&mut self, channel: Channel, value: u16) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced()
-                    .ccr(channel.raw())
-                    .modify(|w| w.set_ccr(value));
-            }
-
-            fn get_max_compare_value(&self) -> u16 {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced().arr().read().arr()
-            }
-        }
-
-        impl CaptureCompare16bitInstance for crate::peripherals::$inst {
-
-        }
-
-        impl crate::pwm::sealed::ComplementaryCaptureCompare16bitInstance for crate::peripherals::$inst {
-            fn set_dead_time_clock_division(&mut self, value: Ckd) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced().cr1().modify(|w| w.set_ckd(value));
-            }
-
-            fn set_dead_time_value(&mut self, value: u8) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced().bdtr().modify(|w| w.set_dtg(value));
-            }
-
-            fn enable_complementary_channel(&mut self, channel: Channel, enable: bool) {
-                use crate::timer::sealed::AdvancedControlInstance;
-                Self::regs_advanced()
-                    .ccer()
-                    .modify(|w| w.set_ccne(channel.raw(), enable));
-            }
-        }
-
-        impl ComplementaryCaptureCompare16bitInstance for crate::peripherals::$inst {
-
-        }
-    };
-}
-
-pin_trait!(Channel1Pin, CaptureCompare16bitInstance);
-pin_trait!(Channel1ComplementaryPin, CaptureCompare16bitInstance);
-pin_trait!(Channel2Pin, CaptureCompare16bitInstance);
-pin_trait!(Channel2ComplementaryPin, CaptureCompare16bitInstance);
-pin_trait!(Channel3Pin, CaptureCompare16bitInstance);
-pin_trait!(Channel3ComplementaryPin, CaptureCompare16bitInstance);
-pin_trait!(Channel4Pin, CaptureCompare16bitInstance);
-pin_trait!(Channel4ComplementaryPin, CaptureCompare16bitInstance);
-pin_trait!(ExternalTriggerPin, CaptureCompare16bitInstance);
-pin_trait!(BreakInputPin, CaptureCompare16bitInstance);
-pin_trait!(BreakInputComparator1Pin, CaptureCompare16bitInstance);
-pin_trait!(BreakInputComparator2Pin, CaptureCompare16bitInstance);
-pin_trait!(BreakInput2Pin, CaptureCompare16bitInstance);
-pin_trait!(BreakInput2Comparator1Pin, CaptureCompare16bitInstance);
-pin_trait!(BreakInput2Comparator2Pin, CaptureCompare16bitInstance);
diff --git a/embassy-stm32/src/qspi/mod.rs b/embassy-stm32/src/qspi/mod.rs
index 31b676088..32382fb28 100644
--- a/embassy-stm32/src/qspi/mod.rs
+++ b/embassy-stm32/src/qspi/mod.rs
@@ -2,7 +2,7 @@
 
 pub mod enums;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use enums::*;
 
 use crate::dma::Transfer;
diff --git a/embassy-stm32/src/rcc/f4.rs b/embassy-stm32/src/rcc/f4.rs
index b84470440..7aa9f0fd2 100644
--- a/embassy-stm32/src/rcc/f4.rs
+++ b/embassy-stm32/src/rcc/f4.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use stm32_metapac::rcc::vals::{Mco1, Mco2, Mcopre};
 
 use super::sealed::RccPeripheral;
diff --git a/embassy-stm32/src/rcc/h7.rs b/embassy-stm32/src/rcc/h7.rs
index 7e5cd0d1a..bbc0e0831 100644
--- a/embassy-stm32/src/rcc/h7.rs
+++ b/embassy-stm32/src/rcc/h7.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 pub use pll::PllConfig;
 use stm32_metapac::rcc::vals::{Mco1, Mco2};
 
diff --git a/embassy-stm32/src/rcc/l4.rs b/embassy-stm32/src/rcc/l4.rs
index 8a9b4adbf..dc5f55d0c 100644
--- a/embassy-stm32/src/rcc/l4.rs
+++ b/embassy-stm32/src/rcc/l4.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use stm32_metapac::rcc::regs::Cfgr;
 use stm32_metapac::rcc::vals::{Lsedrv, Mcopre, Mcosel};
 
diff --git a/embassy-stm32/src/rcc/mod.rs b/embassy-stm32/src/rcc/mod.rs
index 886fc0b93..4ae65d3e6 100644
--- a/embassy-stm32/src/rcc/mod.rs
+++ b/embassy-stm32/src/rcc/mod.rs
@@ -78,6 +78,14 @@ pub struct Clocks {
 /// The existence of this value indicates that the clock configuration can no longer be changed
 static mut CLOCK_FREQS: MaybeUninit<Clocks> = MaybeUninit::uninit();
 
+#[cfg(stm32wb)]
+/// RCC initialization function
+pub(crate) unsafe fn init(config: Config) {
+    set_freqs(compute_clocks(&config));
+
+    configure_clocks(&config);
+}
+
 /// Sets the clock frequencies
 ///
 /// Safety: Sets a mutable global.
diff --git a/embassy-stm32/src/rcc/wb.rs b/embassy-stm32/src/rcc/wb.rs
index e6123821a..4322b950a 100644
--- a/embassy-stm32/src/rcc/wb.rs
+++ b/embassy-stm32/src/rcc/wb.rs
@@ -1,6 +1,5 @@
-use crate::pac::RCC;
-use crate::rcc::{set_freqs, Clocks};
-use crate::time::Hertz;
+use crate::rcc::Clocks;
+use crate::time::{khz, mhz, Hertz};
 
 /// Most of clock setup is copied from stm32l0xx-hal, and adopted to the generated PAC,
 /// and with the addition of the init function to configure a system clock.
@@ -13,11 +12,94 @@ pub const HSI_FREQ: Hertz = Hertz(16_000_000);
 /// LSI speed
 pub const LSI_FREQ: Hertz = Hertz(32_000);
 
-/// System clock mux source
 #[derive(Clone, Copy)]
-pub enum ClockSrc {
-    HSE(Hertz),
-    HSI16,
+pub enum HsePrescaler {
+    NotDivided,
+    Div2,
+}
+
+impl From<HsePrescaler> for bool {
+    fn from(value: HsePrescaler) -> Self {
+        match value {
+            HsePrescaler::NotDivided => false,
+            HsePrescaler::Div2 => true,
+        }
+    }
+}
+
+pub struct Hse {
+    pub prediv: HsePrescaler,
+
+    pub frequency: Hertz,
+}
+
+/// System clock mux source
+#[derive(Clone, Copy, PartialEq)]
+pub enum Sysclk {
+    /// MSI selected as sysclk
+    MSI,
+    /// HSI selected as sysclk
+    HSI,
+    /// HSE selected as sysclk
+    HSE,
+    /// PLL selected as sysclk
+    Pll,
+}
+
+impl From<Sysclk> for u8 {
+    fn from(value: Sysclk) -> Self {
+        match value {
+            Sysclk::MSI => 0b00,
+            Sysclk::HSI => 0b01,
+            Sysclk::HSE => 0b10,
+            Sysclk::Pll => 0b11,
+        }
+    }
+}
+
+#[derive(Clone, Copy, PartialEq)]
+pub enum PllSource {
+    Hsi,
+    Msi,
+    Hse,
+}
+
+impl From<PllSource> for u8 {
+    fn from(value: PllSource) -> Self {
+        match value {
+            PllSource::Msi => 0b01,
+            PllSource::Hsi => 0b10,
+            PllSource::Hse => 0b11,
+        }
+    }
+}
+
+pub enum Pll48Source {
+    PllSai,
+    Pll,
+    Msi,
+    Hsi48,
+}
+
+pub struct PllMux {
+    /// Source clock selection.
+    pub source: PllSource,
+
+    /// PLL pre-divider (DIVM). Must be between 1 and 63.
+    pub prediv: u8,
+}
+
+pub struct Pll {
+    /// PLL multiplication factor. Must be between 4 and 512.
+    pub mul: u16,
+
+    /// PLL P division factor. If None, PLL P output is disabled. Must be between 1 and 128.
+    /// On PLL1, it must be even (in particular, it cannot be 1.)
+    pub divp: Option<u16>,
+    /// PLL Q division factor. If None, PLL Q output is disabled. Must be between 1 and 128.
+    pub divq: Option<u16>,
+    /// PLL R division factor. If None, PLL R output is disabled. Must be between 1 and 128.
+    pub divr: Option<u16>,
 }
 
 /// AHB prescaler
@@ -84,86 +166,250 @@ impl Into<u8> for AHBPrescaler {
 
 /// Clocks configutation
 pub struct Config {
-    pub mux: ClockSrc,
-    pub ahb_pre: AHBPrescaler,
+    pub hse: Option<Hse>,
+    pub lse: Option<Hertz>,
+    pub sys: Sysclk,
+    pub mux: Option<PllMux>,
+    pub pll48: Option<Pll48Source>,
+
+    pub pll: Option<Pll>,
+    pub pllsai: Option<Pll>,
+
+    pub ahb1_pre: AHBPrescaler,
+    pub ahb2_pre: AHBPrescaler,
+    pub ahb3_pre: AHBPrescaler,
     pub apb1_pre: APBPrescaler,
     pub apb2_pre: APBPrescaler,
 }
 
+pub const WPAN_DEFAULT: Config = Config {
+    hse: Some(Hse {
+        frequency: mhz(32),
+        prediv: HsePrescaler::NotDivided,
+    }),
+    lse: Some(khz(32)),
+    sys: Sysclk::Pll,
+    mux: Some(PllMux {
+        source: PllSource::Hse,
+        prediv: 2,
+    }),
+    pll48: None,
+
+    pll: Some(Pll {
+        mul: 12,
+        divp: Some(3),
+        divq: Some(4),
+        divr: Some(3),
+    }),
+    pllsai: None,
+
+    ahb1_pre: AHBPrescaler::NotDivided,
+    ahb2_pre: AHBPrescaler::Div2,
+    ahb3_pre: AHBPrescaler::NotDivided,
+    apb1_pre: APBPrescaler::NotDivided,
+    apb2_pre: APBPrescaler::NotDivided,
+};
+
 impl Default for Config {
     #[inline]
     fn default() -> Config {
         Config {
-            mux: ClockSrc::HSI16,
-            ahb_pre: AHBPrescaler::NotDivided,
+            hse: None,
+            lse: None,
+            sys: Sysclk::HSI,
+            mux: None,
+            pll48: None,
+            pll: None,
+            pllsai: None,
+
+            ahb1_pre: AHBPrescaler::NotDivided,
+            ahb2_pre: AHBPrescaler::NotDivided,
+            ahb3_pre: AHBPrescaler::NotDivided,
             apb1_pre: APBPrescaler::NotDivided,
             apb2_pre: APBPrescaler::NotDivided,
         }
     }
 }
 
-pub(crate) unsafe fn init(config: Config) {
-    let (sys_clk, sw) = match config.mux {
-        ClockSrc::HSI16 => {
-            // Enable HSI16
-            RCC.cr().write(|w| w.set_hsion(true));
-            while !RCC.cr().read().hsirdy() {}
+pub(crate) fn compute_clocks(config: &Config) -> Clocks {
+    let hse_clk = config.hse.as_ref().map(|hse| match hse.prediv {
+        HsePrescaler::NotDivided => hse.frequency,
+        HsePrescaler::Div2 => hse.frequency / 2u32,
+    });
 
-            (HSI_FREQ.0, 0x01)
+    let mux_clk = config.mux.as_ref().map(|pll_mux| {
+        (match pll_mux.source {
+            PllSource::Hse => hse_clk.unwrap(),
+            PllSource::Hsi => HSI_FREQ,
+            _ => unreachable!(),
+        } / pll_mux.prediv)
+    });
+
+    let (pll_r, _pll_q, _pll_p) = match &config.pll {
+        Some(pll) => {
+            let pll_vco = mux_clk.unwrap() * pll.mul as u32;
+
+            (
+                pll.divr.map(|divr| pll_vco / divr),
+                pll.divq.map(|divq| pll_vco / divq),
+                pll.divp.map(|divp| pll_vco / divp),
+            )
         }
-        ClockSrc::HSE(freq) => {
-            // Enable HSE
-            RCC.cr().write(|w| w.set_hseon(true));
-            while !RCC.cr().read().hserdy() {}
+        None => (None, None, None),
+    };
 
-            (freq.0, 0x02)
+    let sys_clk = match config.sys {
+        Sysclk::HSE => hse_clk.unwrap(),
+        Sysclk::HSI => HSI_FREQ,
+        Sysclk::Pll => pll_r.unwrap(),
+        _ => unreachable!(),
+    };
+
+    let ahb1_clk = match config.ahb1_pre {
+        AHBPrescaler::NotDivided => sys_clk,
+        pre => {
+            let pre: u8 = pre.into();
+            let pre = 1u32 << (pre as u32 - 7);
+            sys_clk / pre
         }
     };
 
-    RCC.cfgr().modify(|w| {
-        w.set_sw(sw.into());
-        w.set_hpre(config.ahb_pre.into());
-        w.set_ppre1(config.apb1_pre.into());
-        w.set_ppre2(config.apb2_pre.into());
-    });
+    let ahb2_clk = match config.ahb2_pre {
+        AHBPrescaler::NotDivided => sys_clk,
+        pre => {
+            let pre: u8 = pre.into();
+            let pre = 1u32 << (pre as u32 - 7);
+            sys_clk / pre
+        }
+    };
 
-    let ahb_freq: u32 = match config.ahb_pre {
+    let ahb3_clk = match config.ahb3_pre {
         AHBPrescaler::NotDivided => sys_clk,
         pre => {
             let pre: u8 = pre.into();
-            let pre = 1 << (pre as u32 - 7);
+            let pre = 1u32 << (pre as u32 - 7);
             sys_clk / pre
         }
     };
 
-    let (apb1_freq, apb1_tim_freq) = match config.apb1_pre {
-        APBPrescaler::NotDivided => (ahb_freq, ahb_freq),
+    let (apb1_clk, apb1_tim_clk) = match config.apb1_pre {
+        APBPrescaler::NotDivided => (ahb1_clk, ahb1_clk),
         pre => {
             let pre: u8 = pre.into();
             let pre: u8 = 1 << (pre - 3);
-            let freq = ahb_freq / pre as u32;
-            (freq, freq * 2)
+            let freq = ahb1_clk / pre as u32;
+            (freq, freq * 2u32)
         }
     };
 
-    let (apb2_freq, apb2_tim_freq) = match config.apb2_pre {
-        APBPrescaler::NotDivided => (ahb_freq, ahb_freq),
+    let (apb2_clk, apb2_tim_clk) = match config.apb2_pre {
+        APBPrescaler::NotDivided => (ahb1_clk, ahb1_clk),
         pre => {
             let pre: u8 = pre.into();
             let pre: u8 = 1 << (pre - 3);
-            let freq = ahb_freq / pre as u32;
-            (freq, freq * 2)
+            let freq = ahb1_clk / pre as u32;
+            (freq, freq * 2u32)
+        }
+    };
+
+    Clocks {
+        sys: sys_clk,
+        ahb1: ahb1_clk,
+        ahb2: ahb2_clk,
+        ahb3: ahb3_clk,
+        apb1: apb1_clk,
+        apb2: apb2_clk,
+        apb1_tim: apb1_tim_clk,
+        apb2_tim: apb2_tim_clk,
+    }
+}
+
+pub(crate) fn configure_clocks(config: &Config) {
+    let pwr = crate::pac::PWR;
+    let rcc = crate::pac::RCC;
+
+    let needs_hsi = if let Some(pll_mux) = &config.mux {
+        pll_mux.source == PllSource::Hsi
+    } else {
+        false
+    };
+
+    if needs_hsi || config.sys == Sysclk::HSI {
+        rcc.cr().modify(|w| {
+            w.set_hsion(true);
+        });
+
+        while !rcc.cr().read().hsirdy() {}
+    }
+
+    match &config.lse {
+        Some(_) => {
+            rcc.cfgr().modify(|w| w.set_stopwuck(true));
+
+            pwr.cr1().modify(|w| w.set_dbp(true));
+            pwr.cr1().modify(|w| w.set_dbp(true));
+
+            rcc.bdcr().modify(|w| w.set_lseon(true));
+        }
+        _ => {}
+    }
+
+    match &config.hse {
+        Some(hse) => {
+            rcc.cr().modify(|w| {
+                w.set_hsepre(hse.prediv.into());
+                w.set_hseon(true);
+            });
+
+            while !rcc.cr().read().hserdy() {}
+        }
+        _ => {}
+    }
+
+    match &config.mux {
+        Some(pll_mux) => {
+            rcc.pllcfgr().modify(|w| {
+                w.set_pllm(pll_mux.prediv);
+                w.set_pllsrc(pll_mux.source.into());
+            });
         }
+        _ => {}
     };
 
-    set_freqs(Clocks {
-        sys: Hertz(sys_clk),
-        ahb1: Hertz(ahb_freq),
-        ahb2: Hertz(ahb_freq),
-        ahb3: Hertz(ahb_freq),
-        apb1: Hertz(apb1_freq),
-        apb2: Hertz(apb2_freq),
-        apb1_tim: Hertz(apb1_tim_freq),
-        apb2_tim: Hertz(apb2_tim_freq),
+    match &config.pll {
+        Some(pll) => {
+            rcc.pllcfgr().modify(|w| {
+                w.set_plln(pll.mul as u8);
+                pll.divp.map(|divp| {
+                    w.set_pllpen(true);
+                    w.set_pllp((divp - 1) as u8)
+                });
+                pll.divq.map(|divq| {
+                    w.set_pllqen(true);
+                    w.set_pllq((divq - 1) as u8)
+                });
+                pll.divr.map(|divr| {
+                    // w.set_pllren(true);
+                    w.set_pllr((divr - 1) as u8);
+                });
+            });
+
+            rcc.cr().modify(|w| w.set_pllon(true));
+
+            while !rcc.cr().read().pllrdy() {}
+        }
+        _ => {}
+    }
+
+    rcc.cfgr().modify(|w| {
+        w.set_sw(config.sys.into());
+        w.set_hpre(config.ahb1_pre.into());
+        w.set_ppre1(config.apb1_pre.into());
+        w.set_ppre2(config.apb2_pre.into());
+    });
+
+    rcc.extcfgr().modify(|w| {
+        w.set_c2hpre(config.ahb2_pre.into());
+        w.set_shdhpre(config.ahb3_pre.into());
     });
 }
diff --git a/embassy-stm32/src/rng.rs b/embassy-stm32/src/rng.rs
index b2faec53d..27415c2d7 100644
--- a/embassy-stm32/src/rng.rs
+++ b/embassy-stm32/src/rng.rs
@@ -3,7 +3,7 @@
 use core::future::poll_fn;
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use rand_core::{CryptoRng, RngCore};
 
diff --git a/embassy-stm32/src/rtc/mod.rs b/embassy-stm32/src/rtc/mod.rs
index 12a2ac795..323be3187 100644
--- a/embassy-stm32/src/rtc/mod.rs
+++ b/embassy-stm32/src/rtc/mod.rs
@@ -15,7 +15,7 @@ pub use self::datetime::{DateTime, DayOfWeek, Error as DateTimeError};
 #[cfg_attr(any(rtc_v3, rtc_v3u5), path = "v3.rs")]
 mod _version;
 pub use _version::*;
-use embassy_hal_common::Peripheral;
+use embassy_hal_internal::Peripheral;
 
 /// Errors that can occur on methods on [RtcClock]
 #[derive(Clone, Debug, PartialEq, Eq)]
diff --git a/embassy-stm32/src/sdmmc/mod.rs b/embassy-stm32/src/sdmmc/mod.rs
index 698292bff..6b532363c 100644
--- a/embassy-stm32/src/sdmmc/mod.rs
+++ b/embassy-stm32/src/sdmmc/mod.rs
@@ -6,8 +6,8 @@ use core::marker::PhantomData;
 use core::ops::{Deref, DerefMut};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use embassy_sync::waitqueue::AtomicWaker;
 use sdio_host::{BusWidth, CardCapacity, CardStatus, CurrentState, SDStatus, CID, CSD, OCR, SCR};
 
@@ -225,6 +225,9 @@ const DMA_TRANSFER_OPTIONS: crate::dma::TransferOptions = crate::dma::TransferOp
     mburst: crate::dma::Burst::Incr4,
     flow_ctrl: crate::dma::FlowControl::Peripheral,
     fifo_threshold: Some(crate::dma::FifoThreshold::Full),
+    circular: false,
+    half_transfer_ir: false,
+    complete_transfer_ir: true,
 };
 #[cfg(all(sdmmc_v1, not(dma)))]
 const DMA_TRANSFER_OPTIONS: crate::dma::TransferOptions = crate::dma::TransferOptions {
diff --git a/embassy-stm32/src/spi/mod.rs b/embassy-stm32/src/spi/mod.rs
index d5f63f84e..bdf3c85b0 100644
--- a/embassy-stm32/src/spi/mod.rs
+++ b/embassy-stm32/src/spi/mod.rs
@@ -4,7 +4,7 @@ use core::ptr;
 
 use embassy_embedded_hal::SetConfig;
 use embassy_futures::join::join;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 pub use embedded_hal_02::spi::{Mode, Phase, Polarity, MODE_0, MODE_1, MODE_2, MODE_3};
 
 use crate::dma::{slice_ptr_parts, word, Transfer};
diff --git a/embassy-stm32/src/pwm/complementary_pwm.rs b/embassy-stm32/src/timer/complementary_pwm.rs
index 4d64d005c..64bb32c39 100644
--- a/embassy-stm32/src/pwm/complementary_pwm.rs
+++ b/embassy-stm32/src/timer/complementary_pwm.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use stm32_metapac::timer::vals::Ckd;
 
 use super::simple_pwm::*;
diff --git a/embassy-stm32/src/timer/mod.rs b/embassy-stm32/src/timer/mod.rs
index a92f854ec..286ab556f 100644
--- a/embassy-stm32/src/timer/mod.rs
+++ b/embassy-stm32/src/timer/mod.rs
@@ -1,3 +1,6 @@
+pub mod complementary_pwm;
+pub mod simple_pwm;
+
 use stm32_metapac::timer::vals;
 
 use crate::interrupt;
@@ -44,24 +47,122 @@ pub(crate) mod sealed {
         fn regs_advanced() -> crate::pac::timer::TimAdv;
     }
 
-    #[cfg(hrtim_v1)]
-    pub trait HighResolutionControlInstance: RccPeripheral {
-        type Interrupt: interrupt::typelevel::Interrupt;
+    pub trait CaptureCompare16bitInstance: GeneralPurpose16bitInstance {
+        /// Global output enable. Does not do anything on non-advanced timers.
+        fn enable_outputs(&mut self, enable: bool);
+
+        fn set_output_compare_mode(&mut self, channel: Channel, mode: OutputCompareMode);
+
+        fn enable_channel(&mut self, channel: Channel, enable: bool);
+
+        fn set_compare_value(&mut self, channel: Channel, value: u16);
+
+        fn get_max_compare_value(&self) -> u16;
+    }
+
+    pub trait ComplementaryCaptureCompare16bitInstance: CaptureCompare16bitInstance {
+        fn set_dead_time_clock_division(&mut self, value: vals::Ckd);
+
+        fn set_dead_time_value(&mut self, value: u8);
+
+        fn enable_complementary_channel(&mut self, channel: Channel, enable: bool);
+    }
+
+    pub trait CaptureCompare32bitInstance: GeneralPurpose32bitInstance {
+        fn set_output_compare_mode(&mut self, channel: Channel, mode: OutputCompareMode);
+
+        fn enable_channel(&mut self, channel: Channel, enable: bool);
+
+        fn set_compare_value(&mut self, channel: Channel, value: u32);
+
+        fn get_max_compare_value(&self) -> u32;
+    }
+}
+
+#[derive(Clone, Copy)]
+pub enum Channel {
+    Ch1,
+    Ch2,
+    Ch3,
+    Ch4,
+}
+
+impl Channel {
+    pub fn raw(&self) -> usize {
+        match self {
+            Channel::Ch1 => 0,
+            Channel::Ch2 => 1,
+            Channel::Ch3 => 2,
+            Channel::Ch4 => 3,
+        }
+    }
+}
+
+#[derive(Clone, Copy)]
+pub enum OutputCompareMode {
+    Frozen,
+    ActiveOnMatch,
+    InactiveOnMatch,
+    Toggle,
+    ForceInactive,
+    ForceActive,
+    PwmMode1,
+    PwmMode2,
+}
 
-        fn regs() -> crate::pac::hrtim::Hrtim;
+impl From<OutputCompareMode> for stm32_metapac::timer::vals::Ocm {
+    fn from(mode: OutputCompareMode) -> Self {
+        match mode {
+            OutputCompareMode::Frozen => stm32_metapac::timer::vals::Ocm::FROZEN,
+            OutputCompareMode::ActiveOnMatch => stm32_metapac::timer::vals::Ocm::ACTIVEONMATCH,
+            OutputCompareMode::InactiveOnMatch => stm32_metapac::timer::vals::Ocm::INACTIVEONMATCH,
+            OutputCompareMode::Toggle => stm32_metapac::timer::vals::Ocm::TOGGLE,
+            OutputCompareMode::ForceInactive => stm32_metapac::timer::vals::Ocm::FORCEINACTIVE,
+            OutputCompareMode::ForceActive => stm32_metapac::timer::vals::Ocm::FORCEACTIVE,
+            OutputCompareMode::PwmMode1 => stm32_metapac::timer::vals::Ocm::PWMMODE1,
+            OutputCompareMode::PwmMode2 => stm32_metapac::timer::vals::Ocm::PWMMODE2,
+        }
     }
 }
 
+pub trait Basic16bitInstance: sealed::Basic16bitInstance + 'static {}
+
 pub trait GeneralPurpose16bitInstance: sealed::GeneralPurpose16bitInstance + 'static {}
 
 pub trait GeneralPurpose32bitInstance: sealed::GeneralPurpose32bitInstance + 'static {}
 
 pub trait AdvancedControlInstance: sealed::AdvancedControlInstance + 'static {}
 
-#[cfg(hrtim_v1)]
-pub trait HighResolutionControlInstance: sealed::HighResolutionControlInstance + 'static {}
+pub trait CaptureCompare16bitInstance:
+    sealed::CaptureCompare16bitInstance + GeneralPurpose16bitInstance + 'static
+{
+}
 
-pub trait Basic16bitInstance: sealed::Basic16bitInstance + 'static {}
+pub trait ComplementaryCaptureCompare16bitInstance:
+    sealed::ComplementaryCaptureCompare16bitInstance + AdvancedControlInstance + 'static
+{
+}
+
+pub trait CaptureCompare32bitInstance:
+    sealed::CaptureCompare32bitInstance + CaptureCompare16bitInstance + GeneralPurpose32bitInstance + 'static
+{
+}
+
+pin_trait!(Channel1Pin, CaptureCompare16bitInstance);
+pin_trait!(Channel1ComplementaryPin, CaptureCompare16bitInstance);
+pin_trait!(Channel2Pin, CaptureCompare16bitInstance);
+pin_trait!(Channel2ComplementaryPin, CaptureCompare16bitInstance);
+pin_trait!(Channel3Pin, CaptureCompare16bitInstance);
+pin_trait!(Channel3ComplementaryPin, CaptureCompare16bitInstance);
+pin_trait!(Channel4Pin, CaptureCompare16bitInstance);
+pin_trait!(Channel4ComplementaryPin, CaptureCompare16bitInstance);
+pin_trait!(ExternalTriggerPin, CaptureCompare16bitInstance);
+pin_trait!(BreakInputPin, CaptureCompare16bitInstance);
+pin_trait!(BreakInputComparator1Pin, CaptureCompare16bitInstance);
+pin_trait!(BreakInputComparator2Pin, CaptureCompare16bitInstance);
+pin_trait!(BreakInput2Pin, CaptureCompare16bitInstance);
+pin_trait!(BreakInput2Comparator1Pin, CaptureCompare16bitInstance);
+pin_trait!(BreakInput2Comparator2Pin, CaptureCompare16bitInstance);
 
 #[allow(unused)]
 macro_rules! impl_basic_16bit_timer {
@@ -150,33 +251,94 @@ macro_rules! impl_32bit_timer {
     };
 }
 
+#[allow(unused)]
+macro_rules! impl_compare_capable_16bit {
+    ($inst:ident) => {
+        impl sealed::CaptureCompare16bitInstance for crate::peripherals::$inst {
+            fn enable_outputs(&mut self, _enable: bool) {}
+
+            fn set_output_compare_mode(&mut self, channel: Channel, mode: OutputCompareMode) {
+                use sealed::GeneralPurpose16bitInstance;
+                let r = Self::regs_gp16();
+                let raw_channel: usize = channel.raw();
+                r.ccmr_output(raw_channel / 2)
+                    .modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
+            }
+
+            fn enable_channel(&mut self, channel: Channel, enable: bool) {
+                use sealed::GeneralPurpose16bitInstance;
+                Self::regs_gp16()
+                    .ccer()
+                    .modify(|w| w.set_cce(channel.raw(), enable));
+            }
+
+            fn set_compare_value(&mut self, channel: Channel, value: u16) {
+                use sealed::GeneralPurpose16bitInstance;
+                Self::regs_gp16().ccr(channel.raw()).modify(|w| w.set_ccr(value));
+            }
+
+            fn get_max_compare_value(&self) -> u16 {
+                use sealed::GeneralPurpose16bitInstance;
+                Self::regs_gp16().arr().read().arr()
+            }
+        }
+    };
+}
+
 foreach_interrupt! {
     ($inst:ident, timer, TIM_BASIC, UP, $irq:ident) => {
         impl_basic_16bit_timer!($inst, $irq);
-
-        impl Basic16bitInstance for crate::peripherals::$inst {
-        }
+        impl Basic16bitInstance for crate::peripherals::$inst {}
     };
     ($inst:ident, timer, TIM_GP16, UP, $irq:ident) => {
         impl_basic_16bit_timer!($inst, $irq);
-
-        impl Basic16bitInstance for crate::peripherals::$inst {
-        }
+        impl_compare_capable_16bit!($inst);
+        impl Basic16bitInstance for crate::peripherals::$inst {}
+        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {}
+        impl CaptureCompare16bitInstance for crate::peripherals::$inst {}
 
         impl sealed::GeneralPurpose16bitInstance for crate::peripherals::$inst {
             fn regs_gp16() -> crate::pac::timer::TimGp16 {
                 crate::pac::$inst
             }
         }
-
-        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {
-        }
     };
 
     ($inst:ident, timer, TIM_GP32, UP, $irq:ident) => {
         impl_basic_16bit_timer!($inst, $irq);
+        impl_32bit_timer!($inst);
+        impl_compare_capable_16bit!($inst);
+        impl Basic16bitInstance for crate::peripherals::$inst {}
+        impl CaptureCompare16bitInstance for crate::peripherals::$inst {}
+        impl CaptureCompare32bitInstance for crate::peripherals::$inst {}
+        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {}
+        impl GeneralPurpose32bitInstance for crate::peripherals::$inst {}
+
+        impl sealed::CaptureCompare32bitInstance for crate::peripherals::$inst {
+            fn set_output_compare_mode(
+                &mut self,
+                channel: Channel,
+                mode: OutputCompareMode,
+            ) {
+                use crate::timer::sealed::GeneralPurpose32bitInstance;
+                let raw_channel = channel.raw();
+                Self::regs_gp32().ccmr_output(raw_channel / 2).modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
+            }
 
-        impl Basic16bitInstance for crate::peripherals::$inst {
+            fn enable_channel(&mut self, channel: Channel, enable: bool) {
+                use crate::timer::sealed::GeneralPurpose32bitInstance;
+                Self::regs_gp32().ccer().modify(|w| w.set_cce(channel.raw(), enable));
+            }
+
+            fn set_compare_value(&mut self, channel: Channel, value: u32) {
+                use crate::timer::sealed::GeneralPurpose32bitInstance;
+                Self::regs_gp32().ccr(channel.raw()).modify(|w| w.set_ccr(value));
+            }
+
+            fn get_max_compare_value(&self) -> u32 {
+                use crate::timer::sealed::GeneralPurpose32bitInstance;
+                Self::regs_gp32().arr().read().arr() as u32
+            }
         }
 
         impl sealed::GeneralPurpose16bitInstance for crate::peripherals::$inst {
@@ -184,21 +346,16 @@ foreach_interrupt! {
                 unsafe { crate::pac::timer::TimGp16::from_ptr(crate::pac::$inst.as_ptr()) }
             }
         }
-
-        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {
-        }
-
-        impl_32bit_timer!($inst);
-
-        impl GeneralPurpose32bitInstance for crate::peripherals::$inst {
-        }
     };
 
     ($inst:ident, timer, TIM_ADV, UP, $irq:ident) => {
         impl_basic_16bit_timer!($inst, $irq);
 
-        impl Basic16bitInstance for crate::peripherals::$inst {
-        }
+        impl Basic16bitInstance for crate::peripherals::$inst {}
+        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {}
+        impl CaptureCompare16bitInstance for crate::peripherals::$inst {}
+        impl ComplementaryCaptureCompare16bitInstance for crate::peripherals::$inst {}
+        impl AdvancedControlInstance for crate::peripherals::$inst {}
 
         impl sealed::GeneralPurpose16bitInstance for crate::peripherals::$inst {
             fn regs_gp16() -> crate::pac::timer::TimGp16 {
@@ -206,17 +363,71 @@ foreach_interrupt! {
             }
         }
 
-        impl GeneralPurpose16bitInstance for crate::peripherals::$inst {
-        }
-
         impl sealed::AdvancedControlInstance for crate::peripherals::$inst {
             fn regs_advanced() -> crate::pac::timer::TimAdv {
                 crate::pac::$inst
             }
         }
 
-        impl AdvancedControlInstance for crate::peripherals::$inst {
+        impl sealed::CaptureCompare16bitInstance for crate::peripherals::$inst {
+            fn enable_outputs(&mut self, enable: bool) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                let r = Self::regs_advanced();
+                r.bdtr().modify(|w| w.set_moe(enable));
+            }
+
+            fn set_output_compare_mode(
+                &mut self,
+                channel: Channel,
+                mode: OutputCompareMode,
+            ) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                let r = Self::regs_advanced();
+                let raw_channel: usize = channel.raw();
+                r.ccmr_output(raw_channel / 2)
+                    .modify(|w| w.set_ocm(raw_channel % 2, mode.into()));
+            }
+
+            fn enable_channel(&mut self, channel: Channel, enable: bool) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced()
+                    .ccer()
+                    .modify(|w| w.set_cce(channel.raw(), enable));
+            }
+
+            fn set_compare_value(&mut self, channel: Channel, value: u16) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced()
+                    .ccr(channel.raw())
+                    .modify(|w| w.set_ccr(value));
+            }
+
+            fn get_max_compare_value(&self) -> u16 {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced().arr().read().arr()
+            }
+        }
+
+        impl sealed::ComplementaryCaptureCompare16bitInstance for crate::peripherals::$inst {
+            fn set_dead_time_clock_division(&mut self, value: vals::Ckd) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced().cr1().modify(|w| w.set_ckd(value));
+            }
+
+            fn set_dead_time_value(&mut self, value: u8) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced().bdtr().modify(|w| w.set_dtg(value));
+            }
+
+            fn enable_complementary_channel(&mut self, channel: Channel, enable: bool) {
+                use crate::timer::sealed::AdvancedControlInstance;
+                Self::regs_advanced()
+                    .ccer()
+                    .modify(|w| w.set_ccne(channel.raw(), enable));
+            }
         }
+
+
     };
 
     ($inst:ident, hrtim, HRTIM, MASTER, $irq:ident) => {
diff --git a/embassy-stm32/src/pwm/simple_pwm.rs b/embassy-stm32/src/timer/simple_pwm.rs
index 995f59c23..514796930 100644
--- a/embassy-stm32/src/pwm/simple_pwm.rs
+++ b/embassy-stm32/src/timer/simple_pwm.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::{into_ref, PeripheralRef};
 
 use super::*;
 #[allow(unused_imports)]
diff --git a/embassy-stm32/src/usart/buffered.rs b/embassy-stm32/src/usart/buffered.rs
index 433ad299c..ca117da82 100644
--- a/embassy-stm32/src/usart/buffered.rs
+++ b/embassy-stm32/src/usart/buffered.rs
@@ -2,7 +2,7 @@ use core::future::poll_fn;
 use core::slice;
 use core::task::Poll;
 
-use embassy_hal_common::atomic_ring_buffer::RingBuffer;
+use embassy_hal_internal::atomic_ring_buffer::RingBuffer;
 use embassy_sync::waitqueue::AtomicWaker;
 
 use super::*;
diff --git a/embassy-stm32/src/usart/mod.rs b/embassy-stm32/src/usart/mod.rs
index ea8e525ea..d99034bca 100644
--- a/embassy-stm32/src/usart/mod.rs
+++ b/embassy-stm32/src/usart/mod.rs
@@ -5,8 +5,8 @@ use core::marker::PhantomData;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::drop::OnDrop;
-use embassy_hal_common::{into_ref, PeripheralRef};
+use embassy_hal_internal::drop::OnDrop;
+use embassy_hal_internal::{into_ref, PeripheralRef};
 use futures::future::{select, Either};
 
 use crate::dma::{NoDma, Transfer};
@@ -857,7 +857,7 @@ fn configure(r: Regs, config: &Config, pclk_freq: Hertz, kind: Kind, enable_rx:
         "Using {} oversampling, desired baudrate: {}, actual baudrate: {}",
         oversampling,
         config.baudrate,
-        pclk_freq.0 / div
+        (pclk_freq.0 * mul as u32) / div
     );
 
     r.cr2().write(|w| {
diff --git a/embassy-stm32/src/usart/ringbuffered.rs b/embassy-stm32/src/usart/ringbuffered.rs
index c74d7e092..80261d048 100644
--- a/embassy-stm32/src/usart/ringbuffered.rs
+++ b/embassy-stm32/src/usart/ringbuffered.rs
@@ -2,7 +2,7 @@ use core::future::poll_fn;
 use core::sync::atomic::{compiler_fence, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::PeripheralRef;
+use embassy_hal_internal::PeripheralRef;
 use futures::future::{select, Either};
 
 use super::{clear_interrupt_flags, rdr, sr, BasicInstance, Error, UartRx};
diff --git a/embassy-stm32/src/usb/usb.rs b/embassy-stm32/src/usb/usb.rs
index ecdd1d0b8..cef196355 100644
--- a/embassy-stm32/src/usb/usb.rs
+++ b/embassy-stm32/src/usb/usb.rs
@@ -5,7 +5,7 @@ use core::marker::PhantomData;
 use core::sync::atomic::{AtomicBool, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::into_ref;
+use embassy_hal_internal::into_ref;
 use embassy_sync::waitqueue::AtomicWaker;
 use embassy_usb_driver as driver;
 use embassy_usb_driver::{
diff --git a/embassy-stm32/src/usb_otg/usb.rs b/embassy-stm32/src/usb_otg/usb.rs
index fd0e22adf..348f0f79d 100644
--- a/embassy-stm32/src/usb_otg/usb.rs
+++ b/embassy-stm32/src/usb_otg/usb.rs
@@ -3,7 +3,7 @@ use core::marker::PhantomData;
 use core::sync::atomic::{AtomicBool, AtomicU16, Ordering};
 use core::task::Poll;
 
-use embassy_hal_common::{into_ref, Peripheral};
+use embassy_hal_internal::{into_ref, Peripheral};
 use embassy_sync::waitqueue::AtomicWaker;
 use embassy_usb_driver::{
     self, Bus as _, Direction, EndpointAddress, EndpointAllocError, EndpointError, EndpointIn, EndpointInfo,
diff --git a/embassy-stm32/src/wdg/mod.rs b/embassy-stm32/src/wdg/mod.rs
index b03e81d6e..eafd03364 100644
--- a/embassy-stm32/src/wdg/mod.rs
+++ b/embassy-stm32/src/wdg/mod.rs
@@ -1,6 +1,6 @@
 use core::marker::PhantomData;
 
-use embassy_hal_common::{into_ref, Peripheral};
+use embassy_hal_internal::{into_ref, Peripheral};
 use stm32_metapac::iwdg::vals::{Key, Pr};
 
 use crate::rcc::LSI_FREQ;
diff --git a/examples/nrf-rtos-trace/Cargo.toml b/examples/nrf-rtos-trace/Cargo.toml
index 30b67b7b2..068474e7a 100644
--- a/examples/nrf-rtos-trace/Cargo.toml
+++ b/examples/nrf-rtos-trace/Cargo.toml
@@ -34,3 +34,6 @@ log = { version = "0.4.17", optional = true }
 [[bin]]
 name = "rtos_trace"
 required-features = ["nightly"]
+
+[profile.release]
+debug = 2
diff --git a/examples/nrf52840-rtic/Cargo.toml b/examples/nrf52840-rtic/Cargo.toml
index ded3b7db8..715f1ecfe 100644
--- a/examples/nrf52840-rtic/Cargo.toml
+++ b/examples/nrf52840-rtic/Cargo.toml
@@ -19,3 +19,6 @@ cortex-m = { version = "0.7.6", features = ["inline-asm", "critical-section-sing
 cortex-m-rt = "0.7.0"
 panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
+
+[profile.release]
+debug = 2
diff --git a/examples/nrf52840/Cargo.toml b/examples/nrf52840/Cargo.toml
index 9b41ec5ab..780aaeac2 100644
--- a/examples/nrf52840/Cargo.toml
+++ b/examples/nrf52840/Cargo.toml
@@ -57,5 +57,5 @@ embedded-hal-async = { version = "0.2.0-alpha.2", optional = true }
 num-integer = { version = "0.1.45", default-features = false }
 microfft = "0.5.0"
 
-[patch.crates-io]
-lora-phy = { git = "https://github.com/embassy-rs/lora-phy", rev = "ad289428fd44b02788e2fa2116445cc8f640a265" }
+[profile.release]
+debug = 2
diff --git a/examples/nrf5340/Cargo.toml b/examples/nrf5340/Cargo.toml
index f1d45f336..b0e51dcf4 100644
--- a/examples/nrf5340/Cargo.toml
+++ b/examples/nrf5340/Cargo.toml
@@ -53,3 +53,6 @@ rand = { version = "0.8.4", default-features = false }
 embedded-storage = "0.3.0"
 usbd-hid = "0.6.0"
 serde = { version = "1.0.136", default-features = false }
+
+[profile.release]
+debug = 2
diff --git a/examples/rp/Cargo.toml b/examples/rp/Cargo.toml
index c812cb3ee..8c61dc5e1 100644
--- a/examples/rp/Cargo.toml
+++ b/examples/rp/Cargo.toml
@@ -53,7 +53,4 @@ pio = "0.2.1"
 rand = { version = "0.8.5", default-features = false }
 
 [profile.release]
-debug = true
-
-[patch.crates-io]
-lora-phy = { git = "https://github.com/embassy-rs/lora-phy", rev = "ad289428fd44b02788e2fa2116445cc8f640a265" }
+debug = 2
diff --git a/examples/rp/src/bin/pio_async.rs b/examples/rp/src/bin/pio_async.rs
index c001d6440..a6d6144be 100644
--- a/examples/rp/src/bin/pio_async.rs
+++ b/examples/rp/src/bin/pio_async.rs
@@ -8,7 +8,6 @@ use embassy_executor::Spawner;
 use embassy_rp::bind_interrupts;
 use embassy_rp::peripherals::PIO0;
 use embassy_rp::pio::{Common, Config, InterruptHandler, Irq, Pio, PioPin, ShiftDirection, StateMachine};
-use embassy_rp::relocate::RelocatedProgram;
 use fixed::traits::ToFixed;
 use fixed_macro::types::U56F8;
 use {defmt_rtt as _, panic_probe as _};
@@ -29,9 +28,8 @@ fn setup_pio_task_sm0<'a>(pio: &mut Common<'a, PIO0>, sm: &mut StateMachine<'a,
         ".wrap",
     );
 
-    let relocated = RelocatedProgram::new(&prg.program);
     let mut cfg = Config::default();
-    cfg.use_program(&pio.load_program(&relocated), &[]);
+    cfg.use_program(&pio.load_program(&prg.program), &[]);
     let out_pin = pio.make_pio_pin(pin);
     cfg.set_out_pins(&[&out_pin]);
     cfg.set_set_pins(&[&out_pin]);
@@ -65,9 +63,8 @@ fn setup_pio_task_sm1<'a>(pio: &mut Common<'a, PIO0>, sm: &mut StateMachine<'a,
         ".wrap",
     );
 
-    let relocated = RelocatedProgram::new(&prg.program);
     let mut cfg = Config::default();
-    cfg.use_program(&pio.load_program(&relocated), &[]);
+    cfg.use_program(&pio.load_program(&prg.program), &[]);
     cfg.clock_divider = (U56F8!(125_000_000) / 2000).to_fixed();
     cfg.shift_in.auto_fill = true;
     cfg.shift_in.direction = ShiftDirection::Right;
@@ -96,9 +93,8 @@ fn setup_pio_task_sm2<'a>(pio: &mut Common<'a, PIO0>, sm: &mut StateMachine<'a,
         "irq 3 [15]",
         ".wrap",
     );
-    let relocated = RelocatedProgram::new(&prg.program);
     let mut cfg = Config::default();
-    cfg.use_program(&pio.load_program(&relocated), &[]);
+    cfg.use_program(&pio.load_program(&prg.program), &[]);
     cfg.clock_divider = (U56F8!(125_000_000) / 2000).to_fixed();
     sm.set_config(&cfg);
 }
diff --git a/examples/rp/src/bin/pio_dma.rs b/examples/rp/src/bin/pio_dma.rs
index 9ab72e1f3..86e5017ac 100644
--- a/examples/rp/src/bin/pio_dma.rs
+++ b/examples/rp/src/bin/pio_dma.rs
@@ -8,7 +8,6 @@ use embassy_executor::Spawner;
 use embassy_futures::join::join;
 use embassy_rp::peripherals::PIO0;
 use embassy_rp::pio::{Config, InterruptHandler, Pio, ShiftConfig, ShiftDirection};
-use embassy_rp::relocate::RelocatedProgram;
 use embassy_rp::{bind_interrupts, Peripheral};
 use fixed::traits::ToFixed;
 use fixed_macro::types::U56F8;
@@ -46,9 +45,8 @@ async fn main(_spawner: Spawner) {
         ".wrap",
     );
 
-    let relocated = RelocatedProgram::new(&prg.program);
     let mut cfg = Config::default();
-    cfg.use_program(&common.load_program(&relocated), &[]);
+    cfg.use_program(&common.load_program(&prg.program), &[]);
     cfg.clock_divider = (U56F8!(125_000_000) / U56F8!(10_000)).to_fixed();
     cfg.shift_in = ShiftConfig {
         auto_fill: true,
diff --git a/examples/rp/src/bin/pio_hd44780.rs b/examples/rp/src/bin/pio_hd44780.rs
index 8aedd24b6..d80c5c24b 100644
--- a/examples/rp/src/bin/pio_hd44780.rs
+++ b/examples/rp/src/bin/pio_hd44780.rs
@@ -14,7 +14,6 @@ use embassy_rp::pio::{
     Config, Direction, FifoJoin, InterruptHandler, Pio, PioPin, ShiftConfig, ShiftDirection, StateMachine,
 };
 use embassy_rp::pwm::{self, Pwm};
-use embassy_rp::relocate::RelocatedProgram;
 use embassy_rp::{bind_interrupts, into_ref, Peripheral, PeripheralRef};
 use embassy_time::{Duration, Instant, Timer};
 use {defmt_rtt as _, panic_probe as _};
@@ -127,9 +126,8 @@ impl<'l> HD44780<'l> {
 
         sm0.set_pin_dirs(Direction::Out, &[&rs, &rw, &e, &db4, &db5, &db6, &db7]);
 
-        let relocated = RelocatedProgram::new(&prg.program);
         let mut cfg = Config::default();
-        cfg.use_program(&common.load_program(&relocated), &[&e]);
+        cfg.use_program(&common.load_program(&prg.program), &[&e]);
         cfg.clock_divider = 125u8.into();
         cfg.set_out_pins(&[&db4, &db5, &db6, &db7]);
         cfg.shift_out = ShiftConfig {
@@ -201,9 +199,8 @@ impl<'l> HD44780<'l> {
             "#
         );
 
-        let relocated = RelocatedProgram::new(&prg.program);
         let mut cfg = Config::default();
-        cfg.use_program(&common.load_program(&relocated), &[&e]);
+        cfg.use_program(&common.load_program(&prg.program), &[&e]);
         cfg.clock_divider = 8u8.into(); // ~64ns/insn
         cfg.set_jmp_pin(&db7);
         cfg.set_set_pins(&[&rs, &rw]);
diff --git a/examples/rp/src/bin/pio_uart.rs b/examples/rp/src/bin/pio_uart.rs
new file mode 100644
index 000000000..5fddbe292
--- /dev/null
+++ b/examples/rp/src/bin/pio_uart.rs
@@ -0,0 +1,392 @@
+//! This example shows how to use the PIO module in the RP2040 chip to implement a duplex UART.
+//! The PIO module is a very powerful peripheral that can be used to implement many different
+//! protocols. It is a very flexible state machine that can be programmed to do almost anything.
+//!
+//! This example opens up a USB device that implements a CDC ACM serial port. It then uses the
+//! PIO module to implement a UART that is connected to the USB serial port. This allows you to
+//! communicate with a device connected to the RP2040 over USB serial.
+
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+#![feature(async_fn_in_trait)]
+
+use defmt::{info, panic, trace};
+use embassy_executor::Spawner;
+use embassy_futures::join::{join, join3};
+use embassy_rp::bind_interrupts;
+use embassy_rp::peripherals::{PIO0, USB};
+use embassy_rp::pio::InterruptHandler as PioInterruptHandler;
+use embassy_rp::usb::{Driver, Instance, InterruptHandler};
+use embassy_sync::blocking_mutex::raw::NoopRawMutex;
+use embassy_sync::pipe::Pipe;
+use embassy_usb::class::cdc_acm::{CdcAcmClass, Receiver, Sender, State};
+use embassy_usb::driver::EndpointError;
+use embassy_usb::{Builder, Config};
+use embedded_io::asynch::{Read, Write};
+use {defmt_rtt as _, panic_probe as _};
+
+use crate::uart::PioUart;
+use crate::uart_rx::PioUartRx;
+use crate::uart_tx::PioUartTx;
+
+bind_interrupts!(struct Irqs {
+    USBCTRL_IRQ => InterruptHandler<USB>;
+    PIO0_IRQ_0 => PioInterruptHandler<PIO0>;
+});
+
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    info!("Hello there!");
+
+    let p = embassy_rp::init(Default::default());
+
+    // Create the driver, from the HAL.
+    let driver = Driver::new(p.USB, Irqs);
+
+    // Create embassy-usb Config
+    let mut config = Config::new(0xc0de, 0xcafe);
+    config.manufacturer = Some("Embassy");
+    config.product = Some("PIO UART example");
+    config.serial_number = Some("12345678");
+    config.max_power = 100;
+    config.max_packet_size_0 = 64;
+
+    // Required for windows compatibility.
+    // https://developer.nordicsemi.com/nRF_Connect_SDK/doc/1.9.1/kconfig/CONFIG_CDC_ACM_IAD.html#help
+    config.device_class = 0xEF;
+    config.device_sub_class = 0x02;
+    config.device_protocol = 0x01;
+    config.composite_with_iads = true;
+
+    // Create embassy-usb DeviceBuilder using the driver and config.
+    // It needs some buffers for building the descriptors.
+    let mut device_descriptor = [0; 256];
+    let mut config_descriptor = [0; 256];
+    let mut bos_descriptor = [0; 256];
+    let mut control_buf = [0; 64];
+
+    let mut state = State::new();
+
+    let mut builder = Builder::new(
+        driver,
+        config,
+        &mut device_descriptor,
+        &mut config_descriptor,
+        &mut bos_descriptor,
+        &mut control_buf,
+    );
+
+    // Create classes on the builder.
+    let class = CdcAcmClass::new(&mut builder, &mut state, 64);
+
+    // Build the builder.
+    let mut usb = builder.build();
+
+    // Run the USB device.
+    let usb_fut = usb.run();
+
+    // PIO UART setup
+    let uart = PioUart::new(9600, p.PIO0, p.PIN_4, p.PIN_5);
+    let (mut uart_tx, mut uart_rx) = uart.split();
+
+    // Pipe setup
+    let usb_pipe: Pipe<NoopRawMutex, 20> = Pipe::new();
+    let mut usb_pipe_writer = usb_pipe.writer();
+    let mut usb_pipe_reader = usb_pipe.reader();
+
+    let uart_pipe: Pipe<NoopRawMutex, 20> = Pipe::new();
+    let mut uart_pipe_writer = uart_pipe.writer();
+    let mut uart_pipe_reader = uart_pipe.reader();
+
+    let (mut usb_tx, mut usb_rx) = class.split();
+
+    // Read + write from USB
+    let usb_future = async {
+        loop {
+            info!("Wait for USB connection");
+            usb_rx.wait_connection().await;
+            info!("Connected");
+            let _ = join(
+                usb_read(&mut usb_rx, &mut uart_pipe_writer),
+                usb_write(&mut usb_tx, &mut usb_pipe_reader),
+            )
+            .await;
+            info!("Disconnected");
+        }
+    };
+
+    // Read + write from UART
+    let uart_future = join(
+        uart_read(&mut uart_rx, &mut usb_pipe_writer),
+        uart_write(&mut uart_tx, &mut uart_pipe_reader),
+    );
+
+    // Run everything concurrently.
+    // If we had made everything `'static` above instead, we could do this using separate tasks instead.
+    join3(usb_fut, usb_future, uart_future).await;
+}
+
+struct Disconnected {}
+
+impl From<EndpointError> for Disconnected {
+    fn from(val: EndpointError) -> Self {
+        match val {
+            EndpointError::BufferOverflow => panic!("Buffer overflow"),
+            EndpointError::Disabled => Disconnected {},
+        }
+    }
+}
+
+/// Read from the USB and write it to the UART TX pipe
+async fn usb_read<'d, T: Instance + 'd>(
+    usb_rx: &mut Receiver<'d, Driver<'d, T>>,
+    uart_pipe_writer: &mut embassy_sync::pipe::Writer<'_, NoopRawMutex, 20>,
+) -> Result<(), Disconnected> {
+    let mut buf = [0; 64];
+    loop {
+        let n = usb_rx.read_packet(&mut buf).await?;
+        let data = &buf[..n];
+        trace!("USB IN: {:x}", data);
+        uart_pipe_writer.write(data).await;
+    }
+}
+
+/// Read from the USB TX pipe and write it to the USB
+async fn usb_write<'d, T: Instance + 'd>(
+    usb_tx: &mut Sender<'d, Driver<'d, T>>,
+    usb_pipe_reader: &mut embassy_sync::pipe::Reader<'_, NoopRawMutex, 20>,
+) -> Result<(), Disconnected> {
+    let mut buf = [0; 64];
+    loop {
+        let n = usb_pipe_reader.read(&mut buf).await;
+        let data = &buf[..n];
+        trace!("USB OUT: {:x}", data);
+        usb_tx.write_packet(&data).await?;
+    }
+}
+
+/// Read from the UART and write it to the USB TX pipe
+async fn uart_read(
+    uart_rx: &mut PioUartRx<'_>,
+    usb_pipe_writer: &mut embassy_sync::pipe::Writer<'_, NoopRawMutex, 20>,
+) -> ! {
+    let mut buf = [0; 64];
+    loop {
+        let n = uart_rx.read(&mut buf).await.expect("UART read error");
+        if n == 0 {
+            continue;
+        }
+        let data = &buf[..n];
+        trace!("UART IN: {:x}", buf);
+        usb_pipe_writer.write(data).await;
+    }
+}
+
+/// Read from the UART TX pipe and write it to the UART
+async fn uart_write(
+    uart_tx: &mut PioUartTx<'_>,
+    uart_pipe_reader: &mut embassy_sync::pipe::Reader<'_, NoopRawMutex, 20>,
+) -> ! {
+    let mut buf = [0; 64];
+    loop {
+        let n = uart_pipe_reader.read(&mut buf).await;
+        let data = &buf[..n];
+        trace!("UART OUT: {:x}", data);
+        let _ = uart_tx.write(&data).await;
+    }
+}
+
+mod uart {
+    use embassy_rp::peripherals::PIO0;
+    use embassy_rp::pio::{Pio, PioPin};
+    use embassy_rp::Peripheral;
+
+    use crate::uart_rx::PioUartRx;
+    use crate::uart_tx::PioUartTx;
+    use crate::Irqs;
+
+    pub struct PioUart<'a> {
+        tx: PioUartTx<'a>,
+        rx: PioUartRx<'a>,
+    }
+
+    impl<'a> PioUart<'a> {
+        pub fn new(
+            baud: u64,
+            pio: impl Peripheral<P = PIO0> + 'a,
+            tx_pin: impl PioPin,
+            rx_pin: impl PioPin,
+        ) -> PioUart<'a> {
+            let Pio {
+                mut common, sm0, sm1, ..
+            } = Pio::new(pio, Irqs);
+
+            let tx = PioUartTx::new(&mut common, sm0, tx_pin, baud);
+            let rx = PioUartRx::new(&mut common, sm1, rx_pin, baud);
+
+            PioUart { tx, rx }
+        }
+
+        pub fn split(self) -> (PioUartTx<'a>, PioUartRx<'a>) {
+            (self.tx, self.rx)
+        }
+    }
+}
+
+mod uart_tx {
+    use core::convert::Infallible;
+
+    use embassy_rp::gpio::Level;
+    use embassy_rp::peripherals::PIO0;
+    use embassy_rp::pio::{Common, Config, Direction, FifoJoin, PioPin, ShiftDirection, StateMachine};
+    use embedded_io::asynch::Write;
+    use embedded_io::Io;
+    use fixed::traits::ToFixed;
+    use fixed_macro::types::U56F8;
+
+    pub struct PioUartTx<'a> {
+        sm_tx: StateMachine<'a, PIO0, 0>,
+    }
+
+    impl<'a> PioUartTx<'a> {
+        pub fn new(
+            common: &mut Common<'a, PIO0>,
+            mut sm_tx: StateMachine<'a, PIO0, 0>,
+            tx_pin: impl PioPin,
+            baud: u64,
+        ) -> Self {
+            let prg = pio_proc::pio_asm!(
+                r#"
+                .side_set 1 opt
+
+                ; An 8n1 UART transmit program.
+                ; OUT pin 0 and side-set pin 0 are both mapped to UART TX pin.
+
+                    pull       side 1 [7]  ; Assert stop bit, or stall with line in idle state
+                    set x, 7   side 0 [7]  ; Preload bit counter, assert start bit for 8 clocks
+                bitloop:                   ; This loop will run 8 times (8n1 UART)
+                    out pins, 1            ; Shift 1 bit from OSR to the first OUT pin
+                    jmp x-- bitloop   [6]  ; Each loop iteration is 8 cycles.
+            "#
+            );
+            let tx_pin = common.make_pio_pin(tx_pin);
+            sm_tx.set_pins(Level::High, &[&tx_pin]);
+            sm_tx.set_pin_dirs(Direction::Out, &[&tx_pin]);
+
+            let mut cfg = Config::default();
+
+            cfg.set_out_pins(&[&tx_pin]);
+            cfg.use_program(&common.load_program(&prg.program), &[&tx_pin]);
+            cfg.shift_out.auto_fill = false;
+            cfg.shift_out.direction = ShiftDirection::Right;
+            cfg.fifo_join = FifoJoin::TxOnly;
+            cfg.clock_divider = (U56F8!(125_000_000) / (8 * baud)).to_fixed();
+            sm_tx.set_config(&cfg);
+            sm_tx.set_enable(true);
+
+            Self { sm_tx }
+        }
+
+        pub async fn write_u8(&mut self, data: u8) {
+            self.sm_tx.tx().wait_push(data as u32).await;
+        }
+    }
+
+    impl Io for PioUartTx<'_> {
+        type Error = Infallible;
+    }
+
+    impl Write for PioUartTx<'_> {
+        async fn write(&mut self, buf: &[u8]) -> Result<usize, Infallible> {
+            for byte in buf {
+                self.write_u8(*byte).await;
+            }
+            Ok(buf.len())
+        }
+    }
+}
+
+mod uart_rx {
+    use core::convert::Infallible;
+
+    use embassy_rp::gpio::Level;
+    use embassy_rp::peripherals::PIO0;
+    use embassy_rp::pio::{Common, Config, Direction, FifoJoin, PioPin, ShiftDirection, StateMachine};
+    use embedded_io::asynch::Read;
+    use embedded_io::Io;
+    use fixed::traits::ToFixed;
+    use fixed_macro::types::U56F8;
+
+    pub struct PioUartRx<'a> {
+        sm_rx: StateMachine<'a, PIO0, 1>,
+    }
+
+    impl<'a> PioUartRx<'a> {
+        pub fn new(
+            common: &mut Common<'a, PIO0>,
+            mut sm_rx: StateMachine<'a, PIO0, 1>,
+            rx_pin: impl PioPin,
+            baud: u64,
+        ) -> Self {
+            let prg = pio_proc::pio_asm!(
+                r#"
+                ; Slightly more fleshed-out 8n1 UART receiver which handles framing errors and
+                ; break conditions more gracefully.
+                ; IN pin 0 and JMP pin are both mapped to the GPIO used as UART RX.
+
+                start:
+                    wait 0 pin 0        ; Stall until start bit is asserted
+                    set x, 7    [10]    ; Preload bit counter, then delay until halfway through
+                rx_bitloop:                ; the first data bit (12 cycles incl wait, set).
+                    in pins, 1          ; Shift data bit into ISR
+                    jmp x-- rx_bitloop [6] ; Loop 8 times, each loop iteration is 8 cycles
+                    jmp pin good_rx_stop   ; Check stop bit (should be high)
+
+                    irq 4 rel           ; Either a framing error or a break. Set a sticky flag,
+                    wait 1 pin 0        ; and wait for line to return to idle state.
+                    jmp start           ; Don't push data if we didn't see good framing.
+
+                good_rx_stop:              ; No delay before returning to start; a little slack is
+                    push                ; important in case the TX clock is slightly too fast.
+            "#
+            );
+            let mut cfg = Config::default();
+            cfg.use_program(&common.load_program(&prg.program), &[]);
+
+            let rx_pin = common.make_pio_pin(rx_pin);
+            sm_rx.set_pins(Level::High, &[&rx_pin]);
+            cfg.set_in_pins(&[&rx_pin]);
+            cfg.set_jmp_pin(&rx_pin);
+            sm_rx.set_pin_dirs(Direction::In, &[&rx_pin]);
+
+            cfg.clock_divider = (U56F8!(125_000_000) / (8 * baud)).to_fixed();
+            cfg.shift_out.auto_fill = false;
+            cfg.shift_out.direction = ShiftDirection::Right;
+            cfg.fifo_join = FifoJoin::RxOnly;
+            sm_rx.set_config(&cfg);
+            sm_rx.set_enable(true);
+
+            Self { sm_rx }
+        }
+
+        pub async fn read_u8(&mut self) -> u8 {
+            self.sm_rx.rx().wait_pull().await as u8
+        }
+    }
+
+    impl Io for PioUartRx<'_> {
+        type Error = Infallible;
+    }
+
+    impl Read for PioUartRx<'_> {
+        async fn read(&mut self, buf: &mut [u8]) -> Result<usize, Infallible> {
+            let mut i = 0;
+            while i < buf.len() {
+                buf[i] = self.read_u8().await;
+                i += 1;
+            }
+            Ok(i)
+        }
+    }
+}
diff --git a/examples/rp/src/bin/pio_ws2812.rs b/examples/rp/src/bin/pio_ws2812.rs
index 3de2bd48d..bc87016ec 100644
--- a/examples/rp/src/bin/pio_ws2812.rs
+++ b/examples/rp/src/bin/pio_ws2812.rs
@@ -12,7 +12,6 @@ use embassy_rp::peripherals::PIO0;
 use embassy_rp::pio::{
     Common, Config, FifoJoin, Instance, InterruptHandler, Pio, PioPin, ShiftConfig, ShiftDirection, StateMachine,
 };
-use embassy_rp::relocate::RelocatedProgram;
 use embassy_rp::{bind_interrupts, clocks, into_ref, Peripheral, PeripheralRef};
 use embassy_time::{Duration, Timer};
 use fixed::types::U24F8;
@@ -73,8 +72,7 @@ impl<'d, P: Instance, const S: usize, const N: usize> Ws2812<'d, P, S, N> {
         cfg.set_out_pins(&[&out_pin]);
         cfg.set_set_pins(&[&out_pin]);
 
-        let relocated = RelocatedProgram::new(&prg);
-        cfg.use_program(&pio.load_program(&relocated), &[&out_pin]);
+        cfg.use_program(&pio.load_program(&prg), &[&out_pin]);
 
         // Clock config, measured in kHz to avoid overflows
         // TODO CLOCK_FREQ should come from embassy_rp
diff --git a/examples/std/Cargo.toml b/examples/std/Cargo.toml
index 92933ab50..42adede10 100644
--- a/examples/std/Cargo.toml
+++ b/examples/std/Cargo.toml
@@ -23,3 +23,6 @@ clap = { version = "3.0.0-beta.5", features = ["derive"] }
 rand_core = { version = "0.6.3", features = ["std"] }
 heapless = { version = "0.7.5", default-features = false }
 static_cell = { version = "1.1", features = ["nightly"]}
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32c0/Cargo.toml b/examples/stm32c0/Cargo.toml
index 26837abef..8534921ab 100644
--- a/examples/stm32c0/Cargo.toml
+++ b/examples/stm32c0/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32c031c6 to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32c031c6", "memory-x", "unstable-pac", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32c031c6", "memory-x", "unstable-pac", "exti"]  }
 
 defmt = "0.3"
 defmt-rtt = "0.4"
@@ -19,3 +20,6 @@ embedded-hal = "0.2.6"
 panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 heapless = { version = "0.7.5", default-features = false }
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f0/Cargo.toml b/examples/stm32f0/Cargo.toml
index b7b5eaa99..46b6db45c 100644
--- a/examples/stm32f0/Cargo.toml
+++ b/examples/stm32f0/Cargo.toml
@@ -7,6 +7,8 @@ license = "MIT OR Apache-2.0"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
+# Change stm32f091rc to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "memory-x", "stm32f091rc", "time-driver-any", "exti", "unstable-pac"] }
 cortex-m = { version = "0.7.6", features = ["inline-asm", "critical-section-single-core"] }
 cortex-m-rt = "0.7.0"
 defmt = "0.3"
@@ -15,5 +17,7 @@ panic-probe = "0.3"
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "memory-x", "stm32f091rc", "time-driver-any", "exti", "unstable-pac"] }
 static_cell = { version = "1.1", features = ["nightly"]}
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f1/Cargo.toml b/examples/stm32f1/Cargo.toml
index 29cad5b67..5d32992cd 100644
--- a/examples/stm32f1/Cargo.toml
+++ b/examples/stm32f1/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32f103c8 to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f103c8", "unstable-pac", "memory-x", "time-driver-any", "unstable-traits" ]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f103c8", "unstable-pac", "memory-x", "time-driver-any", "unstable-traits" ]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
 
@@ -25,3 +26,6 @@ nb = "1.0.0"
 
 [profile.dev]
 opt-level = "s"
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f2/Cargo.toml b/examples/stm32f2/Cargo.toml
index 652210c7f..9857fb631 100644
--- a/examples/stm32f2/Cargo.toml
+++ b/examples/stm32f2/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32f207zg to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f207zg", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f207zg", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 
 defmt = "0.3"
 defmt-rtt = "0.4"
@@ -20,3 +21,6 @@ panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 heapless = { version = "0.7.5", default-features = false }
 nb = "1.0.0"
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f3/Cargo.toml b/examples/stm32f3/Cargo.toml
index 489d0ff4c..bd594d16a 100644
--- a/examples/stm32f3/Cargo.toml
+++ b/examples/stm32f3/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32f303ze to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f303ze", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f303ze", "unstable-pac", "memory-x", "time-driver-any", "exti"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
 
@@ -24,3 +25,6 @@ heapless = { version = "0.7.5", default-features = false }
 nb = "1.0.0"
 embedded-storage = "0.3.0"
 static_cell = { version = "1.1", features = ["nightly"]}
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f4/Cargo.toml b/examples/stm32f4/Cargo.toml
index c1c821364..fdd3b7d4f 100644
--- a/examples/stm32f4/Cargo.toml
+++ b/examples/stm32f4/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32f429zi to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "unstable-traits", "defmt", "stm32f429zi", "unstable-pac", "memory-x", "time-driver-any", "exti", "embedded-sdmmc", "chrono"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
-embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers", "arch-cortex-m", "executor-thread", "executor-interrupt"] }
+embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "executor-interrupt", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "unstable-traits", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "unstable-traits", "defmt", "stm32f429zi", "unstable-pac", "memory-x", "time-driver-any", "exti", "embedded-sdmmc", "chrono"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 embassy-net = { version = "0.1.0", path = "../../embassy-net", features = ["defmt", "tcp", "dhcpv4", "medium-ethernet", "nightly"] }
 
diff --git a/examples/stm32f4/src/bin/can.rs b/examples/stm32f4/src/bin/can.rs
index 08bed88db..f84f74d30 100644
--- a/examples/stm32f4/src/bin/can.rs
+++ b/examples/stm32f4/src/bin/can.rs
@@ -40,10 +40,13 @@ async fn main(_spawner: Spawner) {
 
     can.as_mut()
         .modify_config()
-        .set_bit_timing(0x001c0003) // http://www.bittiming.can-wiki.info/
         .set_loopback(true) // Receive own frames
         .set_silent(true)
-        .enable();
+        .leave_disabled();
+
+    can.set_bitrate(1_000_000);
+
+    can.enable().await;
 
     let mut i: u8 = 0;
     loop {
diff --git a/examples/stm32f4/src/bin/dac.rs b/examples/stm32f4/src/bin/dac.rs
index 3a6216712..aaedcfecc 100644
--- a/examples/stm32f4/src/bin/dac.rs
+++ b/examples/stm32f4/src/bin/dac.rs
@@ -14,11 +14,11 @@ async fn main(_spawner: Spawner) -> ! {
     info!("Hello World, dude!");
 
     let mut dac = DacCh1::new(p.DAC, NoDma, p.PA4);
+    unwrap!(dac.set_trigger_enable(false));
 
     loop {
         for v in 0..=255 {
             unwrap!(dac.set(Value::Bit8(to_sine_wave(v))));
-            dac.trigger();
         }
     }
 }
diff --git a/examples/stm32f4/src/bin/pwm.rs b/examples/stm32f4/src/bin/pwm.rs
index 7c5902052..4f130c26b 100644
--- a/examples/stm32f4/src/bin/pwm.rs
+++ b/examples/stm32f4/src/bin/pwm.rs
@@ -4,9 +4,9 @@
 
 use defmt::*;
 use embassy_executor::Spawner;
-use embassy_stm32::pwm::simple_pwm::{PwmPin, SimplePwm};
-use embassy_stm32::pwm::Channel;
 use embassy_stm32::time::khz;
+use embassy_stm32::timer::simple_pwm::{PwmPin, SimplePwm};
+use embassy_stm32::timer::Channel;
 use embassy_time::{Duration, Timer};
 use {defmt_rtt as _, panic_probe as _};
 
diff --git a/examples/stm32f4/src/bin/pwm_complementary.rs b/examples/stm32f4/src/bin/pwm_complementary.rs
index a8a68ed6e..8cc2a4117 100644
--- a/examples/stm32f4/src/bin/pwm_complementary.rs
+++ b/examples/stm32f4/src/bin/pwm_complementary.rs
@@ -4,10 +4,10 @@
 
 use defmt::*;
 use embassy_executor::Spawner;
-use embassy_stm32::pwm::complementary_pwm::{ComplementaryPwm, ComplementaryPwmPin};
-use embassy_stm32::pwm::simple_pwm::PwmPin;
-use embassy_stm32::pwm::Channel;
 use embassy_stm32::time::khz;
+use embassy_stm32::timer::complementary_pwm::{ComplementaryPwm, ComplementaryPwmPin};
+use embassy_stm32::timer::simple_pwm::PwmPin;
+use embassy_stm32::timer::Channel;
 use embassy_time::{Duration, Timer};
 use {defmt_rtt as _, panic_probe as _};
 
diff --git a/examples/stm32f7/Cargo.toml b/examples/stm32f7/Cargo.toml
index 84d7b79c5..a379cbbe3 100644
--- a/examples/stm32f7/Cargo.toml
+++ b/examples/stm32f7/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32f767zi to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f767zi", "memory-x", "unstable-pac", "time-driver-any", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32f767zi", "unstable-pac", "time-driver-any", "exti"]  }
 embassy-net = { path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet"] }
 embedded-io = { version = "0.4.0", features = ["async"] }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
@@ -27,3 +28,6 @@ rand_core = "0.6.3"
 critical-section = "1.1"
 embedded-storage = "0.3.0"
 static_cell = { version = "1.1", features = ["nightly"]}
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32f7/build.rs b/examples/stm32f7/build.rs
index 2b5d412a9..8cd32d7ed 100644
--- a/examples/stm32f7/build.rs
+++ b/examples/stm32f7/build.rs
@@ -1,43 +1,5 @@
-//! adapted from https://github.com/stm32-rs/stm32f7xx-hal/blob/master/build.rs
-use std::fs::File;
-use std::io::prelude::*;
-use std::path::PathBuf;
-use std::{env, io};
-
-#[derive(Debug)]
-enum Error {
-    Env(env::VarError),
-    Io(io::Error),
-}
-
-impl From<env::VarError> for Error {
-    fn from(error: env::VarError) -> Self {
-        Self::Env(error)
-    }
-}
-
-impl From<io::Error> for Error {
-    fn from(error: io::Error) -> Self {
-        Self::Io(error)
-    }
-}
-
-fn main() -> Result<(), Error> {
-    println!("cargo:rerun-if-changed=build.rs");
-    println!("cargo:rerun-if-changed=memory.x");
-
-    let out_dir = env::var("OUT_DIR")?;
-    let out_dir = PathBuf::from(out_dir);
-
-    let memory_x = include_bytes!("memory.x").as_ref();
-    File::create(out_dir.join("memory.x"))?.write_all(memory_x)?;
-
-    // Tell Cargo where to find the file.
-    println!("cargo:rustc-link-search={}", out_dir.display());
-
+fn main() {
     println!("cargo:rustc-link-arg-bins=--nmagic");
     println!("cargo:rustc-link-arg-bins=-Tlink.x");
     println!("cargo:rustc-link-arg-bins=-Tdefmt.x");
-
-    Ok(())
 }
diff --git a/examples/stm32f7/memory.x b/examples/stm32f7/memory.x
deleted file mode 100644
index 899f7a4b8..000000000
--- a/examples/stm32f7/memory.x
+++ /dev/null
@@ -1,12 +0,0 @@
-/* For STM32F765,767,768,769,777,778,779 devices */
-MEMORY
-{
-  /* NOTE K = KiBi = 1024 bytes */
-  FLASH : ORIGIN = 0x08000000, LENGTH = 2M
-  RAM : ORIGIN = 0x20000000, LENGTH = 368K + 16K
-}
-
-/* This is where the call stack will be allocated. */
-/* The stack is of the full descending type. */
-/* NOTE Do NOT modify `_stack_start` unless you know what you are doing */
-_stack_start = ORIGIN(RAM) + LENGTH(RAM);
diff --git a/examples/stm32g0/Cargo.toml b/examples/stm32g0/Cargo.toml
index c88282d91..b4dfe3c6b 100644
--- a/examples/stm32g0/Cargo.toml
+++ b/examples/stm32g0/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32g071rb to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g071rb", "memory-x", "unstable-pac", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g071rb", "memory-x", "unstable-pac", "exti"]  }
 
 defmt = "0.3"
 defmt-rtt = "0.4"
@@ -19,3 +20,6 @@ embedded-hal = "0.2.6"
 panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 heapless = { version = "0.7.5", default-features = false }
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32g4/Cargo.toml b/examples/stm32g4/Cargo.toml
index 18bd03c39..cf3e2ce9b 100644
--- a/examples/stm32g4/Cargo.toml
+++ b/examples/stm32g4/Cargo.toml
@@ -5,11 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32g491re to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g491re", "memory-x", "unstable-pac", "exti"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "time-driver-any", "stm32g491re", "memory-x", "unstable-pac", "exti"]  }
-embassy-hal-common = {version = "0.1.0", path = "../../embassy-hal-common" }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 
 defmt = "0.3"
@@ -21,3 +21,6 @@ embedded-hal = "0.2.6"
 panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 heapless = { version = "0.7.5", default-features = false }
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32g4/src/bin/pwm.rs b/examples/stm32g4/src/bin/pwm.rs
index 8f7842ed7..b5a9b9952 100644
--- a/examples/stm32g4/src/bin/pwm.rs
+++ b/examples/stm32g4/src/bin/pwm.rs
@@ -4,9 +4,9 @@
 
 use defmt::*;
 use embassy_executor::Spawner;
-use embassy_stm32::pwm::simple_pwm::{PwmPin, SimplePwm};
-use embassy_stm32::pwm::Channel;
 use embassy_stm32::time::khz;
+use embassy_stm32::timer::simple_pwm::{PwmPin, SimplePwm};
+use embassy_stm32::timer::Channel;
 use embassy_time::{Duration, Timer};
 use {defmt_rtt as _, panic_probe as _};
 
diff --git a/examples/stm32h5/Cargo.toml b/examples/stm32h5/Cargo.toml
index 227bc28b4..51d3bad1f 100644
--- a/examples/stm32h5/Cargo.toml
+++ b/examples/stm32h5/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32h563zi to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32h563zi", "memory-x", "time-driver-any", "exti", "unstable-pac", "unstable-traits"] }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "unstable-traits", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32h563zi", "time-driver-any", "exti", "unstable-pac", "unstable-traits"] }
 embassy-net = { path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet", "unstable-traits", "proto-ipv6"] }
 embedded-io = { version = "0.4.0", features = ["async"] }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32h5/memory.x b/examples/stm32h5/memory.x
deleted file mode 100644
index 456061509..000000000
--- a/examples/stm32h5/memory.x
+++ /dev/null
@@ -1,5 +0,0 @@
-MEMORY
-{
-    FLASH : ORIGIN = 0x08000000, LENGTH = 0x200000
-    RAM   : ORIGIN = 0x20000000, LENGTH =  0x50000
-}
diff --git a/examples/stm32h7/.cargo/config.toml b/examples/stm32h7/.cargo/config.toml
index 5f680dbce..4160bf855 100644
--- a/examples/stm32h7/.cargo/config.toml
+++ b/examples/stm32h7/.cargo/config.toml
@@ -1,5 +1,5 @@
 [target.thumbv7em-none-eabihf]
-runner = 'probe-rs run --chip STM32H743ZITx'
+runner = 'probe-rs run --chip STM32H7A3ZITxQ'
 
 [build]
 target = "thumbv7em-none-eabihf" # Cortex-M4F and Cortex-M7F (with FPU)
diff --git a/examples/stm32h7/Cargo.toml b/examples/stm32h7/Cargo.toml
index 768702fa9..3c1232e67 100644
--- a/examples/stm32h7/Cargo.toml
+++ b/examples/stm32h7/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32h743bi to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32h743bi", "time-driver-any", "exti", "memory-x", "unstable-pac", "unstable-traits"] }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "unstable-traits", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32h743bi", "time-driver-any", "exti", "unstable-pac", "unstable-traits"] }
 embassy-net = { path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet", "unstable-traits", "proto-ipv6"] }
 embedded-io = { version = "0.4.0", features = ["async"] }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
diff --git a/examples/stm32h7/memory.x b/examples/stm32h7/memory.x
deleted file mode 100644
index 026b14b9b..000000000
--- a/examples/stm32h7/memory.x
+++ /dev/null
@@ -1,5 +0,0 @@
-MEMORY
-{
-    FLASH : ORIGIN = 0x8000000, LENGTH = 1024K
-    RAM : ORIGIN = 0x24000000, LENGTH = 384K
-}
diff --git a/examples/stm32h7/src/bin/dac.rs b/examples/stm32h7/src/bin/dac.rs
index 586b4154b..ee078286b 100644
--- a/examples/stm32h7/src/bin/dac.rs
+++ b/examples/stm32h7/src/bin/dac.rs
@@ -21,11 +21,11 @@ fn main() -> ! {
     let p = embassy_stm32::init(config);
 
     let mut dac = DacCh1::new(p.DAC1, NoDma, p.PA4);
+    unwrap!(dac.set_trigger_enable(false));
 
     loop {
         for v in 0..=255 {
             unwrap!(dac.set(Value::Bit8(to_sine_wave(v))));
-            dac.trigger();
         }
     }
 }
diff --git a/examples/stm32h7/src/bin/dac_dma.rs b/examples/stm32h7/src/bin/dac_dma.rs
new file mode 100644
index 000000000..a9cb5d1ed
--- /dev/null
+++ b/examples/stm32h7/src/bin/dac_dma.rs
@@ -0,0 +1,140 @@
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+
+use defmt::*;
+use embassy_executor::Spawner;
+use embassy_stm32::dac::{DacChannel, ValueArray};
+use embassy_stm32::pac::timer::vals::{Mms, Opm};
+use embassy_stm32::peripherals::{TIM6, TIM7};
+use embassy_stm32::rcc::low_level::RccPeripheral;
+use embassy_stm32::time::{mhz, Hertz};
+use embassy_stm32::timer::low_level::Basic16bitInstance;
+use micromath::F32Ext;
+use {defmt_rtt as _, panic_probe as _};
+
+pub type Dac1Type =
+    embassy_stm32::dac::DacCh1<'static, embassy_stm32::peripherals::DAC1, embassy_stm32::peripherals::DMA1_CH3>;
+
+pub type Dac2Type =
+    embassy_stm32::dac::DacCh2<'static, embassy_stm32::peripherals::DAC1, embassy_stm32::peripherals::DMA1_CH4>;
+
+#[embassy_executor::main]
+async fn main(spawner: Spawner) {
+    let mut config = embassy_stm32::Config::default();
+    config.rcc.sys_ck = Some(mhz(400));
+    config.rcc.hclk = Some(mhz(100));
+    config.rcc.pll1.q_ck = Some(mhz(100));
+
+    // Initialize the board and obtain a Peripherals instance
+    let p: embassy_stm32::Peripherals = embassy_stm32::init(config);
+
+    // Obtain two independent channels (p.DAC1 can only be consumed once, though!)
+    let (dac_ch1, dac_ch2) = embassy_stm32::dac::Dac::new(p.DAC1, p.DMA1_CH3, p.DMA1_CH4, p.PA4, p.PA5).split();
+
+    spawner.spawn(dac_task1(dac_ch1)).ok();
+    spawner.spawn(dac_task2(dac_ch2)).ok();
+}
+
+#[embassy_executor::task]
+async fn dac_task1(mut dac: Dac1Type) {
+    let data: &[u8; 256] = &calculate_array::<256>();
+
+    info!("TIM6 frequency is {}", TIM6::frequency());
+    const FREQUENCY: Hertz = Hertz::hz(200);
+
+    // Compute the reload value such that we obtain the FREQUENCY for the sine
+    let reload: u32 = (TIM6::frequency().0 / FREQUENCY.0) / data.len() as u32;
+
+    // Depends on your clock and on the specific chip used, you may need higher or lower values here
+    if reload < 10 {
+        error!("Reload value {} below threshold!", reload);
+    }
+
+    dac.select_trigger(embassy_stm32::dac::Ch1Trigger::Tim6).unwrap();
+    dac.enable_channel().unwrap();
+
+    TIM6::enable();
+    TIM6::regs().arr().modify(|w| w.set_arr(reload as u16 - 1));
+    TIM6::regs().cr2().modify(|w| w.set_mms(Mms::UPDATE));
+    TIM6::regs().cr1().modify(|w| {
+        w.set_opm(Opm::DISABLED);
+        w.set_cen(true);
+    });
+
+    debug!(
+        "TIM6 Frequency {}, Target Frequency {}, Reload {}, Reload as u16 {}, Samples {}",
+        TIM6::frequency(),
+        FREQUENCY,
+        reload,
+        reload as u16,
+        data.len()
+    );
+
+    // Loop technically not necessary if DMA circular mode is enabled
+    loop {
+        info!("Loop DAC1");
+        if let Err(e) = dac.write(ValueArray::Bit8(data), true).await {
+            error!("Could not write to dac: {}", e);
+        }
+    }
+}
+
+#[embassy_executor::task]
+async fn dac_task2(mut dac: Dac2Type) {
+    let data: &[u8; 256] = &calculate_array::<256>();
+
+    info!("TIM7 frequency is {}", TIM7::frequency());
+
+    const FREQUENCY: Hertz = Hertz::hz(600);
+    let reload: u32 = (TIM7::frequency().0 / FREQUENCY.0) / data.len() as u32;
+
+    if reload < 10 {
+        error!("Reload value {} below threshold!", reload);
+    }
+
+    TIM7::enable();
+    TIM7::regs().arr().modify(|w| w.set_arr(reload as u16 - 1));
+    TIM7::regs().cr2().modify(|w| w.set_mms(Mms::UPDATE));
+    TIM7::regs().cr1().modify(|w| {
+        w.set_opm(Opm::DISABLED);
+        w.set_cen(true);
+    });
+
+    dac.select_trigger(embassy_stm32::dac::Ch2Trigger::Tim7).unwrap();
+
+    debug!(
+        "TIM7 Frequency {}, Target Frequency {}, Reload {}, Reload as u16 {}, Samples {}",
+        TIM7::frequency(),
+        FREQUENCY,
+        reload,
+        reload as u16,
+        data.len()
+    );
+
+    if let Err(e) = dac.write(ValueArray::Bit8(data), true).await {
+        error!("Could not write to dac: {}", e);
+    }
+}
+
+fn to_sine_wave(v: u8) -> u8 {
+    if v >= 128 {
+        // top half
+        let r = 3.14 * ((v - 128) as f32 / 128.0);
+        (r.sin() * 128.0 + 127.0) as u8
+    } else {
+        // bottom half
+        let r = 3.14 + 3.14 * (v as f32 / 128.0);
+        (r.sin() * 128.0 + 127.0) as u8
+    }
+}
+
+fn calculate_array<const N: usize>() -> [u8; N] {
+    let mut res = [0; N];
+    let mut i = 0;
+    while i < N {
+        res[i] = to_sine_wave(i as u8);
+        i += 1;
+    }
+    res
+}
diff --git a/examples/stm32h7/src/bin/low_level_timer_api.rs b/examples/stm32h7/src/bin/low_level_timer_api.rs
index d360df085..45b0872b5 100644
--- a/examples/stm32h7/src/bin/low_level_timer_api.rs
+++ b/examples/stm32h7/src/bin/low_level_timer_api.rs
@@ -6,8 +6,8 @@ use defmt::*;
 use embassy_executor::Spawner;
 use embassy_stm32::gpio::low_level::AFType;
 use embassy_stm32::gpio::Speed;
-use embassy_stm32::pwm::*;
 use embassy_stm32::time::{khz, mhz, Hertz};
+use embassy_stm32::timer::*;
 use embassy_stm32::{into_ref, Config, Peripheral, PeripheralRef};
 use embassy_time::{Duration, Timer};
 use {defmt_rtt as _, panic_probe as _};
diff --git a/examples/stm32h7/src/bin/pwm.rs b/examples/stm32h7/src/bin/pwm.rs
index c5c0dd290..adf2ea9ce 100644
--- a/examples/stm32h7/src/bin/pwm.rs
+++ b/examples/stm32h7/src/bin/pwm.rs
@@ -4,9 +4,9 @@
 
 use defmt::*;
 use embassy_executor::Spawner;
-use embassy_stm32::pwm::simple_pwm::{PwmPin, SimplePwm};
-use embassy_stm32::pwm::Channel;
 use embassy_stm32::time::{khz, mhz};
+use embassy_stm32::timer::simple_pwm::{PwmPin, SimplePwm};
+use embassy_stm32::timer::Channel;
 use embassy_stm32::Config;
 use embassy_time::{Duration, Timer};
 use {defmt_rtt as _, panic_probe as _};
diff --git a/examples/stm32l0/Cargo.toml b/examples/stm32l0/Cargo.toml
index 747cec7bf..c325751c6 100644
--- a/examples/stm32l0/Cargo.toml
+++ b/examples/stm32l0/Cargo.toml
@@ -10,10 +10,11 @@ nightly = ["embassy-stm32/nightly", "embassy-time/nightly", "embassy-time/unstab
            "embassy-lora", "lora-phy", "lorawan-device", "lorawan", "embedded-io/async"]
 
 [dependencies]
+# Change stm32l072cz to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["defmt", "stm32l072cz", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["defmt", "stm32l072cz", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-lora = { version = "0.1.0", path = "../../embassy-lora", features = ["time", "defmt"], optional = true }
 lora-phy = { version = "1", optional = true }
 lorawan-device = { version = "0.10.0", default-features = false, features = ["async", "external-lora-phy"], optional = true }
@@ -33,5 +34,5 @@ heapless = { version = "0.7.5", default-features = false }
 embedded-hal = "0.2.6"
 static_cell = "1.1"
 
-[patch.crates-io]
-lora-phy = { git = "https://github.com/embassy-rs/lora-phy", rev = "ad289428fd44b02788e2fa2116445cc8f640a265" }
+[profile.release]
+debug = 2
diff --git a/examples/stm32l1/Cargo.toml b/examples/stm32l1/Cargo.toml
index dcca1cc3d..329d44cac 100644
--- a/examples/stm32l1/Cargo.toml
+++ b/examples/stm32l1/Cargo.toml
@@ -20,3 +20,6 @@ panic-probe = { version = "0.3", features = ["print-defmt"] }
 futures = { version = "0.3.17", default-features = false, features = ["async-await"] }
 heapless = { version = "0.7.5", default-features = false }
 embedded-storage = "0.3.0"
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32l4/Cargo.toml b/examples/stm32l4/Cargo.toml
index c55558518..3b27d8e81 100644
--- a/examples/stm32l4/Cargo.toml
+++ b/examples/stm32l4/Cargo.toml
@@ -5,11 +5,12 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32l4s5vi to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l4s5vi", "memory-x", "time-driver-any", "exti", "unstable-traits", "chrono"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-embedded-hal = { version = "0.1.0", path = "../../embassy-embedded-hal" }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l4s5vi", "time-driver-any", "exti", "unstable-traits", "chrono"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 
 defmt = "0.3"
@@ -26,3 +27,6 @@ heapless = { version = "0.7.5", default-features = false }
 chrono = { version = "^0.4", default-features = false }
 
 micromath = "2.0.0"
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32l4/build.rs b/examples/stm32l4/build.rs
index 30691aa97..8cd32d7ed 100644
--- a/examples/stm32l4/build.rs
+++ b/examples/stm32l4/build.rs
@@ -1,34 +1,4 @@
-//! This build script copies the `memory.x` file from the crate root into
-//! a directory where the linker can always find it at build time.
-//! For many projects this is optional, as the linker always searches the
-//! project root directory -- wherever `Cargo.toml` is. However, if you
-//! are using a workspace or have a more complicated build setup, this
-//! build script becomes required. Additionally, by requesting that
-//! Cargo re-run the build script whenever `memory.x` is changed,
-//! updating `memory.x` ensures a rebuild of the application with the
-//! new memory settings.
-
-use std::env;
-use std::fs::File;
-use std::io::Write;
-use std::path::PathBuf;
-
 fn main() {
-    // Put `memory.x` in our output directory and ensure it's
-    // on the linker search path.
-    let out = &PathBuf::from(env::var_os("OUT_DIR").unwrap());
-    File::create(out.join("memory.x"))
-        .unwrap()
-        .write_all(include_bytes!("memory.x"))
-        .unwrap();
-    println!("cargo:rustc-link-search={}", out.display());
-
-    // By default, Cargo will re-run a build script whenever
-    // any file in the project changes. By specifying `memory.x`
-    // here, we ensure the build script is only re-run when
-    // `memory.x` is changed.
-    println!("cargo:rerun-if-changed=memory.x");
-
     println!("cargo:rustc-link-arg-bins=--nmagic");
     println!("cargo:rustc-link-arg-bins=-Tlink.x");
     println!("cargo:rustc-link-arg-bins=-Tdefmt.x");
diff --git a/examples/stm32l4/memory.x b/examples/stm32l4/memory.x
deleted file mode 100644
index eb87d1b54..000000000
--- a/examples/stm32l4/memory.x
+++ /dev/null
@@ -1,7 +0,0 @@
-MEMORY
-{
-  /* NOTE 1 K = 1 KiBi = 1024 bytes */
-  /* These values correspond to the STM32L4S5 */
-  FLASH : ORIGIN = 0x08000000, LENGTH = 1024K
-  RAM : ORIGIN = 0x20000000, LENGTH = 128K
-}
diff --git a/examples/stm32l4/src/bin/dac.rs b/examples/stm32l4/src/bin/dac.rs
index ade43eb35..0193a248e 100644
--- a/examples/stm32l4/src/bin/dac.rs
+++ b/examples/stm32l4/src/bin/dac.rs
@@ -13,11 +13,11 @@ fn main() -> ! {
     info!("Hello World!");
 
     let mut dac = DacCh1::new(p.DAC1, NoDma, p.PA4);
+    unwrap!(dac.set_trigger_enable(false));
 
     loop {
         for v in 0..=255 {
             unwrap!(dac.set(Value::Bit8(to_sine_wave(v))));
-            dac.trigger();
         }
     }
 }
diff --git a/examples/stm32l5/Cargo.toml b/examples/stm32l5/Cargo.toml
index 54911482e..1afd00398 100644
--- a/examples/stm32l5/Cargo.toml
+++ b/examples/stm32l5/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32l552ze to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l552ze", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32l552ze", "time-driver-any", "exti", "unstable-traits", "memory-x"]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 embassy-net = { version = "0.1.0", path = "../../embassy-net", features = ["defmt", "nightly", "tcp", "dhcpv4", "medium-ethernet"] }
 embassy-futures = { version = "0.1.0", path = "../../embassy-futures" }
@@ -26,3 +27,6 @@ heapless = { version = "0.7.5", default-features = false }
 rand_core = { version = "0.6.3", default-features = false }
 embedded-io = { version = "0.4.0", features = ["async"] }
 static_cell = { version = "1.1", features = ["nightly"]}
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32u5/Cargo.toml b/examples/stm32u5/Cargo.toml
index 835e32940..db251eafe 100644
--- a/examples/stm32u5/Cargo.toml
+++ b/examples/stm32u5/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32u585ai to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32u585ai", "time-driver-any", "memory-x" ]  }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "unstable-pac", "stm32u585ai", "time-driver-any", "memory-x" ]  }
 embassy-usb = { version = "0.1.0", path = "../../embassy-usb", features = ["defmt"] }
 
 defmt = "0.3"
@@ -22,3 +23,6 @@ futures = { version = "0.3.17", default-features = false, features = ["async-awa
 heapless = { version = "0.7.5", default-features = false }
 
 micromath = "2.0.0"
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32wb/Cargo.toml b/examples/stm32wb/Cargo.toml
index 7c0b83e65..1a5aff352 100644
--- a/examples/stm32wb/Cargo.toml
+++ b/examples/stm32wb/Cargo.toml
@@ -5,11 +5,12 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32wb55rg to your chip name in both dependencies, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32wb55rg", "time-driver-any", "memory-x", "exti"]  }
+embassy-stm32-wpan = { version = "0.1.0", path = "../../embassy-stm32-wpan", features = ["defmt", "stm32wb55rg"] }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "defmt", "stm32wb55rg", "time-driver-any", "memory-x", "exti"]  }
-embassy-stm32-wpan = { version = "0.1.0", path = "../../embassy-stm32-wpan", features = ["defmt", "stm32wb55rg"] }
 embassy-net = { version = "0.1.0", path = "../../embassy-net", features = ["defmt", "udp", "medium-ieee802154", "nightly"], optional=true }
 
 defmt = "0.3"
@@ -50,4 +51,7 @@ required-features = ["ble"]
 
 [[bin]] 
 name = "gatt_server"
-required-features = ["ble"]
\ No newline at end of file
+required-features = ["ble"]
+
+[profile.release]
+debug = 2
diff --git a/examples/stm32wl/Cargo.toml b/examples/stm32wl/Cargo.toml
index e2c66f456..48b69c8d0 100644
--- a/examples/stm32wl/Cargo.toml
+++ b/examples/stm32wl/Cargo.toml
@@ -5,10 +5,11 @@ version = "0.1.0"
 license = "MIT OR Apache-2.0"
 
 [dependencies]
+# Change stm32wl55jc-cm4 to your chip name, if necessary.
+embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "unstable-traits", "defmt", "stm32wl55jc-cm4", "time-driver-any", "memory-x", "unstable-pac", "exti", "chrono"] }
 embassy-sync = { version = "0.2.0", path = "../../embassy-sync", features = ["defmt"] }
 embassy-executor = { version = "0.2.0", path = "../../embassy-executor", features = ["nightly", "arch-cortex-m", "executor-thread", "defmt", "integrated-timers"] }
 embassy-time = { version = "0.1.2", path = "../../embassy-time", features = ["nightly", "unstable-traits", "defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
-embassy-stm32 = { version = "0.1.0", path = "../../embassy-stm32", features = ["nightly", "unstable-traits", "defmt", "stm32wl55jc-cm4", "time-driver-any", "memory-x", "unstable-pac", "exti", "chrono"] }
 embassy-embedded-hal = { version = "0.1.0", path = "../../embassy-embedded-hal" }
 embassy-lora = { version = "0.1.0", path = "../../embassy-lora", features = ["stm32wl", "time", "defmt"] }
 lora-phy = { version = "1" }
@@ -27,5 +28,5 @@ futures = { version = "0.3.17", default-features = false, features = ["async-awa
 heapless = { version = "0.7.5", default-features = false }
 chrono = { version = "^0.4", default-features = false }
 
-[patch.crates-io]
-lora-phy = { git = "https://github.com/embassy-rs/lora-phy", rev = "ad289428fd44b02788e2fa2116445cc8f640a265" }
+[profile.release]
+debug = 2
diff --git a/examples/wasm/Cargo.toml b/examples/wasm/Cargo.toml
index 3679e3857..2791cc341 100644
--- a/examples/wasm/Cargo.toml
+++ b/examples/wasm/Cargo.toml
@@ -17,3 +17,6 @@ wasm-bindgen = "0.2"
 web-sys = { version = "0.3", features = ["Document", "Element", "HtmlElement", "Node", "Window" ] }
 log = "0.4.11"
 critical-section = { version = "1.1", features = ["std"] }
+
+[profile.release]
+debug = 2
diff --git a/tests/nrf/src/bin/wifi_esp_hosted_perf.rs b/tests/nrf/src/bin/wifi_esp_hosted_perf.rs
index 398ab9d27..e2adfe0be 100644
--- a/tests/nrf/src/bin/wifi_esp_hosted_perf.rs
+++ b/tests/nrf/src/bin/wifi_esp_hosted_perf.rs
@@ -118,9 +118,9 @@ const WIFI_NETWORK: &str = "EmbassyTest";
 const WIFI_PASSWORD: &str = "V8YxhKt5CdIAJFud";
 
 const TEST_DURATION: usize = 10;
-const TEST_EXPECTED_DOWNLOAD_KBPS: usize = 150;
-const TEST_EXPECTED_UPLOAD_KBPS: usize = 150;
-const TEST_EXPECTED_UPLOAD_DOWNLOAD_KBPS: usize = 150;
+const TEST_EXPECTED_DOWNLOAD_KBPS: usize = 50;
+const TEST_EXPECTED_UPLOAD_KBPS: usize = 50;
+const TEST_EXPECTED_UPLOAD_DOWNLOAD_KBPS: usize = 50;
 const RX_BUFFER_SIZE: usize = 4096;
 const TX_BUFFER_SIZE: usize = 4096;
 const SERVER_ADDRESS: Ipv4Address = Ipv4Address::new(192, 168, 2, 2);
diff --git a/tests/rp/src/bin/cyw43-perf.rs b/tests/rp/src/bin/cyw43-perf.rs
index bc127e2e5..fffdabc9b 100644
--- a/tests/rp/src/bin/cyw43-perf.rs
+++ b/tests/rp/src/bin/cyw43-perf.rs
@@ -48,9 +48,9 @@ async fn main(spawner: Spawner) {
     }
 
     // cyw43 firmware needs to be flashed manually:
-    //     probe-rs download 43439A0.bin     --format bin --chip RP2040 --base-address 0x101c0000
+    //     probe-rs download 43439A0.bin     --format bin --chip RP2040 --base-address 0x101b0000
     //     probe-rs download 43439A0_clm.bin --format bin --chip RP2040 --base-address 0x101f8000
-    let fw = unsafe { core::slice::from_raw_parts(0x101c0000 as *const u8, 224190) };
+    let fw = unsafe { core::slice::from_raw_parts(0x101b0000 as *const u8, 230321) };
     let clm = unsafe { core::slice::from_raw_parts(0x101f8000 as *const u8, 4752) };
 
     let pwr = Output::new(p.PIN_23, Level::Low);
diff --git a/tests/rp/src/bin/pio_irq.rs b/tests/rp/src/bin/pio_irq.rs
index 45004424a..bdea63eaa 100644
--- a/tests/rp/src/bin/pio_irq.rs
+++ b/tests/rp/src/bin/pio_irq.rs
@@ -9,7 +9,6 @@ use embassy_executor::Spawner;
 use embassy_rp::bind_interrupts;
 use embassy_rp::peripherals::PIO0;
 use embassy_rp::pio::{Config, InterruptHandler, Pio};
-use embassy_rp::relocate::RelocatedProgram;
 use {defmt_rtt as _, panic_probe as _};
 
 bind_interrupts!(struct Irqs {
@@ -35,9 +34,8 @@ async fn main(_spawner: Spawner) {
         "irq wait 1",
     );
 
-    let relocated = RelocatedProgram::new(&prg.program);
     let mut cfg = Config::default();
-    cfg.use_program(&common.load_program(&relocated), &[]);
+    cfg.use_program(&common.load_program(&prg.program), &[]);
     sm.set_config(&cfg);
     sm.set_enable(true);
 
diff --git a/tests/rp/src/bin/pio_multi_load.rs b/tests/rp/src/bin/pio_multi_load.rs
new file mode 100644
index 000000000..356f16795
--- /dev/null
+++ b/tests/rp/src/bin/pio_multi_load.rs
@@ -0,0 +1,126 @@
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+#[path = "../common.rs"]
+mod common;
+
+use defmt::info;
+use embassy_executor::Spawner;
+use embassy_rp::bind_interrupts;
+use embassy_rp::peripherals::PIO0;
+use embassy_rp::pio::{Config, InterruptHandler, LoadError, Pio};
+use {defmt_rtt as _, panic_probe as _};
+
+bind_interrupts!(struct Irqs {
+    PIO0_IRQ_0 => InterruptHandler<PIO0>;
+});
+
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    let p = embassy_rp::init(Default::default());
+    let pio = p.PIO0;
+    let Pio {
+        mut common,
+        mut sm0,
+        mut sm1,
+        mut sm2,
+        irq_flags,
+        ..
+    } = Pio::new(pio, Irqs);
+
+    // load with explicit origin works
+    let prg1 = pio_proc::pio_asm!(
+        ".origin 4"
+        "nop",
+        "nop",
+        "nop",
+        "nop",
+        "nop",
+        "nop",
+        "nop",
+        "irq 0",
+        "nop",
+        "nop",
+    );
+    let loaded1 = common.load_program(&prg1.program);
+    assert_eq!(loaded1.origin, 4);
+    assert_eq!(loaded1.wrap.source, 13);
+    assert_eq!(loaded1.wrap.target, 4);
+
+    // load without origin chooses a free space
+    let prg2 = pio_proc::pio_asm!("nop", "nop", "nop", "nop", "nop", "nop", "nop", "irq 1", "nop", "nop",);
+    let loaded2 = common.load_program(&prg2.program);
+    assert_eq!(loaded2.origin, 14);
+    assert_eq!(loaded2.wrap.source, 23);
+    assert_eq!(loaded2.wrap.target, 14);
+
+    // wrapping around the end of program space automatically works
+    let prg3 =
+        pio_proc::pio_asm!("nop", "nop", "nop", "nop", "nop", "nop", "nop", "nop", "nop", "nop", "nop", "irq 2",);
+    let loaded3 = common.load_program(&prg3.program);
+    assert_eq!(loaded3.origin, 24);
+    assert_eq!(loaded3.wrap.source, 3);
+    assert_eq!(loaded3.wrap.target, 24);
+
+    // check that the programs actually work
+    {
+        let mut cfg = Config::default();
+        cfg.use_program(&loaded1, &[]);
+        sm0.set_config(&cfg);
+        sm0.set_enable(true);
+        while !irq_flags.check(0) {}
+        sm0.set_enable(false);
+    }
+    {
+        let mut cfg = Config::default();
+        cfg.use_program(&loaded2, &[]);
+        sm1.set_config(&cfg);
+        sm1.set_enable(true);
+        while !irq_flags.check(1) {}
+        sm1.set_enable(false);
+    }
+    {
+        let mut cfg = Config::default();
+        cfg.use_program(&loaded3, &[]);
+        sm2.set_config(&cfg);
+        sm2.set_enable(true);
+        while !irq_flags.check(2) {}
+        sm2.set_enable(false);
+    }
+
+    // instruction memory is full now. all loads should fail.
+    {
+        let prg = pio_proc::pio_asm!(".origin 0", "nop");
+        match common.try_load_program(&prg.program) {
+            Err(LoadError::AddressInUse(0)) => (),
+            _ => panic!("program loaded when it shouldn't"),
+        };
+
+        let prg = pio_proc::pio_asm!("nop");
+        match common.try_load_program(&prg.program) {
+            Err(LoadError::InsufficientSpace) => (),
+            _ => panic!("program loaded when it shouldn't"),
+        };
+    }
+
+    // freeing some memory should allow further loads though.
+    unsafe {
+        common.free_instr(loaded3.used_memory);
+    }
+    {
+        let prg = pio_proc::pio_asm!(".origin 0", "nop");
+        match common.try_load_program(&prg.program) {
+            Ok(_) => (),
+            _ => panic!("program didn't loaded when it shouldn"),
+        };
+
+        let prg = pio_proc::pio_asm!("nop");
+        match common.try_load_program(&prg.program) {
+            Ok(_) => (),
+            _ => panic!("program didn't loaded when it shouldn"),
+        };
+    }
+
+    info!("Test OK");
+    cortex_m::asm::bkpt();
+}
diff --git a/tests/stm32/Cargo.toml b/tests/stm32/Cargo.toml
index 3007cd1e6..17320649e 100644
--- a/tests/stm32/Cargo.toml
+++ b/tests/stm32/Cargo.toml
@@ -7,11 +7,11 @@ autobins = false
 
 [features]
 stm32f103c8 = ["embassy-stm32/stm32f103c8", "not-gpdma"]     # Blue Pill
-stm32f429zi = ["embassy-stm32/stm32f429zi", "chrono", "can", "not-gpdma"]     # Nucleo "sdmmc"
-stm32g071rb = ["embassy-stm32/stm32g071rb", "not-gpdma"]     # Nucleo
+stm32f429zi = ["embassy-stm32/stm32f429zi", "chrono", "can", "not-gpdma", "dac-adc-pin"]     # Nucleo "sdmmc"
+stm32g071rb = ["embassy-stm32/stm32g071rb", "not-gpdma", "dac-adc-pin"]     # Nucleo
 stm32c031c6 = ["embassy-stm32/stm32c031c6", "not-gpdma"]     # Nucleo
 stm32g491re = ["embassy-stm32/stm32g491re", "not-gpdma"]     # Nucleo
-stm32h755zi = ["embassy-stm32/stm32h755zi-cm7", "not-gpdma"] # Nucleo
+stm32h755zi = ["embassy-stm32/stm32h755zi-cm7", "not-gpdma", "dac-adc-pin"] # Nucleo
 stm32wb55rg = ["embassy-stm32/stm32wb55rg", "not-gpdma", "ble", "mac" ]     # Nucleo
 stm32h563zi = ["embassy-stm32/stm32h563zi"]     # Nucleo
 stm32u585ai = ["embassy-stm32/stm32u585ai"]     # IoT board
@@ -23,6 +23,7 @@ ble = ["dep:embassy-stm32-wpan", "embassy-stm32-wpan/ble"]
 mac = ["dep:embassy-stm32-wpan", "embassy-stm32-wpan/mac"]
 embassy-stm32-wpan = []
 not-gpdma = []
+dac-adc-pin = []
 
 [dependencies]
 teleprobe-meta = "1"
@@ -42,6 +43,7 @@ cortex-m-rt = "0.7.0"
 embedded-hal = "0.2.6"
 embedded-hal-1 = { package = "embedded-hal", version = "=1.0.0-alpha.11" }
 embedded-hal-async = { version = "=0.2.0-alpha.2" }
+micromath = "2.0.0"
 panic-probe = { version = "0.3.0", features = ["print-defmt"] }
 rand_core = { version = "0.6", default-features = false }
 rand_chacha = { version = "0.3", default-features = false }
@@ -56,6 +58,11 @@ path = "src/bin/can.rs"
 required-features = [ "can",]
 
 [[bin]]
+name = "dac"
+path = "src/bin/dac.rs"
+required-features = [ "dac-adc-pin",]
+
+[[bin]]
 name = "gpio"
 path = "src/bin/gpio.rs"
 required-features = []
diff --git a/tests/stm32/src/bin/dac.rs b/tests/stm32/src/bin/dac.rs
new file mode 100644
index 000000000..67a7d5b59
--- /dev/null
+++ b/tests/stm32/src/bin/dac.rs
@@ -0,0 +1,81 @@
+#![no_std]
+#![no_main]
+#![feature(type_alias_impl_trait)]
+
+// required-features: dac-adc-pin
+
+#[path = "../common.rs"]
+mod common;
+use common::*;
+use defmt::assert;
+use embassy_executor::Spawner;
+use embassy_stm32::adc::Adc;
+use embassy_stm32::dac::{DacCh1, DacChannel, Value};
+use embassy_stm32::dma::NoDma;
+use embassy_time::{Delay, Duration, Timer};
+use {defmt_rtt as _, panic_probe as _};
+
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    // Initialize the board and obtain a Peripherals instance
+    let p: embassy_stm32::Peripherals = embassy_stm32::init(config());
+
+    #[cfg(feature = "stm32f429zi")]
+    let dac_peripheral = p.DAC;
+
+    #[cfg(any(feature = "stm32h755zi", feature = "stm32g071rb"))]
+    let dac_peripheral = p.DAC1;
+
+    let mut dac: DacCh1<'_, _, NoDma> = DacCh1::new(dac_peripheral, NoDma, p.PA4);
+    unwrap!(dac.set_trigger_enable(false));
+
+    let mut adc = Adc::new(p.ADC1, &mut Delay);
+
+    #[cfg(feature = "stm32h755zi")]
+    let normalization_factor = 256;
+    #[cfg(any(feature = "stm32f429zi", feature = "stm32g071rb"))]
+    let normalization_factor: i32 = 16;
+
+    unwrap!(dac.set(Value::Bit8(0)));
+    // Now wait a little to obtain a stable value
+    Timer::after(Duration::from_millis(30)).await;
+    let offset = adc.read(&mut unsafe { embassy_stm32::Peripherals::steal() }.PA4);
+
+    for v in 0..=255 {
+        // First set the DAC output value
+        let dac_output_val = to_sine_wave(v);
+        unwrap!(dac.set(Value::Bit8(dac_output_val)));
+
+        // Now wait a little to obtain a stable value
+        Timer::after(Duration::from_millis(30)).await;
+
+        // Need to steal the peripherals here because PA4 is obviously in use already
+        let measured = adc.read(&mut unsafe { embassy_stm32::Peripherals::steal() }.PA4);
+        // Calibrate and normalize the measurement to get close to the dac_output_val
+        let measured_normalized = ((measured as i32 - offset as i32) / normalization_factor) as i16;
+
+        info!("value / measured: {} / {}", dac_output_val, measured_normalized);
+
+        // The deviations are quite enormous but that does not matter since this is only a quick test
+        assert!((dac_output_val as i16 - measured_normalized).abs() < 15);
+    }
+
+    info!("Test OK");
+    cortex_m::asm::bkpt();
+}
+
+use core::f32::consts::PI;
+
+use micromath::F32Ext;
+
+fn to_sine_wave(v: u8) -> u8 {
+    if v >= 128 {
+        // top half
+        let r = PI * ((v - 128) as f32 / 128.0);
+        (r.sin() * 128.0 + 127.0) as u8
+    } else {
+        // bottom half
+        let r = PI + PI * (v as f32 / 128.0);
+        (r.sin() * 128.0 + 127.0) as u8
+    }
+}
diff --git a/tests/stm32/src/bin/wpan_ble.rs b/tests/stm32/src/bin/wpan_ble.rs
index 3ad8aca4e..452da77a4 100644
--- a/tests/stm32/src/bin/wpan_ble.rs
+++ b/tests/stm32/src/bin/wpan_ble.rs
@@ -12,6 +12,7 @@ use common::*;
 use embassy_executor::Spawner;
 use embassy_stm32::bind_interrupts;
 use embassy_stm32::ipcc::{Config, ReceiveInterruptHandler, TransmitInterruptHandler};
+use embassy_stm32::rcc::WPAN_DEFAULT;
 use embassy_stm32_wpan::hci::host::uart::UartHci;
 use embassy_stm32_wpan::hci::host::{AdvertisingFilterPolicy, EncryptionKey, HostHci, OwnAddressType};
 use embassy_stm32_wpan::hci::types::AdvertisingType;
@@ -40,7 +41,10 @@ async fn run_mm_queue(memory_manager: mm::MemoryManager) {
 
 #[embassy_executor::main]
 async fn main(spawner: Spawner) {
-    let p = embassy_stm32::init(config());
+    let mut config = config();
+    config.rcc = WPAN_DEFAULT;
+
+    let p = embassy_stm32::init(config);
     info!("Hello World!");
 
     let config = Config::default();
diff --git a/tests/stm32/src/bin/wpan_mac.rs b/tests/stm32/src/bin/wpan_mac.rs
index b04a19ee9..7eab2fd38 100644
--- a/tests/stm32/src/bin/wpan_mac.rs
+++ b/tests/stm32/src/bin/wpan_mac.rs
@@ -10,6 +10,7 @@ use common::*;
 use embassy_executor::Spawner;
 use embassy_stm32::bind_interrupts;
 use embassy_stm32::ipcc::{Config, ReceiveInterruptHandler, TransmitInterruptHandler};
+use embassy_stm32::rcc::WPAN_DEFAULT;
 use embassy_stm32_wpan::mac::commands::{AssociateRequest, GetRequest, ResetRequest, SetRequest};
 use embassy_stm32_wpan::mac::event::MacEvent;
 use embassy_stm32_wpan::mac::typedefs::{
@@ -31,7 +32,10 @@ async fn run_mm_queue(memory_manager: mm::MemoryManager) {
 
 #[embassy_executor::main]
 async fn main(spawner: Spawner) {
-    let p = embassy_stm32::init(config());
+    let mut config = config();
+    config.rcc = WPAN_DEFAULT;
+
+    let p = embassy_stm32::init(config);
     info!("Hello World!");
 
     let config = Config::default();
diff --git a/tests/stm32/src/common.rs b/tests/stm32/src/common.rs
index 3d2a9b8ef..ca5cb43ac 100644
--- a/tests/stm32/src/common.rs
+++ b/tests/stm32/src/common.rs
@@ -33,6 +33,7 @@ pub fn config() -> Config {
     {
         config.rcc.sys_ck = Some(Hertz(400_000_000));
         config.rcc.pll1.q_ck = Some(Hertz(100_000_000));
+        config.rcc.adc_clock_source = embassy_stm32::rcc::AdcClockSource::PerCk;
     }
 
     #[cfg(feature = "stm32u585ai")]