43 files changed, 955 insertions, 377 deletions

diff --git a/embassy-nrf/CHANGELOG.md b/embassy-nrf/CHANGELOG.md
index be79bde5d..cfb040ef5 100644
--- a/embassy-nrf/CHANGELOG.md
+++ b/embassy-nrf/CHANGELOG.md
@@ -28,6 +28,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - changed: `gpiote::InputChannel::wait()` now ensures events are seen as soon as the function is called, even if the future is not polled
 - bugfix: use correct flash size for nRF54l
 - changed: add workaround for anomaly 66 on nrf52
+- added: expose PPI events available on SPIS peripheral
 
 ## 0.8.0 - 2025-09-30
 
diff --git a/embassy-nrf/src/spis.rs b/embassy-nrf/src/spis.rs
index 96a9c0ae0..6f837c317 100644
--- a/embassy-nrf/src/spis.rs
+++ b/embassy-nrf/src/spis.rs
@@ -17,6 +17,7 @@ use crate::gpio::{self, AnyPin, OutputDrive, Pin as GpioPin, SealedPin as _, con
 use crate::interrupt::typelevel::Interrupt;
 use crate::pac::gpio::vals as gpiovals;
 use crate::pac::spis::vals;
+use crate::ppi::Event;
 use crate::util::slice_in_ram_or;
 use crate::{interrupt, pac};
 
@@ -334,6 +335,20 @@ impl<'d> Spis<'d> {
         Ok((n_rx, n_tx))
     }
 
+    /// Returns the ACQUIRED event, for use with PPI.
+    ///
+    /// This event will fire when the semaphore is acquired.
+    pub fn event_acquired(&self) -> Event<'d> {
+        Event::from_reg(self.r.events_acquired())
+    }
+
+    /// Returns the END event, for use with PPI.
+    ///
+    /// This event will fire when the slave transaction is complete.
+    pub fn event_end(&self) -> Event<'d> {
+        Event::from_reg(self.r.events_end())
+    }
+
     async fn async_inner(&mut self, rx: &mut [u8], tx: &[u8]) -> Result<(usize, usize), Error> {
         match self.async_inner_from_ram(rx, tx).await {
             Ok(n) => Ok(n),
diff --git a/embassy-rp/CHANGELOG.md b/embassy-rp/CHANGELOG.md
index 4b0d738a7..7480b729f 100644
--- a/embassy-rp/CHANGELOG.md
+++ b/embassy-rp/CHANGELOG.md
@@ -18,6 +18,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Add reset_to_usb_boot for rp235x ([#4705](https://github.com/embassy-rs/embassy/pull/4705))
 - Add fix #4822 in PIO onewire. Change to disable the state machine before setting y register ([#4824](https://github.com/embassy-rs/embassy/pull/4824))
 - Add PIO::Ws2812 color order support
+- Add TX-only, no SCK SPI support
+- Remove atomic-polyfill with critical-section instead ([#4948](https://github.com/embassy-rs/embassy/pull/4948))
 
 ## 0.8.0 - 2025-08-26
 
@@ -115,4 +117,3 @@ Small release fixing a few gnarly bugs, upgrading is strongly recommended.
 - rename the Channel trait to Slice and the PwmPin to PwmChannel
 - i2c: Fix race condition that appears on fast repeated transfers.
 - Add a basic "read to break" function
-
diff --git a/embassy-rp/Cargo.toml b/embassy-rp/Cargo.toml
index 9ad4b47a3..421f0b0f6 100644
--- a/embassy-rp/Cargo.toml
+++ b/embassy-rp/Cargo.toml
@@ -47,7 +47,7 @@ rt = [ "rp-pac/rt" ]
 defmt = ["dep:defmt", "embassy-usb-driver/defmt", "embassy-hal-internal/defmt"]
 ## Enable log support
 log = ["dep:log"]
-## Enable chrono support  
+## Enable chrono support
 chrono = ["dep:chrono"]
 
 ## Configure the [`critical-section`](https://docs.rs/critical-section) crate to use an implementation that is safe for multicore use on rp2040.
@@ -159,7 +159,6 @@ embassy-futures = { version = "0.1.2", path = "../embassy-futures" }
 embassy-hal-internal = { version = "0.3.0", path = "../embassy-hal-internal", features = ["cortex-m", "prio-bits-2"] }
 embassy-embedded-hal = { version = "0.5.0", path = "../embassy-embedded-hal" }
 embassy-usb-driver = { version = "0.2.0", path = "../embassy-usb-driver" }
-atomic-polyfill = "1.0.1"
 defmt = { version = "1.0.1", optional = true }
 log = { version = "0.4.14", optional = true }
 nb = "1.1.0"
diff --git a/embassy-rp/src/pio/mod.rs b/embassy-rp/src/pio/mod.rs
index 92b2c603e..1c370fdfc 100644
--- a/embassy-rp/src/pio/mod.rs
+++ b/embassy-rp/src/pio/mod.rs
@@ -2,10 +2,9 @@
 use core::future::Future;
 use core::marker::PhantomData;
 use core::pin::Pin as FuturePin;
-use core::sync::atomic::{Ordering, compiler_fence};
+use core::sync::atomic::{AtomicU8, AtomicU32, Ordering, compiler_fence};
 use core::task::{Context, Poll};
 
-use atomic_polyfill::{AtomicU8, AtomicU64};
 use embassy_hal_internal::{Peri, PeripheralType};
 use embassy_sync::waitqueue::AtomicWaker;
 use fixed::FixedU32;
@@ -1232,7 +1231,13 @@ impl<'d, PIO: Instance> Common<'d, PIO> {
             w.set_pde(false);
         });
         // we can be relaxed about this because we're &mut here and nothing is cached
-        PIO::state().used_pins.fetch_or(1 << pin.pin_bank(), Ordering::Relaxed);
+        critical_section::with(|_| {
+            let val = PIO::state().used_pins.load(Ordering::Relaxed);
+            PIO::state()
+                .used_pins
+                .store(val | 1 << pin.pin_bank(), Ordering::Relaxed);
+        });
+
         Pin {
             pin: pin.into(),
             pio: PhantomData::default(),
@@ -1404,6 +1409,42 @@ impl<'d, PIO: Instance> Pio<'d, PIO> {
     }
 }
 
+struct AtomicU64 {
+    upper_32: AtomicU32,
+    lower_32: AtomicU32,
+}
+
+impl AtomicU64 {
+    const fn new(val: u64) -> Self {
+        let upper_32 = (val >> 32) as u32;
+        let lower_32 = val as u32;
+
+        Self {
+            upper_32: AtomicU32::new(upper_32),
+            lower_32: AtomicU32::new(lower_32),
+        }
+    }
+
+    fn load(&self, order: Ordering) -> u64 {
+        let (upper, lower) = critical_section::with(|_| (self.upper_32.load(order), self.lower_32.load(order)));
+
+        let upper = (upper as u64) << 32;
+        let lower = lower as u64;
+
+        upper | lower
+    }
+
+    fn store(&self, val: u64, order: Ordering) {
+        let upper_32 = (val >> 32) as u32;
+        let lower_32 = val as u32;
+
+        critical_section::with(|_| {
+            self.upper_32.store(upper_32, order);
+            self.lower_32.store(lower_32, order);
+        });
+    }
+}
+
 /// Representation of the PIO state keeping a record of which pins are assigned to
 /// each PIO.
 // make_pio_pin notionally takes ownership of the pin it is given, but the wrapped pin
@@ -1418,7 +1459,12 @@ pub struct State {
 
 fn on_pio_drop<PIO: Instance>() {
     let state = PIO::state();
-    if state.users.fetch_sub(1, Ordering::AcqRel) == 1 {
+    let users_state = critical_section::with(|_| {
+        let val = state.users.load(Ordering::Acquire) - 1;
+        state.users.store(val, Ordering::Release);
+        val
+    });
+    if users_state == 1 {
         let used_pins = state.used_pins.load(Ordering::Relaxed);
         let null = pac::io::vals::Gpio0ctrlFuncsel::NULL as _;
         for i in 0..crate::gpio::BANK0_PIN_COUNT {
diff --git a/embassy-rp/src/spi.rs b/embassy-rp/src/spi.rs
index d9410e78d..39f128214 100644
--- a/embassy-rp/src/spi.rs
+++ b/embassy-rp/src/spi.rs
@@ -308,6 +308,11 @@ impl<'d, T: Instance> Spi<'d, T, Blocking> {
         )
     }
 
+    /// Create an SPI driver in blocking mode supporting writes only, without SCK pin.
+    pub fn new_blocking_txonly_nosck(inner: Peri<'d, T>, mosi: Peri<'d, impl MosiPin<T> + 'd>, config: Config) -> Self {
+        Self::new_inner(inner, None, Some(mosi.into()), None, None, None, None, config)
+    }
+
     /// Create an SPI driver in blocking mode supporting reads only.
     pub fn new_blocking_rxonly(
         inner: Peri<'d, T>,
@@ -371,6 +376,26 @@ impl<'d, T: Instance> Spi<'d, T, Async> {
         )
     }
 
+    /// Create an SPI driver in async mode supporting DMA write operations only,
+    /// without SCK pin.
+    pub fn new_txonly_nosck(
+        inner: Peri<'d, T>,
+        mosi: Peri<'d, impl MosiPin<T> + 'd>,
+        tx_dma: Peri<'d, impl Channel>,
+        config: Config,
+    ) -> Self {
+        Self::new_inner(
+            inner,
+            None,
+            Some(mosi.into()),
+            None,
+            None,
+            Some(tx_dma.into()),
+            None,
+            config,
+        )
+    }
+
     /// Create an SPI driver in async mode supporting DMA read operations only.
     pub fn new_rxonly(
         inner: Peri<'d, T>,
diff --git a/embassy-rp/src/uart/buffered.rs b/embassy-rp/src/uart/buffered.rs
index 02649ad81..fdb8ce776 100644
--- a/embassy-rp/src/uart/buffered.rs
+++ b/embassy-rp/src/uart/buffered.rs
@@ -1,8 +1,8 @@
 //! Buffered UART driver.
 use core::future::Future;
 use core::slice;
+use core::sync::atomic::{AtomicU8, Ordering};
 
-use atomic_polyfill::AtomicU8;
 use embassy_hal_internal::atomic_ring_buffer::RingBuffer;
 
 use super::*;
@@ -241,7 +241,11 @@ impl BufferedUartRx {
     }
 
     fn get_rx_error(state: &State) -> Option<Error> {
-        let errs = state.rx_error.swap(0, Ordering::Relaxed);
+        let errs = critical_section::with(|_| {
+            let val = state.rx_error.load(Ordering::Relaxed);
+            state.rx_error.store(0, Ordering::Relaxed);
+            val
+        });
         if errs & RXE_OVERRUN != 0 {
             Some(Error::Overrun)
         } else if errs & RXE_BREAK != 0 {
@@ -555,7 +559,10 @@ impl<T: Instance> interrupt::typelevel::Handler<T::Interrupt> for BufferedInterr
                 }
                 let dr = r.uartdr().read();
                 if (dr.0 >> 8) != 0 {
-                    s.rx_error.fetch_or((dr.0 >> 8) as u8, Ordering::Relaxed);
+                    critical_section::with(|_| {
+                        let val = s.rx_error.load(Ordering::Relaxed);
+                        s.rx_error.store(val | ((dr.0 >> 8) as u8), Ordering::Relaxed);
+                    });
                     error = true;
                     // only fill the buffer with valid characters. the current character is fine
                     // if the error is an overrun, but if we add it to the buffer we'll report
diff --git a/embassy-rp/src/uart/mod.rs b/embassy-rp/src/uart/mod.rs
index 8be87a5d2..b7b569dd5 100644
--- a/embassy-rp/src/uart/mod.rs
+++ b/embassy-rp/src/uart/mod.rs
@@ -1,9 +1,9 @@
 //! UART driver.
 use core::future::poll_fn;
 use core::marker::PhantomData;
+use core::sync::atomic::{AtomicU16, Ordering};
 use core::task::Poll;
 
-use atomic_polyfill::{AtomicU16, Ordering};
 use embassy_futures::select::{Either, select};
 use embassy_hal_internal::{Peri, PeripheralType};
 use embassy_sync::waitqueue::AtomicWaker;
@@ -456,7 +456,12 @@ impl<'d> UartRx<'d, Async> {
             transfer,
             poll_fn(|cx| {
                 self.dma_state.rx_err_waker.register(cx.waker());
-                match self.dma_state.rx_errs.swap(0, Ordering::Relaxed) {
+                let rx_errs = critical_section::with(|_| {
+                    let val = self.dma_state.rx_errs.load(Ordering::Relaxed);
+                    self.dma_state.rx_errs.store(0, Ordering::Relaxed);
+                    val
+                });
+                match rx_errs {
                     0 => Poll::Pending,
                     e => Poll::Ready(Uartris(e as u32)),
                 }
@@ -468,7 +473,11 @@ impl<'d> UartRx<'d, Async> {
             Either::First(()) => {
                 // We're here because the DMA finished, BUT if an error occurred on the LAST
                 // byte, then we may still need to grab the error state!
-                Uartris(self.dma_state.rx_errs.swap(0, Ordering::Relaxed) as u32)
+                Uartris(critical_section::with(|_| {
+                    let val = self.dma_state.rx_errs.load(Ordering::Relaxed);
+                    self.dma_state.rx_errs.store(0, Ordering::Relaxed);
+                    val
+                }) as u32)
             }
             Either::Second(e) => {
                 // We're here because we errored, which means this is the error that
@@ -616,7 +625,12 @@ impl<'d> UartRx<'d, Async> {
                 transfer,
                 poll_fn(|cx| {
                     self.dma_state.rx_err_waker.register(cx.waker());
-                    match self.dma_state.rx_errs.swap(0, Ordering::Relaxed) {
+                    let rx_errs = critical_section::with(|_| {
+                        let val = self.dma_state.rx_errs.load(Ordering::Relaxed);
+                        self.dma_state.rx_errs.store(0, Ordering::Relaxed);
+                        val
+                    });
+                    match rx_errs {
                         0 => Poll::Pending,
                         e => Poll::Ready(Uartris(e as u32)),
                     }
@@ -629,7 +643,11 @@ impl<'d> UartRx<'d, Async> {
                 Either::First(()) => {
                     // We're here because the DMA finished, BUT if an error occurred on the LAST
                     // byte, then we may still need to grab the error state!
-                    Uartris(self.dma_state.rx_errs.swap(0, Ordering::Relaxed) as u32)
+                    Uartris(critical_section::with(|_| {
+                        let val = self.dma_state.rx_errs.load(Ordering::Relaxed);
+                        self.dma_state.rx_errs.store(0, Ordering::Relaxed);
+                        val
+                    }) as u32)
                 }
                 Either::Second(e) => {
                     // We're here because we errored, which means this is the error that
diff --git a/embassy-stm32/CHANGELOG.md b/embassy-stm32/CHANGELOG.md
index 0f19b14b3..4c38b0add 100644
--- a/embassy-stm32/CHANGELOG.md
+++ b/embassy-stm32/CHANGELOG.md
@@ -8,6 +8,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased - ReleaseDate
 
 - fix: fix incorrect handling of split interrupts in timer driver
+- feat: allow granular stop for regular usart
+- feat: Add continuous waveform method to SimplePWM
+- change: remove waveform timer method
 - change: low power: store stop mode for dma channels
 - fix: Fixed ADC4 enable() for WBA
 - feat: allow use of anyadcchannel for adc4
@@ -41,7 +44,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - feat: Configurable gpio speed for QSPI
 - feat: derive Clone, Copy and defmt::Format for all *SPI-related configs
 - fix: handle address and data-length errors in OSPI
-- feat: Allow OSPI DMA writes larger than 64kB using chunking
+- feat: Allow OSPI/HSPI/XSPI DMA writes larger than 64kB using chunking
 - feat: More ADC enums for g0 PAC, API change for oversampling, allow separate sample times
 - feat: Add USB CRS sync support for STM32C071
 - fix: RTC register definition for STM32L4P5 and L4Q5 as they use v3 register map.
@@ -52,6 +55,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - feat: stm32/usart: add `eager_reads` option to control if buffered readers return as soon as possible or after more data is available ([#4668](https://github.com/embassy-rs/embassy/pull/4668))
 - feat: stm32/usart: add `de_assertion_time` and `de_deassertion_time` config options
 - change: stm32/uart: BufferedUartRx now returns all available bytes from the internal buffer
+- fix: Properly set the transfer size for OSPI/HSPI/XSPI transfers with word sizes other than 8 bits.
 - fix: stm32/adc: Calculate the ADC prescaler in a way that it allows for the max frequency to be reached
 - fix: Prevent a HardFault crash on STM32H5 devices by changing `uid()` to return `[u8; 12]` by value instead of a reference. (Fixes #2696)
 - change: timer: added output compare values
diff --git a/embassy-stm32/src/dma/dma_bdma.rs b/embassy-stm32/src/dma/dma_bdma.rs
index b46ae2813..adc084474 100644
--- a/embassy-stm32/src/dma/dma_bdma.rs
+++ b/embassy-stm32/src/dma/dma_bdma.rs
@@ -8,8 +8,9 @@ use embassy_sync::waitqueue::AtomicWaker;
 
 use super::ringbuffer::{DmaCtrl, Error, ReadableDmaRingBuffer, WritableDmaRingBuffer};
 use super::word::{Word, WordSize};
-use super::{AnyChannel, BusyChannel, Channel, Dir, Request, STATE};
+use super::{AnyChannel, Channel, Dir, Request, STATE};
 use crate::interrupt::typelevel::Interrupt;
+use crate::rcc::BusyPeripheral;
 use crate::{interrupt, pac};
 
 pub(crate) struct ChannelInfo {
@@ -602,7 +603,7 @@ impl AnyChannel {
 /// DMA transfer.
 #[must_use = "futures do nothing unless you `.await` or poll them"]
 pub struct Transfer<'a> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
 }
 
 impl<'a> Transfer<'a> {
@@ -714,7 +715,7 @@ impl<'a> Transfer<'a> {
         );
         channel.start();
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
         }
     }
 
@@ -818,7 +819,7 @@ impl<'a> DmaCtrl for DmaCtrlImpl<'a> {
 
 /// Ringbuffer for receiving data using DMA circular mode.
 pub struct ReadableRingBuffer<'a, W: Word> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
     ringbuf: ReadableDmaRingBuffer<'a, W>,
 }
 
@@ -855,7 +856,7 @@ impl<'a, W: Word> ReadableRingBuffer<'a, W> {
         );
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
             ringbuf: ReadableDmaRingBuffer::new(buffer),
         }
     }
@@ -974,7 +975,7 @@ impl<'a, W: Word> Drop for ReadableRingBuffer<'a, W> {
 
 /// Ringbuffer for writing data using DMA circular mode.
 pub struct WritableRingBuffer<'a, W: Word> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
     ringbuf: WritableDmaRingBuffer<'a, W>,
 }
 
@@ -1011,7 +1012,7 @@ impl<'a, W: Word> WritableRingBuffer<'a, W> {
         );
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
             ringbuf: WritableDmaRingBuffer::new(buffer),
         }
     }
diff --git a/embassy-stm32/src/dma/gpdma/mod.rs b/embassy-stm32/src/dma/gpdma/mod.rs
index 383c74a78..bfd0570f8 100644
--- a/embassy-stm32/src/dma/gpdma/mod.rs
+++ b/embassy-stm32/src/dma/gpdma/mod.rs
@@ -11,10 +11,10 @@ use linked_list::Table;
 
 use super::word::{Word, WordSize};
 use super::{AnyChannel, Channel, Dir, Request, STATE};
-use crate::dma::BusyChannel;
 use crate::interrupt::typelevel::Interrupt;
 use crate::pac;
 use crate::pac::gpdma::vals;
+use crate::rcc::BusyPeripheral;
 
 pub mod linked_list;
 pub mod ringbuffered;
@@ -409,7 +409,7 @@ impl AnyChannel {
 /// Linked-list DMA transfer.
 #[must_use = "futures do nothing unless you `.await` or poll them"]
 pub struct LinkedListTransfer<'a, const ITEM_COUNT: usize> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
 }
 
 impl<'a, const ITEM_COUNT: usize> LinkedListTransfer<'a, ITEM_COUNT> {
@@ -431,7 +431,7 @@ impl<'a, const ITEM_COUNT: usize> LinkedListTransfer<'a, ITEM_COUNT> {
         channel.start();
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
         }
     }
 
@@ -508,7 +508,7 @@ impl<'a, const ITEM_COUNT: usize> Future for LinkedListTransfer<'a, ITEM_COUNT>
 /// DMA transfer.
 #[must_use = "futures do nothing unless you `.await` or poll them"]
 pub struct Transfer<'a> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
 }
 
 impl<'a> Transfer<'a> {
@@ -629,7 +629,7 @@ impl<'a> Transfer<'a> {
         channel.start();
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
         }
     }
 
diff --git a/embassy-stm32/src/dma/gpdma/ringbuffered.rs b/embassy-stm32/src/dma/gpdma/ringbuffered.rs
index 54e4d5f71..c150d0b95 100644
--- a/embassy-stm32/src/dma/gpdma/ringbuffered.rs
+++ b/embassy-stm32/src/dma/gpdma/ringbuffered.rs
@@ -12,7 +12,8 @@ use super::{AnyChannel, STATE, TransferOptions};
 use crate::dma::gpdma::linked_list::{RunMode, Table};
 use crate::dma::ringbuffer::{DmaCtrl, Error, ReadableDmaRingBuffer, WritableDmaRingBuffer};
 use crate::dma::word::Word;
-use crate::dma::{BusyChannel, Channel, Dir, Request};
+use crate::dma::{Channel, Dir, Request};
+use crate::rcc::BusyPeripheral;
 
 struct DmaCtrlImpl<'a>(Peri<'a, AnyChannel>);
 
@@ -49,7 +50,7 @@ impl<'a> DmaCtrl for DmaCtrlImpl<'a> {
 
 /// Ringbuffer for receiving data using GPDMA linked-list mode.
 pub struct ReadableRingBuffer<'a, W: Word> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
     ringbuf: ReadableDmaRingBuffer<'a, W>,
     table: Table<2>,
     options: TransferOptions,
@@ -70,7 +71,7 @@ impl<'a, W: Word> ReadableRingBuffer<'a, W> {
         let table = Table::<2>::new_ping_pong::<W>(request, peri_addr, buffer, Dir::PeripheralToMemory);
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
             ringbuf: ReadableDmaRingBuffer::new(buffer),
             table,
             options,
@@ -189,7 +190,7 @@ impl<'a, W: Word> Drop for ReadableRingBuffer<'a, W> {
 
 /// Ringbuffer for writing data using GPDMA linked-list mode.
 pub struct WritableRingBuffer<'a, W: Word> {
-    channel: BusyChannel<'a>,
+    channel: BusyPeripheral<Peri<'a, AnyChannel>>,
     ringbuf: WritableDmaRingBuffer<'a, W>,
     table: Table<2>,
     options: TransferOptions,
@@ -210,7 +211,7 @@ impl<'a, W: Word> WritableRingBuffer<'a, W> {
         let table = Table::<2>::new_ping_pong::<W>(request, peri_addr, buffer, Dir::MemoryToPeripheral);
 
         Self {
-            channel: BusyChannel::new(channel),
+            channel: BusyPeripheral::new(channel),
             ringbuf: WritableDmaRingBuffer::new(buffer),
             table,
             options,
diff --git a/embassy-stm32/src/dma/mod.rs b/embassy-stm32/src/dma/mod.rs
index 4becc2d87..efb324fa6 100644
--- a/embassy-stm32/src/dma/mod.rs
+++ b/embassy-stm32/src/dma/mod.rs
@@ -3,14 +3,12 @@
 
 #[cfg(any(bdma, dma))]
 mod dma_bdma;
-use core::ops;
 
 #[cfg(any(bdma, dma))]
 pub use dma_bdma::*;
 
 #[cfg(gpdma)]
 pub(crate) mod gpdma;
-use embassy_hal_internal::Peri;
 #[cfg(gpdma)]
 pub use gpdma::ringbuffered::*;
 #[cfg(gpdma)]
@@ -27,9 +25,10 @@ pub(crate) use util::*;
 pub(crate) mod ringbuffer;
 pub mod word;
 
-use embassy_hal_internal::{PeripheralType, impl_peripheral};
+use embassy_hal_internal::{Peri, PeripheralType, impl_peripheral};
 
 use crate::interrupt;
+use crate::rcc::StoppablePeripheral;
 
 /// The direction of a DMA transfer.
 #[derive(Debug, Copy, Clone, PartialEq, Eq)]
@@ -48,6 +47,13 @@ pub type Request = u8;
 #[cfg(not(any(dma_v2, bdma_v2, gpdma, dmamux)))]
 pub type Request = ();
 
+impl<'a> StoppablePeripheral for Peri<'a, AnyChannel> {
+    #[cfg(feature = "low-power")]
+    fn stop_mode(&self) -> crate::rcc::StopMode {
+        self.stop_mode
+    }
+}
+
 pub(crate) trait SealedChannel {
     #[cfg(not(stm32n6))]
     fn id(&self) -> u8;
@@ -103,44 +109,6 @@ macro_rules! dma_channel_impl {
     };
 }
 
-pub(crate) struct BusyChannel<'a> {
-    channel: Peri<'a, AnyChannel>,
-}
-
-impl<'a> BusyChannel<'a> {
-    pub fn new(channel: Peri<'a, AnyChannel>) -> Self {
-        #[cfg(feature = "low-power")]
-        critical_section::with(|cs| {
-            crate::rcc::increment_stop_refcount(cs, channel.stop_mode);
-        });
-
-        Self { channel }
-    }
-}
-
-impl<'a> Drop for BusyChannel<'a> {
-    fn drop(&mut self) {
-        #[cfg(feature = "low-power")]
-        critical_section::with(|cs| {
-            crate::rcc::decrement_stop_refcount(cs, self.stop_mode);
-        });
-    }
-}
-
-impl<'a> ops::Deref for BusyChannel<'a> {
-    type Target = Peri<'a, AnyChannel>;
-
-    fn deref(&self) -> &Self::Target {
-        &self.channel
-    }
-}
-
-impl<'a> ops::DerefMut for BusyChannel<'a> {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        &mut self.channel
-    }
-}
-
 /// Type-erased DMA channel.
 pub struct AnyChannel {
     pub(crate) id: u8,
diff --git a/embassy-stm32/src/hspi/mod.rs b/embassy-stm32/src/hspi/mod.rs
index 69baa708e..1d3560678 100644
--- a/embassy-stm32/src/hspi/mod.rs
+++ b/embassy-stm32/src/hspi/mod.rs
@@ -391,7 +391,7 @@ impl<'d, T: Instance, M: PeriMode> Hspi<'d, T, M> {
         while T::REGS.sr().read().busy() {}
 
         T::REGS.cr().modify(|w| {
-            w.set_fmode(0.into());
+            w.set_fmode(FunctionalMode::IndirectWrite.into());
         });
 
         // Configure alternate bytes
@@ -498,7 +498,8 @@ impl<'d, T: Instance, M: PeriMode> Hspi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -537,7 +538,8 @@ impl<'d, T: Instance, M: PeriMode> Hspi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         T::REGS
             .cr()
@@ -767,7 +769,8 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -782,16 +785,18 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.blocking_wait();
+            transfer.blocking_wait();
+        }
 
         finish_dma(T::REGS);
 
@@ -807,21 +812,24 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(FunctionalMode::IndirectWrite.into()));
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .write(buf, T::REGS.dr().as_ptr() as *mut W, Default::default())
-        };
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .write(chunk, T::REGS.dr().as_ptr() as *mut W, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.blocking_wait();
+            transfer.blocking_wait();
+        }
 
         finish_dma(T::REGS);
 
@@ -837,7 +845,8 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -852,16 +861,18 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.await;
+            transfer.await;
+        }
 
         finish_dma(T::REGS);
 
@@ -877,21 +888,25 @@ impl<'d, T: Instance> Hspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(FunctionalMode::IndirectWrite.into()));
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .write(buf, T::REGS.dr().as_ptr() as *mut W, Default::default())
-        };
+        // TODO: implement this using a LinkedList DMA to offload the whole transfer off the CPU.
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .write(chunk, T::REGS.dr().as_ptr() as *mut W, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.await;
+            transfer.await;
+        }
 
         finish_dma(T::REGS);
 
diff --git a/embassy-stm32/src/i2c/v2.rs b/embassy-stm32/src/i2c/v2.rs
index 61e550ad4..b2ba94e21 100644
--- a/embassy-stm32/src/i2c/v2.rs
+++ b/embassy-stm32/src/i2c/v2.rs
@@ -1075,8 +1075,6 @@ impl<'d, IM: MasterMode> I2c<'d, Async, IM> {
 
     /// Write.
     pub async fn write(&mut self, address: u8, write: &[u8]) -> Result<(), Error> {
-        #[cfg(all(feature = "low-power", stm32wlex))]
-        let _device_busy = crate::low_power::DeviceBusy::new_stop1();
         let timeout = self.timeout();
         if write.is_empty() {
             self.write_internal(address.into(), write, true, timeout)
@@ -1091,8 +1089,6 @@ impl<'d, IM: MasterMode> I2c<'d, Async, IM> {
     ///
     /// The buffers are concatenated in a single write transaction.
     pub async fn write_vectored(&mut self, address: Address, write: &[&[u8]]) -> Result<(), Error> {
-        #[cfg(all(feature = "low-power", stm32wlex))]
-        let _device_busy = crate::low_power::DeviceBusy::new_stop1();
         let timeout = self.timeout();
 
         if write.is_empty() {
@@ -1124,8 +1120,6 @@ impl<'d, IM: MasterMode> I2c<'d, Async, IM> {
 
     /// Read.
     pub async fn read(&mut self, address: u8, buffer: &mut [u8]) -> Result<(), Error> {
-        #[cfg(all(feature = "low-power", stm32wlex))]
-        let _device_busy = crate::low_power::DeviceBusy::new_stop1();
         let timeout = self.timeout();
 
         if buffer.is_empty() {
@@ -1138,8 +1132,6 @@ impl<'d, IM: MasterMode> I2c<'d, Async, IM> {
 
     /// Write, restart, read.
     pub async fn write_read(&mut self, address: u8, write: &[u8], read: &mut [u8]) -> Result<(), Error> {
-        #[cfg(all(feature = "low-power", stm32wlex))]
-        let _device_busy = crate::low_power::DeviceBusy::new_stop1();
         let timeout = self.timeout();
 
         if write.is_empty() {
@@ -1165,9 +1157,6 @@ impl<'d, IM: MasterMode> I2c<'d, Async, IM> {
     ///
     /// [transaction contract]: embedded_hal_1::i2c::I2c::transaction
     pub async fn transaction(&mut self, addr: u8, operations: &mut [Operation<'_>]) -> Result<(), Error> {
-        #[cfg(all(feature = "low-power", stm32wlex))]
-        let _device_busy = crate::low_power::DeviceBusy::new_stop1();
-
         if operations.is_empty() {
             return Err(Error::ZeroLengthTransfer);
         }
diff --git a/embassy-stm32/src/low_power.rs b/embassy-stm32/src/low_power.rs
index bd8290da0..cdf3323fb 100644
--- a/embassy-stm32/src/low_power.rs
+++ b/embassy-stm32/src/low_power.rs
@@ -51,7 +51,7 @@ use embassy_executor::*;
 
 use crate::interrupt;
 pub use crate::rcc::StopMode;
-use crate::rcc::{RCC_CONFIG, REFCOUNT_STOP1, REFCOUNT_STOP2, decrement_stop_refcount, increment_stop_refcount};
+use crate::rcc::{BusyPeripheral, RCC_CONFIG, REFCOUNT_STOP1, REFCOUNT_STOP2};
 use crate::time_driver::get_driver;
 
 const THREAD_PENDER: usize = usize::MAX;
@@ -59,7 +59,9 @@ const THREAD_PENDER: usize = usize::MAX;
 static mut EXECUTOR_TAKEN: bool = false;
 
 /// Prevent the device from going into the stop mode if held
-pub struct DeviceBusy(StopMode);
+pub struct DeviceBusy {
+    _stop_mode: BusyPeripheral<StopMode>,
+}
 
 impl DeviceBusy {
     /// Create a new DeviceBusy with stop1.
@@ -74,19 +76,9 @@ impl DeviceBusy {
 
     /// Create a new DeviceBusy.
     pub fn new(stop_mode: StopMode) -> Self {
-        critical_section::with(|cs| {
-            increment_stop_refcount(cs, stop_mode);
-        });
-
-        Self(stop_mode)
-    }
-}
-
-impl Drop for DeviceBusy {
-    fn drop(&mut self) {
-        critical_section::with(|cs| {
-            decrement_stop_refcount(cs, self.0);
-        });
+        Self {
+            _stop_mode: BusyPeripheral::new(stop_mode),
+        }
     }
 }
 
diff --git a/embassy-stm32/src/ospi/mod.rs b/embassy-stm32/src/ospi/mod.rs
index 592a8594a..2d5dbd95a 100644
--- a/embassy-stm32/src/ospi/mod.rs
+++ b/embassy-stm32/src/ospi/mod.rs
@@ -451,7 +451,7 @@ impl<'d, T: Instance, M: PeriMode> Ospi<'d, T, M> {
         }
 
         T::REGS.cr().modify(|w| {
-            w.set_fmode(0.into());
+            w.set_fmode(vals::FunctionalMode::INDIRECT_WRITE);
         });
 
         // Configure alternate bytes
@@ -577,7 +577,8 @@ impl<'d, T: Instance, M: PeriMode> Ospi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -616,7 +617,8 @@ impl<'d, T: Instance, M: PeriMode> Ospi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         T::REGS
             .cr()
@@ -1153,7 +1155,8 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -1168,16 +1171,18 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.blocking_wait();
+            transfer.blocking_wait();
+        }
 
         finish_dma(T::REGS);
 
@@ -1193,13 +1198,14 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(vals::FunctionalMode::INDIRECT_WRITE));
 
         // TODO: implement this using a LinkedList DMA to offload the whole transfer off the CPU.
-        for chunk in buf.chunks(0xFFFF) {
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
             let transfer = unsafe {
                 self.dma
                     .as_mut()
@@ -1226,7 +1232,8 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -1241,16 +1248,18 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.await;
+            transfer.await;
+        }
 
         finish_dma(T::REGS);
 
@@ -1266,13 +1275,14 @@ impl<'d, T: Instance> Ospi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(vals::FunctionalMode::INDIRECT_WRITE));
 
         // TODO: implement this using a LinkedList DMA to offload the whole transfer off the CPU.
-        for chunk in buf.chunks(0xFFFF) {
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
             let transfer = unsafe {
                 self.dma
                     .as_mut()
diff --git a/embassy-stm32/src/rcc/mod.rs b/embassy-stm32/src/rcc/mod.rs
index 85434fa83..1dd634cfe 100644
--- a/embassy-stm32/src/rcc/mod.rs
+++ b/embassy-stm32/src/rcc/mod.rs
@@ -4,6 +4,7 @@
 #![allow(missing_docs)] // TODO
 
 use core::mem::MaybeUninit;
+use core::ops;
 
 mod bd;
 pub use bd::*;
@@ -112,7 +113,7 @@ pub fn clocks<'a>(_rcc: &'a crate::Peri<'a, crate::peripherals::RCC>) -> &'a Clo
 }
 
 #[cfg(feature = "low-power")]
-pub(crate) fn increment_stop_refcount(_cs: CriticalSection, stop_mode: StopMode) {
+fn increment_stop_refcount(_cs: CriticalSection, stop_mode: StopMode) {
     match stop_mode {
         StopMode::Standby => {}
         StopMode::Stop2 => unsafe {
@@ -125,7 +126,7 @@ pub(crate) fn increment_stop_refcount(_cs: CriticalSection, stop_mode: StopMode)
 }
 
 #[cfg(feature = "low-power")]
-pub(crate) fn decrement_stop_refcount(_cs: CriticalSection, stop_mode: StopMode) {
+fn decrement_stop_refcount(_cs: CriticalSection, stop_mode: StopMode) {
     match stop_mode {
         StopMode::Standby => {}
         StopMode::Stop2 => unsafe {
@@ -182,6 +183,12 @@ pub enum StopMode {
     Standby,
 }
 
+#[cfg(feature = "low-power")]
+type BusyRccPeripheral = BusyPeripheral<StopMode>;
+
+#[cfg(not(feature = "low-power"))]
+type BusyRccPeripheral = ();
+
 impl RccInfo {
     /// Safety:
     /// - `reset_offset_and_bit`, if set, must correspond to valid xxxRST bit
@@ -234,9 +241,6 @@ impl RccInfo {
             }
         }
 
-        #[cfg(feature = "low-power")]
-        increment_stop_refcount(_cs, self.stop_mode);
-
         // set the xxxRST bit
         let reset_ptr = self.reset_ptr();
         if let Some(reset_ptr) = reset_ptr {
@@ -292,9 +296,6 @@ impl RccInfo {
             }
         }
 
-        #[cfg(feature = "low-power")]
-        decrement_stop_refcount(_cs, self.stop_mode);
-
         // clear the xxxEN bit
         let enable_ptr = self.enable_ptr();
         unsafe {
@@ -303,16 +304,63 @@ impl RccInfo {
         }
     }
 
+    #[allow(dead_code)]
+    pub(crate) fn increment_stop_refcount_with_cs(&self, _cs: CriticalSection) {
+        #[cfg(feature = "low-power")]
+        increment_stop_refcount(_cs, self.stop_mode);
+    }
+
+    #[allow(dead_code)]
+    pub(crate) fn increment_stop_refcount(&self) {
+        #[cfg(feature = "low-power")]
+        critical_section::with(|cs| self.increment_stop_refcount_with_cs(cs))
+    }
+
+    #[allow(dead_code)]
+    pub(crate) fn decrement_stop_refcount_with_cs(&self, _cs: CriticalSection) {
+        #[cfg(feature = "low-power")]
+        decrement_stop_refcount(_cs, self.stop_mode);
+    }
+
+    #[allow(dead_code)]
+    pub(crate) fn decrement_stop_refcount(&self) {
+        #[cfg(feature = "low-power")]
+        critical_section::with(|cs| self.decrement_stop_refcount_with_cs(cs))
+    }
+
     // TODO: should this be `unsafe`?
     pub(crate) fn enable_and_reset(&self) {
+        critical_section::with(|cs| {
+            self.enable_and_reset_with_cs(cs);
+            self.increment_stop_refcount_with_cs(cs);
+        })
+    }
+
+    #[allow(dead_code)]
+    pub(crate) fn enable_and_reset_without_stop(&self) {
         critical_section::with(|cs| self.enable_and_reset_with_cs(cs))
     }
 
     // TODO: should this be `unsafe`?
     pub(crate) fn disable(&self) {
+        critical_section::with(|cs| {
+            self.disable_with_cs(cs);
+            self.decrement_stop_refcount_with_cs(cs);
+        })
+    }
+
+    // TODO: should this be `unsafe`?
+    #[allow(dead_code)]
+    pub(crate) fn disable_without_stop(&self) {
         critical_section::with(|cs| self.disable_with_cs(cs))
     }
 
+    #[allow(dead_code)]
+    pub(crate) fn block_stop(&self) -> BusyRccPeripheral {
+        #[cfg(feature = "low-power")]
+        BusyPeripheral::new(self.stop_mode)
+    }
+
     fn reset_ptr(&self) -> Option<*mut u32> {
         if self.reset_offset_or_0xff != 0xff {
             Some(unsafe { (RCC.as_ptr() as *mut u32).add(self.reset_offset_or_0xff as _) })
@@ -326,6 +374,53 @@ impl RccInfo {
     }
 }
 
+pub(crate) trait StoppablePeripheral {
+    #[cfg(feature = "low-power")]
+    #[allow(dead_code)]
+    fn stop_mode(&self) -> StopMode;
+}
+
+#[cfg(feature = "low-power")]
+impl<'a> StoppablePeripheral for StopMode {
+    fn stop_mode(&self) -> StopMode {
+        *self
+    }
+}
+
+pub(crate) struct BusyPeripheral<T: StoppablePeripheral> {
+    peripheral: T,
+}
+
+impl<T: StoppablePeripheral> BusyPeripheral<T> {
+    pub fn new(peripheral: T) -> Self {
+        #[cfg(feature = "low-power")]
+        critical_section::with(|cs| increment_stop_refcount(cs, peripheral.stop_mode()));
+
+        Self { peripheral }
+    }
+}
+
+impl<T: StoppablePeripheral> Drop for BusyPeripheral<T> {
+    fn drop(&mut self) {
+        #[cfg(feature = "low-power")]
+        critical_section::with(|cs| decrement_stop_refcount(cs, self.peripheral.stop_mode()));
+    }
+}
+
+impl<T: StoppablePeripheral> ops::Deref for BusyPeripheral<T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        &self.peripheral
+    }
+}
+
+impl<T: StoppablePeripheral> ops::DerefMut for BusyPeripheral<T> {
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        &mut self.peripheral
+    }
+}
+
 #[allow(unused)]
 mod util {
     use crate::time::Hertz;
diff --git a/embassy-stm32/src/timer/complementary_pwm.rs b/embassy-stm32/src/timer/complementary_pwm.rs
index 6d4c70dff..77f19a37b 100644
--- a/embassy-stm32/src/timer/complementary_pwm.rs
+++ b/embassy-stm32/src/timer/complementary_pwm.rs
@@ -220,9 +220,11 @@ impl<'d, T: AdvancedInstance4Channel> ComplementaryPwm<'d, T> {
     ///
     /// Note:
     /// you will need to provide corresponding TIMx_UP DMA channel to use this method.
-    #[inline(always)]
     pub async fn waveform_up(&mut self, dma: Peri<'_, impl super::UpDma<T>>, channel: Channel, duty: &[u16]) {
-        self.inner.waveform_up(dma, channel, duty).await
+        self.inner.enable_channel(channel, true);
+        self.inner.enable_update_dma(true);
+        self.inner.setup_update_dma(dma, channel, duty).await;
+        self.inner.enable_update_dma(false);
     }
 
     /// Generate a multichannel sequence of PWM waveforms using DMA triggered by timer update events.
@@ -254,7 +256,6 @@ impl<'d, T: AdvancedInstance4Channel> ComplementaryPwm<'d, T> {
     /// Also be aware that embassy timers use one of timers internally. It is possible to
     /// switch this timer by using `time-driver-timX` feature.
     ///
-    #[inline(always)]
     pub async fn waveform_up_multi_channel(
         &mut self,
         dma: Peri<'_, impl super::UpDma<T>>,
@@ -262,15 +263,11 @@ impl<'d, T: AdvancedInstance4Channel> ComplementaryPwm<'d, T> {
         ending_channel: Channel,
         duty: &[u16],
     ) {
+        self.inner.enable_update_dma(true);
         self.inner
-            .waveform_up_multi_channel(dma, starting_channel, ending_channel, duty)
+            .setup_update_dma_burst(dma, starting_channel, ending_channel, duty)
             .await;
-    }
-
-    /// Generate a sequence of PWM waveform
-    #[inline(always)]
-    pub async fn waveform<C: TimerChannel>(&mut self, dma: Peri<'_, impl super::Dma<T, C>>, duty: &[u16]) {
-        self.inner.waveform(dma, duty).await;
+        self.inner.enable_update_dma(false);
     }
 }
 
diff --git a/embassy-stm32/src/timer/low_level.rs b/embassy-stm32/src/timer/low_level.rs
index f0105ece8..aba08081f 100644
--- a/embassy-stm32/src/timer/low_level.rs
+++ b/embassy-stm32/src/timer/low_level.rs
@@ -13,9 +13,10 @@ use embassy_hal_internal::Peri;
 pub use stm32_metapac::timer::vals::{FilterValue, Mms as MasterMode, Sms as SlaveMode, Ts as TriggerSource};
 
 use super::*;
+use crate::dma::{Transfer, WritableRingBuffer};
 use crate::pac::timer::vals;
+use crate::rcc;
 use crate::time::Hertz;
-use crate::{dma, rcc};
 
 /// Input capture mode.
 #[derive(Clone, Copy)]
@@ -659,29 +660,88 @@ impl<'d, T: GeneralInstance4Channel> Timer<'d, T> {
         }
     }
 
+    /// Setup a ring buffer for the channel
+    pub fn setup_ring_buffer<'a>(
+        &mut self,
+        dma: Peri<'a, impl super::UpDma<T>>,
+        channel: Channel,
+        dma_buf: &'a mut [u16],
+    ) -> WritableRingBuffer<'a, u16> {
+        #[allow(clippy::let_unit_value)] // eg. stm32f334
+        let req = dma.request();
+
+        unsafe {
+            use crate::dma::TransferOptions;
+            #[cfg(not(any(bdma, gpdma)))]
+            use crate::dma::{Burst, FifoThreshold};
+
+            let dma_transfer_option = TransferOptions {
+                #[cfg(not(any(bdma, gpdma)))]
+                fifo_threshold: Some(FifoThreshold::Full),
+                #[cfg(not(any(bdma, gpdma)))]
+                mburst: Burst::Incr8,
+                ..Default::default()
+            };
+
+            WritableRingBuffer::new(
+                dma,
+                req,
+                self.regs_1ch().ccr(channel.index()).as_ptr() as *mut u16,
+                dma_buf,
+                dma_transfer_option,
+            )
+        }
+    }
+
     /// Generate a sequence of PWM waveform
     ///
     /// Note:
     /// you will need to provide corresponding TIMx_UP DMA channel to use this method.
-    pub async fn waveform_up(&mut self, dma: Peri<'_, impl super::UpDma<T>>, channel: Channel, duty: &[u16]) {
+    pub fn setup_update_dma<'a>(
+        &mut self,
+        dma: Peri<'a, impl super::UpDma<T>>,
+        channel: Channel,
+        duty: &'a [u16],
+    ) -> Transfer<'a> {
         #[allow(clippy::let_unit_value)] // eg. stm32f334
         let req = dma.request();
 
-        let original_update_dma_state = self.get_update_dma_state();
+        unsafe {
+            #[cfg(not(any(bdma, gpdma)))]
+            use crate::dma::{Burst, FifoThreshold};
+            use crate::dma::{Transfer, TransferOptions};
 
-        if !original_update_dma_state {
-            self.enable_update_dma(true);
-        }
+            let dma_transfer_option = TransferOptions {
+                #[cfg(not(any(bdma, gpdma)))]
+                fifo_threshold: Some(FifoThreshold::Full),
+                #[cfg(not(any(bdma, gpdma)))]
+                mburst: Burst::Incr8,
+                ..Default::default()
+            };
 
-        self.waveform_helper(dma, req, channel, duty).await;
+            match self.bits() {
+                TimerBits::Bits16 => Transfer::new_write(
+                    dma,
+                    req,
+                    duty,
+                    self.regs_1ch().ccr(channel.index()).as_ptr() as *mut u16,
+                    dma_transfer_option,
+                ),
+                #[cfg(not(any(stm32l0)))]
+                TimerBits::Bits32 => {
+                    #[cfg(not(any(bdma, gpdma)))]
+                    panic!("unsupported timer bits");
 
-        // Since DMA is closed before timer update event trigger DMA is turn off,
-        // this can almost always trigger a DMA FIFO error.
-        //
-        // optional TODO:
-        // clean FEIF after disable UDE
-        if !original_update_dma_state {
-            self.enable_update_dma(false);
+                    #[cfg(any(bdma, gpdma))]
+                    Transfer::new_write(
+                        dma,
+                        req,
+                        duty,
+                        self.regs_1ch().ccr(channel.index()).as_ptr() as *mut u32,
+                        dma_transfer_option,
+                    )
+                }
+            }
         }
     }
 
@@ -714,13 +774,13 @@ impl<'d, T: GeneralInstance4Channel> Timer<'d, T> {
     /// Also be aware that embassy timers use one of timers internally. It is possible to
     /// switch this timer by using `time-driver-timX` feature.
     ///
-    pub async fn waveform_up_multi_channel(
+    pub fn setup_update_dma_burst<'a>(
         &mut self,
-        dma: Peri<'_, impl super::UpDma<T>>,
+        dma: Peri<'a, impl super::UpDma<T>>,
         starting_channel: Channel,
         ending_channel: Channel,
-        duty: &[u16],
-    ) {
+        duty: &'a [u16],
+    ) -> Transfer<'a> {
         let cr1_addr = self.regs_gp16().cr1().as_ptr() as u32;
         let start_ch_index = starting_channel.index();
         let end_ch_index = ending_channel.index();
@@ -738,11 +798,6 @@ impl<'d, T: GeneralInstance4Channel> Timer<'d, T> {
         #[allow(clippy::let_unit_value)] // eg. stm32f334
         let req = dma.request();
 
-        let original_update_dma_state = self.get_update_dma_state();
-        if !original_update_dma_state {
-            self.enable_update_dma(true);
-        }
-
         unsafe {
             #[cfg(not(any(bdma, gpdma)))]
             use crate::dma::{Burst, FifoThreshold};
@@ -763,115 +818,9 @@ impl<'d, T: GeneralInstance4Channel> Timer<'d, T> {
                 self.regs_gp16().dmar().as_ptr() as *mut u16,
                 dma_transfer_option,
             )
-            .await
-        };
-
-        if !original_update_dma_state {
-            self.enable_update_dma(false);
         }
     }
 
-    /// Generate a sequence of PWM waveform
-    pub async fn waveform<C: TimerChannel>(&mut self, dma: Peri<'_, impl super::Dma<T, C>>, duty: &[u16]) {
-        use crate::pac::timer::vals::Ccds;
-
-        #[allow(clippy::let_unit_value)] // eg. stm32f334
-        let req = dma.request();
-
-        let cc_channel = C::CHANNEL;
-
-        let original_cc_dma_on_update = self.get_cc_dma_selection() == Ccds::ON_UPDATE;
-        let original_cc_dma_enabled = self.get_cc_dma_enable_state(cc_channel);
-
-        // redirect CC DMA request onto Update Event
-        if !original_cc_dma_on_update {
-            self.set_cc_dma_selection(Ccds::ON_UPDATE)
-        }
-
-        if !original_cc_dma_enabled {
-            self.set_cc_dma_enable_state(cc_channel, true);
-        }
-
-        self.waveform_helper(dma, req, cc_channel, duty).await;
-
-        // Since DMA is closed before timer Capture Compare Event trigger DMA is turn off,
-        // this can almost always trigger a DMA FIFO error.
-        //
-        // optional TODO:
-        // clean FEIF after disable UDE
-        if !original_cc_dma_enabled {
-            self.set_cc_dma_enable_state(cc_channel, false);
-        }
-
-        if !original_cc_dma_on_update {
-            self.set_cc_dma_selection(Ccds::ON_COMPARE)
-        }
-    }
-
-    async fn waveform_helper(
-        &mut self,
-        dma: Peri<'_, impl dma::Channel>,
-        req: dma::Request,
-        channel: Channel,
-        duty: &[u16],
-    ) {
-        let original_duty_state = self.get_compare_value(channel);
-        let original_enable_state = self.get_channel_enable_state(channel);
-
-        if !original_enable_state {
-            self.enable_channel(channel, true);
-        }
-
-        unsafe {
-            #[cfg(not(any(bdma, gpdma)))]
-            use crate::dma::{Burst, FifoThreshold};
-            use crate::dma::{Transfer, TransferOptions};
-
-            let dma_transfer_option = TransferOptions {
-                #[cfg(not(any(bdma, gpdma)))]
-                fifo_threshold: Some(FifoThreshold::Full),
-                #[cfg(not(any(bdma, gpdma)))]
-                mburst: Burst::Incr8,
-                ..Default::default()
-            };
-
-            match self.bits() {
-                TimerBits::Bits16 => {
-                    Transfer::new_write(
-                        dma,
-                        req,
-                        duty,
-                        self.regs_1ch().ccr(channel.index()).as_ptr() as *mut u16,
-                        dma_transfer_option,
-                    )
-                    .await
-                }
-                #[cfg(not(any(stm32l0)))]
-                TimerBits::Bits32 => {
-                    #[cfg(not(any(bdma, gpdma)))]
-                    panic!("unsupported timer bits");
-
-                    #[cfg(any(bdma, gpdma))]
-                    Transfer::new_write(
-                        dma,
-                        req,
-                        duty,
-                        self.regs_1ch().ccr(channel.index()).as_ptr() as *mut u32,
-                        dma_transfer_option,
-                    )
-                    .await
-                }
-            };
-        };
-
-        // restore output compare state
-        if !original_enable_state {
-            self.enable_channel(channel, false);
-        }
-
-        self.set_compare_value(channel, original_duty_state);
-    }
-
     /// Get capture value for a channel.
     pub fn get_capture_value(&self, channel: Channel) -> u32 {
         self.get_compare_value(channel)
diff --git a/embassy-stm32/src/timer/mod.rs b/embassy-stm32/src/timer/mod.rs
index 804d1ef37..3fa363881 100644
--- a/embassy-stm32/src/timer/mod.rs
+++ b/embassy-stm32/src/timer/mod.rs
@@ -12,6 +12,7 @@ pub mod low_level;
 pub mod one_pulse;
 pub mod pwm_input;
 pub mod qei;
+pub mod ringbuffered;
 pub mod simple_pwm;
 
 use crate::interrupt;
diff --git a/embassy-stm32/src/timer/ringbuffered.rs b/embassy-stm32/src/timer/ringbuffered.rs
new file mode 100644
index 000000000..e8f97bf59
--- /dev/null
+++ b/embassy-stm32/src/timer/ringbuffered.rs
@@ -0,0 +1,169 @@
+//! RingBuffered PWM driver.
+
+use core::mem::ManuallyDrop;
+use core::task::Waker;
+
+use super::low_level::Timer;
+use super::{Channel, GeneralInstance4Channel};
+use crate::dma::WritableRingBuffer;
+use crate::dma::ringbuffer::Error;
+
+/// A PWM channel that uses a DMA ring buffer for continuous waveform generation.
+///
+/// This allows you to continuously update PWM duty cycles via DMA without blocking the CPU.
+/// The ring buffer enables smooth, uninterrupted waveform generation by automatically cycling
+/// through duty cycle values stored in memory.
+///
+/// You can write new duty cycle values to the ring buffer while it's running, enabling
+/// dynamic waveform generation for applications like motor control, LED dimming, or audio output.
+///
+/// # Example
+/// ```ignore
+/// let mut channel = pwm.ch1().into_ring_buffered_channel(dma_ch, &mut buffer);
+/// channel.start(); // Start DMA transfer
+/// channel.write(&[100, 200, 300]).ok(); // Update duty cycles
+/// ```
+pub struct RingBufferedPwmChannel<'d, T: GeneralInstance4Channel> {
+    timer: ManuallyDrop<Timer<'d, T>>,
+    ring_buf: WritableRingBuffer<'d, u16>,
+    channel: Channel,
+}
+
+impl<'d, T: GeneralInstance4Channel> RingBufferedPwmChannel<'d, T> {
+    pub(crate) fn new(
+        timer: ManuallyDrop<Timer<'d, T>>,
+        channel: Channel,
+        ring_buf: WritableRingBuffer<'d, u16>,
+    ) -> Self {
+        Self {
+            timer,
+            ring_buf,
+            channel,
+        }
+    }
+
+    /// Start the ring buffer operation.
+    ///
+    /// You must call this after creating it for it to work.
+    pub fn start(&mut self) {
+        self.ring_buf.start()
+    }
+
+    /// Clear all data in the ring buffer.
+    pub fn clear(&mut self) {
+        self.ring_buf.clear()
+    }
+
+    /// Write elements directly to the raw buffer. This can be used to fill the buffer before starting the DMA transfer.
+    pub fn write_immediate(&mut self, buf: &[u16]) -> Result<(usize, usize), Error> {
+        self.ring_buf.write_immediate(buf)
+    }
+
+    /// Write elements from the ring buffer
+    /// Return a tuple of the length written and the length remaining in the buffer
+    pub fn write(&mut self, buf: &[u16]) -> Result<(usize, usize), Error> {
+        self.ring_buf.write(buf)
+    }
+
+    /// Write an exact number of elements to the ringbuffer.
+    pub async fn write_exact(&mut self, buffer: &[u16]) -> Result<usize, Error> {
+        self.ring_buf.write_exact(buffer).await
+    }
+
+    /// Wait for any ring buffer write error.
+    pub async fn wait_write_error(&mut self) -> Result<usize, Error> {
+        self.ring_buf.wait_write_error().await
+    }
+
+    /// The current length of the ringbuffer
+    pub fn len(&mut self) -> Result<usize, Error> {
+        self.ring_buf.len()
+    }
+
+    /// The capacity of the ringbuffer
+    pub const fn capacity(&self) -> usize {
+        self.ring_buf.capacity()
+    }
+
+    /// Set a waker to be woken when at least one byte is send.
+    pub fn set_waker(&mut self, waker: &Waker) {
+        self.ring_buf.set_waker(waker)
+    }
+
+    /// Request the DMA to reset. The configuration for this channel will not be preserved.
+    ///
+    /// This doesn't immediately stop the transfer, you have to wait until is_running returns false.
+    pub fn request_reset(&mut self) {
+        self.ring_buf.request_reset()
+    }
+
+    /// Request the transfer to pause, keeping the existing configuration for this channel.
+    /// To restart the transfer, call [`start`](Self::start) again.
+    ///
+    /// This doesn't immediately stop the transfer, you have to wait until is_running returns false.
+    pub fn request_pause(&mut self) {
+        self.ring_buf.request_pause()
+    }
+
+    /// Return whether DMA is still running.
+    ///
+    /// If this returns false, it can be because either the transfer finished, or it was requested to stop early with request_stop.
+    pub fn is_running(&mut self) -> bool {
+        self.ring_buf.is_running()
+    }
+
+    /// Stop the DMA transfer and await until the buffer is empty.
+    ///
+    /// This disables the DMA transfer's circular mode so that the transfer stops when all available data has been written.
+    ///
+    /// This is designed to be used with streaming output data such as the I2S/SAI or DAC.
+    pub async fn stop(&mut self) {
+        self.ring_buf.stop().await
+    }
+
+    /// Enable the given channel.
+    pub fn enable(&mut self) {
+        self.timer.enable_channel(self.channel, true);
+    }
+
+    /// Disable the given channel.
+    pub fn disable(&mut self) {
+        self.timer.enable_channel(self.channel, false);
+    }
+
+    /// Check whether given channel is enabled
+    pub fn is_enabled(&self) -> bool {
+        self.timer.get_channel_enable_state(self.channel)
+    }
+
+    /// Get max duty value.
+    ///
+    /// This value depends on the configured frequency and the timer's clock rate from RCC.
+    pub fn max_duty_cycle(&self) -> u16 {
+        let max = self.timer.get_max_compare_value();
+        assert!(max < u16::MAX as u32);
+        max as u16 + 1
+    }
+
+    /// Set the output polarity for a given channel.
+    pub fn set_polarity(&mut self, polarity: super::low_level::OutputPolarity) {
+        self.timer.set_output_polarity(self.channel, polarity);
+    }
+
+    /// Set the output compare mode for a given channel.
+    pub fn set_output_compare_mode(&mut self, mode: super::low_level::OutputCompareMode) {
+        self.timer.set_output_compare_mode(self.channel, mode);
+    }
+}
+
+/// A group of four [`SimplePwmChannel`]s, obtained from [`SimplePwm::split`].
+pub struct RingBufferedPwmChannels<'d, T: GeneralInstance4Channel> {
+    /// Channel 1
+    pub ch1: RingBufferedPwmChannel<'d, T>,
+    /// Channel 2
+    pub ch2: RingBufferedPwmChannel<'d, T>,
+    /// Channel 3
+    pub ch3: RingBufferedPwmChannel<'d, T>,
+    /// Channel 4
+    pub ch4: RingBufferedPwmChannel<'d, T>,
+}
diff --git a/embassy-stm32/src/timer/simple_pwm.rs b/embassy-stm32/src/timer/simple_pwm.rs
index 6c9ef17e0..484e9fd81 100644
--- a/embassy-stm32/src/timer/simple_pwm.rs
+++ b/embassy-stm32/src/timer/simple_pwm.rs
@@ -4,6 +4,7 @@ use core::marker::PhantomData;
 use core::mem::ManuallyDrop;
 
 use super::low_level::{CountingMode, OutputCompareMode, OutputPolarity, Timer};
+use super::ringbuffered::RingBufferedPwmChannel;
 use super::{Ch1, Ch2, Ch3, Ch4, Channel, GeneralInstance4Channel, TimerChannel, TimerPin};
 use crate::Peri;
 #[cfg(gpio_v2)]
@@ -158,6 +159,33 @@ impl<'d, T: GeneralInstance4Channel> SimplePwmChannel<'d, T> {
     pub fn set_output_compare_mode(&mut self, mode: OutputCompareMode) {
         self.timer.set_output_compare_mode(self.channel, mode);
     }
+
+    /// Convert this PWM channel into a ring-buffered PWM channel.
+    ///
+    /// This allows continuous PWM waveform generation using a DMA ring buffer.
+    /// The ring buffer enables dynamic updates to the PWM duty cycle without blocking.
+    ///
+    /// # Arguments
+    /// * `tx_dma` - The DMA channel to use for transferring duty cycle values
+    /// * `dma_buf` - The buffer to use as a ring buffer (must be non-empty and <= 65535 elements)
+    ///
+    /// # Panics
+    /// Panics if `dma_buf` is empty or longer than 65535 elements.
+    pub fn into_ring_buffered_channel(
+        mut self,
+        tx_dma: Peri<'d, impl super::UpDma<T>>,
+        dma_buf: &'d mut [u16],
+    ) -> RingBufferedPwmChannel<'d, T> {
+        assert!(!dma_buf.is_empty() && dma_buf.len() <= 0xFFFF);
+
+        self.timer.enable_update_dma(true);
+
+        RingBufferedPwmChannel::new(
+            unsafe { self.timer.clone_unchecked() },
+            self.channel,
+            self.timer.setup_ring_buffer(tx_dma, self.channel, dma_buf),
+        )
+    }
 }
 
 /// A group of four [`SimplePwmChannel`]s, obtained from [`SimplePwm::split`].
@@ -316,9 +344,11 @@ impl<'d, T: GeneralInstance4Channel> SimplePwm<'d, T> {
     /// You will need to provide corresponding `TIMx_UP` DMA channel to use this method.
     /// Also be aware that embassy timers use one of timers internally. It is possible to
     /// switch this timer by using `time-driver-timX` feature.
-    #[inline(always)]
     pub async fn waveform_up(&mut self, dma: Peri<'_, impl super::UpDma<T>>, channel: Channel, duty: &[u16]) {
-        self.inner.waveform_up(dma, channel, duty).await;
+        self.inner.enable_channel(channel, true);
+        self.inner.enable_update_dma(true);
+        self.inner.setup_update_dma(dma, channel, duty).await;
+        self.inner.enable_update_dma(false);
     }
 
     /// Generate a multichannel sequence of PWM waveforms using DMA triggered by timer update events.
@@ -350,7 +380,6 @@ impl<'d, T: GeneralInstance4Channel> SimplePwm<'d, T> {
     /// Also be aware that embassy timers use one of timers internally. It is possible to
     /// switch this timer by using `time-driver-timX` feature.
     ///
-    #[inline(always)]
     pub async fn waveform_up_multi_channel(
         &mut self,
         dma: Peri<'_, impl super::UpDma<T>>,
@@ -358,15 +387,11 @@ impl<'d, T: GeneralInstance4Channel> SimplePwm<'d, T> {
         ending_channel: Channel,
         duty: &[u16],
     ) {
+        self.inner.enable_update_dma(true);
         self.inner
-            .waveform_up_multi_channel(dma, starting_channel, ending_channel, duty)
+            .setup_update_dma_burst(dma, starting_channel, ending_channel, duty)
             .await;
-    }
-
-    /// Generate a sequence of PWM waveform
-    #[inline(always)]
-    pub async fn waveform<C: TimerChannel>(&mut self, dma: Peri<'_, impl super::Dma<T, C>>, duty: &[u16]) {
-        self.inner.waveform(dma, duty).await;
+        self.inner.enable_update_dma(false);
     }
 }
 
diff --git a/embassy-stm32/src/usart/mod.rs b/embassy-stm32/src/usart/mod.rs
index 0e7da634d..8047d6005 100644
--- a/embassy-stm32/src/usart/mod.rs
+++ b/embassy-stm32/src/usart/mod.rs
@@ -491,6 +491,8 @@ impl<'d> UartTx<'d, Async> {
 
     /// Initiate an asynchronous UART write
     pub async fn write(&mut self, buffer: &[u8]) -> Result<(), Error> {
+        let _ = self.info.rcc.block_stop();
+
         let r = self.info.regs;
 
         half_duplex_set_rx_tx_before_write(&r, self.duplex == Duplex::Half(HalfDuplexReadback::Readback));
@@ -508,6 +510,8 @@ impl<'d> UartTx<'d, Async> {
 
     /// Wait until transmission complete
     pub async fn flush(&mut self) -> Result<(), Error> {
+        let _ = self.info.rcc.block_stop();
+
         flush(&self.info, &self.state).await
     }
 }
@@ -569,7 +573,7 @@ impl<'d, M: Mode> UartTx<'d, M> {
         let state = self.state;
         state.tx_rx_refcount.store(1, Ordering::Relaxed);
 
-        info.rcc.enable_and_reset();
+        info.rcc.enable_and_reset_without_stop();
 
         info.regs.cr3().modify(|w| {
             w.set_ctse(self.cts.is_some());
@@ -726,6 +730,8 @@ impl<'d> UartRx<'d, Async> {
 
     /// Initiate an asynchronous UART read
     pub async fn read(&mut self, buffer: &mut [u8]) -> Result<(), Error> {
+        let _ = self.info.rcc.block_stop();
+
         self.inner_read(buffer, false).await?;
 
         Ok(())
@@ -733,6 +739,8 @@ impl<'d> UartRx<'d, Async> {
 
     /// Initiate an asynchronous read with idle line detection enabled
     pub async fn read_until_idle(&mut self, buffer: &mut [u8]) -> Result<usize, Error> {
+        let _ = self.info.rcc.block_stop();
+
         self.inner_read(buffer, true).await
     }
 
@@ -1004,7 +1012,7 @@ impl<'d, M: Mode> UartRx<'d, M> {
             .eager_reads
             .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
 
-        info.rcc.enable_and_reset();
+        info.rcc.enable_and_reset_without_stop();
 
         info.regs.cr3().write(|w| {
             w.set_rtse(self.rts.is_some());
@@ -1143,7 +1151,7 @@ fn drop_tx_rx(info: &Info, state: &State) {
         refcount == 1
     });
     if is_last_drop {
-        info.rcc.disable();
+        info.rcc.disable_without_stop();
     }
 }
 
@@ -1506,7 +1514,7 @@ impl<'d, M: Mode> Uart<'d, M> {
             .eager_reads
             .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
 
-        info.rcc.enable_and_reset();
+        info.rcc.enable_and_reset_without_stop();
 
         info.regs.cr3().write(|w| {
             w.set_rtse(self.rx.rts.is_some());
diff --git a/embassy-stm32/src/usart/ringbuffered.rs b/embassy-stm32/src/usart/ringbuffered.rs
index bac570d27..cc5224b69 100644
--- a/embassy-stm32/src/usart/ringbuffered.rs
+++ b/embassy-stm32/src/usart/ringbuffered.rs
@@ -117,6 +117,8 @@ impl<'d> UartRx<'d, Async> {
         let rx = unsafe { self.rx.as_ref().map(|x| x.clone_unchecked()) };
         let rts = unsafe { self.rts.as_ref().map(|x| x.clone_unchecked()) };
 
+        info.rcc.increment_stop_refcount();
+
         // Don't disable the clock
         mem::forget(self);
 
@@ -324,6 +326,7 @@ impl<'d> RingBufferedUartRx<'d> {
 
 impl Drop for RingBufferedUartRx<'_> {
     fn drop(&mut self) {
+        self.info.rcc.decrement_stop_refcount();
         self.stop_uart();
         self.rx.as_ref().map(|x| x.set_as_disconnected());
         self.rts.as_ref().map(|x| x.set_as_disconnected());
diff --git a/embassy-stm32/src/xspi/mod.rs b/embassy-stm32/src/xspi/mod.rs
index a80a2692b..466e1a9b4 100644
--- a/embassy-stm32/src/xspi/mod.rs
+++ b/embassy-stm32/src/xspi/mod.rs
@@ -420,9 +420,9 @@ impl<'d, T: Instance, M: PeriMode> Xspi<'d, T, M> {
             return Err(XspiError::InvalidCommand);
         }
 
-        T::REGS.cr().modify(|w| {
-            w.set_fmode(0.into());
-        });
+        T::REGS
+            .cr()
+            .modify(|w| w.set_fmode(Fmode::from_bits(XspiMode::IndirectWrite.into())));
 
         // Configure alternate bytes
         if let Some(ab) = command.alternate_bytes {
@@ -538,8 +538,8 @@ impl<'d, T: Instance, M: PeriMode> Xspi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        // self.configure_command(&transaction, Some(buf.len()))?;
-        self.configure_command(&transaction, Some(buf.len())).unwrap();
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -578,7 +578,8 @@ impl<'d, T: Instance, M: PeriMode> Xspi<'d, T, M> {
             w.set_dmaen(false);
         });
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         T::REGS
             .cr()
@@ -1145,7 +1146,8 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -1160,16 +1162,18 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.blocking_wait();
+            transfer.blocking_wait();
+        }
 
         finish_dma(T::REGS);
 
@@ -1185,21 +1189,24 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(Fmode::from_bits(XspiMode::IndirectWrite.into())));
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .write(buf, T::REGS.dr().as_ptr() as *mut W, Default::default())
-        };
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .write(chunk, T::REGS.dr().as_ptr() as *mut W, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.blocking_wait();
+            transfer.blocking_wait();
+        }
 
         finish_dma(T::REGS);
 
@@ -1215,7 +1222,8 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
 
         let current_address = T::REGS.ar().read().address();
         let current_instruction = T::REGS.ir().read().instruction();
@@ -1230,16 +1238,18 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
             T::REGS.ar().write(|v| v.set_address(current_address));
         }
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .read(T::REGS.dr().as_ptr() as *mut W, buf, Default::default())
-        };
+        for chunk in buf.chunks_mut(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .read(T::REGS.dr().as_ptr() as *mut W, chunk, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.await;
+            transfer.await;
+        }
 
         finish_dma(T::REGS);
 
@@ -1255,21 +1265,25 @@ impl<'d, T: Instance> Xspi<'d, T, Async> {
         // Wait for peripheral to be free
         while T::REGS.sr().read().busy() {}
 
-        self.configure_command(&transaction, Some(buf.len()))?;
+        let transfer_size_bytes = buf.len() * W::size().bytes();
+        self.configure_command(&transaction, Some(transfer_size_bytes))?;
         T::REGS
             .cr()
             .modify(|v| v.set_fmode(Fmode::from_bits(XspiMode::IndirectWrite.into())));
 
-        let transfer = unsafe {
-            self.dma
-                .as_mut()
-                .unwrap()
-                .write(buf, T::REGS.dr().as_ptr() as *mut W, Default::default())
-        };
+        // TODO: implement this using a LinkedList DMA to offload the whole transfer off the CPU.
+        for chunk in buf.chunks(0xFFFF / W::size().bytes()) {
+            let transfer = unsafe {
+                self.dma
+                    .as_mut()
+                    .unwrap()
+                    .write(chunk, T::REGS.dr().as_ptr() as *mut W, Default::default())
+            };
 
-        T::REGS.cr().modify(|w| w.set_dmaen(true));
+            T::REGS.cr().modify(|w| w.set_dmaen(true));
 
-        transfer.await;
+            transfer.await;
+        }
 
         finish_dma(T::REGS);
 
diff --git a/embassy-time-driver/CHANGELOG.md b/embassy-time-driver/CHANGELOG.md
index cdd432437..4951f8c3e 100644
--- a/embassy-time-driver/CHANGELOG.md
+++ b/embassy-time-driver/CHANGELOG.md
@@ -11,7 +11,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## 0.2.1 - 2025-08-26
 
 - Allow inlining on time driver boundary
-- add 133MHz tick rate to support PR2040 @ 133MHz when `TIMERx`'s `SOURCE` is set to `SYSCLK`
+- Add 133MHz tick rate to support PR2040 @ 133MHz when `TIMERx`'s `SOURCE` is set to `SYSCLK`
+- Add 375KHz tick rate support
 
 ## 0.2.0 - 2025-01-02
 
diff --git a/embassy-time-driver/Cargo.toml b/embassy-time-driver/Cargo.toml
index a52e82433..cbb6168b9 100644
--- a/embassy-time-driver/Cargo.toml
+++ b/embassy-time-driver/Cargo.toml
@@ -118,6 +118,8 @@ tick-hz-256_000 = []
 tick-hz-262_144 = []
 ## 320.0kHz Tick Rate
 tick-hz-320_000 = []
+## 375.0kHz Tick Rate
+tick-hz-375_000 = []
 ## 512.0kHz Tick Rate
 tick-hz-512_000 = []
 ## 524.288kHz Tick Rate
diff --git a/embassy-time-driver/gen_tick.py b/embassy-time-driver/gen_tick.py
index 080434457..3cb6552df 100644
--- a/embassy-time-driver/gen_tick.py
+++ b/embassy-time-driver/gen_tick.py
@@ -1,5 +1,4 @@
 import os
-from glob import glob
 
 abspath = os.path.abspath(__file__)
 dname = os.path.dirname(abspath)
@@ -22,6 +21,8 @@ for i in range(1, 30):
     ticks.append(10 * i * 1_000_000)
 for i in range(15, 50):
     ticks.append(20 * i * 1_000_000)
+
+ticks.append(375 * 1000)
 ticks.append(133 * 1_000_000)
 
 seen = set()
diff --git a/embassy-time-driver/src/tick.rs b/embassy-time-driver/src/tick.rs
index 5059e1628..247ec9ab3 100644
--- a/embassy-time-driver/src/tick.rs
+++ b/embassy-time-driver/src/tick.rs
@@ -74,6 +74,8 @@ pub const TICK_HZ: u64 = 256_000;
 pub const TICK_HZ: u64 = 262_144;
 #[cfg(feature = "tick-hz-320_000")]
 pub const TICK_HZ: u64 = 320_000;
+#[cfg(feature = "tick-hz-375_000")]
+pub const TICK_HZ: u64 = 375_000;
 #[cfg(feature = "tick-hz-512_000")]
 pub const TICK_HZ: u64 = 512_000;
 #[cfg(feature = "tick-hz-524_288")]
@@ -358,6 +360,7 @@ pub const TICK_HZ: u64 = 5_242_880_000;
     feature = "tick-hz-256_000",
     feature = "tick-hz-262_144",
     feature = "tick-hz-320_000",
+    feature = "tick-hz-375_000",
     feature = "tick-hz-512_000",
     feature = "tick-hz-524_288",
     feature = "tick-hz-640_000",
diff --git a/embassy-time/CHANGELOG.md b/embassy-time/CHANGELOG.md
index 4a50da8ef..17f8a3837 100644
--- a/embassy-time/CHANGELOG.md
+++ b/embassy-time/CHANGELOG.md
@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased - ReleaseDate
 
 - Add as_nanos and from_nanos where missing
+- Added 375KHz tick rate support
 
 ## 0.5.0 - 2025-08-26
 
diff --git a/embassy-time/Cargo.toml b/embassy-time/Cargo.toml
index 05614dbf5..a7ed51e78 100644
--- a/embassy-time/Cargo.toml
+++ b/embassy-time/Cargo.toml
@@ -178,6 +178,8 @@ tick-hz-256_000 = ["embassy-time-driver/tick-hz-256_000"]
 tick-hz-262_144 = ["embassy-time-driver/tick-hz-262_144"]
 ## 320.0kHz Tick Rate
 tick-hz-320_000 = ["embassy-time-driver/tick-hz-320_000"]
+## 375.0kHz Tick Rate
+tick-hz-375_000 = ["embassy-time-driver/tick-hz-375_000"]
 ## 512.0kHz Tick Rate
 tick-hz-512_000 = ["embassy-time-driver/tick-hz-512_000"]
 ## 524.288kHz Tick Rate
diff --git a/embassy-usb/CHANGELOG.md b/embassy-usb/CHANGELOG.md
index cfb1bf021..3dd71ffbc 100644
--- a/embassy-usb/CHANGELOG.md
+++ b/embassy-usb/CHANGELOG.md
@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Unreleased - ReleaseDate
 
 - Add support for USB HID Boot Protocol Mode 
+- Bump usbd-hid from 0.8.1 to 0.9.0
 
 ## 0.5.1 - 2025-08-26
 
diff --git a/embassy-usb/Cargo.toml b/embassy-usb/Cargo.toml
index aeb7392f1..3d1e005e4 100644
--- a/embassy-usb/Cargo.toml
+++ b/embassy-usb/Cargo.toml
@@ -69,5 +69,5 @@ heapless = "0.8"
 embedded-io-async = "0.6.1"
 
 # for HID
-usbd-hid = { version = "0.8.1", optional = true }
+usbd-hid = { version = "0.9.0", optional = true }
 ssmarshal = { version = "1.0", default-features = false, optional = true }
diff --git a/examples/nrf52840/Cargo.toml b/examples/nrf52840/Cargo.toml
index a026d6352..1fe3d2419 100644
--- a/examples/nrf52840/Cargo.toml
+++ b/examples/nrf52840/Cargo.toml
@@ -28,7 +28,7 @@ cortex-m-rt = "0.7.0"
 panic-probe = { version = "1.0.0", features = ["print-defmt"] }
 rand = { version = "0.9.0", default-features = false }
 embedded-storage = "0.3.1"
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 serde = { version = "1.0.136", default-features = false }
 embedded-hal = { version = "1.0" }
 embedded-hal-async = { version = "1.0" }
diff --git a/examples/nrf5340/Cargo.toml b/examples/nrf5340/Cargo.toml
index 4dcbdd715..97efe58e8 100644
--- a/examples/nrf5340/Cargo.toml
+++ b/examples/nrf5340/Cargo.toml
@@ -23,7 +23,7 @@ cortex-m = { version = "0.7.6", features = ["inline-asm", "critical-section-sing
 cortex-m-rt = "0.7.0"
 panic-probe = { version = "1.0.0", features = ["print-defmt"] }
 embedded-storage = "0.3.1"
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 serde = { version = "1.0.136", default-features = false }
 
 [profile.release]
diff --git a/examples/rp/Cargo.toml b/examples/rp/Cargo.toml
index 640addb28..9d7d99259 100644
--- a/examples/rp/Cargo.toml
+++ b/examples/rp/Cargo.toml
@@ -45,7 +45,7 @@ display-interface = "0.5.0"
 byte-slice-cast = { version = "1.2.0", default-features = false }
 smart-leds = "0.4.0"
 heapless = "0.8"
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 
 embedded-hal-1 = { package = "embedded-hal", version = "1.0" }
 embedded-hal-async = "1.0"
diff --git a/examples/rp235x/Cargo.toml b/examples/rp235x/Cargo.toml
index 39a4f421a..ad396275b 100644
--- a/examples/rp235x/Cargo.toml
+++ b/examples/rp235x/Cargo.toml
@@ -46,7 +46,7 @@ display-interface = "0.5.0"
 byte-slice-cast = { version = "1.2.0", default-features = false }
 smart-leds = "0.3.0"
 heapless = "0.8"
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 
 embedded-hal-1 = { package = "embedded-hal", version = "1.0" }
 embedded-hal-async = "1.0"
diff --git a/examples/stm32f4/Cargo.toml b/examples/stm32f4/Cargo.toml
index d06b7505c..b4555045a 100644
--- a/examples/stm32f4/Cargo.toml
+++ b/examples/stm32f4/Cargo.toml
@@ -32,7 +32,7 @@ critical-section = "1.1"
 nb = "1.0.0"
 embedded-storage = "0.3.1"
 micromath = "2.0.0"
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 static_cell = "2"
 chrono = { version = "^0.4", default-features = false}
 
diff --git a/examples/stm32f7/src/bin/pwm.rs b/examples/stm32f7/src/bin/pwm.rs
new file mode 100644
index 000000000..b071eb597
--- /dev/null
+++ b/examples/stm32f7/src/bin/pwm.rs
@@ -0,0 +1,61 @@
+#![no_std]
+#![no_main]
+
+use defmt::*;
+use embassy_executor::Spawner;
+use embassy_stm32::Config;
+use embassy_stm32::gpio::OutputType;
+use embassy_stm32::time::{Hertz, mhz};
+use embassy_stm32::timer::simple_pwm::{PwmPin, SimplePwm};
+use embassy_time::Timer;
+use {defmt_rtt as _, panic_probe as _};
+
+// If you are trying this and your USB device doesn't connect, the most
+// common issues are the RCC config and vbus_detection
+//
+// See https://embassy.dev/book/#_the_usb_examples_are_not_working_on_my_board_is_there_anything_else_i_need_to_configure
+// for more information.
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    info!("Hello World!");
+
+    let mut config = Config::default();
+    {
+        use embassy_stm32::rcc::*;
+        config.rcc.hse = Some(Hse {
+            freq: Hertz(8_000_000),
+            mode: HseMode::Bypass,
+        });
+        config.rcc.pll_src = PllSource::HSE;
+        config.rcc.pll = Some(Pll {
+            prediv: PllPreDiv::DIV4,
+            mul: PllMul::MUL200,
+            divp: Some(PllPDiv::DIV2), // 8mhz / 4 * 200 / 2 = 200Mhz
+            divq: Some(PllQDiv::DIV4), // 8mhz / 4 * 200 / 4 = 100Mhz
+            divr: None,
+        });
+        config.rcc.ahb_pre = AHBPrescaler::DIV1;
+        config.rcc.apb1_pre = APBPrescaler::DIV4;
+        config.rcc.apb2_pre = APBPrescaler::DIV2;
+        config.rcc.sys = Sysclk::PLL1_P;
+    }
+    let p = embassy_stm32::init(config);
+    let ch1_pin = PwmPin::new(p.PE9, OutputType::PushPull);
+    let mut pwm = SimplePwm::new(p.TIM1, Some(ch1_pin), None, None, None, mhz(1), Default::default());
+    let mut ch1 = pwm.ch1();
+    ch1.enable();
+
+    info!("PWM initialized");
+    info!("PWM max duty {}", ch1.max_duty_cycle());
+
+    loop {
+        ch1.set_duty_cycle_fully_off();
+        Timer::after_millis(300).await;
+        ch1.set_duty_cycle_fraction(1, 4);
+        Timer::after_millis(300).await;
+        ch1.set_duty_cycle_fraction(1, 2);
+        Timer::after_millis(300).await;
+        ch1.set_duty_cycle(ch1.max_duty_cycle() - 1);
+        Timer::after_millis(300).await;
+    }
+}
diff --git a/examples/stm32f7/src/bin/pwm_ringbuffer.rs b/examples/stm32f7/src/bin/pwm_ringbuffer.rs
new file mode 100644
index 000000000..4d191ac13
--- /dev/null
+++ b/examples/stm32f7/src/bin/pwm_ringbuffer.rs
@@ -0,0 +1,153 @@
+#![no_std]
+#![no_main]
+
+use defmt::*;
+use embassy_executor::Spawner;
+use embassy_stm32::Config;
+use embassy_stm32::gpio::OutputType;
+use embassy_stm32::time::mhz;
+use embassy_stm32::timer::simple_pwm::{PwmPin, SimplePwm};
+use embassy_time::Timer;
+use {defmt_rtt as _, panic_probe as _};
+
+// If you are trying this and your USB device doesn't connect, the most
+// common issues are the RCC config and vbus_detection
+//
+// See https://embassy.dev/book/#_the_usb_examples_are_not_working_on_my_board_is_there_anything_else_i_need_to_configure
+// for more information.
+#[embassy_executor::main]
+async fn main(_spawner: Spawner) {
+    info!("PWM Ring Buffer Example");
+
+    let mut config = Config::default();
+    {
+        use embassy_stm32::rcc::*;
+        use embassy_stm32::time::Hertz;
+        config.rcc.hse = Some(Hse {
+            freq: Hertz(8_000_000),
+            mode: HseMode::Bypass,
+        });
+        config.rcc.pll_src = PllSource::HSE;
+        config.rcc.pll = Some(Pll {
+            prediv: PllPreDiv::DIV4,
+            mul: PllMul::MUL200,
+            divp: Some(PllPDiv::DIV2), // 8mhz / 4 * 200 / 2 = 200Mhz
+            divq: Some(PllQDiv::DIV4), // 8mhz / 4 * 200 / 4 = 100Mhz
+            divr: None,
+        });
+        config.rcc.ahb_pre = AHBPrescaler::DIV1;
+        config.rcc.apb1_pre = APBPrescaler::DIV4;
+        config.rcc.apb2_pre = APBPrescaler::DIV2;
+        config.rcc.sys = Sysclk::PLL1_P;
+    }
+    let p = embassy_stm32::init(config);
+
+    // Initialize PWM on TIM1
+    let ch1_pin = PwmPin::new(p.PE9, OutputType::PushPull);
+    let ch2_pin = PwmPin::new(p.PE11, OutputType::PushPull);
+    let mut pwm = SimplePwm::new(
+        p.TIM1,
+        Some(ch1_pin),
+        Some(ch2_pin),
+        None,
+        None,
+        mhz(1),
+        Default::default(),
+    );
+
+    // Use channel 1 for static PWM at 50%
+    let mut ch1 = pwm.ch1();
+    ch1.enable();
+    ch1.set_duty_cycle_fraction(1, 2);
+    info!("Channel 1 (PE9/D6): Static 50% duty cycle");
+
+    // Get max duty from channel 1 before converting channel 2
+    let max_duty = ch1.max_duty_cycle();
+    info!("PWM max duty: {}", max_duty);
+
+    // Create a DMA ring buffer for channel 2
+    const BUFFER_SIZE: usize = 128;
+    static mut DMA_BUFFER: [u16; BUFFER_SIZE] = [0u16; BUFFER_SIZE];
+    let dma_buffer = unsafe { &mut *core::ptr::addr_of_mut!(DMA_BUFFER) };
+
+    // Pre-fill buffer with initial sine wave using lookup table approach
+    for i in 0..BUFFER_SIZE {
+        // Simple sine approximation using triangle wave
+        let phase = (i * 256) / BUFFER_SIZE;
+        let sine_approx = if phase < 128 {
+            phase as u16 * 2
+        } else {
+            (255 - phase) as u16 * 2
+        };
+        dma_buffer[i] = (sine_approx as u32 * max_duty as u32 / 256) as u16;
+    }
+
+    // Convert channel 2 to ring-buffered PWM
+    let mut ring_pwm = pwm.ch1().into_ring_buffered_channel(p.DMA2_CH5, dma_buffer);
+
+    info!("Ring buffer capacity: {}", ring_pwm.capacity());
+
+    // Pre-write some initial data to the buffer before starting
+    info!("Pre-writing initial waveform data...");
+
+    ring_pwm.write(&[0; BUFFER_SIZE]).unwrap();
+
+    // Enable the PWM channel output
+    ring_pwm.enable();
+
+    // Start the DMA ring buffer
+    ring_pwm.start();
+    info!("Channel 2 (PE11/D5): Ring buffered sine wave started");
+
+    // Give DMA time to start consuming
+    Timer::after_millis(10).await;
+
+    // Continuously update the waveform
+    let mut phase: f32 = 0.0;
+    let mut amplitude: f32 = 1.0;
+    let mut amplitude_direction = -0.05;
+
+    loop {
+        // Generate new waveform data with varying amplitude
+        let mut new_data = [0u16; 32];
+        for i in 0..new_data.len() {
+            // Triangle wave approximation for sine
+            let pos = ((i as u32 + phase as u32) * 4) % 256;
+            let sine_approx = if pos < 128 {
+                pos as u16 * 2
+            } else {
+                (255 - pos) as u16 * 2
+            };
+            let scaled = (sine_approx as u32 * (amplitude * 256.0) as u32) / (256 * 256);
+            new_data[i] = ((scaled * max_duty as u32) / 256) as u16;
+        }
+
+        // Write new data to the ring buffer
+        match ring_pwm.write_exact(&new_data).await {
+            Ok(_remaining) => {}
+            Err(e) => {
+                info!("Write error: {:?}", e);
+            }
+        }
+
+        // Update phase for animation effect
+        phase += 2.0;
+        if phase >= 64.0 {
+            phase = 0.0;
+        }
+
+        // Vary amplitude for breathing effect
+        amplitude += amplitude_direction;
+        if amplitude <= 0.2 || amplitude >= 1.0 {
+            amplitude_direction = -amplitude_direction;
+        }
+
+        // Log buffer status periodically
+        if (phase as u32) % 10 == 0 {
+            match ring_pwm.len() {
+                Ok(len) => info!("Ring buffer fill: {}/{}", len, ring_pwm.capacity()),
+                Err(_) => info!("Error reading buffer length"),
+            }
+        }
+    }
+}
diff --git a/examples/stm32g4/Cargo.toml b/examples/stm32g4/Cargo.toml
index 8bbeb594c..d1c19582b 100644
--- a/examples/stm32g4/Cargo.toml
+++ b/examples/stm32g4/Cargo.toml
@@ -13,7 +13,7 @@ embassy-executor = { path = "../../embassy-executor", features = ["arch-cortex-m
 embassy-time = { path = "../../embassy-time", features = ["defmt", "defmt-timestamp-uptime", "tick-hz-32_768"] }
 embassy-usb = { path = "../../embassy-usb", features = ["defmt"] }
 embassy-futures = { path = "../../embassy-futures" }
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 
 defmt = "1.0.1"
 defmt-rtt = "1.0.0"
diff --git a/examples/stm32l5/Cargo.toml b/examples/stm32l5/Cargo.toml
index b6158c854..586b00836 100644
--- a/examples/stm32l5/Cargo.toml
+++ b/examples/stm32l5/Cargo.toml
@@ -14,7 +14,7 @@ embassy-time = { version = "0.5.0", path = "../../embassy-time", features = ["de
 embassy-usb = { version = "0.5.1", path = "../../embassy-usb", features = ["defmt"] }
 embassy-net = { version = "0.7.1", path = "../../embassy-net", features = ["defmt",  "tcp", "dhcpv4", "medium-ethernet"] }
 embassy-futures = { version = "0.1.2", path = "../../embassy-futures" }
-usbd-hid = "0.8.1"
+usbd-hid = "0.9.0"
 
 defmt = "1.0.1"
 defmt-rtt = "1.0.0"