5 files changed, 239 insertions, 64 deletions

diff --git a/embassy-hal-internal/src/atomic_ring_buffer.rs b/embassy-hal-internal/src/atomic_ring_buffer.rs
index 00b7a1249..8c3889b85 100644
--- a/embassy-hal-internal/src/atomic_ring_buffer.rs
+++ b/embassy-hal-internal/src/atomic_ring_buffer.rs
@@ -133,6 +133,18 @@ impl RingBuffer {
         self.len.load(Ordering::Relaxed)
     }
 
+    /// Return number of items available to read.
+    pub fn available(&self) -> usize {
+        let end = self.end.load(Ordering::Relaxed);
+        let len = self.len.load(Ordering::Relaxed);
+        let start = self.start.load(Ordering::Relaxed);
+        if end >= start {
+            end - start
+        } else {
+            2 * len - start + end
+        }
+    }
+
     /// Check if buffer is full.
     pub fn is_full(&self) -> bool {
         let len = self.len.load(Ordering::Relaxed);
@@ -142,6 +154,11 @@ impl RingBuffer {
         self.wrap(start + len) == end
     }
 
+    /// Check if buffer is at least half full.
+    pub fn is_half_full(&self) -> bool {
+        self.available() >= self.len.load(Ordering::Relaxed) / 2
+    }
+
     /// Check if buffer is empty.
     pub fn is_empty(&self) -> bool {
         let start = self.start.load(Ordering::Relaxed);
@@ -394,6 +411,7 @@ mod tests {
             rb.init(b.as_mut_ptr(), 4);
 
             assert_eq!(rb.is_empty(), true);
+            assert_eq!(rb.is_half_full(), false);
             assert_eq!(rb.is_full(), false);
 
             rb.writer().push(|buf| {
@@ -406,6 +424,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), true);
 
             rb.writer().push(|buf| {
@@ -415,6 +434,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), true);
 
             rb.reader().pop(|buf| {
@@ -424,6 +444,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), false);
 
             rb.reader().pop(|buf| {
@@ -432,6 +453,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), false);
 
             rb.reader().pop(|buf| {
@@ -447,6 +469,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), true);
+            assert_eq!(rb.is_half_full(), false);
             assert_eq!(rb.is_full(), false);
 
             rb.reader().pop(|buf| {
@@ -460,14 +483,28 @@ mod tests {
                 1
             });
 
+            assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), false);
+            assert_eq!(rb.is_full(), false);
+
             rb.writer().push(|buf| {
                 assert_eq!(3, buf.len());
                 buf[0] = 11;
-                buf[1] = 12;
-                2
+                1
+            });
+
+            assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
+            assert_eq!(rb.is_full(), false);
+
+            rb.writer().push(|buf| {
+                assert_eq!(2, buf.len());
+                buf[0] = 12;
+                1
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), false);
 
             rb.writer().push(|buf| {
@@ -477,6 +514,7 @@ mod tests {
             });
 
             assert_eq!(rb.is_empty(), false);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), true);
         }
     }
@@ -490,6 +528,7 @@ mod tests {
             rb.init(b.as_mut_ptr(), b.len());
 
             assert_eq!(rb.is_empty(), true);
+            assert_eq!(rb.is_half_full(), true);
             assert_eq!(rb.is_full(), true);
 
             rb.writer().push(|buf| {
diff --git a/embassy-stm32/CHANGELOG.md b/embassy-stm32/CHANGELOG.md
index 1443472f5..a6ee5c4b8 100644
--- a/embassy-stm32/CHANGELOG.md
+++ b/embassy-stm32/CHANGELOG.md
@@ -27,6 +27,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - fix: Cut down the capabilities of the STM32L412 and L422 RTC as those are missing binary timer mode and underflow interrupt.
 - fix: Allow configuration of the internal pull up/down resistors on the pins for the Qei peripheral, as well as the Qei decoder mode.
 - feat: stm32/rcc/mco: Added support for IO driver strength when using Master Clock Out IO. This changes signature on Mco::new taking a McoConfig struct ([#4679](https://github.com/embassy-rs/embassy/pull/4679))
+- feat: derive Clone, Copy and defmt::Format for all SPI-related configs
+- 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
 
 ## 0.4.0 - 2025-08-26
 
diff --git a/embassy-stm32/src/usart/buffered.rs b/embassy-stm32/src/usart/buffered.rs
index c734eed49..10dc02334 100644
--- a/embassy-stm32/src/usart/buffered.rs
+++ b/embassy-stm32/src/usart/buffered.rs
@@ -1,7 +1,7 @@
 use core::future::poll_fn;
 use core::marker::PhantomData;
 use core::slice;
-use core::sync::atomic::{AtomicBool, AtomicU8, Ordering};
+use core::sync::atomic::{AtomicBool, AtomicU8, AtomicUsize, Ordering};
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
@@ -68,8 +68,15 @@ unsafe fn on_interrupt(r: Regs, state: &'static State) {
             // FIXME: Should we disable any further RX interrupts when the buffer becomes full.
         }
 
-        if !state.rx_buf.is_empty() {
-            state.rx_waker.wake();
+        let eager = state.eager_reads.load(Ordering::Relaxed);
+        if eager > 0 {
+            if state.rx_buf.available() >= eager {
+                state.rx_waker.wake();
+            }
+        } else {
+            if state.rx_buf.is_half_full() {
+                state.rx_waker.wake();
+            }
         }
     }
 
@@ -132,6 +139,7 @@ pub(super) struct State {
     tx_done: AtomicBool,
     tx_rx_refcount: AtomicU8,
     half_duplex_readback: AtomicBool,
+    eager_reads: AtomicUsize,
 }
 
 impl State {
@@ -144,6 +152,7 @@ impl State {
             tx_done: AtomicBool::new(true),
             tx_rx_refcount: AtomicU8::new(0),
             half_duplex_readback: AtomicBool::new(false),
+            eager_reads: AtomicUsize::new(0),
         }
     }
 }
@@ -419,6 +428,9 @@ impl<'d> BufferedUart<'d> {
         let state = T::buffered_state();
         let kernel_clock = T::frequency();
 
+        state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
         state.half_duplex_readback.store(
             config.duplex == Duplex::Half(HalfDuplexReadback::Readback),
             Ordering::Relaxed,
@@ -456,6 +468,9 @@ impl<'d> BufferedUart<'d> {
         let info = self.rx.info;
         let state = self.rx.state;
         state.tx_rx_refcount.store(2, Ordering::Relaxed);
+        state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
 
         info.rcc.enable_and_reset();
 
@@ -527,6 +542,11 @@ impl<'d> BufferedUart<'d> {
     pub fn set_config(&mut self, config: &Config) -> Result<(), ConfigError> {
         reconfigure(self.rx.info, self.rx.kernel_clock, config)?;
 
+        self.rx
+            .state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
+
         self.rx.info.regs.cr1().modify(|w| {
             w.set_rxneie(true);
             w.set_idleie(true);
@@ -553,24 +573,30 @@ impl<'d> BufferedUartRx<'d> {
         poll_fn(move |cx| {
             let state = self.state;
             let mut rx_reader = unsafe { state.rx_buf.reader() };
-            let data = rx_reader.pop_slice();
+            let mut buf_len = 0;
+            let mut data = rx_reader.pop_slice();
 
-            if !data.is_empty() {
-                let len = data.len().min(buf.len());
-                buf[..len].copy_from_slice(&data[..len]);
+            while !data.is_empty() && buf_len < buf.len() {
+                let data_len = data.len().min(buf.len() - buf_len);
+                buf[buf_len..buf_len + data_len].copy_from_slice(&data[..data_len]);
+                buf_len += data_len;
 
                 let do_pend = state.rx_buf.is_full();
-                rx_reader.pop_done(len);
+                rx_reader.pop_done(data_len);
 
                 if do_pend {
                     self.info.interrupt.pend();
                 }
 
-                return Poll::Ready(Ok(len));
+                data = rx_reader.pop_slice();
             }
 
-            state.rx_waker.register(cx.waker());
-            Poll::Pending
+            if buf_len != 0 {
+                Poll::Ready(Ok(buf_len))
+            } else {
+                state.rx_waker.register(cx.waker());
+                Poll::Pending
+            }
         })
         .await
     }
@@ -579,21 +605,24 @@ impl<'d> BufferedUartRx<'d> {
         loop {
             let state = self.state;
             let mut rx_reader = unsafe { state.rx_buf.reader() };
-            let data = rx_reader.pop_slice();
+            let mut buf_len = 0;
+            let mut data = rx_reader.pop_slice();
 
-            if !data.is_empty() {
-                let len = data.len().min(buf.len());
-                buf[..len].copy_from_slice(&data[..len]);
+            while !data.is_empty() && buf_len < buf.len() {
+                let data_len = data.len().min(buf.len() - buf_len);
+                buf[buf_len..buf_len + data_len].copy_from_slice(&data[..data_len]);
+                buf_len += data_len;
 
                 let do_pend = state.rx_buf.is_full();
-                rx_reader.pop_done(len);
+                rx_reader.pop_done(data_len);
 
                 if do_pend {
                     self.info.interrupt.pend();
                 }
 
-                return Ok(len);
+                data = rx_reader.pop_slice();
             }
+            return Ok(buf_len);
         }
     }
 
@@ -633,6 +662,10 @@ impl<'d> BufferedUartRx<'d> {
     pub fn set_config(&mut self, config: &Config) -> Result<(), ConfigError> {
         reconfigure(self.info, self.kernel_clock, config)?;
 
+        self.state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
+
         self.info.regs.cr1().modify(|w| {
             w.set_rxneie(true);
             w.set_idleie(true);
diff --git a/embassy-stm32/src/usart/mod.rs b/embassy-stm32/src/usart/mod.rs
index ff211e0c9..0d2d86aca 100644
--- a/embassy-stm32/src/usart/mod.rs
+++ b/embassy-stm32/src/usart/mod.rs
@@ -4,7 +4,7 @@
 
 use core::future::poll_fn;
 use core::marker::PhantomData;
-use core::sync::atomic::{compiler_fence, AtomicU8, Ordering};
+use core::sync::atomic::{compiler_fence, AtomicU8, AtomicUsize, Ordering};
 use core::task::Poll;
 
 use embassy_embedded_hal::SetConfig;
@@ -185,6 +185,12 @@ pub enum ConfigError {
     RxOrTxNotEnabled,
     /// Data bits and parity combination not supported
     DataParityNotSupported,
+    /// DE assertion time too high
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    DeAssertionTimeTooHigh,
+    /// DE deassertion time too high
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    DeDeassertionTimeTooHigh,
 }
 
 #[non_exhaustive]
@@ -206,6 +212,21 @@ pub struct Config {
     /// If false: the error is ignored and cleared
     pub detect_previous_overrun: bool,
 
+    /// If `None` (the default) then read-like calls on `BufferedUartRx` and `RingBufferedUartRx`
+    /// typically only wake/return after line idle or after the buffer is at least half full
+    /// (for `BufferedUartRx`) or the DMA buffer is written at the half or full positions
+    /// (for `RingBufferedUartRx`), though it may also wake/return earlier in some circumstances.
+    ///
+    /// If `Some(n)` then such reads are also woken/return as soon as at least `n` words are
+    /// available in the buffer, in addition to waking/returning when the conditions described
+    /// above are met. `Some(0)` is treated as `None`. Setting this for `RingBufferedUartRx`
+    /// will trigger an interrupt for every received word to check the buffer level, which may
+    /// impact performance at high data rates.
+    ///
+    /// Has no effect on plain `Uart` or `UartRx` reads, which are specified to either
+    /// return a single word, a full buffer, or after line idle.
+    pub eager_reads: Option<usize>,
+
     /// Set this to true if the line is considered noise free.
     /// This will increase the receiver’s tolerance to clock deviations,
     /// but will effectively disable noise detection.
@@ -239,6 +260,14 @@ pub struct Config {
     /// Set the pin configuration for the DE pin.
     pub de_config: OutputConfig,
 
+    /// Set DE assertion time before the first start bit, 0-31 16ths of a bit period.
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    pub de_assertion_time: u8,
+
+    /// Set DE deassertion time after the last stop bit, 0-31 16ths of a bit period.
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    pub de_deassertion_time: u8,
+
     // private: set by new_half_duplex, not by the user.
     duplex: Duplex,
 }
@@ -270,6 +299,7 @@ impl Default for Config {
             parity: Parity::ParityNone,
             // historical behavior
             detect_previous_overrun: false,
+            eager_reads: None,
             #[cfg(not(usart_v1))]
             assume_noise_free: false,
             #[cfg(any(usart_v3, usart_v4))]
@@ -283,6 +313,10 @@ impl Default for Config {
             tx_config: OutputConfig::PushPull,
             rts_config: OutputConfig::PushPull,
             de_config: OutputConfig::PushPull,
+            #[cfg(not(any(usart_v1, usart_v2)))]
+            de_assertion_time: 0,
+            #[cfg(not(any(usart_v1, usart_v2)))]
+            de_deassertion_time: 0,
             duplex: Duplex::Full,
         }
     }
@@ -966,6 +1000,9 @@ impl<'d, M: Mode> UartRx<'d, M> {
         let info = self.info;
         let state = self.state;
         state.tx_rx_refcount.store(1, Ordering::Relaxed);
+        state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
 
         info.rcc.enable_and_reset();
 
@@ -982,6 +1019,9 @@ impl<'d, M: Mode> UartRx<'d, M> {
 
     /// Reconfigure the driver
     pub fn set_config(&mut self, config: &Config) -> Result<(), ConfigError> {
+        self.state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
         reconfigure(self.info, self.kernel_clock, config)
     }
 
@@ -1462,6 +1502,9 @@ impl<'d, M: Mode> Uart<'d, M> {
         let info = self.rx.info;
         let state = self.rx.state;
         state.tx_rx_refcount.store(2, Ordering::Relaxed);
+        state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
 
         info.rcc.enable_and_reset();
 
@@ -1690,6 +1733,16 @@ fn configure(
         return Err(ConfigError::RxOrTxNotEnabled);
     }
 
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    let dem = r.cr3().read().dem();
+
+    #[cfg(not(any(usart_v1, usart_v2)))]
+    if config.de_assertion_time > 31 {
+        return Err(ConfigError::DeAssertionTimeTooHigh);
+    } else if config.de_deassertion_time > 31 {
+        return Err(ConfigError::DeDeassertionTimeTooHigh);
+    }
+
     // UART must be disabled during configuration.
     r.cr1().modify(|w| {
         w.set_ue(false);
@@ -1738,6 +1791,20 @@ fn configure(
             w.set_re(enable_rx);
         }
 
+        #[cfg(not(any(usart_v1, usart_v2)))]
+        if dem {
+            w.set_deat(if over8 {
+                config.de_assertion_time / 2
+            } else {
+                config.de_assertion_time
+            });
+            w.set_dedt(if over8 {
+                config.de_assertion_time / 2
+            } else {
+                config.de_assertion_time
+            });
+        }
+
         // configure word size and parity, since the parity bit is inserted into the MSB position,
         // it increases the effective word size
         match (config.parity, config.data_bits) {
@@ -2022,6 +2089,7 @@ struct State {
     rx_waker: AtomicWaker,
     tx_waker: AtomicWaker,
     tx_rx_refcount: AtomicU8,
+    eager_reads: AtomicUsize,
 }
 
 impl State {
@@ -2030,6 +2098,7 @@ impl State {
             rx_waker: AtomicWaker::new(),
             tx_waker: AtomicWaker::new(),
             tx_rx_refcount: AtomicU8::new(0),
+            eager_reads: AtomicUsize::new(0),
         }
     }
 }
diff --git a/embassy-stm32/src/usart/ringbuffered.rs b/embassy-stm32/src/usart/ringbuffered.rs
index 5f4e87834..27071fb31 100644
--- a/embassy-stm32/src/usart/ringbuffered.rs
+++ b/embassy-stm32/src/usart/ringbuffered.rs
@@ -26,9 +26,9 @@ use crate::Peri;
 /// contain enough bytes to fill the buffer passed by the caller of
 /// the function, or is empty.
 ///
-/// Waiting for bytes operates in one of two modes, depending on
-/// the behavior of the sender and the size of the buffer passed
-/// to the function:
+/// Waiting for bytes operates in one of three modes, depending on
+/// the behavior of the sender, the size of the buffer passed
+/// to the function, and the configuration:
 ///
 /// - If the sender sends intermittently, the 'idle line'
 /// condition will be detected when the sender stops, and any
@@ -47,7 +47,11 @@ use crate::Peri;
 /// interrupt when those specific buffer addresses have been
 /// written.
 ///
-/// In both cases this will result in variable latency due to the
+/// - If `eager_reads` is enabled in `config`, the UART interrupt
+/// is enabled on all data reception and the call will only wait
+/// for at least one byte to be available before returning.
+///
+/// In the first two cases this will result in variable latency due to the
 /// buffering effect. For example, if the baudrate is 2400 bps, and
 /// the configuration is 8 data bits, no parity bit, and one stop bit,
 /// then a byte will be received every ~4.16ms. If the ring buffer is
@@ -68,15 +72,10 @@ use crate::Peri;
 /// sending, but would be falsely triggered in the worst-case
 /// buffer delay scenario.
 ///
-/// Note: This latency is caused by the limited capabilities of the
-/// STM32 DMA controller; since it cannot generate an interrupt when
-/// it stores a byte into an empty ring buffer, or in any other
-/// configurable conditions, it is not possible to take notice of the
-/// contents of the ring buffer more quickly without introducing
-/// polling. As a result the latency can be reduced by calling the
-/// read functions repeatedly with smaller buffers to receive the
-/// available bytes, as each call to a read function will explicitly
-/// check the ring buffer for available bytes.
+/// Note: Enabling `eager_reads` with `RingBufferedUartRx` will enable
+/// an UART RXNE interrupt, which will cause an interrupt to occur on
+/// every received data byte. The data is still copied using DMA, but
+/// there is nevertheless additional processing overhead for each byte.
 pub struct RingBufferedUartRx<'d> {
     info: &'static Info,
     state: &'static State,
@@ -133,6 +132,9 @@ impl<'d> UartRx<'d, Async> {
 impl<'d> RingBufferedUartRx<'d> {
     /// Reconfigure the driver
     pub fn set_config(&mut self, config: &Config) -> Result<(), ConfigError> {
+        self.state
+            .eager_reads
+            .store(config.eager_reads.unwrap_or(0), Ordering::Relaxed);
         reconfigure(self.info, self.kernel_clock, config)
     }
 
@@ -148,8 +150,8 @@ impl<'d> RingBufferedUartRx<'d> {
         let r = self.info.regs;
         // clear all interrupts and DMA Rx Request
         r.cr1().modify(|w| {
-            // disable RXNE interrupt
-            w.set_rxneie(false);
+            // use RXNE only when returning reads early
+            w.set_rxneie(self.state.eager_reads.load(Ordering::Relaxed) > 0);
             // enable parity interrupt if not ParityNone
             w.set_peie(w.pce());
             // enable idle line interrupt
@@ -248,39 +250,67 @@ impl<'d> RingBufferedUartRx<'d> {
     async fn wait_for_data_or_idle(&mut self) -> Result<(), Error> {
         compiler_fence(Ordering::SeqCst);
 
-        // Future which completes when idle line is detected
-        let s = self.state;
-        let uart = poll_fn(|cx| {
-            s.rx_waker.register(cx.waker());
-
-            compiler_fence(Ordering::SeqCst);
-
-            if check_idle_and_errors(self.info.regs)? {
-                // Idle line is detected
-                Poll::Ready(Ok(()))
-            } else {
-                Poll::Pending
-            }
-        });
+        loop {
+            // Future which completes when idle line is detected
+            let s = self.state;
+            let mut uart_init = false;
+            let uart = poll_fn(|cx| {
+                s.rx_waker.register(cx.waker());
+
+                compiler_fence(Ordering::SeqCst);
+
+                // We may have been woken by IDLE or, if eager_reads is set, by RXNE.
+                // However, DMA will clear RXNE, so we can't check directly, and because
+                // the other future borrows `ring_buf`, we can't check `len()` here either.
+                // Instead, return from this future and we'll check the length afterwards.
+                let eager = s.eager_reads.load(Ordering::Relaxed) > 0;
+
+                let idle = check_idle_and_errors(self.info.regs)?;
+                if idle || (eager && uart_init) {
+                    // Idle line is detected, or eager reads is set and some data is available.
+                    Poll::Ready(Ok(idle))
+                } else {
+                    uart_init = true;
+                    Poll::Pending
+                }
+            });
 
-        let mut dma_init = false;
-        // Future which completes when the DMA controller indicates it
-        // has written to the ring buffer's middle byte, or last byte
-        let dma = poll_fn(|cx| {
-            self.ring_buf.set_waker(cx.waker());
+            let mut dma_init = false;
+            // Future which completes when the DMA controller indicates it
+            // has written to the ring buffer's middle byte, or last byte
+            let dma = poll_fn(|cx| {
+                self.ring_buf.set_waker(cx.waker());
 
-            let status = match dma_init {
-                false => Poll::Pending,
-                true => Poll::Ready(()),
-            };
+                let status = match dma_init {
+                    false => Poll::Pending,
+                    true => Poll::Ready(()),
+                };
 
-            dma_init = true;
-            status
-        });
+                dma_init = true;
+                status
+            });
 
-        match select(uart, dma).await {
-            Either::Left((result, _)) => result,
-            Either::Right(((), _)) => Ok(()),
+            match select(uart, dma).await {
+                // UART woke with line idle
+                Either::Left((Ok(true), _)) => {
+                    return Ok(());
+                }
+                // UART woke without idle or error: word received
+                Either::Left((Ok(false), _)) => {
+                    let eager = self.state.eager_reads.load(Ordering::Relaxed);
+                    if eager > 0 && self.ring_buf.len().unwrap_or(0) >= eager {
+                        return Ok(());
+                    } else {
+                        continue;
+                    }
+                }
+                // UART woke with error
+                Either::Left((Err(e), _)) => {
+                    return Err(e);
+                }
+                // DMA woke
+                Either::Right(((), _)) => return Ok(()),
+            }
         }
     }