diff --git a/embassy-stm32/src/dma_traits.rs b/embassy-stm32/src/dma_traits.rs
index 8f1a9f40..518e10dc 100644
--- a/embassy-stm32/src/dma_traits.rs
+++ b/embassy-stm32/src/dma_traits.rs
@@ -1,4 +1,5 @@
 use core::future::Future;
+use embassy::util::Unborrow;
 
 pub trait WriteDma<T> {
     type WriteDmaFuture<'a>: Future<Output = ()> + 'a
@@ -19,3 +20,17 @@ pub trait ReadDma<T> {
     where
         T: 'a;
 }
+
+pub trait NoDmaMarker {}
+
+pub struct NoDma;
+
+impl NoDmaMarker for NoDma {}
+
+unsafe impl Unborrow for NoDma {
+    type Target = NoDma;
+
+    unsafe fn unborrow(self) -> Self::Target {
+        self
+    }
+}
diff --git a/embassy-stm32/src/usart/v3.rs b/embassy-stm32/src/usart/v3.rs
index 0071c597..6c9696c7 100644
--- a/embassy-stm32/src/usart/v3.rs
+++ b/embassy-stm32/src/usart/v3.rs
@@ -6,20 +6,30 @@ use embassy_extras::unborrow;
 use crate::pac::usart::{regs, vals};
 
 use super::*;
+use core::future::Future;
+use futures::TryFutureExt;
 
-pub struct Uart<'d, T: Instance> {
+use crate::dma_traits::NoDma;
+use crate::dma_traits::NoDmaMarker;
+
+#[allow(dead_code)]
+pub struct Uart<'d, T: Instance, TxDma = NoDma, RxDma = NoDma> {
     inner: T,
     phantom: PhantomData<&'d mut T>,
+    tx_dma: TxDma,
+    rx_dma: RxDma,
 }
 
-impl<'d, T: Instance> Uart<'d, T> {
+impl<'d, T: Instance, TxDma, RxDma> Uart<'d, T, TxDma, RxDma> {
     pub fn new(
         inner: impl Unborrow<Target = T>,
         rx: impl Unborrow<Target = impl RxPin<T>>,
         tx: impl Unborrow<Target = impl TxPin<T>>,
+        tx_dma: impl Unborrow<Target = TxDma>,
+        rx_dma: impl Unborrow<Target = RxDma>,
         config: Config,
     ) -> Self {
-        unborrow!(inner, rx, tx);
+        unborrow!(inner, rx, tx, tx_dma, rx_dma);
 
         // Uncomment once we find all of the H7's UART clocks.
         T::enable();
@@ -54,11 +64,17 @@ impl<'d, T: Instance> Uart<'d, T> {
         Self {
             inner,
             phantom: PhantomData,
+            tx_dma,
+            rx_dma,
         }
     }
 
     #[cfg(any(dma, dmamux))]
-    pub async fn write_dma(&mut self, ch: &mut impl TxDma<T>, buffer: &[u8]) -> Result<(), Error> {
+    pub async fn write_dma(&mut self, buffer: &[u8]) -> Result<(), Error>
+    where
+        TxDma: crate::usart::TxDma<T>,
+    {
+        let ch = &mut self.tx_dma;
         unsafe {
             self.inner.regs().cr3().modify(|reg| {
                 reg.set_dmat(true);
@@ -99,7 +115,11 @@ impl<'d, T: Instance> Uart<'d, T> {
     }
 }
 
-impl<'d, T: Instance> embedded_hal::blocking::serial::Write<u8> for Uart<'d, T> {
+impl<'d, T: Instance, TxDma, RxDma> embedded_hal::blocking::serial::Write<u8>
+    for Uart<'d, T, TxDma, RxDma>
+where
+    TxDma: NoDmaMarker,
+{
     type Error = Error;
     fn bwrite_all(&mut self, buffer: &[u8]) -> Result<(), Self::Error> {
         unsafe {
@@ -119,3 +139,15 @@ impl<'d, T: Instance> embedded_hal::blocking::serial::Write<u8> for Uart<'d, T>
         Ok(())
     }
 }
+
+// rustfmt::skip because intellij removes the 'where' claus on the associated type.
+#[rustfmt::skip]
+impl<'d, T: Instance, TxDma, RxDma> embassy_traits::uart::Write for Uart<'d, T, TxDma, RxDma>
+    where TxDma: crate::usart::TxDma<T>
+{
+    type WriteFuture<'a> where Self: 'a = impl Future<Output = Result<(), embassy_traits::uart::Error>>;
+
+    fn write<'a>(&'a mut self, buf: &'a [u8]) -> Self::WriteFuture<'a> {
+        self.write_dma(buf).map_err(|_| embassy_traits::uart::Error::Other)
+    }
+}
diff --git a/examples/stm32l4/src/bin/usart_dma.rs b/examples/stm32l4/src/bin/usart_dma.rs
index cc630e0d..1239bf9c 100644
--- a/examples/stm32l4/src/bin/usart_dma.rs
+++ b/examples/stm32l4/src/bin/usart_dma.rs
@@ -17,22 +17,23 @@ use embassy_stm32::usart::{Config, Uart};
 use example_common::*;
 use heapless::String;
 use stm32l4::stm32l4x5 as pac;
+use embassy_stm32::dma_traits::NoDma;
+use embassy_traits::uart::Write as AsyncWrite;
+use embassy::io::AsyncWriteExt;
 
 #[embassy::task]
 async fn main_task() {
     let mut p = embassy_stm32::init(Default::default());
 
     let config = Config::default();
-    let mut usart = Uart::new(p.UART4, p.PA1, p.PA0, config);
+    let mut usart = Uart::new(p.UART4, p.PA1, p.PA0, p.DMA1_3, NoDma, config);
 
     for n in 0u32.. {
         let mut s: String<128> = String::new();
         core::write!(&mut s, "Hello DMA World {}!\r\n", n).unwrap();
 
-        usart
-            .write_dma(&mut p.DMA1_3, s.as_bytes())
-            .await
-            .unwrap();
+        usart.write( s.as_bytes() ).await;
+
         info!("wrote DMA");
     }
 }