diff --git a/embassy-stm32/src/dma/bdma.rs b/embassy-stm32/src/dma/bdma.rs
index d262caeb6..fbd753a71 100644
--- a/embassy-stm32/src/dma/bdma.rs
+++ b/embassy-stm32/src/dma/bdma.rs
@@ -1,6 +1,7 @@
 #![macro_use]
 
 use core::future::Future;
+use core::sync::atomic::{fence, Ordering};
 use core::task::Poll;
 
 use embassy::interrupt::{Interrupt, InterruptExt};
@@ -61,6 +62,9 @@ pub(crate) unsafe fn do_transfer(
 
         // Wait for the transfer to complete when it was ongoing.
         while ch.cr().read().en() {}
+
+        // "Subsequent reads and writes cannot be moved ahead of preceding reads."
+        fence(Ordering::Acquire);
     });
 
     #[cfg(dmamux)]
@@ -72,6 +76,9 @@ pub(crate) unsafe fn do_transfer(
             .modify(|w| w.set_cs(channel_number as _, request))
     });
 
+    // "Preceding reads and writes cannot be moved past subsequent writes."
+    fence(Ordering::Release);
+
     ch.par().write_value(peri_addr as u32);
     ch.mar().write_value(mem_addr as u32);
     ch.ndtr().write(|w| w.set_ndt(mem_len as u16));
diff --git a/embassy-stm32/src/dma/dma.rs b/embassy-stm32/src/dma/dma.rs
index 9bf083de4..bce9656d1 100644
--- a/embassy-stm32/src/dma/dma.rs
+++ b/embassy-stm32/src/dma/dma.rs
@@ -1,4 +1,5 @@
 use core::future::Future;
+use core::sync::atomic::{fence, Ordering};
 use core::task::Poll;
 
 use embassy::interrupt::{Interrupt, InterruptExt};
@@ -64,11 +65,17 @@ pub(crate) unsafe fn do_transfer(
 
         // Wait for the transfer to complete when it was ongoing.
         while ch.cr().read().en() {}
+
+        // "Subsequent reads and writes cannot be moved ahead of preceding reads."
+        fence(Ordering::Acquire);
     });
 
     #[cfg(dmamux)]
     super::dmamux::configure_dmamux(dmamux_regs, dmamux_ch_num, request);
 
+    // "Preceding reads and writes cannot be moved past subsequent writes."
+    fence(Ordering::Release);
+
     unsafe {
         ch.par().write_value(peri_addr as u32);
         ch.m0ar().write_value(mem_addr as u32);