diff --git a/embassy-stm32/src/dma/bdma.rs b/embassy-stm32/src/dma/bdma.rs
index e6ce05b7..6160b9f4 100644
--- a/embassy-stm32/src/dma/bdma.rs
+++ b/embassy-stm32/src/dma/bdma.rs
@@ -78,8 +78,7 @@ foreach_dma_channel! {
                 );
             }
 
-            unsafe fn start_write_repeated<W: Word>(&mut self, _request: Request, repeated: W, count: usize, reg_addr: *mut W, options: TransferOptions) {
-                let buf = [repeated];
+            unsafe fn start_write_repeated<W: Word>(&mut self, _request: Request, repeated: *const W, count: usize, reg_addr: *mut W, options: TransferOptions) {
                 low_level_api::start_transfer(
                     pac::$dma_peri,
                     $channel_num,
@@ -87,7 +86,7 @@ foreach_dma_channel! {
                     _request,
                     vals::Dir::FROMMEMORY,
                     reg_addr as *const u32,
-                    buf.as_ptr() as *mut u32,
+                    repeated as *mut u32,
                     count,
                     false,
                     vals::Size::from(W::bits()),
diff --git a/embassy-stm32/src/dma/dma.rs b/embassy-stm32/src/dma/dma.rs
index 97a3df08..fec60f70 100644
--- a/embassy-stm32/src/dma/dma.rs
+++ b/embassy-stm32/src/dma/dma.rs
@@ -102,15 +102,14 @@ foreach_dma_channel! {
                 )
             }
 
-            unsafe fn start_write_repeated<W: Word>(&mut self, request: Request, repeated: W, count: usize, reg_addr: *mut W, options: TransferOptions) {
-                let buf = [repeated];
+            unsafe fn start_write_repeated<W: Word>(&mut self, request: Request, repeated: *const W, count: usize, reg_addr: *mut W, options: TransferOptions) {
                 low_level_api::start_transfer(
                     pac::$dma_peri,
                     $channel_num,
                     request,
                     vals::Dir::MEMORYTOPERIPHERAL,
                     reg_addr as *const u32,
-                    buf.as_ptr() as *mut u32,
+                    repeated as *mut u32,
                     count,
                     false,
                     vals::Size::from(W::bits()),
diff --git a/embassy-stm32/src/dma/gpdma.rs b/embassy-stm32/src/dma/gpdma.rs
index bde8c3ef..d252cef3 100644
--- a/embassy-stm32/src/dma/gpdma.rs
+++ b/embassy-stm32/src/dma/gpdma.rs
@@ -75,15 +75,14 @@ foreach_dma_channel! {
                 )
             }
 
-            unsafe fn start_write_repeated<W: Word>(&mut self, request: Request, repeated: W, count: usize, reg_addr: *mut W, options: TransferOptions) {
-                let buf = [repeated];
+            unsafe fn start_write_repeated<W: Word>(&mut self, request: Request, repeated: *const W, count: usize, reg_addr: *mut W, options: TransferOptions) {
                 low_level_api::start_transfer(
                     pac::$dma_peri,
                     $channel_num,
                     request,
                     low_level_api::Dir::MemoryToPeripheral,
                     reg_addr as *const u32,
-                    buf.as_ptr() as *mut u32,
+                    repeated as *mut u32,
                     count,
                     false,
                     W::bits(),
diff --git a/embassy-stm32/src/dma/mod.rs b/embassy-stm32/src/dma/mod.rs
index 74bce6aa..b51e0d40 100644
--- a/embassy-stm32/src/dma/mod.rs
+++ b/embassy-stm32/src/dma/mod.rs
@@ -59,7 +59,7 @@ pub(crate) mod sealed {
         unsafe fn start_write_repeated<W: super::Word>(
             &mut self,
             request: Request,
-            repeated: W,
+            repeated: *const W,
             count: usize,
             reg_addr: *mut W,
             options: TransferOptions,
@@ -246,7 +246,7 @@ mod transfers {
     pub fn write_repeated<'a, W: Word>(
         channel: impl Peripheral<P = impl Channel> + 'a,
         request: Request,
-        repeated: W,
+        repeated: *const W,
         count: usize,
         reg_addr: *mut W,
     ) -> impl Future<Output = ()> + 'a {
diff --git a/embassy-stm32/src/spi/mod.rs b/embassy-stm32/src/spi/mod.rs
index ab4352a5..5b81c791 100644
--- a/embassy-stm32/src/spi/mod.rs
+++ b/embassy-stm32/src/spi/mod.rs
@@ -95,13 +95,10 @@ impl<'d, T: Instance, Tx, Rx> Spi<'d, T, Tx, Rx> {
         into_ref!(peri, sck, mosi, miso);
         unsafe {
             sck.set_as_af(sck.af_num(), AFType::OutputPushPull);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             sck.set_speed(crate::gpio::Speed::VeryHigh);
             mosi.set_as_af(mosi.af_num(), AFType::OutputPushPull);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             mosi.set_speed(crate::gpio::Speed::VeryHigh);
             miso.set_as_af(miso.af_num(), AFType::Input);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             miso.set_speed(crate::gpio::Speed::VeryHigh);
         }
 
@@ -129,10 +126,8 @@ impl<'d, T: Instance, Tx, Rx> Spi<'d, T, Tx, Rx> {
         into_ref!(sck, miso);
         unsafe {
             sck.set_as_af(sck.af_num(), AFType::OutputPushPull);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             sck.set_speed(crate::gpio::Speed::VeryHigh);
             miso.set_as_af(miso.af_num(), AFType::Input);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             miso.set_speed(crate::gpio::Speed::VeryHigh);
         }
 
@@ -160,10 +155,8 @@ impl<'d, T: Instance, Tx, Rx> Spi<'d, T, Tx, Rx> {
         into_ref!(sck, mosi);
         unsafe {
             sck.set_as_af(sck.af_num(), AFType::OutputPushPull);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             sck.set_speed(crate::gpio::Speed::VeryHigh);
             mosi.set_as_af(mosi.af_num(), AFType::OutputPushPull);
-            #[cfg(any(spi_v2, spi_v3, spi_v4))]
             mosi.set_speed(crate::gpio::Speed::VeryHigh);
         }
 
@@ -474,7 +467,7 @@ impl<'d, T: Instance, Tx, Rx> Spi<'d, T, Tx, Rx> {
         let tx_request = self.txdma.request();
         let tx_dst = T::REGS.tx_ptr();
         let clock_byte = 0x00u8;
-        let tx_f = crate::dma::write_repeated(&mut self.txdma, tx_request, clock_byte, clock_byte_count, tx_dst);
+        let tx_f = crate::dma::write_repeated(&mut self.txdma, tx_request, &clock_byte, clock_byte_count, tx_dst);
 
         unsafe {
             set_txdmaen(T::REGS, true);
@@ -772,10 +765,13 @@ fn finish_dma(regs: Regs) {
         #[cfg(not(any(spi_v3, spi_v4)))]
         while regs.sr().read().bsy() {}
 
+        // Disable the spi peripheral
         regs.cr1().modify(|w| {
             w.set_spe(false);
         });
 
+        // The peripheral automatically disables the DMA stream on completion without error,
+        // but it does not clear the RXDMAEN/TXDMAEN flag in CR2.
         #[cfg(not(any(spi_v3, spi_v4)))]
         regs.cr2().modify(|reg| {
             reg.set_txdmaen(false);