diff --git a/hal/src/rtl872x/usart_hal.cpp b/hal/src/rtl872x/usart_hal.cpp
index acd31451d7..54354a9516 100644
--- a/hal/src/rtl872x/usart_hal.cpp
+++ b/hal/src/rtl872x/usart_hal.cpp
@@ -288,6 +288,7 @@ class Usart {
         txBuffer_.reset();
         curTxCount_ = 0;
         transmitting_ = false;
+        busy_ = false;
         receiving_ = false;
         state_ = HAL_USART_STATE_DISABLED;
         return SYSTEM_ERROR_NONE;
@@ -300,6 +301,7 @@ class Usart {
         dataInFlight(true /* commit */);
         CHECK(disable(false));
         transmitting_ = false;
+        busy_ = false;
         receiving_ = 0;
         rxBuffer_.prune();
         txBuffer_.reset();
@@ -339,12 +341,12 @@ class Usart {
     ssize_t flush() {
         CHECK_TRUE(isEnabled(), SYSTEM_ERROR_INVALID_STATE);
         while (true) {
-            while (transmitting_) {
+            while (transmitting_ || busy_) {
                 // FIXME: busy loop
             }
             {
                 AtomicBlock atomic(this);
-                if (!isEnabled() || txBuffer_.empty()) {
+                if ((!isEnabled() || txBuffer_.empty()) && !busy_) {
                     break;
                 }
             }
@@ -470,6 +472,7 @@ class Usart {
         if (event & HAL_USART_PVT_EVENT_WRITABLE) {
             if (space() <= 0) {
                 TxLock lk(this);
+                xEventGroupClearBits(evGroup_, HAL_USART_PVT_EVENT_WRITABLE);
                 UART_INTConfig(uartInstance, RUART_IER_ETBEI, ENABLE);
                 // Temporarily disable TX DMA, otherwise, the TX FIFO won't be empty until all data is transferred by DMA.
                 UART_TXDMACmd(uartInstance, DISABLE);
@@ -572,6 +575,9 @@ class Usart {
                 } else {
                     UART_TXDMACmd(uartInstance, ENABLE);
                     BaseType_t yield = pdFALSE;
+                    if (!uart->transmitting_) {
+                        uart->busy_ = false; // All bytes sent if no new DMA transfers are in progress
+                    }
                     if (xEventGroupSetBitsFromISR(uart->evGroup_, HAL_USART_PVT_EVENT_WRITABLE, &yield) != pdFAIL) {
                         portYIELD_FROM_ISR(yield);
                     }
@@ -618,6 +624,7 @@ class Usart {
               curTxCount_(0),
               state_(HAL_USART_STATE_DISABLED),
               transmitting_(false),
+              busy_(false),
               receiving_(false),
               configured_(false),
               index_(index),
@@ -831,6 +838,8 @@ class Usart {
                 curTxCount_ = consumable;
                 GDMA_Init(txDmaInitStruct_.GDMA_Index, txDmaInitStruct_.GDMA_ChNum, &txDmaInitStruct_);
                 GDMA_Cmd(txDmaInitStruct_.GDMA_Index, txDmaInitStruct_.GDMA_ChNum, ENABLE);
+                busy_ = true;
+                UART_INTConfig(uartInstance, RUART_IER_ETBEI, ENABLE);
             } else {
                 // LOG UART doesn't support DMA transmission
                 consumable = std::min(MAX_UART_FIFO_SIZE, consumable);
@@ -982,6 +991,7 @@ class Usart {
 
     volatile hal_usart_state_t state_;
     volatile bool transmitting_;
+    volatile bool busy_;
     volatile bool receiving_;
     bool configured_;