diff --git a/cores/arduino/HardwareSerial.cpp b/cores/arduino/HardwareSerial.cpp
index a06a6275eb..ff9ca2d2a5 100644
--- a/cores/arduino/HardwareSerial.cpp
+++ b/cores/arduino/HardwareSerial.cpp
@@ -441,7 +441,7 @@ void HardwareSerial::begin(unsigned long baud, byte config)
 void HardwareSerial::end()
 {
   // wait for transmission of outgoing data
-  flush();
+  flush(TX_TIMEOUT);
 
   uart_deinit(&_serial);
 
@@ -487,20 +487,25 @@ int HardwareSerial::availableForWrite(void)
   return tail - head - 1;
 }
 
-void HardwareSerial::flush()
+void HardwareSerial::flush(uint32_t timeout)
 {
   // If we have never written a byte, no need to flush. This special
   // case is needed since there is no way to force the TXC (transmit
   // complete) bit to 1 during initialization
-  if (!_written) {
-    return;
-  }
-
-  while ((_serial.tx_head != _serial.tx_tail)) {
-    // nop, the interrupt handler will free up space for us
+  if (_written) {
+    uint32_t tickstart = HAL_GetTick();
+    while ((_serial.tx_head != _serial.tx_tail)) {
+      // the interrupt handler will free up space for us
+      // Only manage timeout if any
+      if ((timeout != 0) && ((HAL_GetTick() - tickstart) >= timeout)) {
+        // clear any transmit data
+        _serial.tx_head = _serial.tx_tail;
+        break;
+      }
+    }
+    // If we get here, nothing is queued anymore (DRIE is disabled) and
+    // the hardware finished transmission (TXC is set).
   }
-  // If we get here, nothing is queued anymore (DRIE is disabled) and
-  // the hardware finished transmission (TXC is set).
 }
 
 size_t HardwareSerial::write(const uint8_t *buffer, size_t size)
diff --git a/cores/arduino/HardwareSerial.h b/cores/arduino/HardwareSerial.h
index 54ebfe2af3..b0089bab19 100644
--- a/cores/arduino/HardwareSerial.h
+++ b/cores/arduino/HardwareSerial.h
@@ -125,7 +125,7 @@ class HardwareSerial : public Stream {
     virtual int peek(void);
     virtual int read(void);
     int availableForWrite(void);
-    virtual void flush(void);
+    virtual void flush(uint32_t timeout = 0);
     virtual size_t write(uint8_t);
     inline size_t write(unsigned long n)
     {
diff --git a/cores/arduino/stm32/uart.h b/cores/arduino/stm32/uart.h
index 9d2da47549..2a8537f45c 100644
--- a/cores/arduino/stm32/uart.h
+++ b/cores/arduino/stm32/uart.h
@@ -86,7 +86,9 @@ struct serial_s {
 };
 
 /* Exported constants --------------------------------------------------------*/
+#ifndef TX_TIMEOUT
 #define TX_TIMEOUT  1000
+#endif
 
 #if !defined(RCC_USART1CLKSOURCE_HSI)
 /* Some series like C0 have 2 derivated clock from HSI: HSIKER (for peripherals)