diff --git a/cores/arduino/HardwareSerial.cpp b/cores/arduino/HardwareSerial.cpp
index 122c45ea65..0c377f9eaf 100644
--- a/cores/arduino/HardwareSerial.cpp
+++ b/cores/arduino/HardwareSerial.cpp
@@ -334,15 +334,23 @@ void HardwareSerial::_rx_complete_irq(serial_t *obj)
   }
 }
 
-// Actual interrupt handlers //////////////////////////////////////////////////////////////
+// Actual interrupt handlers //////////////////////////////////////////////////
 
 int HardwareSerial::_tx_complete_irq(serial_t *obj)
 {
-  // If interrupts are enabled, there must be more data in the output
-  // buffer. Send the next byte
+  size_t remaining_data;
+  // previous HAL transfer is finished, move tail pointer accordingly
   obj->tx_tail = (obj->tx_tail + obj->tx_size) % SERIAL_TX_BUFFER_SIZE;
 
-  if (obj->tx_head == obj->tx_tail) {
+  // If buffer is not empty (head != tail), send remaining data
+  if (obj->tx_head != obj->tx_tail) {
+    remaining_data = (SERIAL_TX_BUFFER_SIZE + obj->tx_head - obj->tx_tail)
+                     % SERIAL_TX_BUFFER_SIZE;
+    // Limit the next transmission to the buffer end
+    // because HAL is not able to manage rollover
+    obj->tx_size = min(remaining_data,
+                       (size_t)(SERIAL_TX_BUFFER_SIZE - obj->tx_tail));
+    uart_attach_tx_callback(obj, _tx_complete_irq, obj->tx_size);
     return -1;
   }
 
@@ -482,9 +490,10 @@ void HardwareSerial::flush()
 
 size_t HardwareSerial::write(const uint8_t *buffer, size_t size)
 {
-  tx_buffer_index_t i;
-  size_t size_tmp;
+  size_t size_intermediate;
   size_t ret = size;
+  size_t available = availableForWrite();
+  size_t available_till_buffer_end = SERIAL_TX_BUFFER_SIZE - _serial.tx_head;
 
   _written = true;
   if (isHalfDuplex()) {
@@ -494,49 +503,46 @@ size_t HardwareSerial::write(const uint8_t *buffer, size_t size)
     }
   }
 
-  // If necessary split transfert till end of TX buffer
-  while (_serial.tx_head + size > SERIAL_TX_BUFFER_SIZE) {
-    size_t size_intermediate = SERIAL_TX_BUFFER_SIZE - _serial.tx_head;
-
-    write(buffer, size_intermediate);
-    size -= size_intermediate;
-    buffer += size_intermediate;
+  // If the output buffer is full, there's nothing for it other than to
+  // wait for the interrupt handler to free space
+  while (!availableForWrite()) {
+    // nop, the interrupt handler will free up space for us
   }
 
-  // Here size if less or equal to SERIAL_TX_BUFFER_SIZE, but SERIAL_TX_BUFFER_SIZE is not possible as tx_head = tx_tail is ambiguous empty or full
-  if (size == SERIAL_TX_BUFFER_SIZE) {
-    size_t size_intermediate = SERIAL_TX_BUFFER_SIZE - 1;
-
+  // HAL doesn't manage rollover, so split transfer till end of TX buffer
+  // Also, split transfer according to available space in buffer
+  while ((size > available_till_buffer_end) || (size > available)) {
+    size_intermediate = min(available, available_till_buffer_end);
     write(buffer, size_intermediate);
     size -= size_intermediate;
     buffer += size_intermediate;
+    available = availableForWrite();
+    available_till_buffer_end = SERIAL_TX_BUFFER_SIZE - _serial.tx_head;
   }
 
-  size_tmp = size;
-
-  while (size_tmp) {
-    i = (_serial.tx_head + 1) % SERIAL_TX_BUFFER_SIZE;
-
-
-    // If the output buffer is full, there's nothing for it other than to
-    // wait for the interrupt handler to empty it a bit
-    while (i == _serial.tx_tail) {
-      // nop, the interrupt handler will free up space for us
-    }
-    _serial.tx_buff[_serial.tx_head] = *buffer;
-    _serial.tx_head = i;
-    size_tmp --;
-    buffer ++;
-  }
-
-  while ((_serial.tx_head != (_serial.tx_tail + size) % SERIAL_TX_BUFFER_SIZE)) {
-    // nop, previous transfert no yet completed
+  // Copy data to buffer. Take into account rollover if necessary.
+  if (_serial.tx_head + size <= SERIAL_TX_BUFFER_SIZE) {
+    memcpy(&_serial.tx_buff[_serial.tx_head], buffer, size);
+    size_intermediate = size;
+  } else {
+    // memcpy till end of buffer then continue memcpy from beginning of buffer
+    size_intermediate = SERIAL_TX_BUFFER_SIZE - _serial.tx_head;
+    memcpy(&_serial.tx_buff[_serial.tx_head], buffer, size_intermediate);
+    memcpy(&_serial.tx_buff[0], buffer + size_intermediate,
+           size - size_intermediate);
   }
 
-  _serial.tx_size = size;
+  // Data are copied to buffer, move head pointer accordingly
+  _serial.tx_head = (_serial.tx_head + size) % SERIAL_TX_BUFFER_SIZE;
 
+  // Transfer data with HAL only is there is no TX data transfer ongoing
+  // otherwise, data transfer will be done asynchronously from callback
   if (!serial_tx_active(&_serial)) {
-    uart_attach_tx_callback(&_serial, _tx_complete_irq, size);
+    // note: tx_size correspond to size of HAL data transfer,
+    // not the total amount of data in the buffer.
+    // To compute size of data in buffer compare head and tail
+    _serial.tx_size = size_intermediate;
+    uart_attach_tx_callback(&_serial, _tx_complete_irq, size_intermediate);
   }
 
   /* There is no real error management so just return transfer size requested*/