Memory leak when calling SerialPort.Write() loop

[jeffer80]

In my project, a file about 1MB should be sent via the serial port. So I call the SerialPort.Write in a for loop.
After the loop is executed several times, in my project is 20, the internal memory will be used up.
Based on the logs, each loop will occupy 2408 bytes.

for (int i = 0; i < packNum; i++)
{
NativeMemory.GetMemoryInfo(NativeMemory.MemoryType.Internal, out totalMem, out totalFree, out freeBlock);
Debug.WriteLine($"Internal Total Mem {totalMem}, Free Mem {totalFree}, Free Block {freeBlock}");
sp.Write(b, 0, b.length);
}

log output:
Internal Total Mem 259987, Free Mem 6423, Free Block 4608
Internal Total Mem 259987, Free Mem 4015, Free Block 2304
Internal Total Mem 259987, Free Mem 1607, Free Block 336

[alberk8]

Can you try to run the loop without calling the sp.Write, see if there is still a memory leak.

[jeffer80]

Can you try to run the loop without calling the sp.Write, see if there is still a memory leak.

If the sp.Write dosen't be called, there will be no memeory leak.
And I have tried different package size, the coccupied memory always is 2408 bytes in each loop.
If the memeory wasn't used up in the loop, it will be released after the loop.

[alberk8]

Can you post a github repo a sample project to replicate the issue?

[jeffer80]

I created a simple project that only calls the sp.Write() looply, it works well.
In my project, many features are available: PWM, BLE, MQTT, and several long-running threads.
I think some tasks cause the low-priority task to be unable to free the resources, but I'm not sure which one causes the issue.

Following the AI's advice, I changed the native code like below.
It creates one task only that handles the TX data. It actually worked.

void UartTxWorkerTask_sys(void *pvParameters)
{
// other codes...

while (true)
{
    xSemaphoreTake(palUart->TxTriggerSem, portMAX_DELAY);

    if (palUart->TxOngoingCount == 0)
        break;

    uart_write_bytes(palUart->UartNum, (const char *)palUart->TxBuffer, palUart->TxOngoingCount);

    // set event flag for COM OUT
    Events_Set(SYSTEM_EVENT_FLAG_COM_OUT);
}

// other codes...
}

HRESULT Library_sys_io_ser_native_System_IO_Ports_SerialPort::Write___VOID__SZARRAY_U1__I4__I4(CLR_RT_StackFrame &stack)
{
// other codes...
// if (xTaskCreate(UartTxWorkerTask_sys, task_name, 2048, palUart, 12, NULL) != pdPASS)
// {
// ESP_LOGE(TAG, "Failed to start UART TX task");
// NANOCLR_SET_AND_LEAVE(CLR_E_FAIL);
// }

      xSemaphoreGive(palUart->TxTriggerSem);

// other codes...
}

[alberk8]

Is it possible to wait for uart to finish writing (uart_wait_tx_done()) before the SerialPorts::Write returns?

[josesimoes]

No, we have to use the execution engine wait for events pattern.

[josesimoes]

@jeffer80 fix on the way. Your comments above pointed me in the right direction to find the root cause. PR in the pipeline.

[josesimoes]

Fixed in nanoframework/nf-interpreter#3277