I have attached a trace file showing the error.
When ethernetif_input() receives multiple frames in its for() loop, the 2nd frame is assigned the 1st frame's buffer so the 1st frame is processed twice. Then every following frame in the receive loop uses the wrong receive buffer, the previous message buffer.. In the case when I sent 10 UDP frames that were received within the for() loop, the 1st frame is correct, the 2nd is a duplicate of the 1st, then remaining are assigned the wrong buffer, and the last frame was never received. This is very repeatable. It happens whenever 2 or more frames are received by ethernetif_input() . Just being connected to a busy Ethernet is sufficient to receive multiple frames.
The attached file is too long to post inline.
One more issue to consider in the ethernetif_input() for() loop:
void ethernetif_input( void const * argument )
{
struct pbuf *p;
struct netif *netif = (struct netif *) argument;
for( ;; )
{
if( osSemaphoreWait( RxPktSemaphore, TIME_WAITING_FOR_INPUT ) == osOK )
{
do
{
p = low_level_input( netif );
if (p != NULL)
{
if (netif->input( p, netif) != ERR_OK )
{
pbuf_free(p);
}
}
/* Build Rx descriptor to be ready for next data reception */
HAL_ETH_BuildRxDescriptors(&heth);
}while(p!=NULL);
The call to netif->input( p, netif) only places the p into the tcpip_input() mbox. The Rx_Buff in the payload has not been processed. When HAL_ETH_BuildRxDescriptors(&heth) is called, the Rx_Buff is returned to DMA OWN, and the DMA can reuse the Rx_Buff before or while the IP stack is processing the contents, which can lead to corruption.
