[lwip-users] Timing problem on UDP sending

[Muhamad Ikhwan Ismail]

Hi guys,

I have observed the following :

1.I do send() on a UDP socket, which then post the reference to data into the socket layer.
The application thread then blocks on the semaphore op_completed.

2.The TCPIP thread then checks the mailbox and starts preparing the packet and queue the reference in
the Ethernet buffer descriptor ring for transmission.Since the ring buffer descriptor is rather full the packet
cannot be transmitted at that exact moment by the MAC.

3. The TCPIP thread returns to the point where the semaphore is triggered. It then blocks and there
is a task change and the application thread is now running.

4. The application thread now rewrites the same data which was not yet sent. Hence the previous data packet are corrupted.

This only happens when we have high frequency sending and only with UDP since TCP copies all the data into the TCPIP stack.
I dont know if this qualifies as a bug, I just wanted to point it out to you guys that this could happen. The workaround can easily
be done on the application layer of course.

My implementation is lwip1.3.0 on PowerPC processor

M Ikhwan Ismail

---

See how Windows Mobile brings your life together—at home, work, or on the go. See Now