[En-Nut-Discussion] TCP/IP problems with httpd

[Ulrich Prinz]

Hi Thiago!


Am 10.01.2011 19:57, schrieb Thiago A. Corrêa:
> Hi,
>
>      I'm doing some developments with EVK1105 and our custom board and
> I'm having some problems with the TCP stack.

Same do I on the AT91SAM7X-EK...
>
>      Using the httpd example, I'm able to use nutdisc to enumerate the
> board.

I didn't check nutdisc, to be honest.

> The gateway data is completely backwards, but the rest is
> solid. I can see that it's even able to obtain a lease from the DHCP,
> but no TCP connection seams to work.

With me it is most time the other way round. In high traffic conditions, 
the board never gets a DHCP. In low traffic conditions it works, 
including TCP.

> If I telnet to port 80 I get a
> timeout on the PC console. I can see thru breakpoints and the debugger
> that the network driver seems to be working properly, data goes up
> even into the TCP state machine, and the connection seams to be
> accepted inside user code.
>      Pings also doesn't get replies, but somehow, UDP looks OK.
>
In low traffic conditions, ping works, up to some thousands of packets. 
The higher the traffic is, the less pings are returning until the system 
hangs. I can shoot any Enthernut with a simple adaptive or fat ping 
(ping -a / ping -f).

>      I'm having some trouble pinpointing the exact problem. Since the
> platform specific code (AVR32) looks ok, does anyone else see this
> problem in their platform?
>      Any ideas what could be wrong? Timers? Something funky else where?

It is definitive the upper networks stack. In the company network no 
ethernut installation works as all of them crash in no time after 
connecting to the network.

If you now confirm, that it is not the at91_emac stack, but the avr32 
ones too, then we can be sure that it must be in the higher layers. If 
we keep in mind the fact, that data is sometimes going through, it must 
be something very difficult... I didn't figure it out till today, but 
hadn't so much time in the last weeks, as I hoped to have.

I bet on a race condition or a memory leak somewhere from arp/icmp upwards.

If only one could jump in and help. Ole, Harald, couldn't we concentrate 
on that thing and nail that thing down before any next release and other 
updates and improvements? I think for an ETHERnut a working stack is 
sort of essential... :)

Kind regards
Ulrich