[tulip] Re: problem with tulip card ceasing to function - requires ifup/ifdown to fix

[Alexander Haväng]

Hi people,

We are having the exact same problem as the one reported here. We also have the
4 port DLink card acting as a traffic shaper and firewall.
When the system is attacked with a lot of small packets, first the TX process is stopped,
and refuses to forward packets, after a few more seconds the RX process is also stopped.

No errors or warnings are reported from the driver (the tulip beta one from Becker).
Other tulip drivers (such as the one in 2.4.16), and the 1.1.8 development driver die the
same way. We hacked the 1.1.8 driver to use automatic polling, and that worked a lot better.

We suspect there might be something in the driver that makes it unaware of rx/tx problems.

If we leave the max_interupt_work at 25, and set debug to max, we see a lot of "too much work during interupt"
messages, and increasing that variable gets rid of the warning, but not the problem.

Once the packetstorm is over, one would think the driver to start working again.. but it
does not.

The packet storm in this case is a single machine puking small (64 bytes) UDP packets as fast as
it can. With a working driver, this acchieves 95+ kpps, and with the latest driver from Donald, we
see throughputs at 20 kpps for 2-3 seconds, then it dies and never comes back.

At times like this one would like Donald to sit in the chair beside us... ;)

We will investigate further, and see if we can come up with a patch against the latest
tulip driver.

Regards
Alexander Haväng, CTO Netintact