Continuing sis900 problem.

[Rob Landley]

If you unplug the cat 5 from an sis900 card and plug it back in, the card
starts holding on to packets for updwards of 15 seconds.

PING mail.trommello.org (208.185.217.233) from 10.0.2.253 : 56(84) bytes of data.
64 bytes from trommello.org (208.185.217.233): icmp_seq=3 ttl=242 time=80.168 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=0 ttl=242 time=3.481 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=1 ttl=242 time=2.489 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=2 ttl=242 time=1.491 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=4 ttl=242 time=80.174 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=5 ttl=242 time=85.992 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=6 ttl=242 time=81.517 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=7 ttl=242 time=2.000 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=8 ttl=242 time=1.000 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=15 ttl=242 time=86.527 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=16 ttl=242 time=90.386 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=17 ttl=242 time=80.170 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=18 ttl=242 time=90.185 msec
64 bytes from trommello.org (208.185.217.233): icmp_seq=9 ttl=242 time=9.599 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=10 ttl=242 time=8.598 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=11 ttl=242 time=7.599 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=12 ttl=242 time=6.599 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=13 ttl=242 time=5.599 sec
64 bytes from trommello.org (208.185.217.233): icmp_seq=14 ttl=242 time=4.599 sec

This is 100% reproducible on the hardware I've got.  Compile the driver in,
bring the interface up, unplug the cat 5 cable, plug it back in, device is in
la-la land.  (Or maybe the network stack is.  I dunno.)

Last time I mentioned this, other people said they were seeing a similar problem.
Donald Becker's website points you to a SiS URL (404) for support on the driver.

Still using a 2.4.18 kernel...

Rob