I had a flaky network problem where high loads (fast downloads from tux.org or sun.com) would cause a hang or complete drop of all network connections on the box. Originally, I had plugged in two CAT5 cables hoping that it would be one of the gbit ethernet ports but that had no effect.

When the network went down I was unable to ping or do anything network related until someone physically walked to the box and did a service network restart. It was quite annoying since this box was in a remote location.

The error message that appeared in /var/log/messages was:

Mar 17 16:45:14 {servername} kernel: tg3: tg3_stop_block timed out, ofs=3400 enable_bit=2

Eventually, I found (with help of a coworker) that the box was reachable when it was ‘down’ on the local subnet. I could ssh in from a box on the same subnet and fix the problem. Not a bad workaround but the crashing was fairly easy to produce. About 1 or 2 minutes of ~300KB (bytes, not bits) of fast downloading would cause the “Broadcom Corporation NetXtreme BCM5700 Gigabit Ethernet (rev 12)” to die.

This was running Fedora Core 1. I had a similar box (a poweredge 1750) that was running RedHat9 working fine. I was very confused because the linux kernels were very close in version numbers.

I googled and found many people with the same problem. In the past I’ve fixed problems like this by using a generic driver like `tulip’ or something. I eventually tried a different network driver in this vein of thought. My /etc/modules.conf file had

alias eth0 tg3 alias eth1 tg3

At the top, so I tried depmod e1000 which spit out a bunch of IRQ errors meaning that the driver couldn’t find the appropriate hardware. I tried bcm5700 which linux didn’t have compiled in by default. I wasn’t about to do a whole kernel recompile for one stupid driver. I tried e100 and it returned no errors. This was happy.

lsmod showed that e100 was loaded although showing “0” in the “Used By” column. That’s ok. So maybe I can use the e100 driver for eth0 or eth1 … but this still didn’t explain why I was able to get to the box (on eth1) on the same subnet.

Then I found the answer which as of now seems to be the answer. netstat -nr shows this {ips changed to protect the innocent}:

[root@server /]# netstat -nr Kernel IP routing table Destination Gateway Genmask Flags MSS Window irtt Iface U 0 0 0 eth1 U 0 0 0 eth1 U 0 0 0 eth1 U 0 0 0 lo UG 0 0 0 eth0

Note that (the default route) is specifically going through eth0. Now it makes sense. So I changed alias eth0 tg3 -> alias eth0 e100, ran depmod -a, rebooted a few times and heavy downloads have been running smoothly for 27 minutes now.

I’ll update this if this solution turns evil. :} For now, the e100 is metal. \m/