[E1000-devel] NIC not seeing ARP replies and general strange behaviour

[Kyrios <ky...@gm...>]

Hi,

we have bought 4 servers with Supermicro X7DVL-E Motherbards. This
motherboard uses the 5000P chipset and has 2 Onboard NICs.

After some time (5minutes to 5 days) no more connectivity to the servers is
possible. I have noticed that the other servers in the network don't get Arp
replies from the servers.
Strange thing is that from the servers itself (via console) networking is
possible to other nodes. What definitely helped was not using the second NIC
at all. When only using eth0 the servers worked for at least 2 weeks.

I already tried the following:
 - 3 different switches
 - Putting the NICs on different VLANs (also this shouldn't be required
since they are in different Subnets)
 - Updated the e1000 Modules from Debian Etch 2.6.18 to the latest Version
7.5.x
 - Enabled and disabled some Watchdog function in the BIOS

Also I've noticed that unloading and reloading the module makes the NICs
respond to ARP replies again.

The vendor of the servers sent me a link without any further details about
what this firmware does.
ftp://ftp.supermicro.com/CDR-SIMIPMI_1.10_for_SIM_IPMI/EEPROM/FML

I've also found a thread/post by someone else on the Freebsd Mailing list.
This guy seemed to have the exact same problemh
ttp://lists.freebsd.org/pipermail/freebsd-stable/2007-March/033450.html


So here are my questions:

1. Does this sound like an Interrupt problem? What are the suggested module
options for that NIC(s) ? I'm currently using defaults.
2. Could this be some power saving "feature" ?
3. Could this "eeprom-fix" by the vendor solve the problem? If so... is it
possible to path the EEPROM out of linux with the eeprom file found on the
ftp server? With ethtool?


Bye and big thanks for any input
Thorsten



-- 
... black holes are where god divided by zero.