ARP problem with 6.2-STABLE Intel PRO/1000 NIC, latest em driver

[Mark Costlow]

The Machine:

I have a dual Xeon 5130 machine, Supermicro motherboard, with
the 82563EB NIC.  From dmesg:

CPU: Intel(R) Xeon(R) CPU            5130  @ 2.00GHz (2000.08-MHz 686-class CPU)
cpu0: <ACPI CPU> on acpi0
em0: <Intel(R) PRO/1000 Network Connection Version - 6.2.9> port 0x2000-0x201f mem 0xda000000-0xda01ffff irq 18 at device 0.0 on pci4

The machine has 4G RAM and a 3ware 9000 series RAID controller with 2 drives.

pciconf -l says:

em0 at pci4:0:0:   class=0x020000 card=0x000015d9 chip=0x10968086 rev=0x01 hdr=0x00
em1 at pci4:0:1:   class=0x020000 card=0x000015d9 chip=0x10968086 rev=0x01 hdr=0x00


The symptom:

The machine boots OK, but can only intermittently make netork connections.
Eventually determined that it seems to only see a few ARP packets, so
it's falling out of other machines' ARP tables, and is often unable to
see the replies to its own ARP requests.  It does see SOME ARPs
though.  When it is able to communicate with another machine, it
does not appear to drop any packets between them (e.g. I scp'd a 500M file
at 300Mbps to this machine).

When I run "tcpdump -n arp" I see a few ARPs, but not many.  In a 1-minute
period, I saw 3 ARP who-has/reply packets.  On a different machine on
the same ethernet switch, I saw 225 who-has/reply packets in the same
1-minute period.

I've tried different cables, and a different switch.  I started with
6.2-RELEASE, and then went to 6.2-STABLE on 3/3/07 to get the latest
em driver fixes.  I've used SMP and GENERIC kernels.  I get the same
results in all cases.

There are no firewall rules installed.

I plugged in a USB ethernet adapter (realtek), and it works straight away.
"tcpdump -n arp" sees the same noise as other machines on that LAN.

I read through the recent threads on the em driver, but didn't see any
reported symptoms like this.  Has anyone seen anything like this?  Got
any hints for me?  Am I doing something stupid?  Did I leave out any
useful information about my configuration?

Thanks,

Mark
-- 
Mark Costlow    | Southwest Cyberport | Fax:   +1-505-232-7975
cheeks at swcp.com | Web:   www.swcp.com | Voice: +1-505-232-7992

abq-strange.com -- Interesting photos taken in Albuquerque, NM
                   Last post: Art Is OK...And Dangerous - 2007-03-02 10:27:17