CentOS - Nov 2007 - Intel 1000/PRO GT (e1000 driver) and "Detect Tx Unit Hang" error with 4GB RAM

My system configuration:
ASUS M2A-VM motherboard
AMD Athlon 64 X2 4200+ 2.2 GHz
4x A-DATA 1GB DDR2 800 memory
2x Intel 10/100/1000 Pro/1000 GT Desktop Network Adapter
2x Seagate Barracuda 250GB HD (RAID 1, software RAID)
CentOS 5 x86_64; Kernel 2.6.23 (custom built); Version 7.6.9.2 e1000 driver

The symptoms of this problem are outlined at:

http://e1000.sourceforge.net/wiki/index.php/Issues[1]
http://e1000.sourceforge.net/wiki/in...p/Tx_Unit_Hang[2]

Last night I started experiencing the "Detected Tx Unit Hang" problem
with the Intel
e1000 NIC. This happened after I upgraded my system to 4GB RAM  
(previously 2GB). I have 2
of these cards in the system. I updated the Linux kernel to 2.6.23 and  
I downloaded from
Sourceforge and installed the most recent stable version of the e1000  
driver for Linux,
version 7.6.9.2. I still experiencing the "Detected Tx Unit Hang"  
message. I had to
recompile the kernel because upgrading to 4GB with the current kernel  
for CentOS 5
(2.6.18.8-1) causes an error, ata1: softreset failed (1st FIS failed),  
which results in a
kernel panic. Upgrading the kernel to 2.6.23 fixed that problem, but  
now I have a problem
with my network cards.

Searching around, I found posts saying that disabling acpi with the  
kernel options
"acpi=off noacpi" would fix it, but it did not. I tried added explicit
modprobe options
for the driver in /etc/modprobe.conf (options e1000 XsumRX=0  
Speed=1000 Duplex=2
InterruptThrottleRate=0 FlowControl=3 RxDescriptors=4096  
TxDescriptors=4096 RxIntDelay=0
TxIntDelay=0). Still no change. Still getting experiencing the problem.

I then tried another suggestion I found in a forum discussion `ethtool  
-K eth0 tso=off`.
Seems to have had no effect on the problem.

This problem occurs immediately when the system is trying to bring the  
device up. I
cannot even get to a point to try sending traffic over the network  
interface because it
never negotiates an IP address from DHCP. If I specify a static IP  
address, the address
is assigned, but I still experience the problem, and I cannot even  
ping another host.

Now, if I reduce the amount of RAM to 3GB or less, everything works  
fine! So, this leads
me to believe that my kernel and driver are configured, compiled, and  
functioning
correctly. It also leads me to believe that there are no problems with  
the network cards.
So, I though perhaps a bad memory module, but no matter which 3  
modules of the 4 I leave
in, I get the same results. Everything works fine until I add the 4th module.

Then I found an article on Intel's site saying that some older EEPROM  
have the power
management option turned on, and that could cause the problem. So, I  
downloaded the
script that would fix the bit in the EEPROM (turning off power  
management). The script
says that it does not apply to my version of the EEPROM. When I run  
`ethtool -e
(eth0|eth1)` I do not have the bit on 0x0010 that is set to "de", so I
must believe that
the script is correct in assessing that it does not apply to my NICs.

So, I thought that perhaps my power supply could be the problem.  
Perhaps the PSU doesn't
supply enough power to power everything when I add the 4th memory  
module. It is just a
generic 300W PSU that came with the case (I have new 500W PSUs on the  
way). So, I pulled
out one of the NICs and disconnected the DVD drive. That is about all  
I can eliminate.
Reducing the hardware installed made no difference.

I am running the 64-bit kernel, so I should have no trouble supporting  
the 4GM RAM,
correct?.

Now, I am out of ideas, and I seem to have hit a brick wall. One of  
the things that
disturbs me is that all of the articles I have found concerning this  
problem are dated
1-2 years ago.

Can anyone offer me any assistance?

Links:
------
[1] http://e1000.sourceforge.net/wiki/index.php/Issues
[2] http://e1000.sourceforge.net/wiki/in...p/Tx_Unit_Hang


----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.centos.org/pipermail/centos/attachments/20071109/393805ba/attachment.html>

On Nov 9, 2007 9:57 AM, Steven Alexson <steve at alexson.org> wrote:
>
> Can anyone offer me any assistance?
Doesn't help fix your immediate issue, but the e1000 driver has some
substantial updates coming in the 5.1 kernel. It's entirely possible
that this will fix your issue.


-- 
During times of universal deceit, telling the truth becomes a revolutionary act.
George Orwell

Hmmm...any chance that you can ellaborate on what updates will be
included with 5.1. Not sure if it will be of practical interest
anymore (I have 4 new network cards...non-Intel..on the way) since I
have spent far more time on this problem than I should have. But, from
a curiousity perspective, I am curious.

If I could resolve this problem and use the Intel cards, that would be
ideal. Then I could return the 4 cards I ordered today and save a bit
of money. If waiting the couple of weeks (projected) that it takes for
5.1 to roll out, that might be worth it.

Thanks for the insight.


----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.