Network broken on HP e-Vectra

Hi,
I've hoped to use my little HP e-Vectra to run some Linux services on but
have come across a puzzling problem.

I've tried two distributions, Ubuntu 6.06, and Centos 4.4, and both have the
same problem. The card looks to be operating properly according to the
dmesg, but I can't ping out of the box (No route to host), and I can't ping
in (No response).

If I check /proc/interrupts, I can see the interrupt counts for eth0
climbing up.

If this were the end of the issue, I'd just write the box off as faulty and
move on - except that if I load a windows via a dual boot partition, the
networking is just fine. In fact, once I've run windows, and the
networking's going, I can then restart and start Linux and the network
works! The network stays working until I power down the box.

According to the dmesg, the box is loading the Donald Becker 3c59x ethernet
driver (the documentation says it's a 3c905c).

My take on this is that the windows code is initiating something in the
3c905 that the Linux code isn't. Sadly, I have no idea where to go from
here.

Please can anyone help with something other that 'plug in a new network
card' (no slots in this box) or 'check your routing tables'.

thanks,
Russell

Russell Cutcliffe wrote:
> Hi,
> I've hoped to use my little HP e-Vectra to run some Linux services on but
> have come across a puzzling problem.
>
> I've tried two distributions, Ubuntu 6.06, and Centos 4.4, and both have the
> same problem. The card looks to be operating properly according to the
> dmesg, but I can't ping out of the box (No route to host), and I can't ping
> in (No response).
>
> If I check /proc/interrupts, I can see the interrupt counts for eth0
> climbing up.
>
> If this were the end of the issue, I'd just write the box off as faulty and
> move on - except that if I load a windows via a dual boot partition, the
> networking is just fine. In fact, once I've run windows, and the
> networking's going, I can then restart and start Linux and the network
> works! The network stays working until I power down the box.
>
> According to the dmesg, the box is loading the Donald Becker 3c59x ethernet
> driver (the documentation says it's a 3c905c).
>
> My take on this is that the windows code is initiating something in the
> 3c905 that the Linux code isn't. Sadly, I have no idea where to go from
> here.

I would do two things: (1) extract the 3c59x items from the dmesg output ('dmesg | grep 3c59x' ?)
and the output of 'ifconfig eth0' for the two cases, and (2) post them to netdev@vger.kernel.org -
the Linux network mailing list. You should also include your kernel version (uname -r).

Larry

On Sun, 15 Jul 2007, in the Usenet newsgroup comp.os.linux.networking, in
article <nTkmi.7171$4A1.5210@news-server.bigpond.net.au>, Russell Cutcliffe
wrote:

>I've hoped to use my little HP e-Vectra to run some Linux services on
>but have come across a puzzling problem.

Web Results 1 - 10 of about 441 for HP e-Vectra Linux networking.
(0.25 seconds)

That _seems_ to suggest that others have gotten it working, though I
don't see any mention of network problems in a VERY QUICK scan of
these resources. I don't see the device listed in the Mobile-Guide
from the LDP either.

>I've tried two distributions, Ubuntu 6.06, and Centos 4.4, and both
>have the same problem. The card looks to be operating properly
>according to the dmesg, but I can't ping out of the box (No route to
>host), and I can't ping in (No response).
>
>If I check /proc/interrupts, I can see the interrupt counts for eth0
>climbing up.

Well, I'm really glad that you can see this - but you seem to be on the
other side of the pond, and it's hard to see the display from here. The
"no route to host" message means that the box can not determine the
hardware address using ARP - which _usually_ means a card or wiring
problem - that's probably not the case here. The incrementing interrupt
count is also a good sign.

>If this were the end of the issue, I'd just write the box off as faulty
>and move on - except that if I load a windows via a dual boot partition,
>the networking is just fine. In fact, once I've run windows, and the
>networking's going, I can then restart and start Linux and the network
>works! The network stays working until I power down the box.

Huge clue. The card is not being initialized properly during boot.
When you bring it up under windoze, that O/S is somehow able to kick
the hardware in the right place, and things work. Your Linux setup is
not doing the same thing - what ever that may be. Normally, this is
just the opposite - you boot in windoze, then shut down - and Linux
can't use the card, while booting directly into Linux makes it work
fine. That problem is caused by windoze setting the card into a power
saving mode that the Linux driver isn't aware of. This was a big and
frequent problem with RealTek cards. The solution there was to set up
the BIOS correctly.

>According to the dmesg, the box is loading the Donald Becker 3c59x
>ethernet driver (the documentation says it's a 3c905c).

That's what google says.

>My take on this is that the windows code is initiating something in
>the 3c905 that the Linux code isn't. Sadly, I have no idea where to
>go from here.

If I'm reading Becker's notes correctly, this card has an mii interface
and is therefore able to be queried using 'mii-tools'. I'd suggest
using that to look at the card when Linux is booted directly (and
won't work), and when re-booting from windoze (with the card now
working correctly).

>Please can anyone help with something other that 'plug in a new
>network card' (no slots in this box) or 'check your routing tables'.

How about showing that section of the boot messages that specifically
relate to the card and the loading of the driver (about 5-6 lines)
and the output of /sbin/ifconfig eth0. You might not be seeing
something that is obvious to others. Actually, looking at
/var/log/messages and comparing the 'work/no-work' boot messages
relating to the card might also offer clues.

Old guy

I've done some more with this since I posted the original message.

I found some documentation for the module that suggested I could set options
during the modprobe.
http://www.mjmwired.net/kernel/Docum...ing/vortex.txt

I added option=0, which supposedly pegs the network at 10baseT, and it
works. The option I want, option=0x204 (100baseT, FD) does nothing useful.

I eventually put a windows 2000 on the machine, and ran the 3com diagnostic,
force-setting the card's eeprom to 100baseT, full duplex. Back in Linux,
the system starts up OK, but appears stuck in 10base. Running mii-tool at
this point just appears to upset the switch, as we start getting lots of Rx
errors.

Russell Cutcliffe <wusl.a@gmail.com> wrote:
> I've done some more with this since I posted the original message.

> I found some documentation for the module that suggested I could set options
> during the modprobe.
> http://www.mjmwired.net/kernel/Docum...ing/vortex.txt

> I added option=0, which supposedly pegs the network at 10baseT, and it
> works. The option I want, option=0x204 (100baseT, FD) does nothing useful.
^^^^^^^^^^^^
Probably just a typo but that should be options=0x204 . If all else
fails and you want to try another module here is a link to a source
which claims to support the 3c905c and includes a useful readme.txt
that explains options as well as a short howto for compiling it:

http://support.3com.com/infodeli/too...90x-102.tar.gz

The readme is also available separately here:

http://support.3com.com/infodeli/too...nux/readme.txt

(I don't have the card but your post was, well, interesting.)

Regards-
--
Clifford Kite
/* The wealth of a nation is created by the productive labor of its
* citizens. */

Well,
I've found out how to make things work on my little e-Vectra, and I thought
I'd post what I'd done in hopes that someone else somewhere might benefit.

I don't claim that I know exactly what's going on, but I've found that if I
use mii-tool to force my settings to 100baseTx-HD, then remove and reload
the module, I can get the system to work reliably in 100MB mode.

The commands:

mii-tool -F 100baseTx-HD eth1
rmmod 3c59x
modprobe 3c59x

and it all works.

Any word from the gurus on why this would be?

BTW, thanks to the gents who replied to my original post, your replies
provided much food for thought...

"Russell Cutcliffe" <wusl.a@gmail.com> wrote in message
news:nTkmi.7171$4A1.5210@news-server.bigpond.net.au...
> Hi,
> I've hoped to use my little HP e-Vectra to run some Linux services on but
> have come across a puzzling problem.
>
> I've tried two distributions, Ubuntu 6.06, and Centos 4.4, and both have
> the same problem. The card looks to be operating properly according to
> the dmesg, but I can't ping out of the box (No route to host), and I can't
> ping in (No response).
>
> If I check /proc/interrupts, I can see the interrupt counts for eth0
> climbing up.
>
> If this were the end of the issue, I'd just write the box off as faulty
> and move on - except that if I load a windows via a dual boot partition,
> the networking is just fine. In fact, once I've run windows, and the
> networking's going, I can then restart and start Linux and the network
> works! The network stays working until I power down the box.
>
> According to the dmesg, the box is loading the Donald Becker 3c59x
> ethernet driver (the documentation says it's a 3c905c).
>
> My take on this is that the windows code is initiating something in the
> 3c905 that the Linux code isn't. Sadly, I have no idea where to go from
> here.
>
> Please can anyone help with something other that 'plug in a new network
> card' (no slots in this box) or 'check your routing tables'.
>
> thanks,
> Russell
>

Well, spoke too soon.

I hasdn't tested the behaviour through a power down.

When I powering up my machine this morning, it still wouldn't work.

What I'm considering doing now is to boot into Win2000, so that the card
sets up OK, then chaining in Linux.

I know this is a networking thread, but if anyone has suggestions on how to
automate the double startup, please let me know...



"Russell Cutcliffe" <wusl.a@gmail.com> wrote in message
news:ph%si.15924$4A1.10736@news-server.bigpond.net.au...
> Well,
> I've found out how to make things work on my little e-Vectra, and I
> thought I'd post what I'd done in hopes that someone else somewhere might
> benefit.
>
> I don't claim that I know exactly what's going on, but I've found that if
> I use mii-tool to force my settings to 100baseTx-HD, then remove and
> reload the module, I can get the system to work reliably in 100MB mode.
>
> The commands:
>
> mii-tool -F 100baseTx-HD eth1
> rmmod 3c59x
> modprobe 3c59x
>
> and it all works.
>
> Any word from the gurus on why this would be?
>
> BTW, thanks to the gents who replied to my original post, your replies
> provided much food for thought...
>
> "Russell Cutcliffe" <wusl.a@gmail.com> wrote in message
> news:nTkmi.7171$4A1.5210@news-server.bigpond.net.au...
>> Hi,
>> I've hoped to use my little HP e-Vectra to run some Linux services on but
>> have come across a puzzling problem.
>>
>> I've tried two distributions, Ubuntu 6.06, and Centos 4.4, and both have
>> the same problem. The card looks to be operating properly according to
>> the dmesg, but I can't ping out of the box (No route to host), and I
>> can't ping in (No response).
>>
>> If I check /proc/interrupts, I can see the interrupt counts for eth0
>> climbing up.
>>
>> If this were the end of the issue, I'd just write the box off as faulty
>> and move on - except that if I load a windows via a dual boot partition,
>> the networking is just fine. In fact, once I've run windows, and the
>> networking's going, I can then restart and start Linux and the network
>> works! The network stays working until I power down the box.
>>
>> According to the dmesg, the box is loading the Donald Becker 3c59x
>> ethernet driver (the documentation says it's a 3c905c).
>>
>> My take on this is that the windows code is initiating something in the
>> 3c905 that the Linux code isn't. Sadly, I have no idea where to go from
>> here.
>>
>> Please can anyone help with something other that 'plug in a new network
>> card' (no slots in this box) or 'check your routing tables'.
>>
>> thanks,
>> Russell
>>
>
>