RHEL WS 2.1, latest kernel, network wont' start

[fred_118]

Installed RH's latest update kernel (2.4.9-e.57) and now the network won't start at boot. After boot, doing "modprobe eth0" elicits a usage error (can't show it to you right now, I've got a different kernel booted, and neglected to save a copy first) with some complaint implying that it couldn't find the network hardware. I installed the latest PSP for RHEL 2.1, a couple of weeks before installing the kernel update, without trouble. when running the new kernel, tried installing it again and it did't make any difference.

If I boot the prior kernel (-49), as now, everything seems fine.

Clues please?

[Steven E. Protter]

Sometimes they make mistakes.

Report it to bugzilla and continue using the old kernel.

SEP

[xyko_1]

Hi Fred,

Do you know if you have installed an specific nic driver after the general installation of RedHat ? Sometimes it's necessary to some nics. I allways have to install a Bradcom driver BCM5700 after instaling RedHat AS in my DL/ML servers.
If you did so, you have to reinstal the nic's driver for the new kernel. Have attention also on modules.conf to see if the nic module is correctly defined.

Regards,
Xyko

[fred smith_3]

no, I don't believe there were any special drivers installed apart from whatever was installed by the RH installation. I've certainly upgraded the kernel (using RH released errata only) multiple times in the one-year-plus interval since installing the machine without having any such trouble.

I neglected to mention in the original post that this is a DL320 G2.

Also, the PSP (Proliant Support Pack) I mentioned installing is: psp-7.11.rhel21.i386.en.tar.gz

[xyko_1]

Hi Fred,

do you have more then one nic on your server ?

Sometimes (very unusual, but..) the order the nics are discovered changes and eth0 becames eth1 and vice-versa.

regards,
xyko

[fred smith_3]

yeah, a DL320 G2 has two built in interface ports. But neither one will work. "modprobe eth0" fails. I didn't think to try it for eth1 since I never use eth1 for anything in the current installation. lsmod shows no network driver loaded. eth0 is not started, and obviously then "ifconfig eth0 up" does not work. The loadable module file DOES exist in the file system.

Simply rebooting the previous kernel allows it to work. Weird.

[xyko_1]

Fred,

2 things :

1 - boot with the new kernel and see if lspci (or cat /proc/pci) reports the existence of the two nics.
2 - if lspci reports two nics, try to change the network cable to the other nic and restart network.

If lspci does not report nics then ..... I guess that is not possible that the new kernel doesn't have network support at all, then, for a reason that I dont't know, your new kernel is not recognising your nics.

What are your nics ?

regards,
Xyko

[fred smith_3]

It's hard to imagine that a standard RH errata update kernel would have broken networking, but I suppose stranger things have happened....

ok, I'll try that. It may be a few hours before I can do the reboot, though.

In the meantime, here's some info from the kernel that IS working.

from dmesg:
Broadcom Gigabit Ethernet Driver bcm5700 with Broadcom NIC Extension (NICE) ver. 7.2.24e (04/23/04)
eth0: HP NC7760 Gigabit Server Adapter found at mem f6fd0000, IRQ 7, node addr 000bcdaf1a49
eth0: Broadcom BCM5703 Integrated Copper transceiver found
eth0: Scatter-gather ON, 64-bit DMA ON, Tx Checksum ON, Rx Checksum ON
eth1: HP NC7760 Gigabit Server Adapter found at mem f6fc0000, IRQ 10, node addr 000bcdaf164c
eth1: Broadcom BCM5703 Integrated Copper transceiver found
eth1: Scatter-gather ON, 64-bit DMA ON, Tx Checksum ON, Rx Checksum ON
bcm5700: eth0 NIC Link is UP, 10 Mbps half duplex

I did look thru the dmesg output when the non-working kernel was booted, and I don't remember seeing that stuff in it, but I'll look again when I get a chance to reboot it.

from lspci:
00:05.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5702X Gigabit Ethernet (rev 02)
00:06.0 Ethernet controller: Broadcom Corporation NetXtreme BCM5702X Gigabit Ethernet (rev 02)

[xyko_1]

Ok Fred,
let's wait for the next reboot.

But one more thing :

did you see that your nic is working 10 mbts half duplex ?

bcm5700: eth0 NIC Link is UP, 10 Mbps half duplex

Is it correct ?

If not, verify the configuration of your switch and garantee that the corresponding port is configured auto-negotiation. That's the common configuration for linux servers.

Xyko

[fred smith_3]

Yes, that's correct. it's a small office (9 people) with only a 10mbps lan, using hubs rather than switches.

[fred smith_3]

OK, I just rebooted it and gathered together a bunch of output from varous programs (see attachment).

The lines I showed in an earlier posting, taken from dmesg, do not exist in the dmesg when booting the new kernel. lspci shows the hardware, but at boot time there is no apparent attempt to bring up eth0. modprobe eth0 does elicit errors (see attachment), and if you watch carefully when it boots some of that same text scrolls by quickly.

the only thing I see in these diagnostics that's particularly interesting is that in the /lib/modules/xxx tree for the new kernel, the bcm5700.o file is quite a lot smaller than the one in the same location for the kernel that works.

[xyko_1 · Accepted solution]

Fred,

I suggest you to rebuild bcm5700 module from the newest source avaiable for the new kernel.

http://www.broadcom.com/drivers/driver-sla.php?driver=570x-Linux

It seems that the module is not good.

regards,
xyko

[xyko_1]

Fred,

From Broadcom's FAQ site.

94.) How can I install BCM5700 Linux driver for my running Kernel?
You must install the kernel source tree that matches your running kernel (the output of uname -r). Try the following for RedHat distributions:

1. Insert the first CD and double-click "start here"
2. Double-click on "system setting"
3. Double-click on "Add or Remove Packages"
4. Select "Development Tools" and "Kernel Development"
5. Just follow the instruction and add the different CD's as requested

regards,
xyko

[fred_118]

Well, I "fixed" it, for some definitions of the word. I avoided the problem by reverting to the tg3 driver, which is now working.

I don't know what limitations tg3 has, maybe it limits me to 10mBps, but that's OK since that's the speed of my network anyway.

I did subsequently get the latest driver from broadcom's web site and built it and it did end up being quite a bit bigger than the one I had before, so I guess it was defective. Probably would work if I put the one I made back in place. but since it is doing what I need it to do the way it is now configured I think I'll leave well-enough alone.

Thanks muchly for all the help, I do appreciate it!

[xyko_1]

Ok Fred,

tg3 will do what you need, I guess, but Broadcon says that its module do a better job.

If you think forum's conversation helped you to solve your problem, please assign some points to Steve (and me :-)). It's just to make the game more funny.

Oh yeah, and close the thread also.

Regards,
Xyko

[fred_118]

With Xyko's help I've found that somehow the bcm5700 module was defective. I "fixed" it by reverting to the tg3 module. I also rebuilt (but did not install) the bcm5700 module, and it came out a significantly different size than the one that didn't work (similar to one that did work in an older kernel) so it likely would work if I went to the trouble to reinstall it. Thanks Xyko for the help!