disabling nics using boot options.

Are there any error messages from the driver?

Bill

I find this in /var/log/messages

Jan  9 20:09:51 compute-1-1 kernel: Error seeking in /dev/kmem
Jan  9 20:09:51 compute-1-1 kernel: Symbol #tg3, value 001bc000
Jan  9 20:09:51 compute-1-1 kernel: Error adding kernel module table entry.

and this in dmesg:

tg3.c:v0.99 (Jun 11, 2002)
PCI: Found IRQ 56 for device 20:02.0
divert: allocating divert_blk for eth1
eth1: Tigon3 [partno(BCM95700A6) rev 0105 PHY(5701)] (PCI:66MHz:64-bit) 10/100/1
000BaseT Ethernet 00:30:6e:38:99:07
PCI: Found IRQ 65 for device 80:01.0
divert: allocating divert_blk for eth2
eth2: Tigon3 [partno(A6825-60101) rev 0105 PHY(5701)] (PCI:66MHz:64-bit) 10/100/
1000BaseT Ethernet 00:30:6e:2b:cc:05
eth2: Link is up at 1000 Mbps, full duplex.
eth2: Flow control is off for TX and off for RX.
ACPI: Power Button (FF) [PWRF]
ACPI: Sleep Button (FF) [SLPF]
eth2: Link is up at 1000 Mbps, full duplex.
eth2: Flow control is off for TX and off for RX.

I don't really know if these are real error messages.

r.

  The Computer Center, University of Tromsø, N-9037 TROMSØ, Norway.
	      phone:+47 77 64 41 07, fax:+47 77 64 41 00
     Roy Dragseth, High Performance Computing System Administrator
	 Direct call: +47 77 64 62 56. email: [email protected]

I've done some more investigation into this matter, it seems like the whole problem stems from a timing issue:

I reconfigured the Gbit interface to use a static ip-adress, but the problem
persisted: The interface comes up seemingly all right during the init process in the normal bootup sequence, but the subsequent ypbind fail to find any ypservers.

So, the problem doesn't have anything to do with dhcp, at least, so it seems to me. But, if I put in a sleep 60 after ifup in the /etc/init.d/network script, everything is ok.

To me, an illiterate when it comes to drivers and network hardware, it seems like the network card (or driver) needs some time to get ready to serve after beeing configured. Again, the problem only occurs when running Gbit, it never shows up when I connect to a 100Mbit switch.

Well, I think I've pulled enough hair out of my head over this. I'll just have to bite the bullet and rig all the machines with an extra cable and use the 100Mbit interface for the network installs. Another day of stretching cables, sigh...

A big thanks to all who have replied to me with suggestions, especially Bill.

r.

  The Computer Center, University of Tromsø, N-9037 TROMSØ, Norway.
	      phone:+47 77 64 41 07, fax:+47 77 64 41 00
     Roy Dragseth, High Performance Computing System Administrator
	 Direct call: +47 77 64 62 56. email: [email protected]

IIRC, dhclient forks and lets the child/fork do the actual work. With sleep 60, seems the new process would run immediately. Perhaps the problem is an issue with the dhclient child-process not getting executed before something else attempts to use the interface?
You have time to try "sleep 1" or "sleep 5"?

hmmm...I don't know anything about Gbit vs 100Bt negotiations, but maybe the Gbit negotiation is just taking longer?

grant

Well, as I said in a later mail the problem maninifests itself even if I use static ip.

No, I haven't.

Yes, that migth be the case.

r.

  The Computer Center, University of Tromsø, N-9037 TROMSØ, Norway.
	      phone:+47 77 64 41 07, fax:+47 77 64 41 00
     Roy Dragseth, High Performance Computing System Administrator
	 Direct call: +47 77 64 62 56. email: [email protected]