Red Hat Bugzilla – Bug 160797
Install fails; dhcp reply not recognized
Last modified: 2007-11-30 17:11:08 EST
From Bugzilla Helper:
User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.7.6) Gecko/20050323 Firefox/1.0.2 Fedora/1.0.2-1.3.1
Description of problem:
I'm trying to install FC4 on an IBM HS20 Bladeserver. I'm doing a network install but the network connecion doesn't come up. The NIC is a broadcom, using the tg3 driver.
On the DHCP I can see the incoming request and the offer made, but the machine says it did not get an offer back. Hard setting the IP address gets me into the next screen, but then the installation server can't be found/contacted, so the network link is not ok either.
Installing RH EL4 doesn't have this problem so it's not an hardware issue, but in the installer.
I will try an PXE install as a workaround an maybe to pin down the problem a bit more.
Version-Release number of selected component (if applicable):
Steps to Reproduce:
1. Install FC4 using the boot.iso CD, boot options: linux nousbstorage askmethod
2. Press enter on all questions and select either eth0/eth1 to use on install
3. Choose DHCP (or fixed IP) for configuration
4a. The DHCP offer fails an returns you to the same screen
4b. The fixed IP gets you into the server identification screen but fails after this; unable to find the server
Actual Results: Installation stops due to lack of network connections
Expected Results: Networking should be up on the selected NIC and the installation server should have been found.
I've now succesfuly installed FC4 on an IBM blade (HS20) and all NICs are
working ok. So the problem is related to the installation (maybe even the
I had to do a harddisk install since installing from CD doesn't work due to the
fact that it's usb based and only works booting with "linux nousbstorage".
I haven't found time to try the PXE install since I will have to change my DHCP
configuration. I may also try an install via disk 1, to see whether it's the
boot.iso or "all" iso images.
Sorry for making it a "pump" issue at first, since it isn't. As I already
mentioned it also happens without DHCP. I just got confused by all the options
Please attach the results of running "sysreport"...thanks!
Created attachment 116612 [details]
sysreport for IBM HS20 Blade - FC4
Installation is "minimal" plus Xen stuff. Sysreport run on Domain0.
Is there any chance that you have STP (i.e. 802.1d spanning tree) enabled on
the switch port connected to your HS20? That can introduce a delay long
enough to cause network installations to fail.
As far as I know, Spanning Tree is disabled (both switches have been reset to
factory default before installation of the Bladecenter in the datacenter). Also,
network installs with FC3 and CentOS 4 work without a problem.
Should check the switches for STP just to be sure -- many of the switching
products I worked on in the past would default to having STP enabled on all
ports as a guard against carelessly creating bridging loops.
Beyond that, I would also suggest disabling auto-negotiation on that device,
forcing the link to 1000/Full or whatever is appropriate for your network.
For good measure, I would also suggest passing ethool="speed 1000 duplex
full" (modified as appropriate for you network) on the kernel command line
when booting the installer. This should cause anaconda to force the link
configuration at its end as well.
Please try all of the above and post the results...thanks!
I hope to have time to perform the requested tests this week. Since the system
is not "at my desk" but in a datacenter 40 km away it's not a 10 min. thing.
Besides that, one of the blades is running a production environment so I'm
restricted in what I can do with respect to changing settings etc.
I was able to do a quick "reference" test last week and proved that for instance
a network install (incl. DHCP based IP) is no problem with CentOS 3.5. I had
similar results in the past with Fedora 3.
I'll give you updated results asap.
Closed due to lack of response. Please reopen when the requested information
We're having similar problems on our Bladeservers. The installer will start up
(but only provided 'nousbstorage' is passed to the kernel) but cannot get an IP
address, and if one is manually assigned, the network is still nonfunctional.
RedHat 9 installed on the same nodes with no problems.
STP was enabled by default on our switch, so I disabled that, but that has not
fixed the problem.
eth0 is reported as 1000/full by the installer, so this appears to be working
Ben, do the NICs work post-install?
Have you tried booting with "acpi=off" or "acpi=noirq"?
Yes, the NICs work just fine after installation (I tried a hard disk install).
This is using the 2.6.12-1.1456_FC4smp kernel.
Booting the installer with either "acpi=off" or "acpi=noirq" also allows it to
get an IP address, so this appears to fix the problem for us.