IBM X306 (ServerRaid7e) and Linux Centos 4.3

It was no fun, and I hope I will never experiance again such bad setup.

Summery: IBM X306. The X306, X206 and some of the others are equipped with ServerRaid7e Sata controller. These controllers lack Linux drivers, and thus, make me a sad person. Drivers are available from IBM web site to RHEL 3 Update 4 (Similar to Centos 3.4), and on a very rare occasion, for RHEL 4, and RHEL 4 Update 1.

Centos 4.3 is equivalent to RHEL 4 Update 3, so it wasn’t quite that.

Stage zero: Come ready.

I’ve came as ready as one can be. Equipped with a laptop capable of serving bootp requests, NFS images of both the 32bit and the 64bit version of Centos 4.3, I was as ready as I can. Discovered that tfptd alone (Debian version) was not enough (or could not boot IBM’s PXE, in this case), and had to replace it to tftpd-hpa, which worked correctly.

Stage one: Asses the problem. The Problem was assessed, and I’ve understood that no disks were available for installation. The kernel was unable to access local disks. Bad.

Stage two: Find a solution.

With Internet access, I was able to identify drivers in IBM site, but they were for RHEL3 only. Some forum (can’t remember link, sorry) led me to a hidden part of IBM’s web site, which included a driver for RHEL 4 Update 1. I’ve downloaded it.

Stage Three: Work harder.

To load a vendor module available for a specific kernel version, you need to have that specific version. I’ve had Centos 4.3, and required the kernel for Centos 4.1. I’ve visited Centos old download source, and was able to download the kernel version for Centos 4.1 (kernel 2.6.9-11.EL.x86_64.rpm), and the required kernel and initrd image for installation.

I’ve started installation with the following command:

linux dd=nfs:192.168.0.1:/mnt/Source/40k8690.img

After a short boot sequence, an error message appread, claiming I did not install the correct version. Of course I did not, I’ve used an older kernel and initrd!

I’ve downloaded the old stage2.img file from cdroot/Centos/base, and replaced the current one.

There was a problem with IBM’s supplied drivers, or the one obtained from the net, so I’ve looked into IBM’s CDROM, the one supplied with the server, and found the driver there. I will add it here, just in case – both RPM and DriverDisk (dd) image.

Finally I’ve managed to install the server, after rather long time in the server room.

One note – do yourselved a favour and replace the kernel package to the one of Centos 4.1 before you reboot the server at the end of the installation. It can save you both rescue install, and both troubles with RPM.

For some reason, I could not install the alternate (older) kernel. It failed to install because it was older, and failed with error when used with "–force". I’ve had to insert the files manuall. It required some tweeking:

I’ve used "rpm2cpio kernel….rpm | cpio -id" to extract the files, and then moved them to the respective directories. I’ve had to do a similar trick for IBM’s drivers, becuase they failed, for some reason, the post-install script. They have created an entry for the raid controller in /etc/modprobe.conf, and I’ve only had to recreate the initrd file. A command similar to this:

mv /boot/initrd-2.6.9-11.EL.img /boot/initrd-2.6.9-11.EL.img.old

initrd -o /boot/initrd-2.6.9-11.EL.img 2.6.9-11.EL

did the trick. Adding the correct entry to /etc/grub.conf fixed it all up. I was able to boot the newly installed system.

I don’t know what are the implications, but I did not dare letting Kudzu change these settings, when it claimed to have found a raid controller. I just ordered it to ignore it and never ask again.