I am trying to compile kernel 2.6.28 on my 12.2 machine, and it ended with kernel panic (can't find root device "301"). When i went back to my 2.6.27.9, it worked just fine.

I used the old 2.6.27 config and ran make oldconfig and i answered all the new features, but most of the cases, i compiled it as a module (mostly are drivers) or even no (mostly experimental features). I wonder why this happened, because i used the same method to compile previous kernel up to 2.6.27 and it always worked.

Any ideas?

Thanks before

My .config file are here: http://pastebin.ca/1299976

Theres a recent trend in the newer kernels of splitting things up into little bits for your entertainment only. Stuff like USB-HID input is now split from a single configuration option to a multiplitudal confusing wave of "logitech devices", "microsoft devices", "blah blah devices", same with wifi and webcam drivers. I suppose this might be happening to whatever driver you might have needed for your root device to happen.

Nevertheless even if you do get 2.6.28 working, you still will need a new sysklogd package as well.

Did you make sure to compile in your filesystem driver? (not a module) Also, do you use an initrd? You might need to make one.

Did you make sure that support for your root filesystem type was compiled in?

Maybe you have good reasons for using 2.6.28, but you might be interested to know that the 2.6.27 kernel has been 'adopted' by Adrian Bunk as the new Long term Support kernel, so for the next 2-3 years he will be maintaining it with security updates and other fixes -but only things that do not break things which were aready fixed. This means that, in contrast to the main kernel branch, each release should be better and better and you should have none of the problems you just described....
Adrians last LTS kernel was the 2.6.16 which he carried all the way to 2.6.16.62. It has been far and away the most stable, secure 2.6 kernel ever. and 2.6.27 will now get the same treatment!

I loved the 2.6.16 kernel, I used it for years. Thanks for the tip on 2.6.27.

Yes, i have added all the filesystem i used as a static modules, because it worked on the previous kernel (2.6.27.x). Perhaps the sysklogd problem is the root problem of my case.

I don't use initrd and it's been a while since i used initrd. The last kernel i used with initrd was 2.6.15 if i'm not mistaken

Oh well, i'm sticking to 2.6.27.x for some time then

Thanks for the information guys

I had issues like this, on some of my machines, and not only with this kernel, until I realized that the systems on which I had problems were dual-boot, with a Windows partition, and that on recent kernel, support for PC-Bios Partitions were not on by default in the vanilla kernel sources.
Adding it fixed the problem.
I also have an issue like that on a system with RAID partitions, but I haven't checked yet if adding this module to my kernel fixed it.

I haven't had any problems with 2.6.28 on newer or older machines, except once when I loaded it on a SATA-based laptop (that I didn't know was SATA-based until I actually looked at the drive!) where I got lots of similar errors.

It seems that some of the changes include SATA drives popping up as SCSI instead of IDE or vice-versa. You can easily diagnose this by looking at the messages on the console when you boot - you should be able to spot your partitions easily and then compare to the one given in the kernel panic.

As a quick test, you can add "root=/dev/hda1" or "root=/dev/sda1" or whatever the new kernel detects your root partition as on the kernel command line to see if it makes a difference. You'll probably find that your fstab is messed up too, but you should be able to get to single-user mode at least once you have found the right device name for the root partition.

Incidentally, you can use "/dev/root" as a device name inside /etc/fstab (I think you might need udev, but I can't remember) to always refer to the device that the kernel has mounted on / no matter what's it's called. This allows me to, for instance, run a dual-boot Windows-Linux machine where the Linux detects the drive as SATA (therefore SCSI) and get full SATA benefit but so that under Windows, I can run QEMU (with the -snapshot parameter!) which can use the same partition directly as an IDE drive and it works just the same - one thinks it's /dev/hda, one thinks it's /dev/sda but they both just work and the only place where either device is mentioned is on the LILO command line.

There are also options to stop it happening under certain circumstances (e.g. hda=noprobe, etc.). Because of the SCSI/IDE thing, if you have a SATA drive, unless you bundle SCSI modules into your initrd or hardcode them into the kernel (Y in make oldconfig), then you won't get ANYTHING show up at all. It was great fun finding that out after having managed to successfully avoid SCSI-kernels for years...

"you can use "/dev/root" as a device name inside /etc/fstab" This is exactly what I was recently advocating and that hsa always been th case -you don't need udev for that. this lets you always be able to boot simply by specifying the correct device to the bootloader -otherwise the boot param has to match what the fstab says.

Ledow: I find that removing "ATA/ATAPI/MFM/RLL support" from the kernel helps with that.

I agree as well. I have a notebook that has a SATA HD and DVD-ROM drive and this helps both of them get picked up properly by the kernel at boot time without any extra configuration.

i have tried to use /dev/hda1 and /dev/sda1 in the "root=" option during LILO screen, but still failed. Probably i should try removing the ATA/ATAPI/MFM/RLL support

rvdboom: My machine does not use dual boot. It's completely Slackware

larryhaja: Mine is still IDE

Where to get those LTS kernel ?

Source from here:

http://www.kernel.org/pub/linux/kernel/v2.6/

Bill.