kernelslacker ([info]kernelslacker) wrote,
@ 2006-12-18 22:29:00
Previous Entry  Add to memories!  Tell a Friend!  Next Entry
Current music:Junkie XL - Angels (feat. Gary Numan)
Entry tags:fedora

Fedora kernel-fu.
Been a busy week or two for kernel stuff in Fedora.

Currently, the rawhide kernel is based on 2.6.20rc1. Leap-frogging 2.6.19 completely (FC5 & 6 are still on 2.6.18.6 until the next update, when they'll move to .19.1 or whatever is current at that time).

I spent lots of time last week beating the rawhide kernel into a shape where it may actually boot for some people. It's been a bit of a challenge.
First, the big change is the migration away from the crusty old parallel ATA drivers to shiny new ones that use the same libata infrastructure as the SATA drivers. A side effect of this is that /dev/hda becomes /dev/sda. This isn't a problem if you're using 'mount by label' (which has been the default in Fedora since forever). If you aren't, well, it's going to be fun.

Further complications abound by some mkinitrd changes that were necessary. Another side-effect of moving to libata was that now it's possible to return to userspace after a 'modprobe my-pata-driver.ko' before the driver has even discovered the disks attached to the controller. Oops. This meant we fell over quickly after failing to find volume groups. For now, a hacky solution has been put into mkinitrd which just spins until the disks announce themselves (with a timeout of a few seconds).

As if that wasn't enough, it seems that 2.6.19 brought about another fun change where RAID volumes aren't automatically started. More mkinitrd hacking needed to make that happen. (In the interim, I'll probably back out the kernel change).

More oddness abound when after trying to reproduce one failure, I noticed that
mkinitrd was being prevented from doing its thing by SELinux policy. Oops. This meant a severely truncated initrd was being created, which unsurprisingly meant "you don't get to boot" next time around. Fixed policy should be in rawhide soon.

Finally, some head-scratching currently going on as to what to do about grub's device.map file which has a reference to the disk to install the bootloader on. This could prove to be fun if /dev/hda is now /dev/sda, and a grub update appears.

There are a number of other "doesn't boot" bugs that still need diagnosing. It's going to take a while to get on top of this. Thankfully FC7test1 is still a way off, so there's some hope at least.

I'm really nervous about pushing out 2.6.19 as an update for FC5/FC6 users, so I'm going to take my time on this one. It probably won't be out until the new year at the soonest. (Even in the unlikely event that 2.6.20 appears before then). Trying to build a universal kernel that works on as much hardware as possible is really, really hard at times.




Create an Account
Forgot your login?
Login w/ OpenID
English • Español • Deutsch • Русский…