Sat, 28 Mar 2009

An update on GRUB2

Some time ago I wrote about the the state of GRUB2 and a milestone on getting it boot my Apple PowerBook G4 without manual intervention. More than a year later, GRUB2 has changed and improved a lot, as the community keeps growing and patches and ideas are continously being posted.

Some months and commits after my previous post, GRUB broke again on Apple OpenFirmware and I'd get dropped to OF console, the amount of commits since the last known working version and the current SVN was quite big, and although I was able to narrow it to a few suspicious changes, I had no time to bisect it properly, and sadly I had to go back to yaboot for a while.

But procrastinating sometimes helps, and when I should have been writing and studying, on December I gave GRUB a new try on my laptop to see if a few important changes to memory allocation would have changed anything. And it did! So after fighting quite a few problems, I was able to report partial success to grub-devel.

Again, getting GRUB installed correctly was a bit challenging and needed some hackery, due to incorrectly generated, and the linux module mysteriously not getting loaded. Luckily, Michel Dänzer found out that this was due to a bug in sort ordering in the HFS module, which broke the lookup of files with underscores like _linux.mod, and for which he posted a possible fix by taking Linux's table of character ordering, which is a blob of hex values.

GRUB developers didn't seem too happy about applying the patch: they argument that a blob like that should be well documented or written in some other more readable way, and there's a possible problem with the mix of Linux GPLv2 and GRUB GPLv3+ codebases, if a table of data like what Michel posted is actually copyrightable. The discussion ended up dying and nothing was done... until Pavel Roskin picked it up weeks later and posted a new patch, based on hfsutils GPLv2+ code, which addressed these issues. The new patch seems to have a few issues, which makes it fail as before, but hopefully it'll be fixed soon.

Additionally, I wasn't able to boot using UUIDs as the search commands fails to detect the correct boot device on my system (but not on Michel's), so I had to disable that in /etc/default/grub.

To workaround the linux module loading bug while the patch is fixed, I just added this ugly hack to /etc/grub.d/09_local_prelinux:

#! /bin/sh -e

# Work-around for bugs in the hfs module which makes the load of
# linux.mod fail.

cat << EOF
insmod (hd,3)/usr/lib/grub/powerpc-ieee1275/_linux.mod
insmod linux

This is enough to get the initrd and linux commands available. However, update-grub will still add search commands to your menu entries even if you disabled UUID support; I can't understand why, but I know it breaks on my PowerBook due to some OF rarety. Just removing the line from the menu entry will leave me with a working config that boots without any manual editing at GRUB prompt.

The latest GRUB snapshot in Debian fixes the issue, but adds one last issue: update-grub will fail due to some gfxterm detection code, a workaround is to replace an exit 1 with exit 0 when this happens in /etc/grub.d/00_header.

On the “weird architectures” front, it's worth noting that this month Dave Miller popped up on the list and started posting patches to fix the rotten SPARC port, and I think it's safe to assume that it'll be on an usable state really soon. Impressive!