Bug#544155: [grub-pc] grub update renders system unbootable (unaligned pointer 0x700ba)

[Felix Zielcke]

Am Samstag, den 29.08.2009, 12:46 +0200 schrieb Martin Steigerwald:
> Am Samstag 29 August 2009 schrieben Sie:
> 
> > Am Samstag, den 29.08.2009, 11:14 +0200 schrieb Martin Steigerwald:
> 
> > > I just let aptitude update grub-pc to 1.96+20090826-3. I do not know
> > > the earlier version, but I guess it was one version before, cause if
> > > I still had the squeeze one it wouldn't have updated to the unstable
> > > automatically.
> > >
> > > On booting up today I got:
> > >
> > > [Linux-bzImage, setup=0x2c00, size=0x25a070]
> > > vga=792 is deprecated. Use set gfxpayload=1024x768x24,1024x768 before
> > > linux command instead.
> > > unaligned pointer 0x700ba
> > > Aborted. Press any key to exit.Operating system not found
> > >
> > > I also tried with another kernel with the same result.
> >
> > With what kernel?
> 
> vmlinuz-2.6.29.6-tp42-toi-3.0.1-01085-gf325e4a

So another custom kernel with some non offical patches.
Maybe they just didn't test it with the 32bit boot protocol the kernel
now have (mainly for EFI).
Unfortunately we can only check the boot protocol number the kernel
officialy supports or maybe the file name but that isn't reliable.
The *0317-1 version didn't use yet the 32bit Linux loader as default.

Please attach your kernel .config. Then I test if this happens with an
official kernel with that config too or it that seems to be related to
this other patches.

> 
> > > Then I booted into grml, overwrote the swap partition and recreated
> > > it via mkswap in order to make sure the TuxOnIce snapshot is
> > > invalidated to avoid data loss in next boot.
> > >
> > > Then I chrooted to my debian system after bind mounting /dev, /proc,
> > > /sys to it, downgraded grub-pc to 1.96+20090317-1 from squeeze and
> > > ran grub- install.
> > >
> > > I thought about running grub-install for the grub-pc from unstable,
> > > but I read the current sid version might be broken anyway[1], thus I
> > > thought I'd better downgrade to the squeeze one in order to get a
> > > working system quickly again. I could try updating to the sid one
> > > again, without running grub-install manually and then I run it
> > > manually and see whether it fixes the issue. So you could know
> > > whether its grub itself or a missing grub- install.
> >
> > If grub-install doestn't get run then /boot/grub still has the old grub
> > and so nothing actually changed, except if grub-mkconfig generates a
> > different grub.cfg.
> 
> So since something changed it means that grub-install has been run?

It should be run, else it can happen that you can't boot your system.
For example we added `--no-floppy' to the search lines, but an older
GRUB 2 doestn't understand it and so prints an error.

> > > Anyway the expected result is: After an update of grub-pc package the
> > > system boots up manually. Thus if a new version is incompatible to
> > > the old stage 1 the upgrade process has to force a grub-install IMHO
> > > or at least strongly recommend it. I don't remember seeing anything
> > > about this in debconf questions. And if a new version of grub-pc
> > > doesn't work on my ThinkPad T42 it shouldn´t enter testing ;)
> >
> > We have already a debconf prompt for running grub-install.
> > If I reverse the if of the postinst correctly, then it gets shown
> > if /boot/grub/stage2 or /boot/grub/menu.lst doestn't exist
> > or /boot/grub/core.img exists.
> > So at least if grub-install from grub2 were run once.
> > Debconf prompts are by design only shown once, if you then want to see
> > it again you'd need to run `dpkg-reconfigure grub-pc'.
> 
> I don't exactly get this. So I can try dpkg-reconfigure grub-pc, say it 
> should run grub-install everytime grub-pc is updated and be done with it?

It should have already asked you this. But maybe you just didn't notice
or whatever. In that case `dpkg-reconfigure grub-pc' just asks again.
Just choose the device(s) where you want to have grub-install run and
you don't need to think any longer about this as long as the devices
don't change.

> > > This is on an IBM ThinkPad T42.
> > >
> > > [1] http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=540125#75
> >
> > I should have mentioned that I meant the 1.96+20090825-1 and 1.96
> > +20090826-1 at that time currently in sid and not the 1.96+20090826-3.
> 
> So I should try the unstable one again?
> 
> What would be your directions?

You can also try the current squeeze version which is 1.96+20090808-1
That uses the new Linux loader too.
With linux16 and initrd16 you can try out the old one. That should work.

-- 
Felix Zielcke
Proud Debian Maintainer