Boot fails on qemu. "Cannot find root file system!"

[MitchellNCharity]

Build 465 fails to boot inside of qemu.

olpc-redhat-stream-development-build-465-20070625_0119-devel_ext3.img
x86_64 fc6 qemu current (0.8.2).

I believe 445 is the last build which worked. 443 works. 445 is said to work. More recent builds have failed. At least one of them on an unknown 32-bit platform.

 [983061.152239] hda: 985057 sectors (504 MB) w/256KiB Cache, CHS=993/255/63, (U)DMA
 [983061.206658] hda: cache flushes supported
 [983061.259795]  hda: hda1
 no root yet, udev rule will write symlink...
 
 waiting up to 60 seconds before dropping to emergency shell.

The wait times out with "WARNING: Cannot find root file system!".

[MitchellNCharity]

Shortened summary.

[MitchellNCharity]

Some additional notes from #sugar/#olpc:

Screenshot: http://www.vendian.org/mncharity/olpc/no_kernel_panic.png

It is not a kernel panic.

In the emergency shell, mount -L OLPCRoot /mnt works. And looks normal.

If one uses the emergency shell to ln -s /dev/hda1 /dev/root , the continued boot encounters:

[1064799.848014] BUG: unable to handle kernel NULL pointer dereference at virtual address 00000020

I'm attaching the boot log. The "no root yet..." and shell interaction occurs just after " hda: hda1".

[MitchellNCharity]

boot log mentioned in "06/27/07 15:54:06 changed by MitchellNCharity".

[MitchellNCharity]

boot log showing both "kernel NULL pointer dereference" and later hang after "Mounting other filesystems:".

[MitchellNCharity]

Adding root=/dev/hda1 to grub's kernel line prevents "Cannot find root file system!", and gets you directly to the "BUG: unable to handle kernel NULL pointer dereference at virtual address 00000020", and the later hang after "Mounting other filesystems:".

(discovery credit to j5)

[MitchellNCharity]

Booting to init level 2, and disabling the netfs boot scripts, avoids the "Mounting other filesystems:" hang.

mv /etc/rc.d/rc3.d/S25netfs /etc/rc.d/rc3.d/DISABLED-S25netfs mv /etc/rc.d/rc5.d/S25netfs /etc/rc.d/rc5.d/DISABLED-S25netfs

This workaround, combined with the "root=/dev/hda1" workaround, permits booting build 481.

The kernel null pointer error remains, but doesn't prevent boot.

Now that a workaround exists, perhaps the priority of this ticket should be downgraded?

[cscott]

This bug occurs for me for the stock Fedora Core 7 LiveCD as well (http://torrent.fedoraproject.org/torrents//Fedora-7-Live-i686.torrent) launching with:

qemu -hda hard-drive.img -cdrom Fedora-7-Live-i686/Fedora-7-Live-i686.iso -boot d -m 256

So this bug appears not to be OLPC-specific.

[cscott]

I believe this is http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=431829 "ata_piix does not work with the PIIX IDE controller provided"; further discussed at http://www.mail-archive.com/qemu-devel@nongnu.org/msg10716.html

I'm guessing our recent kernels began using ata_piix for IDE, which broke qemu. This is a qemu bug, not our fault. (Some versions of) kvm seem to work fine.

[MitchellNCharity]

Build 494 works. It boots, and there is no BUG in the boot log.

[cscott]

Replying to MitchellNCharity:

Build 494 works. It boots, and there is no BUG in the boot log.

I don't think debian's qemu has the correct fix to make this work yet. I'm not going to reopen the bug, but be aware that it may not work for everyone. (Debian's kvm has the patch to make it Work For Me, but I have to use 'kvm -no-kvm' since there seems to be an unrelated bug which otherwise prevents me from getting from grub to the boot penguin.)

[dilinger]

Rather than keeping the bug open, please document it in the wiki; people are more likely to see that when they're attempting to get images working with qemu.

[dilinger]

Er, rather than considering reopening the bug. Need more coffee..