Red Hat Bugzilla – Bug 1096358
System fails to boot with "unaligned pointer" error after live install
Last modified: 2014-05-22 10:51:21 EDT
Created attachment 894140 [details]
screenshot of virtual system failing to boot
Created a new livecd-lxde based on the "fresh as of 20140509 rawhide" [the ks and pkglist are attached]
booted up on qemu-kvm virtual system and ran /usr/bin/liveinst
selected custom partitioning, deleted everything and did auto btrfs config
installed and then rebooted ... oops
Created attachment 894143 [details]
pkglist in lxde-livecd
Created attachment 894145 [details]
kickstart file used to build the livecd iso
I will keep this virtual system around a while in case you need something off it.
I believe you should be able to re-create the problem as it happens everytime for me.
satellit (Thomas Gilliard) has reported seeing this too, on VMs and bare metal: https://fedoraproject.org/wiki/Test_Results:Fedora_21_Rawhide_2014_05_Install#Live_image . I confirm seeing it in a Rawhide-on-Rawhide VM (using a live image composed with today's Rawhide).
Nominating as an Alpha blocker, criterion https://fedoraproject.org/wiki/Fedora_21_Alpha_Release_Criteria#Expected_installed_system_boot_behavior - "A system installed with a release-blocking desktop must boot to a log in screen where it is possible to log in to a working desktop using a user account created during installation or a 'first boot' utility." It doesn't seem to do so, at least when installing from a live image.
1. After the install, running grub2-install and grub2-mkconfig results inm a bootable system ... but should not need to do that.
2. Before being "fixed", the /boot/grub2/grug.cfg file is strange ... the rescue entry is first and the regular kernel is second ... never seen that before.
If the grubby package is one where my patch is included and DEBUG is enabled (or even if my pach is not included but DEBUG is enabled), we should be able to see if grubby is involved since the messages it spues are captured in anaconda logs.
I have an image with a debug-enabled grubby here, I'll try and attach the relevant anaconda logs later tonight.
Created attachment 895010 [details]
program.log from a live install with grubby debugging enabled
Here's program.log from an install affected by the bug, with grubby debugging enabled. This is with a grubby build with Gene's patch for https://bugzilla.redhat.com/show_bug.cgi?id=1094489 applied, but I don't believe that patch is relevant here, reports indicate the bug's affected stock images too - just attaching this for the debug output.
pjones has been working on this, but it doesn't look like an easy fix at all. We're currently having trouble identifying the problem.
just for the record, here's some of the data we've found on affected / not affected configurations:
1) No non-live images seem to be affected by this.
2) Builds from both 'fedora-livecd-desktop.ks' (the old "Desktop" kickstart) and 'fedora-live-workstation.ks' (the new "Workstation" kickstart) seem to be affected.
3) SoaS images do *not* appear to be affected: grub2 runs successfully after an install from Fedora-Live-SoaS-x86_64-rawhide-20140520.iso , but not from Fedora-Live-Workstation-x86_64-rawhide-20140520.iso .
4) Not currently sure about KDE, Xfce etc.
5) the issue showed up some time between 2014-05-05 and 2014-05-07, according to satellit's testing.
on 2014-05-05, there was a new kernel build: kernel-3.15.0-0.rc3.git5.3.fc21 . On 2014-05-06, there was a new anaconda build: anaconda-21.35-1.fc21 . Possibly significant changelog entry: "install: Move Payload postInstall() after bootloader (walters)". There was another new kernel build: kernel-3.15.0-0.rc4.git0.1.fc21 . On 2014-05-07, there was another new kernel build: kernel-3.15.0-0.rc4.git1.1.fc21 . there was also a new gcc - gcc-4.9.0-3.fc21 - but I don't believe the new kernel would've been built with the new gcc.
also affects Fedora-Live-KDE-x86_64-rawhide-20140520.iso .
Created attachment 897840 [details]
strace output from grub2-install on an affected system
As requested by pjones, this is the output of:
strace -s64 -v -f grub2-install /dev/sda > grub2-install.strace 2>&1
from rescue mode on an affected system (installed from 05-20 KDE live), after chrooting and deleting /boot/grub2/i386-pc and /boot/grub2/grubenv . the bug is still present after doing this. Will attach similar output from a non-affected install (SoaS) for comparison.
Created attachment 897842 [details]
strace output from grub2-install on a non-affected system
Here's the same output from a non-affected case (an install of the 05-20 SoaS image).
http://koji.fedoraproject.org/koji/buildinfo?buildID=518131 seems to resolve this, for me. Built a Workstation-ish live image to test, boot of installed system succeeeds. Also tested an SoaS-ish live image still works.