Update to 10.04 2.6.32-25 as KVM Guest no longer boot

[bofh999]

Binary package hint: qemu-kvm

Maybe have to be moved
Anway

KVM Guest 10.04 SERVER updated to 2.6.32-25
After update Guest can no longer boot (freezes at freeing memory...) if you send an crtl alt delete its shows an entry for freeing the md devices and stop there - thats it

Workaround was to reconfig grub to load older Kernel (if someone else having that problem - mount the kvm image somwhere on the host edit /boot/grub/menue.list should be selfdecribing)

Anyway that an simple update breaks ubuntu from booting (even as an virtual guest) is really really really not good

Its reproduceable (i reinstalled a second kvm on the same host - runs fine - after aptitude update / upgrade / reboot same effect)

[Mathias Gug (mathiaz)]

Thank you for taking the time to report this bug and helping to make Ubuntu better.

What is the software version running on the host?

[bofh999 (gdmr)]

Host is Debian 5.0 with all latest Updates.

BTW Status should be critical. I See a lot of People running into same issue.

BTW2, if the host makes a difference then its even more concerning since kvm should be full virtualisation the combination host version and guest kernel shouldnt make a difference

[Serge Hallyn (serge-hallyn)]

I can't reproduce this on a maverick host. Are you using a 32- or 64-bit
guest? What about the host?

You filed this against Ubuntu's qemu package, but you said you are using
Debian 5.0 host. So isn't this bug actually against both the Debian qemu
package and the Ubuntu lucid kernel?

[bofh999 (gdmr)]

both host and guest are 64 bit.

i said it have to be moved maybe but wasnt shure where to fill it exactly.
anyway
it seems to happen when you run quemu/kvm on the host (at least 64 bit debian) and guest 10.04lts 64 bit guest and make latest kernel update.
its everytime reproduceable
its happen now more and more. the only reason not much more are affected is maybe because of the nature of that thing.
server as guests are commonly not rebootet (or most users try to prevent a reboot of course :-)

this week im gonna testinstall a 10.04 as a host and try it again with a guest

but still if the host (different distri but always using qemu/kvm) makes a difference its still very interresting - i mean with a full virtualisation system the guest kernel should run or should not run all the time. what else is the point in using full virtualisation (looks were running into similar issues paravirtualisation has

[Serge Hallyn (serge-hallyn)]

> i said it have to be moved maybe but wasnt shure where to fill it exactly.

Thanks, I didn't know what you'd meant by that. I've marked this as
affecting the kernel. Leann, I'm afraid I'm not sure how to mark it as
affecting precisely the Lucid kernel.

> but still if the host (different distri but always using qemu/kvm) makes a difference its still very interresting

Oh, I remember in 2000 having to grab skunkworks patches to vmware to
make the latest kernel boot on it :)

[Alexandre Derumier (aderumier-odiso)]

I've the same problem, but only on amd opteron host. (opteron 6174).

No problem with intel xeon.

Host system is proxmox, basicaly, debian lenny with squeeze kernel (2.6.32), qemu 0.12.5 or qemu 0.13.

other users reports the same problem

http://forum.proxmox.com/threads/5167-Problem-booting-Ubuntu-Server-Lucid-2.6.32-25-kernels

[Clayton Kramer (clayton-kramer)]

We have also run into the problem on our AMD Opteron KVM hosts and I'm the author of the Proxmox post. The problem seems related to SMP. Setting the guest to run with only one core works.

I see that there was a SMP kernel update in the 2.6.32-25 revision. Could this have caused a regression?

http://changelogs.ubuntu.com/changelogs/pool/main/l/linux/linux_2.6.32-26.48/changelog

* can: Fix SJA1000 command register writes on SMP systems
    - LP: #607824

[Clayton Kramer (clayton-kramer)]

Here's a link to the Proxmox sources for their deb packages.

ftp://download.proxmox.com/sources/

[Imre Gergely (cemc)]

I've tried it on an Ubuntu 10.04 64bit host, with qemu-kvm 0.12.3+noroms-0ubuntu9.2 , and a Lucid 64bit guest with kernel 2.6.32-26. Host is an AMD Phenom II X3 720, not exactly an Opteron but it's AMD.

What I observed is when starting the guest from a powered off position it boots much faster as opposed to doing a reboot after it started. After every reboot it kinda seems freezed for a good couple of seconds after the kernel loads (removing "quiet" helps to see the kernel messages).
Seems like on every reboot it just gets slower and slower. After like 4-5 reboots, it takes 15-20 seconds to get over the kernel and go on with the boot process.

This doesn't happen with a single CPU on the guest.

Not sure if it helps or not, but seems a bit strange. Maybe you want to check this, does it really lock up, or it just takes a while to boot?

[Imre Gergely (cemc)]

It seems to me the boot time after the reboot is related to the uptime somehow... I'm up to 60second "freeze" now on one VM running a 10.04 64bit guest with 3 CPUs. The kernel loads, then there's 60 seconds silence, when nothing happens, and then it continues to boot.
If I stop the vm and restart it, the boot goes very fast. Strange indeed.

[bofh999 (gdmr)]

Its with a 10.04 server 64 bit Guest on 3 cores (dont know with one)
host is debian 5 latest updates

And it defently freezes not for 60 sek, not for 60 min. forever (or better as long the system get electric power lol)

Btw i turned off splash etc as described bevore it freezes at freeing memory... nothing else and nothing else in the logs