Bug 120973

Summary:	Installation of SuSE 10.0 kernel panics when swap partition is on software raid
Product:	[openSUSE] SUSE LINUX 10.0	Reporter:	Ian Rogers <ian.rogers>
Component:	AutoYaST	Assignee:	Pavel Machek <pavel>
Status:	RESOLVED DUPLICATE	QA Contact:	E-mail List <qa-bugs>
Severity:	Normal
Priority:	P5 - None	CC:	behlert, hare
Version:	Final	Keywords:	Install
Target Milestone:	---
Hardware:	i686
OS:	SUSE Other
Whiteboard:
Found By:	Customer	Services Priority:
Business Priority:		Blocker:	---
Marketing QA Status:	---	IT Deployment:	---

Description Ian Rogers 2005-10-07 13:38:49 UTC

RAID-0 is commonly used to improve performance of swap partitions. The SuSE 10
set up configures the swap partition to be the resume partition. If the swap
partition is RAID then this results in a kernel panic after the first CD
installs and the system reboots. To remedy the problem the user must choose the
"fail safe" kernel option or edit the kernel arguments and remove the
"resume=/dev/mdX" option (and possibly replace it with "noresume").

Comment 1 Uwe Gansert 2005-10-10 13:59:53 UTC

I don't expect this to be autoyast related.
Kernel Panic - Hubert?

Comment 2 Hubert Mantel 2005-10-10 16:04:38 UTC

I already have a similar report but still I'm unsure what is actually meant:
Does it mean a regular reboot fails or a resume? I think the latter since
removing the "resume=" parameter makes a difference.

Comment 3 Ian Rogers 2005-10-10 16:11:43 UTC

Thanks for looking into this. Its the former of your options. A regular reboot
causes a kernel panic so the installation fails after the first CD. This
behaviour also happened with RC1. To fix it you can either choose failsafe or
edit the resume line to noresume. I have not tried a suspend/resume on this
machine and doubt it would work without the resume line for grub.

Comment 4 Forgotten User ZhJd0F0L3x 2005-10-10 18:06:17 UTC

i can easily imagine a bug in the "resume from initrd"-code when swap is on a
RAID  device. Maybe Hannes can shed some light on this.

Comment 5 Pavel Machek 2005-10-10 19:13:30 UTC

Do you have actual messages? Can you verify if /dev/mdX is set up correctly at
the time when resume is attempted by initrd code?

[swap on raid seems like very strange thing to do. Unless someone audited raid
code, it might randomly deadlock etc under high load. We may want to discourage
that combination.]

Comment 6 Ian Rogers 2005-10-11 08:56:00 UTC

Thanks for your help. Sorry but I have no boot messages, I'd need a camera of
VMWare to capture them - there isn't any specific debug information for this
problem anwyay. The software RAID is correctly configured and detected, without
the resume statement the kernel will successfully boot the machine. The only
issue is when resume is specified, it causes a kernel panic if want is pointed
to by the resume argument is a software RAID. I don't believe there is any
deadlock issue - many people have used this configuration for many years.
According to:

http://www.tldp.org/HOWTO/Software-RAID-HOWTO-2.html

You may want to discourage swap on software RAID as they say the kernel can do
swap striping without it anyway. If I'd known this I may well have partitioned
the disks differently. Thanks again.

Comment 7 Forgotten User ZhJd0F0L3x 2005-10-11 09:39:58 UTC

anyway - we should not crash during normal boot, this needs to be fixed.

Comment 8 Pavel Machek 2005-10-22 18:26:04 UTC

I thought I replied on this one already?!

can you try to boot into single user mode, setup raid manually, and try to resume system by hand? It should complain that swap has no suspend signature, but not panic.

Comment 9 Pavel Machek 2005-10-22 18:32:38 UTC

Aha, now I see... see the other bug, there's better descirption there.

*** This bug has been marked as a duplicate of 121829 ***