Won't boot after upgrade FC2 to FC3 on Dell 2650, LVM problem?

[Aleksandar Milivojevic]

Aleksandar Milivojevic wrote:
> Deron Meranda wrote:
> 
>> I have a Dell PowerEdge 2650 (dual Xeon) which was running FC2.  I
>> upgraded to FC3 and after the first reboot it failed to start the
>> kernel.  It looks like it is having trouble reading or mounting the
>> filesystems.  The upgrade process itself went flawlessly.  Also it was
>> initially installed/built with FC2, and no non-Fedora software had
>> ever been loaded on it.
> 
> 
> I've seen the same problem on some systems when I was playing around 
> with LVM.  The only difference my for-fun machines were initial 
> inistall.  When booted into rescue mode, I can mount root file system 
> (from LVM volume).  When booting normaly, kernel fails to find LVM 
> volumes (and thus no root file system to mount that leads to kernel panic).
> 
> Have you got any replies on this off-list, or maybe managed to figure it 
> out yourself?

OK, I found why it hasn't worked in my setup.  It was timing issue 
during execution of init script from initrd image.

What happened was that device driver for my SCSI card (sym53c8xx) was 
rather slow in detecting disk drives.  Init script would load the SCSI 
driver, and than go after loading raid1 and dm-* drivers, and detecting 
logical volumes.  The sym53c8xx didn't had enough time to find and 
report disk drives, and the rest was chain reaction: raid1 devices were 
not configured (since raid1 module couldn't see any disks), and LVM 
couldn't find any volumes (since the mirrors were not setup).

This also explains why everything was OK when booting from rescue CD, 
but not when booting normally from disk.  When booting from rescue CD, 
sym53c8xx is first loaded, and than after some user interaction, raid1 
and dm-* modules are loaded.  "user interaction" part gave sym53c8xx 
enough time to detect and report attached disks before raid1 and dm-* 
modules were loaded.  No such luck when booting off disk.

I made rather dirty workaround (inserted "sleep 30") after "insmod 
53c8xx" line in init script.  This gave 53c8xx driver enough time to 
detect disks, and now everything works.

Now I need to remember that I'll have to rebuild initrd manually every 
time I update the kernel...

What SCSI controller do you have on your motherboard, and how disks are 
connected?  It might be that part of my problem was the fact that disks 
were connected to second SCSI controller (first SCSI controller was faulty).

Does anybody know if "insmod" is supposed to wait until driver 
initializes itself (in this case, finds all devices attached to SCSI 
controller).  Or is there race issue with using initrd images (that most 
people are lucky not to hit)?

-- 
Aleksandar Milivojevic <amilivojevic at pbl.ca>    Pollard Banknote Limited
Systems Administrator                           1499 Buffalo Place
Tel: (204) 474-2323 ext 276                     Winnipeg, MB  R3T 1L7