[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#601782: installation report - beta1 installer fails with kernel error - drivers/md/md.c:6192, invalid opcode

On Fri, Oct 29, 2010 at 01:22:16PM -0400, Clyde E. Kunkel wrote:
> I am not using mdadm.  The installer provided it and I used it.  I
> simply at tty-2 during the install attempt did a mdadm --detail /dev/md*
> in order to provide as much info as possible.  The debian software
> provided mdadm.  The results of mdadm --detail do not show a degraded
> array.  Also, I notice that the partitioner correctly displays the md
> and the vgs/lvs before the kernel bugchecks.
> I do have ubuntu installed on the same configuration and it uses dmraid,
> AFAICT.  I also have suse and fedora installed and they both use mdadm
> and I have not seen any raid degradation.
> I am also aware that consensus seems to be to not use bios raid since
> you can achieve the same with mdadm.  In fact, on my other test machine,
> that is the way I have configured a raid 10 set serving as a pv for
> multiple LV installations.  That said, since there are many mobos with
> bios raid, I think it prudent to test that configuration to ensure the
> software is robust.
> If mdraid is what debian is using, perhaps a switch to mdadm would be
> prudent.

Well mdraid and dmraid are NOT the same thing.  The mdraid almost
certainly is using the disks differently than the BIOS raid.  That alone
could explain the crash.

Debian's installer does NOT support dmraid, although there is experimental
support by passing a dmraid=true or something option at boot time.

dmraid is nowhere near feature complete enough on many raid types for
production use.  I was rather unpleasantly surprised when a disk failure
in a raid1 made debian unbootable on dmraid.  You have to replace the
broken disk and rebuild under windows to get dmraid to work with it again.
That's a lot less robust than mdraid which has no issues with disk
failures and can be rebuilt while running linux.

Now if the various chipset makers would document their raid format so
that open source could support them, then dmraid might be functional,
but they don't in general.

Len Sorensen

Reply to: