MD Oops on boot with 2.6.2-rc1-mm3

MD Oops on boot with 2.6.2-rc1-mm3

[Jonathan Kamens]

I get an Oops on boot with 2.6.2-rc1-mm3, trying to boot from a RAID1
MD root partition with two disks in the array; the Oops apparently
causes the raid array not to be assembled, so the boot stops.
Unfortunately, I don't have a serial console so I can't capture the
complete Oops and run ksymoops on it, but here's the information I was
able to gather:

* I have raid=noautodetect turned on, so detection happens from within
  linuxrc in my initrd.  If I get rid of raid=noautodetect, then the
  initial autodetect works fine, but I still get an Oops in the
  autodect run within linuxrc; its call trace looks slightly
  different.

* The kernel was able to successfully start one of my other RAID
  partitions immediately before the Oops which caused md0 to fail to
  start: "raid1: raid set md1 active with 2 out of 2 mirrors".

Here's what I was able to transcribe (I couldn't write down all the
hex numbers that ksymoops would have been able to interpret, but
perhaps what I was able to capture will be helpful):

Unable to handle kernel NULL pointer dereference at virtual address 00000008
EIP is at blkdev_reread_part+0x15/0x90
Call Trace:

  iget5_locked
  blkdev_ioctl
  wake_up_inode
  ioctl_by_bdev
  do_md_run
  printk
  bdevname
  autorun_array
  printk
  bdevname
  autorun_devices
  printk
  autostart_arrays
  exact_lock
  kobj_lookup
  md_ioctl
  get_gendisk
  md_open
  do_open
  blkdev_open
  dentry_open
  blkdev_ioctl
  sys_ioctl
  sys_open
  sysenter_part_esp

I can duplicate this Oops, so if there's any other information from it
which I can provide to help debug it, please let me know.

  jik

Re: MD Oops on boot with 2.6.2-rc1-mm3

[Andrew Morton]

Jonathan Kamens <jik@kamens.brookline.ma.us> wrote:
>
> I get an Oops on boot with 2.6.2-rc1-mm3, trying to boot from a RAID1
> MD root partition with two disks in the array; the Oops apparently
> causes the raid array not to be assembled, so the boot stops.

There appears to be a dud raid patch in -mm.  It'll be one of the md-*
patches.

If you have time, could you work out which one?  Ones to start with might be

ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/broken-out/md-02-preferred_minor-fix.patch

and

ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/broken-out/md-06-allow-partitioning.patch

Re: MD Oops on boot with 2.6.2-rc1-mm3

[Jonathan Kamens]

Yes, backing out md-06-allow-partitioning.patch fixed the MD Oops on
boot for me as well.  Thanks, Andrew and Walt, for the quick response!

  jik

Re: MD Oops on boot with 2.6.2-rc1-mm3

[Andrew Morton]

Walt H <waltabbyh@comcast.net> wrote:
>
> > There appears to be a dud raid patch in -mm.  It'll be one of the md-*
> > patches.
> > 
> > If you have time, could you work out which one?  Ones to start with might be
> > 
> > ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/ \
> > broken-out/md-02-preferred_minor-fix.patch
> > 
> > and
> > 
> > ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/ \
> > broken-out/md-06-allow-partitioning.patch
> > 
> > 
> 
> I had a repeatable oops that sounds identical to what Jonathan
> originally reported. Backing out md-06-allow-partitioning.patch fixed
> the oops at boot for me. Thanks,

Thanks.  md-06 and md-07 are in for a bit of a rethink anyway...

Re: MD Oops on boot with 2.6.2-rc1-mm3

[Walt H]

> There appears to be a dud raid patch in -mm.  It'll be one of the md-*
> patches.
> 
> If you have time, could you work out which one?  Ones to start with might be
> 
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/ \
> broken-out/md-02-preferred_minor-fix.patch
> 
> and
> 
> ftp://ftp.kernel.org/pub/linux/kernel/people/akpm/patches/2.6/2.6.2-rc1/2.6.2-rc1-mm3/ \
> broken-out/md-06-allow-partitioning.patch
> 
> 

I had a repeatable oops that sounds identical to what Jonathan
originally reported. Backing out md-06-allow-partitioning.patch fixed
the oops at boot for me. Thanks,

-Walt