SUMMARY: E10000 boot disk booting on a E420 - long & interesting

From: Filipe Litaiff (filipe_litaiff_at_optiglobe.com.br)
Date: 02/26/04

  • Next message: Filipe Litaiff: "VNC Server on Solaris 9 failing - no proper distribution available"
    Date: Thu, 26 Feb 2004 18:15:31 -0300
    To: <sunmanagers@sunmanagers.org>
    
    

    Hi Forumers,

    This will be an incomplete summary since we have not tried anything in
    practice yet, but the steps are pretty outlined.

    Thanks a bunch for the quick and useful replies from

    Derek Olsen
    Matt Ungaro
    Buddy Lumpkin
    Steve Camp
    Anand Chouthai

    as well as the fellas who remembered us that someone is having fun in other
    places than datacerters.

    Original post:

    -----------------------------------------------------------------------------
    -----------------
    Hi fellow managers,

    Does anyone tried to make Solaris 8 boot disk installed on an E1000 bootable
    in other platform sun4u (E6500 in this case) ? Is that possible?

    What we are planning to achieve that:

    - Make a third mirror of an E10000 boot disk (via SDS).
    - Make this disk bootable
    - Deattach the disk. Remove this disk..
    - In a running E420 with a slot free, insert this disk.
    - Mount the root partitions. Remove all the contents of the /devices
    - Remove path_to_ inst
    - Configure sd.conf properly.
    - Boot this disk, forcing reconfiguration.
    - Voila?

    Will it work? Any missing steps?

    Any help is greatly appreciated.

    Filipe.
    _______________________________________________

    Matt Replied:

    "You could cut some of those steps, place the boot disk in, and on first boot
    do a boot -ar. This will reconstruct the path_to_inst and recreate /dev
    entries. In theory, this should work. But theory and reality aren't always
    one in the same.

    Matt"

    That may work with all regular systems, but the problem is that when a system
    installed in a E10k domain boots, it looks for the control board, starts
    daemons like cvcd and other platform related processes.

    Ryan told us a desencouraging story:

    Just an FYI. We tried doing this for Disaster Recovery purposes. Sun
    does not support this method at all. Yes, "technically" it should work.
    You will also need to regenerate your path_to_inst. Make sure you boot
    the system with boot -ar (this should regenerate path_to_inst).

    We tried doing this but were only able to get the system and running in
    single user mode. (SBUS to SBUS arch. or PCI to PCI arch.)

    Oh, you know what...E420...no go. Two different architechtures. E420 is
    PCI and E10K is SBUS. This will most definetly will not work at all.

    It is certainly not hard to figure out that Sun would not bet on it. But an
    E10k can work with both PCI or SBUS, depending on how you configured your
    system boards. And both are "sun4u".

    Derek sent us a useful receipe for making "transplated" boot disk works.

    # Need to get rid of device trees since the will not work from one box
    # to the next. Also need to nuke the path_to_inst file.
    rm -rf /a/dev /a/devices
    rm /a/etc/path_to_inst

    # Need to restore the device tree from what is actually created during
    # the net boot process.
    cd /
    find devices dev -depth -print | cpio -pdm /a

    # Now we need to create a new path_to_inst
    devfsadm -r /a -p /a/etc/path_to_inst

    # Need to recreate dev entries that got blown away
    devfsadm -i random -i ipf -i md -r /a -p /a/etc/path_to_inst

    That would be great, but when dealing with an E10k domain, things get a bit
    complicated...

    Matt Ungaro basically agreed to Derek.

    Buddy sent us his detailed receipt.

    detach the submirror for this disk (third mirror as you describe below)
    mount the underlying disk slice of that disk to /mnt
    edit /mnt/etc/system and uncomment line that sets rootdev
    edit /mnt/etc/vfstab and change lines from metavolumes to /dev/dsk/cxtxdxsx
    run the following commands:

    rm -r /mnt/devices
    rm -r /mnt/dev

    Pre-Solaris 8:

    rm /mnt/etc/path_to_inst

    drvconfig -R /mnt
    devlinks -R /mnt
    disks -R /mnt
    ports -R /mnt

    Solaris 8 or newer:

    devfsadm -p /mnt/etc/path_to_inst -r /mnt

    Now on Solaris 2.6, boot from new disk with -a option and when prompted,
    re-create path_to_inst. Although it is undocumented, -p option to devfsadm
    does this automatically (truss it and you will see). It doesn't work with
    drvconfig though.

    For Solaris 8 or newer, your done. Boot from the new disk.

    Steve gave us his outlook:

    (...)
    I think you may have to install a new bootblock on the "transplanted" disk.
    I am concerned that the kernel ( /kernel/genunix ) is for a Sun4U-1
    architecture
    (E10K), but you are doing to attempt to run it on a Sun4U architecture (E420
    aka
    Ultra80)... and I fear the consequences. Also, anything on that transplanted
    disk that references /platform/<architecture> or /usr/platform/<architecture>
    will be potential problems.

    I am fairly certain that what you propose is "unsupported" by Sun and is the
    kind of thing that would cause me and my fellow call center E10K support
    engineers to shake our head in wonder at the stupidity of our customer(s)
    and wonder just what the hell they were thinking.

    (...)

    Oh, don't forget to disable all /etc/init.d/ startup scripts that attempt
    to start all sorts of E10K-specific daemons etc.

    Great feedback of an evidently experienced professional. Other reply from him
    (about booting only to a single-user mode)

    (...)
    My guess: some E10K specific scripts and/or daemons are stopping the boot
               process; they may be logging errors, or they may be waiting to
               speak to a non-existent control board or SSP. Without
               /var/adm/messages and/or `dmesg' output and/or /var/log/syslog
               log files (among possibly other files)... I can only speculate
               at this point.
    (...)

    An then, Anand Chouthai came along. He told he works with such scenarios for
    living.

    Yes , it will work, One of the steps I see missing is to take out any
    references to ssp. Also it will work 99% times if the original E10 install
    is done Entire distribution with OEM support.

    (...)

    And at last what seems to be the "voila"

    Take the disk which has the E10K image and put it on E420R:
       Boot from cdrom in single user mode ( same OS on the E10K image) or boot
       from jumpstart in single user mode
       mount the disk with the E10000 image on /a
       cd /a
       mv dev dev.dr
       mv devices devices.dr
       mkdir dev devices
       cd /dev
       find . -depth -print | cpio -pdm /a/dev
       cd /devices
       find . -depth -print | cpio -pdm /a/devices
       cd /a/etc
       mv path_to_inst path_to_inst.dr
       cp /tmp/root/etc/path_to_inst .
       change the vfstab to appropriate disk config
       If the E10000 disk was encapsulated then take the rootdev entries out
       from system file
       install bootblk on the disk
       cd /a/etc
       mv ssphostname ssphostname.dr
       init 0
       ok> boot -rasvw
       when it comes to single user mode <ctrl D> to go to multi user mode.

    Hope it works. Good luck.

    Well, thats all. As I mentioned, we didn4t try it yet, but it seems to be the
    way to get the work done. Anyway I will let the list know how it worked.

    Regards,

    Filipe Barbosa Litaiff
    _______________________________________________
    sunmanagers mailing list
    sunmanagers@sunmanagers.org
    http://www.sunmanagers.org/mailman/listinfo/sunmanagers


  • Next message: Filipe Litaiff: "VNC Server on Solaris 9 failing - no proper distribution available"

    Relevant Pages

    • Re: Installation: doesnt see setup files on hard drive?
      ... > disk the BIOS is set to boot from must have an appropriate Master Boot ... I was worried about those points you made above so I removed all drives except the one I want to instll ...
      (microsoft.public.win2000.general)
    • Re: [SLE] Problems with initrd after mkinitrd
      ... A disk can have three real heads and report a dozen or a hundred! ... > natural metric for partitioning drives that use LBA. ... > - One, less frequent, is the failure of some of the cloned drives to boot ...
      (SuSE)
    • Re: Three Linux Operating Systems in One Computer (Is it possible?)
      ... Hard Disk 1 19.42GiB IDE ... The GRUB boot loader is installed on the first hard disk. ... I am planning to combine /dev/hda6 with /devhda1 using the Gnome Partition ...
      (Ubuntu)
    • [HPADM] Summary MirrorDisk/UX 11.00
      ... disk 0 0/0/1/1.2.0 sdisk CLAIMED DEVICE HP 73.4GMAN3735MC ... Use mkboot to place the boot utilities in the boot area and add the AUTO file. ... LVM will resynchronize the new mirror copies. ... All you need to do is ftp it to your server, chmod 700 on the file then run the command with the cxtxdx name ...
      (HP-UX-Admin)
    • Re: Notice of intention to complain
      ... Suppose there actually is a nasty beastie virus on that disk. ... exists and is okay. ... Boot old computer from it's own hard disk. ... I go rummaging again and find a 320MB Panasonic hard disk. ...
      (alt.comp.anti-virus)