Skip to main content

Notes on Oracle VM Server for SPARC


Notes on Oracle VM Server for SPARC
  1. After logical domains are reconfigured by Oracle VM Server for SPARC or after executing the "ldm unbind-domain/ldm bind-domain" command, if the guest domain is operated by the reset(8) command of XSCF firmware before executing the "ldm add-spconfig" command, a guest domain other than what had been specified may be reset. Or, the specified guest domain is not reset. You should save a configuration of the logical domains with the ldm add-spconfig command. Then, to reset the guest domain before saving it, execute the ldm stop command from the control domain, not from the XSCF.
  2. If you specify a logical domain configuration for the next time you start, use the ldm set-spconfig command in place of ldm add-spconfig -r.
    If you use the ldm add-spconfig -r command to specify a logical domain configuration for the next time you start, and operate a guest domain with the reset(8) command of the XSCF firmware, another guest domain may be reset.
  3. If you execute the ldm migrate-domain command with Oracle VM Server for SPARC for live migration, the following problems occur with the XSCF.
    - If you execute the showdomainstatus(8) command, "Unknown" is the status displayed for the migrated guest domain.
    If you execute the ldm add-spconfig command from the migrated control domain to save configuration information, the status will be displayed normally by the showdomainstatus(8) command.

    - After live migration is executed, if showdomainstatus(8) is then executed at the migration source, "Host stopped" is the status displayed for the migrated and non-existent guest domain.

    - When a physical partition (PPAR) is powered off by the poweroff(8) command, not all of the guest domains may be shut down correctly.

    - When a guest domain is reset by the reset(8) command, a guest domain other than the specified one may be reset. If you reset the guest domain, execute from the guest domain, and not from the XSCF.

    - If SNMP is set, the name of the guest domain under trap notification may be incorrect.
  4. The migration with the ldm migrate-domain command is not supported if the logical domain at the migration source is in the OpenBoot PROM state.
    Perform migration with the ldm migrate-domain command after changing the logical domain at the migration source to either of the following states (CR 15858731):
    - Stopped state (bound state)

    - State where Oracle Solaris is running
  5. Be sure to start the ldmd service (svc:/ldoms/ldmd:default) of the control domain.
  1. The ldm list-rsrc-group, remove-core -g, and ldm remove-memory -g commands of Oracle VM Server for SPARC 3.2 are not supported. For details, see the Oracle VM Server for SPARC 3.2 Administration Guide.
  2. From Oracle VM Server for SPARC 3.2 onwards, you can allocate memory to a logical domain in units of 256 MB if the dynamic reconfiguration of the physical partitions (PPAR DR) feature is enabled by the XSCF firmware. On the other hand, if the PPAR DR feature is disabled, you can allocate memory in units of 4 MB.
  3. Suppose that the system disk in a physical partition (PPAR)(PPAR #A) is switched to a system disk that has been used for another PPAR (PPAR #B), in a system with Oracle VM Server for SPARC 3.1 or later. The PPAR #A domain configuration information saved on the XSCF may be overwritten over the PPAR #B domain configuration information.
    Uninstall Oracle VM Server for SPARC from the system disk (PPAR #A) before switching to the system disk used with the other PPAR (PPAR #B) and starting with the new system disk (PPAR #B). After the switching is completed, reinstall Oracle VM Server for SPARC on the system disk.
    The same procedure is needed when using the system disk connected to another unit or when the saved domain configuration information in the XSCF is restored using the restoreconfig(8) command.
  4. Suppose that you deleted a building block (PSB) by executing dynamic reconfiguration of physical partitions (PPAR DR) in a system with Oracle VM Server for SPARC 3.3 or later. Then, a CPU socket constraint in the degraded state may be added to each logical domain. This is not a problem but an operation for saving the use status of the CPU sockets before PPAR DR execution.
    If the CPU socket constraint in the degraded state appears for a logical domain for which the CPU socket constraint is not set, ignore this constraint.
    For a logical domain that uses the CPU socket constraint, PPAR DR done by the deleteboard command places the CPU socket constraint in the degraded state. Also, resources not specified with the CPU socket constraint may be assigned to the logical domain.
    When replacing a building block (PSB) in response to a fault and so on, first add the system board. Then, you can execute the ldm set-socket --restore-degraded command to restore the CPU socket constraint. If you are only deleting the building block, execute the ldm set-socket command as required to set the CPU socket constraint again.
    For details on the CPU socket constraint, see "8.14  Managing Logical Domain Resources Associated with CPU Sockets" in the Fujitsu SPARC M12 and Fujitsu M10/SPARC M10 System Operation and Administration Guide.
  1. - Before building block deletion
# ldm list-socket
CONSTRAINTS
SOCKET
    TENANT  VCPUS CORES SOCKET_ID GROUP
    primary 8     4     4         /BB1
      :
  1. - After building block deletion
# ldm list-socket
------------------------------------------------------------------------------

Notice: the system is running a degraded configuration because some
required resources were removed by Physical DR.
------------------------------------------------------------------------------

CONSTRAINTS
    DOMAIN              SOCKET_ID      STATE
    primary* (degraded) 4              active
    dom00* (degraded)   0, 3, 4, 5, 6  active
SOCKET
    TENANT  VCPUS CORES SOCKET_ID GROUP
    primary 8     4     4         /BB1
      :
-----------------------------------------------------------------------------
  1. In Oracle VM Server for SPARC 3.3 or later, when the logical domain configuration has been restored by the ldm init-system command, resource binding may fail even if the CPU socket constraint is not set. Then, the following message is output.
    [Example] 
# ldm bind-domain XXXX
Not enough free memory in specified FJ sockets to meet the request.
Domain XXXX has FJ socket resource constraints for recovery.
Use 'ldm set-socket socket_id= XXXX' to clear. 
  1. If the above message has been output, clear the CPU socket constraint with the ldm set-socket command as instructed in the message. Then, try again to bind the resources.
    [Example]
# ldm set-socket socket_id= XXXX
# ldm bind-domain XXXX 
  1. When using the vHBA (Virtual SCSI Host Bus Adapters) function assigned to a guest domain, set the guest domain auto-boot to false, and then save the updated logical domain configuration information.
  1. [How to set]
primary# ldm set-variable auto-boot\?=false domain-name
primary# ldm add-spconfig config-name