test

Solaris 7 11/99 Release Notes Supplement for Sun Hardware

Dynamic Reconfiguration

Release notes and other technical information in this section apply only to the Solaris 7 11/99 version of Sun Enterprise 10000 Dynamic Reconfiguration (DR).

General Issues

This section contains general issues that involve DR on the Sun Enterprise 10000 server. You should read this section before you attempt to install or configure DR.

dr-max-mem Utility

The way dr-max-mem is used has changed since the Solaris 2.5.1 release. Please see the appropriate section of the Solaris 7 11/99 version of the Sun Enterprise 10000 Dynamic Reconfiguration User's Guide for a detailed description of this important OBP variable.

DR and Bound User Processes

For Solaris 7 11/99, DR no longer automatically unbinds user processes from CPUs that are being detached. Users are now required to perform this operation themselves before initiating a DR Detach sequence. The DR Drain operation will fail if CPUs are found with bound processes.

Nonpageable Memory Location

For Solaris 7 11/99, the location of nonpageable memory (that is, OBP and kernel memory), the kernel cage, resides in the highest physical address space, which generally is on the highest-numbered system board in the domain. The direction of growth of this area is from highest to lowest physical addresses. Exceptions to this rule do exist. Always use drshow(1M) to verify a board's memory state.

Kernel Cage Location

For Solaris 7 11/99, the location of the kernel cage changed from the lowest numbered system board(s) to the highest numbered board(s). However, some customers require the kernel cage to be located on the lowest numbered system board. If you have this requirement, you can apply Patch 105873-08, as a minimum level, on the SSP to change the location of the cage. Refer to the Patch README file for instructions on how to install the patch. You can obtain the patch from http://sunsolve.corp.sun.com or from your local service provider.

DR and Tape Drives

For Solaris 7 11/99, the st driver, which has been considered DR unsafe in the past, is now DR safe. The driver now waits for existing I/O requests to complete before the drive is suspended.

DR and SunFDDI

If your Sun Enterprise 10000 server contains any SunFDDI(TM) SBus Adapter 5.0 interfaces and you plan to use DR, you must install Sun Patch 104572-05, or the highest revision level available. You can obtain this patch at the following Web site:

DR and the Sun StorEdge A3000

Dynamic Reconfiguration of Sun(TM) StorEdge(TM) A3000 arrays on the Sun Enterprise 10000 running Solaris 2.6 requires RAID Manager 6.01.00, or later, for correct operation.

DR and the Sun StorEdge A5000

DR will be supported on systems configured with Sun StorEdge A5000 devices and the latest StorEdge A5000 patch. Contact your service provider for more information on how to obtain the patch.

For I/O boards connected to the A5000, you must ensure that the FC-AL disk firmware version is ST19171FC 0413 or newer version. For more information, refer to the following Web site:

DR and pln and soc Drivers

If you are using pln and/or soc drivers on your server, you must ensure that the parameters are properly set in the /etc/system file. The file should include the following two entries, exactly as the appear in the example.

set pln:pln_enable_detach_suspend=1

set soc:soc_enable_detach_suspend=1

Known Bugs

This section contains the known bugs in the Sun Enterprise 10000 DR feature of the Solaris 7 11/99 release.

DR Abort Detach May Not Resume All Detached Disks (4010910)

If a DR detach operation fails to detach an sd device and the controller has sd devices that have already been detached, the abort-detach operation does not resume (that is, re-attach) those devices that share the same controller instance and that have already been detached. This situation rarely occurs. The most likely time for it to happen is if the disk drive firmware is locked up.

Workaround: None

dr_daemon Does Not Find Veritas Device Usage (4011123)

Veritas software usage of disks is not displayed by DR applications in their device-usage displays.

Workaround: None

disks(1M) Command Preens Controller Entries and Corrupts the Sun StorEdge A3000 Device Tree (4096048)

The Solaris disks(1M) command preens or deletes symbolic link entries when disk devices are no longer present in an Enterprise 10000 domain. If you have dynamically reconfigured the devices so that they are no longer present, the Symbios-managed device tree and the Solaris-managed device tree can become inconsistent.

Workaround: You must obtain the proper patch from SunSolve or from your service representative. This patch contains fixes that ensure persistent node names under the Solaris operating environment when StorEdge A3000 devices are dynamically reconfigured out of a given Enterprise 10000 domain. The patch ensures that when the same array is brought back into the domain, it inherits the same node name it originally had, thereby maintaining data availability without having to modify system mount points or user application configurations. You must ensure that you do not perform reconfiguration reboot operations (boot -r) in a domain when A3000 devices are present on the system boards.

DR Operation Fails Due to an Incompatibility with the DMP Feature in SEVM (4215683)

DR and the DMP feature in Sun Enterprise Volume Manager (SEVM) are not compatible. If any disks are under DMP control, they cannot be detached.

Workaround: None

Cannot Drain and/or Detach Sun Enterprise Server System Boards that Host QFE Cards (4231845)

The configuration of a boot server for Solaris 2.5.1-based x86 clients results in the execution of rpld daemons. These daemons maintain an active (open) reference to all network devices, whether or not such devices are in use. These active references prevent DR operations from detaching these devices.

Workaround: Remove or rename the /rplboot directory, then shut down and restart NFS with these commands:

# sh /etc/init.d/nfs.server stop

# sh /etc/init.d/nfs.server start

StorEdge A3000 Daemon Cannot Be Suspended (4100212)

The presence of StorEdge A3000 daemons (rdaemon) prevent a DR Detach operation from completing if the operating environment must be quiesced (that is, suspended and resumed) to detach non-pageable memory.

Workaround: Temporarily disable the daemon before the DR operation, then restart the daemon after the DR operation by using the following command:

# sh /etc/rc2.d/S90amon stop

# sh /etc/rc2.d/S90amon start

Unable to Drain Sysboards in Solaris 7 and dr_daemon Segmentation Faults (4229238)

After numerous iterations through a failure handling condition within the DR daemon, the daemon can encounter a segmentation violation (that is, a core dump). Note that this condition occurs during an error handling flow (that is, a DR operation failure).

Workaround: Restart the DR applications on the SSP.

Bugs Fixed

The following important bugs have been fixed since the last release of DR. This list does not include minor bugs, nor does it include those bugs that were fixed in patches.

4083412 - socal and sf need to support DETACH/SUSPEND/RESUME

4090750 - DR panics during detach when dr-max-mem set is high.

4067860 - SUNWcsr on upgrading from Solaris 2.5.1 8/97 to Solaris 2.6 5/98 creates duplicate majors in name_to_major.

4131197 - dr_daemon core dumps with Solaris 2.6 5/98 release.

4133607 - DR operations with mem_board_interleave_ok cause problems.

4134357 - availrmem is not being reduced during Sun Enterprise 10000 memory detach.

4135366 - Sun StorEdge A3000 patch has been changed.

4136544 - Getting "flusher thread" hang during DR drain with official -20 KU patch.