Power6 High-End System Firmware

02/21/11

• HIPER: IBM testing has uncovered a potential undetected data corruption issue when a mobility operation is performed on an AMS (Active Memory Sharing) partition.  The data corruption can occur in rare instances due to a problem in IBM firmware.  This issue was discovered during internal IBM testing, and has not been reported on any customer system. IBM recommends that systems running on EH340_075 or later move to EH350_103 to pick up the fix for this potential problem.  (Firmware levels older than EH340_075 are not exposed to the problem.)
• On systems with a F/C 5803 or 5873 I/O drawer attached, a problem was fixed that caused a partition to crash during a page migration operation.
  
• A problem was fixed that caused a partition to crash with SRC BA330002 after several concurrent installations of system firmware, or partition migrations, without a reboot.
• A problem was fixed that caused AIX licensing issues when migrating a partition from a P6 to a P7 system.
• On systems running IBM i partitions, a problem was fixed that caused SRC BA040030 to be erroneously logged, and a call home to be made, even though the partition booted successfully.
  

10/26/10

System firmware changes that affect all systems

• HIPER:  A problem was fixed that caused the HMC to show the server's status as incomplete, and SRC B7000602 to be logged against SFLPHMCCMDTASK in serviceable events.  This problem can also cause the system to crash when it occurs.
• HIPER:  A problem was fixed that caused repeated reset/reloads of the service processor, and fail-overs, to occur after a hypervisor-initiated reset/reload of the service processor was completed.  That led to loss of communication between the service processor and the hypervisor (indicated by SRC B182951C).
  
• A problem was fixed that caused a CEC node to unexpectedly lose power, and caused a system crash.  Systems that had a service processor role change at server power off are exposed to this issue.
• A problem was fixed that caused SRC B181440D to be erroneously logged.
• The firmware was enhanced to log SRC B181D30B as informational instead of predictive.
  
• The firmware was enhanced to list the attached devices when viewing the adapter information for a partition profile on the HMC GUI.
  
• A problem was fixed that caused the hypervisor to issue almost continuous reset/reload requests to the service processor.

• The firmware was enhanced to support the network installation of the IBM i operating system from the hardware management console (HMC) command line interface (CLI).
• On systems using the IPv6 protocol, a problem was fixed that caused valid link local and unique link local addresses to be erroneously invalidated.  This prevented the port with that address from being used for network boot or network installation.
  

06/30/10

System firmware changes that affect all systems

• DEFERRED: A problem was fixed that could result in a system checkstop while running floating point computations.  Although this is a high-impact problem, it has a very low probability of occurring.
  
• A problem was fixed that caused a call home to be erroneously made with SRC B181E911, and a service processor dump to be taken unnecessarily.
• A problem was fixed that caused the HMC to show a status of "Incomplete" for the managed system, and numerous service processor dumps to be generated.
• The firmware was enhanced to improve the callouts for certain types of processor failures that log SRC B1xxE504.
• The firmware was enhanced to make a hidden error log visible when a failing GX adapter caused errors to be logged by the processor run-time (PRDF) diagnostics.
• The firmware was enhanced to improve the callouts when NVRAM corruption is detected in the bulk power controller's (BPC's) service processor.
• On systems running EH350_xxx firmware, a problem was fixed the prevented the reset/reload bit from being set correctly in a service processor error log entry.
• A problem was fixed that caused SRC B181E617 to be erroneously logged and a service processor dump to be unnecessarily generated.

•  On systems running the IBM i operating system, a problem was fixed that caused a DLPAR move operation with an IOP (I/O processor) and IOA (I/O adapter) to fail intermittently.  The DLPAR operation was successful, but the IOA failed to power on in the new partition.

• A problem was fixed that would cause a concurrent maintenance operation to fail if the HMC was rebooted before the previous CM operation was complete.
• On systems with F/C 5803 or F/C 5873 I/O drawers attached and a boot device in the drawer, a problem was fixed that prevented a partition from booting after the concurrent repair of the GX adapter that connects the 5803 or 5873 drawer to the system, or to the node that contains the GX adapter.
  

03/10/10

System firmware changes that affect all systems

• HIPER: A problem was fixed that caused the system to crash if the server was running AIX and had a  F/C 5802 or 5877 drawer (in a 19" rack), or F/C 5803 or 5873 drawer (in a 24" rack), attached.
• DEFERRED:  This fix corrects the handling of a specific processor instruction sequence that has the potential to result in undetected data errors.  This specific instruction sequence has only been observed in a small number of highly tuned Floating Point intensive applications.  However, it is strongly recommended that this fix be applied to all POWER6 systems.  This fix has the potential to decrease system performance on applications that make extensive use of floating point divide, square root, or estimate instructions.
  
• A problem was fixed that prevented an SRC from being recorded in the service processor dump produced by a host-initiated reset.
• A problem was fixed that prevented the repair of a deconfigured system controller from being completed successfully.
• A problem was fixed that caused SRC 10009135, followed by 10009139, to be erroneously logged.  These SRCs indicate a system power control network (SPCN) loop is being broken, then re-established.
• The firmware was enhanced to allow a temporary threshold reduction for processor unit book interconnect predictive errors.
• A problem was fixed that caused a reset/reload of a network controller.
• A problem was fixed that, under certain rare circumstances, caused a partition to hang when being shut down.
• A problem was fixed that caused the system to hang with SRCs B182953C, B182954C and B17BE434 being logged.
• The firmware was enhanced to detect and handle 12X InfiniBand I/O drawer cabling errors better.
• A problem was fixed that, under certain rare circumstances, caused the system to become unresponsive and appear to hang  when page migration occurred on a PCIe slot.

• A problem was fixed that caused a virtual SCSI or virtual fibre channel adapter to be seen by the operating system as not bootable when it was added to a partition using a dynamic LPAR (DLPAR) operation.
• On systems running IBM i, a problem was fixed that caused booting the operating system from a fibre channel device to fail with SRC 576B8301.
• On systems with a F/C 5802 or 5877 drawer attached, a problem was fixed that could impact the performance of a 4-port Ethernet adapter F/C 5272, 5275, 5279, 5280, 5525, 5526, or 5527 installed in that drawer.
• In partitions running AIX or Linux, a problem was fixed that caused the addition of an I/O slot to a partition using a dynamic LPAR (DLPAR) add operation to fail.
• On systems with shared processors, a problem was fixed that caused the partitions to hang and become unresponsive for very short periods of time.
• A problem was fixed that prevented the IPv6 DHCP address from being displayed on the advanced system management interface (ASMI) network configuration screens when IPv6 and DHCP were enabled.  This only occurred on systems with virtual LAN (VLAN) addresses (such as eth0.30, eth0.31), and when IPv6 addresses were assigned to the eth0.xx interface.
• On systems running redundant VIOS partitions, a problem was fixed that prevented Ethernet traffic from being properly bridged between the two partitions.  This problem also prevented shared Ethernet adapter failover from working correctly.
• A problem was fixed that caused the hypervisor to loop unnecessarily and consume too many processor cycles.  This impacted the performance of the system.
  

• The firmware was enhanced such that if an Ethernet cable is misplugged on a node controller during a concurrent node add operation, the node add operation will be completed successfully.
• A problem was fixed that prevented the concurrent repair of a redundant system controller.
• A problem was fixed that caused unpredictable system behavior if a capacity on demand (CoD) or a virtualization engine technology (VET) activation code was entered and accepted after a node 0 evacuation was done.  The unpredictable machine behavior might also have occurred, if a node 0 evacuation failed, a system dump was taken, and a memory-preserving IPL was then initiated.
• A problem was fixed that caused a concurrent maintenance operation after a node evacuation to fail.  When this problem occurred, the system erroneously states that a platform memory dump is pending.
• A problem was fixed that prevented a concurrent maintenance operation from completing successfully.

10/30/09

New Features and Functions:

• Support for the concurrent repair of a system controller.
• Support for the concurrent removal of 12X-attached 24" I/O drawers.
• Support for a USB-attached half-high 5.25" backup device using a removable hard disk drive (HDD).
• Support for a platform dump that is not disruptive.
• Support for i5/OS multipath storage I/O through VIOS partitions.

• A problem was fixed that might cause a concurrent firmware maintenance (CFM) operation to fail repeatedly, or a concurrent maintenance (CM) operation to fail repeatedly, when a large number of I/O loop errors were being logged during the CFM operation.
• The firmware was enhanced to handle system dumps (SYSDUMPs) larger than 4GB in size.
• On systems running system firmware release EH340 EH340, a problem was fixed that caused a dynamic LPAR (DLPAR) operation on memory to fail until the platform was rebooted.
• The firmware was enhanced to improve the performance of the F/C 5732 , 5735, and 5769 PCI-E adapters. The firmware was enhanced such that SRCs B181F126, B181F127, and B181F129 are correctly logged, and no longer calls home unnecessarily for these SRCs.
• A problem was fixed that caused a repair and verify (R&V) operation on the hardware management console (HMC) to fail with the message "Exception encountered while rendering panel as HTML".
• The firmware was enhanced such that a generic B1817201 SRC will no longer be logged when a cache error occurs on a node controller (NC).  Unique SRCs will now be logged for cache failures, and upper and lower thresholds have been added to the NC cache error logging scheme.
• The firmware was enhanced to improve the field replaceable unit (FRU) callouts for SRCs B1xxC004 and B1xxC005.
• A problem was fixed that might cause the system to crash with SRC B181E504, then SRC B1813909, being logged.
• The firmware was enhanced to more accurately describe the reason memory was deconfigured on the advanced system management interface (ASMI) memory deconfiguration screen.
• The firmware was enhanced such that when a certain type of hardware failure occurs in a bulk power controller (BPC), the appropriate errors will be logged instead of SRCs B1818601 and B1818611, which indicate a firmware failure.
• On systems using the HEA (host Ethernet adapter), also know as the Integrated Virtual Ethernet (IVE) function, a problem was fixed that caused link failures if the HEA was connected to certain third-party Ethernet switches.  A problem causing an unexpected increment in the Pxs_TXIME register, but not affecting network performance, was also fixed.
  

• A problem was fixed that caused SRC B181A494 to be erroneously logged if a concurrent maintenance operation took longer than 60 minutes.
• On systems with 24" I/O drawers, a problem was fixed that might cause a partition to crash, with a system reboot required for recovery, when a F/C 5797 or 5798 drawer was concurrently added.
• On systems with four drawers, a problem was fixed that caused the system controller to perform a reset/reload, which caused a concurrent maintenance operation to fail, on the fourth node (P4).
• A problem was fixed that caused the current replacement of an InfiniBand GX adapter or I/O planar to fail if a partition owned an embedded device on the planar.