Power8 System Firmware

System firmware changes that affect certain systems

• HIPER/NON-PERVASIVE: DEFERRED:  On systems with DDR4 memory installed, a problem was fixed for the handling of data errors in the L4 cache.   If a data error occurs in the L4 cache of the memory buffer on an affected system and it is pushed out to mainline memory, the data error will not be correctly handled.   A data error originating in the L4 cache may result in incorrect data being stored into memory.  The DDR4 DRAM has feature code (FC) EM8S for a 128GB 1600 MHz CDIMM.
  IBM strongly recommends that the customer should plan an outage to install the firmware fix immediately.  Fix activation requires a subsequent platform IPL following the installation of the firmware fix to eliminate any exposure to this issue.
  This problem only exists on the 8408-E8E systems with the DDR4 DRAM memory feature.
  

New features and functions

• Support was added to Live Partition Mobility to allow migrations between partitions at firmware level FW760 and FW840.22 or later.  Previously, migration operations were not allowed between FW760 and FW840 partitions.
• Support for the CoD on the 226W 4.323 GHz eight core processor (CCIN 54E5, F/C EPXF) for the EasyScale offering of the S822 (8284-22A).  This includes Processor Capacity on Demand (CoD) with Elastic (On/Off) Processor CoD and Trial Processor CoD.  Previously, the CoD support for the EasyScale S822 was only available when using  the ten core 3.42 GHz processor (CCIN 54E8, F/C EPXD).

System firmware changes that affect certain systems

• On systems using PowerVM firmware, a problem was fixed for a sequence of two or more Live Partition Mobility migrations that caused a partition to crash with a SRC BA330000 logged (Memory allocation error in partition firmware).  The sequence of LPM migrations that can trigger the partition crash are as follows:
  The original source partition level can be any FW760.xx, FW763.xx, FW770.xx, FW773.xx, FW780.xx, or FW783.xx P7 level or any FW810.xx, FW820.xx, FW830.xx, or FW840.xx P8 level.  It is migrated first to a system running one of the following levels:
  1) FW730.70 or later 730 firmware or
  2) FW740.60 or later 740 firmware
  And then a second migration is needed to a system running one of the following levels:
  1) FW760.00 - FW760.20 or
  2) FW770.00 - FW770.10
  The twice-migrated system partition is now susceptible to the BA330000 partition crash during normal operations until the partition is rebooted.  If an additional LPM migration is done to any firmware level, the thrice-migrated partition is also susceptible to the partition crash until it is rebooted.
  With the fix applied, the susceptible partitions may still log multiple BA330000 errors but there will be no partition crash.  A reboot of the partition will stop the logging of the BA330000 SRC.
  

New features and functions

• Support for a 128GB DDR4 memory DIMM for the E850 (8408-E8E) model .  Memory feature code #EM8S provides the 128GB CDIMM (1600 MHz, 4GBIT DDR4).   Note that DDR4 and DDR3 DIMMs cannot be mixed in the system.
• Support was added for the Stevens6+ option of the internal tray loading DVD-ROM drive with F/C #EU13.  This is an 8X/24X(max) Slimline SATA DVD-ROM Drive.  The Stevens6+ option is a FRU hardware replacement for the Stevens3+.  MTM 7226-1U3 (Oliver)  FC 5757/5762/5763 attaches to IBM Power Systems and lists Stevens6+ as optional for Stevens3+.  If the Stevens6+  DVD drive is installed on the system without the required firmware support, the boot of an AIX partition will fail when the DVD is used as the load source.  Also, an IBM i partition cannot consistently boot from the DVD drive using D-mode IPL.  A SRC C2004130 may be logged for the load source not found error.
• Support for the IBM PCIe3 12GB cache RAID plus SAS dual 4-port 6Gb x8 adapter with feature code #EJ14 and CCIN 57B1.  This adapter is very similar to the #EJ0L SAS adapter, but it uses a second chip in the card to provide more IOPS capacity (significant performance improvement) and can attach more SSD.  This adapter uses integrated flash memory to provide protection of the write cache, without need for batteries, in case of power failure.
• Support for PowerVM vNIC extended to Linux OS Ubuntu 16.04 LE with up to ten vNIC client adapters for each partition.  PowerVM vNIC combines many of the best features of SR-IOV and PowerVM SEA to provide a network solution with options for advanced functions such as Live Partition Mobility along with better performance and I/O efficiency when compared to PowerVM SEA.  In addition PowerVM vNIC provides users with bandwidth control (QoS) capability by leveraging SR-IOV logical ports as the physical interface to the network.
• PowerVM CoD was enhanced to eliminate the yearly Utility CoD renewal on systems using Utility COD.  The Utility CoD usage is already monitoring to make sure systems are running within the prescribed threshold limit of unreported usage, so a yearly customer renewal is not needed to manage the Utility CoD processor usage,
• Support was added to the DHCP client on the service processor for non-random backoff mode needed for Data Center Manageability Interface (DCMI) V1.5  compliance.  By default, the DHCP client does random backoff delays for retries during DHCP discovery.  For DCMI V1.5, non-random backoff delays were introduced as an option.  Disabling the random back-off mode is not required for normal operations, but if wanted, the system administrator can override the default and disable the random back-off mode by sending the “SET DCMI Configuration Parameters” for the random back-off property of the Discovery Configuration parameter.  A value of "0" for the bit means "Disabled".  Or, the DHCP configuration file can be modified to add "random-backoff off", causing the non-random mode for the retry delays to be used during DHCP discovery.
• Support was added for enhanced diagnostics for PowerVM Simplified Remote Restart (SRR) partitions.   This service pack level is recommended when using SRR partitions.  You can learn more about SSR partitions at the IBM Knowledge Center: " http://www.ibm.com/support/knowledgecenter/HW4P4/p8hat/p8hat_createremotereslpar.htm".
• Support was added for auto-correction in the Advanced System Manager Interface (ASMI) for the "Feature Code/Sequence Number" field of the "System Configuration/Program Vital Product Data/System Enclosures" menu selection.  Lower case letters are invalid in the "Feature Code/Sequence Number" field so these are now changed to upper case letters to help form a valid entry.  For example, if  "78c9-001" was entered, it would be changed to "78C9-001".
• Support was added for HTTP Strict Transport Security (HSTS) compliance for The Advanced System Management Interface (ASMI) web connection.  Even without this feature, any attempt to access ASMI with the HTTP protocol was rejected because the service processor firewall blocks port 80 (HTTP).  But enabling HSTS for ASMI prevents HSTS security warnings for the service processor during network scans by security scanner programs such as IBM AppScan.
  

System firmware changes that affect all systems

• DEFERRED:  A problem was fixed in the dynamic ram (DRAM) initialization to update the VREF on the dimms to the optimal settings and to add an additional margin check test to improve the reliability of the DRAM by screening out more marginal dimms before they can result in a run-time memory fault.
  
• A problem was fixed for a degraded PCI link causing a processor core to be guarded if a non-cacheable unit (NCU) store time-out occurred with SRC B113E540 and PRD signature  "(NCUFIR[9]) STORE_TIMEOUT: Store timed out on PB".  With the fix, the processor core is not guarded because of the NCU error.  If this problem occurs and a core is deconfigured. clear the guard record and re-IPL to regain the processor core.  The solution for degraded PCI links is different from the fix for this problem, but a re-IPL of the CEC or a reset of the PCI adapters could help to recover the PCI links from their degraded mode.
• A problem was fixed for an incorrect reduction in FRU callouts for Processor Run-time Diagnostic (PRD) errors after a reference oscillator clock (OSCC) error has been logged.  Hardware resources are not called out and guarded as expected.  Some of the missing PRD data can be found in the secondary SRC of B181BAF5 logged by hardware services.  The callouts that PRD would have made are in the user data of that error log.
• A problem was fixed for a Qualys network scan for security vulnerabilities causing a core dump in the Intelligent Platform Management Interface (IPMI)  process on the service processor with SRC B181EF88.  The error occurs anytime the Qualys scan is run because it sends an invalid IPMI session id that should have been handled and discarded without a core dump.
• A security problem was fixed in OpenSSL for a possible service processor reset on a null pointer de-reference during RSA PPS signature verification. The Common Vulnerabilities and Exposures issue number is CVE-2015-3194.
• A security problem was fixed in the lighttpd server on the service processor, where a remote attacker, while attempting authentication, could insert strings into the lighttpd server log file.  Under normal operations on the service processor, this does not impact anything because the log is disabled by default.  The Common Vulnerabilities and Exposures issue number is CVE-2015-3200.
  
• A problem was fixed for the service processor going to the reset state instead of the termination state when the anchor card is missing or broken.  At the termination state, the Advanced System Manager Interface (ASMI) can be used to collect failure data and debug the problem with the anchor card.
• A problem was fixed for error log entries created by Hostboot not getting written to the error log in some situations.  This can cause hardware detected as failed by Hostboot to not get reported or have a call-home generated.  This problem will occur whenever Hostboot commits a recovered or informational error as its last error log in the current IPL.  In the next IPL,  one or more error logs from Hostboot will be lost.
• A problem was fixed for a service processor failure during a system power off that causes a reset of the service processor.  The service processor is in the correct state for a normal system power on after the error.  The frequency for this error should be low as it is caused by a very rare race condition in the power off process.
• A problem was fixed so that service processor NVRAM bit flips are now detected and reported as predictive errors after a certain threshold of failures have occurred.  The SRCs reported are B151F109 (threshold of NVRAM errors was reached) or B151F10A (a NVRAM address has failed multiple times).  Previously, these normal wear errors in the NVRAM were ignored.  The bit flip is self-corrected and does not cause a problem but a high occurrence of these could mean that a service processor card FRU or system backplane FRU, as called out in the SRC, is in need of service. 
  
• A security problem was fixed in OpenSSL for a possible service processor reset on a null pointer de-reference during SSL certificate management. The Common Vulnerabilities and Exposures issue number is CVE-2016-0797.
  

System firmware changes that affect certain systems

• DEFERRED:  On systems using PowerVM firmware, a performance improvement was made by disabling the Hot/Cold Affinity (HCA) hardware feature, which gathers memory usage statistics for consumption by partition operating system memory management algorithms.  The statistics gathering can, in rare cases, cause performance to degrade.  The workloads that may experience issues are memory-intensive workloads that have little locality of reference and thus cannot take advantage of hardware memory cache.  As a consequence, the problem occurs very infrequently or not at all except for very specific workloads in a HPC environment.  This performance fix requires an IPL of the system to activate it after it is applied.
• On systems using PowerVM firmware and NovaLink co-management of the partitions, a problem was fixed with the Hardware Management Console (HMC) not showing the co-management master name with the HMC lscomgmt command.  The command displayed blank text for the master owner when NovaLink established the master mode.  This problem occurred whenever Novalink powered on and took the master mode that had been released by the HMC.
• On systems using OPAL firmware, a problem was fixed for Enhanced Error Handling (EEH) recoverable errors on network adapters behind a PLX switch having the backplane called out by OPAL instead of the adapter slot.
• On systems with a PowerVM Active Memory Sharing (AMS) partition with AIX  Level 7.2.0.0 or later with Firmware Assisted Dump enabled, a problem was fixed for a Restart Dump operation failing into KDB mode.  If "q" is entered to exit from KDB mode, the partition fails to start.  The AIX partition must be powered off and back on to recover.  The problem can be circumvented by disabling Firmware Assisted Dump (default is enabled in AIX 7.2).
• On a PowerVM system, a problem was fixed for an incorrect date in partitions created with a Simplified Remote Restart-Capable (SRR) attribute where the date is created as Epoch 01/01/1970 (MM/DD/YYYY).  Without the fix, the user must change the partition time of day when starting the partition for the first time to make it correct.  This problem only occurs with SRR partitions.
• On a PowerVM system with licensed Power Integrated Facility for Linux (IFL) processors, a problem was fixed for a system hang that could occur if the system contains both 1) dedicated processor partitions configured to share processors while active and  2) shared processor partitions.  This problem is more likely to occur on a system with a low number of non-IFL processors.
  
• On systems using PowerVM firmware with dedicated processor partitions,  a problem was fixed for the dedicated processor partition becoming intermittently unresponsive. The problem can be circumvented by changing the partition to use shared processors.  This is a follow-on to the fix provided in 840.11 for a different issue for delays in dedicated processor partitions that were caused by low I/O utilization.
• A problem was fixed for transmit time-outs on a Virtual Function (VF) during stressful network traffic, on systems using PCIe adapters in Single Root I/O Virtualization (SR-IOV) shared-mode.  This fix updates adapter firmware to 10.2.252.1918, for the following Feature Codes: EN15, EN16, EN17, EN18, EN0H, EN0J, EL38, EN0M, EN0N, EN0K, EN0L, and EL3C.
  The SR-IOV adapter firmware level update for the shared-mode adapters happens under user control to prevent unexpected temporary outages on the adapters.  A system reboot will update all SR-IOV shared-mode adapters with the new firmware level.  In addition, when an adapter is first set to SR-IOV shared mode, the adapter firmware is updated to the latest level available with the system firmware (and it is also updated automatically during maintenance operations, such as when the adapter is stopped or replaced).  And lastly, selective manual updates of the SR-IOV adapters can be performed using the Hardware Management Console (HMC).  To selectively update the adapter firmware, follow the steps given at the IBM Knowledge Center for using HMC to make the updates:   https://www.ibm.com/support/knowledgecenter/HW4M4/p8efd/p8efd_updating_sriov_firmware.htm.
  Note:  Adapters that are capable of running in SR-IOV mode, but are currently running in dedicated mode and assigned to a partition, can only be updated concurrently by the OS that owns the adapter.
  
• On systems using OPAL firmware, a problem was fixed in the PCI Host Bridge (PHB) to prevent adapter interrupts from being lost when two interrupts come in at the same time.  The lost interrupts could result in a slow down for the workload using the affected adapter.  This fixes a problem seen with some CAPI workloads that have lots of interrupt masking at the same time as a high interrupt load.  However, the fix is not specific to the CAPI adapters.
• On systems using OPAL firmware, a problem was fixed for an extraneous SRC BB822411 being logged during service processor termination occurrences.  This SRC is unrelated to the root cause of the termination and should be ignored.
• On systems using OPAL firmware, a problem was fixed for a incomplete reporting of a Hypervisor Maintenance Interrupt (HMI) to the host Linux OS.  The fix ensures the CPU Processor Identification Register (PIR) is reported correctly instead of having an all zero value.  HMIs are caused by hardware failures occurring in the SLW (sleep winkle image) for processors or in CAPI  (Coherent Accelerator Processor Interface) adapters.  These cause the hypervisor to investigate the cause of the error by reading SCOM registers to isolate the fault and send a HMI.
• On systems using OPAL firmware, support was added to allow the Linux OS to send alphanumeric strings to the operations panel.  The OS program must use a device driver for /dev/oppanel.  The driver implements a 32 character buffer which a user can read/write by accessing the device (/dev/oppanel). This buffer is then displayed on the operator panel display.
• A problem was fixed for the Advanced System Management Interface (ASMI) "System Service Aids/Call-home Setup" menu not being able to clear the old Service center phone numbers.  The blank or null characters are now accepted and can be used to overlay the existing values.  Without the fix, the characters input to clear the phone number field are rejected and replaced with the old values.  The ASMI Call-home option is not available for systems that are managed by the Hardware Management Console (HMC).
• On PowerVM systems using Elastic Capacity on Demand (CoD) (also known as On/Off CoD), a problem was fixed for losing entitlement amounts when upgrading from FW820 or FW830.  If you upgrade to a service pack level that does not have this fix and lose the entitlement, you can get another On/Off (Elastic) CoD Enablement code from IBM Support.  This problem only pertains to the E850 (8408-E8E), E870 (9119-MME), and E880 (9119-MHE) models.
• On IBM Power System S822 (8284-22A) using PowerVM for IBM i partitions, a problem was fixed for the User-based pricing indicator being off.  This was changed to be on.  The IBM i Licensing fees involves a distinction between User-based and non-User-based pricing.  The model S822, for PurePower (IBM i) now shows User-based pricing as required.
  

New features and functions

• Support for PowerVM co-management mode on the Hardware Management Console (HMC). This feature allows the HMC and PowerVM NovaLink to both have a live management connection to the system.  This is different than the traditional dual-HMC model however, and results in some behavior changes in the HMC.  For hardware and service management functions, the HMC works as it does when not in co-management mode.  However, when in co-management mode, only the PowerVM Co-Management Master can make changes to the PowerVM configuration and change the state of the system.  Power System firmware updates must be done using the HMC,  with the HMC as the Co-Management Master.  All management entities (HMC(s) and NovaLink) have read-access to the partition configuration regardless of whether they are the designated master.  Typically NovaLink will be the co-management master, however if a virtualization task or a firmware update is needed,  one can explicitly request master authority for the HMC, perform the action, and then relinquish the authority back to NovaLink.  The minimum firmware and HMC levels for this feature are FW840.11 and HMC V8R8.4.0.1.  If using PowerVC with NovaLink co-management, the minimum level is PowerVC 1.3.0.2.  Please refer to IBM KnowledgeCenter link "http://ibm.biz/novalink-kc" for more information on the PowerVM NovaLink feature and changing the master authority when doing co-management.
  Note:  If a firmware update is attempted from a co-managing HMC that is not in the master role, the update operation will fail with the following message: "Could not start the update because this management console is not the master console.  Check to see if there is another management console program is attached to the target server {0} (HSCF0261E)" along with HMC SRC E302FB11.
• The default setting for the "Enlarged I/O Memory Capacity" feature was disabled on newly manufactured E850, E870 & E880 models to reduce hypervisor memory usage.  Customers of the new systems using PCI adapters that leverage "Enlarged I/O Memory Capacity" will need to explicitly enable this feature for the supported PCI slots, using ASMI Menus while the system is powered off.  Existing systems will not see a change in their current setting.  For existing systems with only AIX and IBM i partitions that do not benefit from this feature, it can be disabled by using the Advanced System Management Interface (ASMI) for the "System Configuration-> I/O Adapter Enlarged Capacity" panel to uncheck the option for the "I/O Adapter Enlarged Adapter Capacity" feature.
  

System firmware changes that affect certain systems

• On systems using PowerVM partitions, a problem was fixed for error recovery from failed Live Partition Mobility (LPM) migrations.  The recovery error is caused by a partition reset that leaves the partition in an unclean state with the following consequences:  1) A retry on the migration for the failed source partition may not not be allowed; and 2) With enough failed migration recovery errors, it is possible that any new migration attempts for any partition will be denied.  This error condition can be cleared by a re-IPL of the system. The partition recovery error after a failed migration  is much more likely to occur for partitions managed by NovaLink but it is still possible to occur for Hardware Management Console (HMC) managed partitions.
  

New features and functions

• Support was added to block a full Hardware Management Console (HMC) connection to the service processor when the HMC is at a lower firmware major and minor release level than the service processor.  In the past, this check was done only for the major version of the firmware release but it now has been extended to the minor release version level as well.  The HMC at the lower firmware level can still make a limited connection to the higher firmware level service processor.  This will put the CEC in a "Version Mismatch" state.  Firmware updates are allowed with the CEC in the "Version Mismatch" state so that the condition can be corrected with either a HMC update or a firmware update of the CEC.
• Support for Processor Capacity on Demand (CoD) for the IBM Power System S822 (8284-22A) that includes Elastic (On/Off) Processor CoD  and Trial Processor CoD.
• Support was removed in the Advanced Systems Management Interface (ASMI) and IPMI for allowing the IBM Power System S822 (8284-22A) to change between OPAL and PowerVM hypervisor modes.  The default for new 8284-22A systems is PowerVM mode and it cannot be changed to OPAL. For existing customers with 8284-22A systems, both hypervisor modes (PowerVM & OPAL) are still available after the firmware is upgraded to 840.10, so they are not affected by the change.
  
• Support was added for a 4-Core 3.02 GHz POWER8 Processor Card with CCIN 54E9 and feature code #EPXK for the S822 (8284-22A), S812L(8247-21L),  and S822L (8247-22L) models.
• Support for PowerVM vNIC with more vNIC client adapters for each partition, up to 10 from a limit of 6 at the FW840.00 level.  PowerVM vNIC combines many of the best features of SR-IOV and PowerVM SEA to provide a network solution with options for advanced functions such as Live Partition Mobility along with better performance and I/O efficiency when compared to PowerVM SEA.  In addition PowerVM vNIC provides users with bandwidth control (QoS) capability by leveraging SR-IOV logical ports as the physical interface to the network.
• Support for the IBM Power System E850 (8408-E8E) with AIX and Linux partitions.
• The default setting for the "Enlarged I/O Memory Capacity" feature was disabled on newly manufactured E850, E870 & E880 models to reduce hypervisor memory usage.  Customers using PCI adapters that leverage "Enlarged I/O Memory Capacity" will need to explicitly enable this feature for the supported PCI slots, using ASMI Menus while the system is powered off.
  

System firmware changes that affect all systems

• A problem was fixed for false errors logs for SRC B181A40F where upper domain fans are incorrectly reported as missing on a reboot of the service processor.  This problem only pertains to the IBM Power System E850 (8408-E8E).
• A problem was fixed for not being able to control all I/O slots for Huge Dynamic DMA Window (HDDW) capability on the IBM Power System E850 (8408-E8E).  There are 13 I/O slots enabled for HDDW on this system but only 8 could be controlled by the Advanced System Management Interface (ASMI)  panel for "I/O Enlarged Capacity".  This prevented enabling all slots to be HDDW enabled, limiting DMA bandwidth on some of the I/O slots.
• A problem was fixed for a system IPL hang at C100C1B0 with SRC 1100D001 when the power supplies have failed to supply the necessary 12-volt output for the system.   The 1100D001 SRC was calling out the planar when it should have called out the power supplies.  With the fix, the system will terminate as needed and call out the power supply for replacement.  One mode of power supply failure that could trigger the hang is sync-FET failures that disrupt the 12-volt output.
• A problem was fixed for a PCIe3 I/O expansion drawer (#EMX0) not getting all error logs reported when its error log queue is full.  In the case where the error log queue is full with 16 entries, only one entry is returned to the hypervisor for reporting.  This error log truncation only occurs during periods of high error activity in the expansion drawer.
• A problem was fixed for the callout of a VPD collection fault and system termination with SRC 11008402 to include the 1.2vcs VRM FRU.  The power good fault fault for the 1.2 volts would be a primary cause of this error.  Without the fix, the VRM is missing in the callout list and only has the VPDPART isolation procedure.
• A problem was fixed for excessive logging of the SRC 11002610 on a power good (pgood) fault when detected by the Digital Power Subsystem Sweep (DPSS).  Multiple pgood interrupts are signaled by the DPSS in the interval between the first pgood failure and the node power down.  A threshold was added to limit the number of error logs for the condition.
• A problem was fixed to speed recovery for VPD collection time-out errors for PCIe resources in an I/O drawer logged with SRC 10009133 during concurrent firmware updates.  With the fix, the hypervisor is notified as soon as the VPD collection has finished so the PCIe resources can report as available .  Without the fix, there is a delay as long as two hours for the recovery to complete.
• A problem was fixed to allow IPMI entity IDs to be used in ipmitool raw commands on the service processor to get the temperature reading.  Without the fix, the DCMI entity IDs have to be used in the raw command for the "Get temperature" function.
• A problem was fixed for a false unrecoverable error (UE) logged for B1822713 when an invalid cooling zone is found during the adjustment of the system fan speeds.  This error can be ignored as it does not represent a problem with the fans.
• A problem was fixed for loss of back-level protection during firmware updates if an anchor card has been replaced.  The Power system manufacturing process sets the minimum code level a system is allowed to have for proper operation.  If a anchor card is replaced, it is possible that the replacement anchor card is one that has the Minimum MIF Level (MinMifLevel) given as "blank",  and this removes the system back-level protection. With the fix, blanks or nulls on the anchor card for this field are handled correctly to preserve the back-level protection.  Systems that have already lost the back-level protection due to anchor card replacement remain vulnerable to a accidental downgrade of code level by operator error, so code updates to a lower level for these systems should only be performed under guidance from IBM Support.  The following command can be run the Advanced Management Management Interface (ASMI) to determine if the system has lost the back-level protection with the presence of "blanks" or ASCII 20 values for MinMifLevel:
  "registry -l cupd/MinMifLevel" with output:
  "cupd/MinMifLevel:
  2020202020202020 2020202020202020 [ ]
  2020202020202020 2020202020202020 [ ]"
• A problem was fixed for a code update error from FW830 to a FW840 level causes temperature sensors to be lost so that the ipmitool command to list the temperature sensors fails with a IPMI program core dump.  If the temperature sensors are already corrupted due to a preceding code update, this fix adds back in the temperature sensors to allow the ipmitool to work for listing the temperature sensors.
• A problem was fixed for a system checkstop caused by a L2 cache least-recently used (LRU) error that should have been a recoverable error for the processor and the cache.  The cache error should not have caused a L2 HW CTL error checkstop.
• A problem was fixed for a re-IPL with power on failure with B181A40F SRC logged for VPD not found for a DIMM FRU.  The DIMM had been moved to another slot or just removed.  In this situation, a IPL of the system from power off will work without errors, but a re-IPL with power on,  such as that done after processing a hardware dump, will fail with the B181A40F.  Power off the system and IPL to recover.  Until the fix is applied, the problem can be circumvented after a DIMM memory move by putting the PNOR flash memory in genesis mode by running the following commands in ASMI with the CEC powered off:
          1) hwsvPnorCmd -c
          2) hwsvPnorCmd -g
  
• A problem was fixed for the service processor becoming inaccessible when having a dynamic IP address and being in DCMI "non-random" mode for DHCP discovery by customer configuration.  The problem can occur intermittently during a AC power on of the system.  If the service processor does not respond on the network, AC power cycle to recover.  Without the fix, the problem can be circumvented by using the DHCP client in the DCMI "random" mode for DHCP discovery, which is the default on the service processor.
• A problem was fixed for a memory initialization error reported with SRC BC8A0506 that terminates the IPL.  This problem is unlikely to occur because it depends on a specific memory location being used by the code load. The system can be recovered from the error by doing another IPL.
  

System firmware changes that affect certain systems

• On PowerVM systems a problem was fixed to address a performance degradation. The problem surfaces under the following conditions:
  1)    There is at least one VIOS or Linux partition that is running with dedicated processors AND
  2)    There is at least one VIOS or Linux partition running with shared processors AND
  3)    There is at least one AIX or IBMi partitions configured with shared processors. 
  If ALL the above conditions are met AND one of the following actions occur,
  1)    VIOS/Linux dedicated processor partition is configured to share processors while active OR
  2)    A dynamic platform optimization operation (HMC 'optmem' command) is performed OR
  3)    Processors are unlicensed via a capacity on demand operation
  there is an exposure for a loss in performance.
  
• On systems using PowerVM firmware, a problem was fixed for PCIe switch recovery to prevent a partition switch failure during the IPL with error logs for SRC B7006A22 and B7006971  reported.  This problem can occur when doing recovery for an informational error on the switch.  If this problem occurs, the partition must be restarted to recover the affected I/O adapters.
• On systems using PowerVM firmware, a problem was fixed for a concurrent FRU exchange of a CAPI  (Coherent Accelerator Processor Interface) adapter for a standard I/O adapter that results in a vary off failure.  If this failure occurs, the system needs to be re-IPLed to fix the adapter.  The trigger for this failure is a dual exchange where the CAPI adapter is exchanged first for a standard (non-like-typed) adapter.  Then an attempt is made to exchange the standard adapter for a CAPI adapter which fails.
• On systems using PowerVM firmware, a problem was fixed for a CAPI  (Coherent Accelerator Processor Interface) device going to a "Defined" state instead of "Available" after a partition boot.  If the CAPI device is doing recovery and logging error data at the time of the partition boot, the error may occur.  To recover from the error, reboot the partition.  With the fix, the hypervisor will wait for the logging of error data from the CAPI device to finish before proceeding with the partition boot.
• On systems using PowerVM firmware, a problem was fixed for a hypervisor adjunct partition failed with "SRC B2009008 LP=32770" for an unexpected SR-IOV adapter configuration.  Without the fix, the system must be re-IPLed to correct the adjunct error.  This error is infrequent and can only occur if an adapter port configuration is being changed at the same time that error recovery is occurring for the adapter.
• On systems using PowerVM firmware and PCIe adapters in SR-IOV mode,  the following problem was addressed with a Broadcom Limited (formerly known as Avago Technologies and Emulex) adapter firmware update to 10.2.252.1913:  Transmit time-outs on a Virtual Function (VF) during stressful network traffic.
• On systems using PowerVM firmware with an invalid P-side or T-side in the firmware, a problem was fixed in the partition firmware Real-Time Abstraction System (RTAS) so that system Vital Product Data (VPD) is returned at least from the valid side instead of returning no VPD data.   This allows AIX host commands such as lsmcode, lsvpd, and lsattr that rely on the VPD data to work to some extent even if there is one bad code side.  Without the fix,  all the VPD data is blocked from the OS until the invalid code side is recovered by either rejecting the firmware update or attempting to update the system firmware again.
• On systems using PowerVM firmware without a HMC (and in Manufacturing Default Configuration (MDC) mode with a single host partition), a problem was fixed for missing dumps of type SYSDUMP. FSPDUMP. LOGDUMP, and RSCDUMP that were not off-loaded to the host OS.  This is an infrequent error caused by a timing error that causes the dump notification signal to the host OS to be lost.  The missing/pending dumps can be retrieved by rebooting the host OS partition.  The rebooted host OS will receive new notifications of the dumps that have to be off-loaded.
• On systems using PowerVM firmware, a problem was fixed for truncation on the memory fields displayed in the Advanced System Management Interface on the COD panels.  ASMI shows three fields of memory called "Installed memory", Permanent memory", and "Inactive memory".  The largest value that can be displayed in the fields was "9999" GB.  This has been expanded to a maximum of "999999" GB for each of the ASMI fields.  The truncation was only in the displayed memory value, not in the actual memory size being used by the system which was correct.
• On systems using PowerVM firmware and a partition using Active memory Sharing (AMS), a problem was fixed for a Live Partition Mobility (LPM) migration of the AMS partition that can hang the hypervisor on the target CEC.  When an AMS partition migrates to the target CEC, a hang condition can occur after processors are resumed on the target CEC, but before the migration operation completes.  The hang will prevent the migration from completing, and will likely require a CEC reboot to recover the hung processors.  For this problem to occur, there needs to be memory page-based activity (e.g. AMS dedup or Pool paging) that occurs exactly at the same time that the Dirty Page Manager's PSR data for that page is being sent to the target CEC.
• On systems using PowerVM firmware, a problem was fixed for PCIe adapter hangs and network traffic error recovery during Live Partition Mobility (LPM) and SR-IOV vNIC (virtual ethernet adapter)  operations.  An error in the PCI Host Bridge (PHB) hardware can persist in the L3 cache and fail all subsequent network traffic through the PHB.  The PHB  error recovery was enhanced to flush the PHB L3 cache to allow network traffic to resume.
• On systems using PowerVM firmware with AIX or Linux partitions with greater than 8TB of memory, a problem was fixed for Dynamic DMA Window (DDW) enabled adapters IPLing into a "Defined" state,  instead of "Available", and unusable with a "0" size DMA window.  If a DDW enabled adapter is plugged into an HDDW (Huge Dynamic DMA Window) slot in a partition with the large memory size, the OS changes the default DMA window to "0" in size.  To prevent this problem, the Advanced System Management Interface (ASMI) in the service processor can be used to set "I/O Enlarged Capacity" to "0" (which is off), and all the DDW enabled adapters will work on the next IPL.
• On systems using OPAL firmware, a problem was fixed for a held PSI link in delayed power off during a reset/reload of the service processor.  This error makes the service processor do a forced recovery of the PSI link on the next IPL.  For this problem, the PSI SRCs and error logs can be ignored as there is no problem in the PSI link.
• On systems using OPAL firmware, a problem was fixed for intermittent errors in the module autoload function in the ibmpowernv driver.  A compatible property "ibm.opal-sensor" was added to implement the fix for a smooth autoload in Linux.
• On systems using OPAL firmware, a problem was fixed for lost console output for serial consoles during power downs and reboots.  If a power down or reboot is detected, the console output buffer is now flushed before proceeding with the operation.
• On systems using OPAL firmware , an informational message was added that OPAL does not support opal-prd since the processor runtime diagnostics (PRD) are handled by the service processor.
• On systems using OPAL firmware, a performance problem was fixed in the OPAL hypervisor PCI Host Bridge (PHB) to prevent the PHB L3 cache from retrying defunct entries in the L3 after an MSI end of information (EOI) has been received.  The cache line is now flushed after updating the P/Q bits in the priority queue.  The situation is improved (and thus performance) by sending a DCBF (Data Cache Block Flush) to force a flush of PHB cache.  This improves interrupt performance, reducing latency per interrupt.  The improvement will vary by workload.
• On systems using OPAL firmware, a problem was fixed for the OPAL hypervisor not releasing the PSI link after a power off of the CEC.  With the PSI link unavailable, the service processor has to forcibly reclaim it on the next IPL, causing erroneous SRCs and error logs for the PSI link when no problem exists.
• On systems using OPAL firmware, a problem was fixed for a infinite loop in the boot of a host OS linux kernel.  Under rare error conditions in the real time clock, a bad error code returned to the host could cause it to get stuck in an infinite loop.
• On systems using PowerVM firmware and NovaLink management of the partitions, a problem was fixed for error recovery for the NovaLink partition in cases where it has gone unresponsive with a heartbeat failure.  Without the fix, the system would have to be re-IPLed.  With the fix, the hypervisor reboots the NovaLink partition to resume normal operations.
• On PowerVM systems with partitions running Linux, a problem was fixed for intermittent hangs following a Live Partition Mobility (LPM) migration of a Linux partition.  A partition migrating from a source system running FW840.00 to a system running any other supported firmware level may become unresponsive and unusable once it arrives on the target system.  The problem only affects Linux partitions and is intermittent.  Only partitions that have previously been migrated to a FW840.00 system are susceptible to a hang on subsequent migration to another system.  If a partition is hung following a LPM migration, it must be rebooted on the target system to resume operations.
• On systems using OPAL firmware, a problem was fixed that prevented multiple NVIDIA Tesla K80 GPUs from being attached to one PCIe adapter.  This prevented using a PCIe attached GPU drawer.  This fix increases the PCIe MMIO (memory-mapped I/O) space to 1 TB from a previous maximum of 64 GB per PHB/PCIe slot.
• On PowerVM systems with dedicated processor partitions with low I/O utilization, the dedicated processor partition may become intermittently unresponsive. The problem can be circumvented by changing the partition to use shared processors.
• On systems using OPAL firmware, a problem was fixed in OPAL to identify the PCI Host Bridge (PHB) on CAPI adapter errors and not always assume PHB0.
• On systems using OPAL firmware, a problem was fixed in the OPAL gard utility to remove gard records after guarded components have been replaced,  Without the fix, Hostboot and the gard utility could be in disagreement on the replaced components, causing some components to still display as guarded after a repair.
• On systems using PowerVM firmware with partitions with very large number of PCIe adapters, a problem was fixed for partitions that would hang because the partition firmware ran out of memory for the OpenFirmware FCode device drivers for PCIe adapters.  With the fix, the hypervisor is able to dynamically increase the memory to accommodate the larger partition configurations of I/O slots and adapters.
• On PowerVM systems with vNIC adapters, a problem was fixed for doing a network boot or install from the adapter using a VLAN tag.  Without the fix, the support is missing for doing a network boot from the VLAN tag from the SMS RIPL menu.
  
• On systems using PowerVM firmware, a problem was fixed for a Live Partition Mobility (LPM) migration of a partition with large memory that had a migration abort when the partition took longer than five minutes to suspend.  This is a rare problem and is triggered by an abnormally slow response time from the migrating partition.  With the fix, the five minute time limit on the suspend operation has been removed.

• POWER8 (and later) servers include an “update access key” that is checked when system firmware updates are applied to the system.  The initial update access keys include an expiration date which is tied to the product warranty. System firmware updates will not be processed if the GA date of the desired firmware level occurred after the update access key’s expiration date.  As these update access keys expire, they need to be replaced using either the Hardware Management Console (HMC) or the Advanced Management Interface (ASMI) on the service processor.  Update access keys can be obtained via the key management website: http://www.ibm.com/servers/eserver/ess/index.wss.
• Support for allowing the PowerVM hypervisor to continue to run when communication between the service processor and platform firmware has been lost and cannot be re-established.  A SRC B1817212 may be logged and any active partitions will continue to run but they will not be able to be managed by the management console.  The partitions can be allowed to run until the next scheduled service window at which time the service processor can be recovered with an AC power cycle or a pin-hole reset from the operator panel.  This error condition would only be seen on a system that had been running with a single service processor (no redundancy for the service processor).

• Support for a HVDC (180-400 VDC) 1400W power supply in a one plus one or two plus two configuration to support redundancy.  Supported in rack models only with F/C EB2N for the S822 (8284-22A), S814(8286-41A), S824(8286-42A), and E850(8404-E8E) models.  And F/C EL1D for the S812L(8247-21L), S822L(8247-22L), and S824L(8247-42L) models.
• Support in the Advanced Systems Management Interface (ASMI) for managing certificates on the service processor with option "System Configuration/Security/Certificate Management".  Certificate management includes 1) Generation of Certificate Signing Request (CSR) 2) Download of CSR and 3) Upload of signed certificates.  For more information on managing certificates, go to the IBM KnowledgeCenter link for "Certificate Management"
  (https://www-01.ibm.com/support/knowledgecenter/P8ESS/p8hby/p8hby_securitycertificate.htm).
• Support for water cooling of the processor module in place of air cooling fins with feature code #ER2C.  The PCIe C5 slot carries the water lines so a PCIe adapter cannot be used there when the water cooling is installed.  This feature is available for the S822 (8284-22A) and S822L (8247-22L) models only.
• Support for a High Frequency Trading policy to speed the processors.  When this policy is enabled, the processor cores are allowed to run at a higher frequency and voltage for better performance.  A new panel was created in the Advanced Systems Management Interface (ASMI) "System Configuration/High Frequency Trading"  to enable and disable this policy.  In PowerVM mode,  this feature applies only to the S822 (8284-22A), S812L (8247-21L), and S822L (8247-22L) models.  In OPAL mode, this feature applies to   S812L (8247-21L) and S822L (8247-22L) with Ubuntu 14.04.3 bare-metal, Ubuntu 15.10 bare-metal, or RHEL 7.2 LE bare-metal.
• Support for enhanced power management on PowerKVM systems with memory throttling and in-band power measurement capability.  This feature applies to S812L (8247-21L) and S822L (8247-22L) models only.
  
• Support for service processor call home of error logs over ethernet (no dial-up modem required).  The call home setup is done through an option on the Advanced System Management Interface called "System Service Aids/Call-Home Setup".  This feature is only available for systems that are not attached to a management console.  For guidance on how to set up the call-home on the service processor, go to the IBM KnowledgeCenter link for "Configuring the call-home policy" (https://www-01.ibm.com/support/knowledgecenter/P8DEA/p8hby/callhomesetup.htm).
• PowerVM support for Support for Coherent Accelerator Processor Interface (CAPI) adapters.  The PCIe3 LP CAPI Accelerator Adapter with F/C #EJ16 is used on the S812L(8247-21L) and S822L (8247-22L)  models  The PCIe3 CAPI FlashSystem Acclerator Adapter with F/C #EJ17  is used on the S814(8286-41A) and S824(8286-42A) models.  The PCIe3 CAPI FlashSystem Accelerator Adapter with F/C #EJ18 is used on the S822(8284-22A), E870(9119-MME), and E880(9119-MHE) models.  This feature does not apply to the S824L (8247-42L) model.
• Support for PCIe3 Expansion Drawer (#EMX0) lower cable failover, using lane reversal mode to bring up the expansion drawer from the top cable.  This eliminates a single point of failure by supporting lane reversal in case of problems with the lower cable.
• Expanded support of Virtual Ethernet Large send from IPv4 to the IPv6 protocol in PowerVM.
• Support for IBM i network install on a IEEE 802.1Q VLAN.  The OS supported levels are IBM i.7.2.TR3 or later.  This feature applies only to S814 (8286-41A), S824(8286-42A), E870 (9119-MME), and E880 (9119-MHE) models.
• Support for PowerVM vNIC with up to six vNIC client adapters for each partition.  PowerVM vNIC combines many of the best features of SR-IOV and PowerVM SEA to provide a network solution with options for advanced functions such as Live Partition Mobility along with better performance and I/O efficiency when compared to PowerVM SEA.  In addition PowerVM vNIC provides users with bandwidth control (QoS) capability by leveraging SR-IOV logical ports as the physical interface to the network.
  Note:  If more than six vNIC client adapters are used in a partition, the partition will run, as there is no check to prevent the extra adapters, but certain operations such as Live Partition Mobility may fail.
  
• Enhanced handling of errors to allow partial data in a Shared Storage Pool (SSP) cluster.  Under partial data error conditions, the management console "Manage PowerVM" gui will correctly show the working VIOS clusters along with information about the broken VIOS clusters, instead of showing no data.
• PowerVM enhanced to support Little Endian (LE)  Linux guest OSes with Nvidia Compute Intensive Accelerator (PCIe attached GPU) with F/C EC47 and EC4B.  These adapters are only supported on the IBM Power System S824L (8247-42L) model.  Little Endian must be used because the Nvidia software stack is only enabled for LE mode.
• Live Partition Mobility (LPM) was enhanced to allow the user to specify VIOS concurrency level overrides.
• Support was added for PowerVM hard compliance enforcement of the Power Integrated Facility for Linux (IFL).  IFL is an optional lower cost per processor core activation for Linux-only workloads on IBM Power Systems.  Power IFL processor cores can be activated that are restricted to running Linux workloads.  In contrast, processor cores that are activated for general-purpose workloads can run any supported operating system.  PowerVM will block partition activation, LPM and DLPAR requests on a system with IFL processors configured if the total entitlement of AIX and IBMi partitions exceeds the amount of licensed general-purpose processors.  For AIX and IBMi partitions configured with uncapped processors, the PowerVM hypervisor will limit the entitlement and uncapped resources consumed to the amount of expensive processors that are currently licensed.
• Support was added to allow Power Enterprise Pools to convert permanently-licensed (static) processors to Pool Processors using a CPOD COD activation code provided by the management console.  Previously, only unlicensed processors were able to become Pool Processors.
• The management console was enhanced to allow a Live Partition Mobility (LPM) if there is a failed VIOS in a redundant pair.  During LPM, if the VIOS is inactive, the management console will use stored configuration information to perform the LPM.
• The firmware update process from the management console and from in-band OS (except for IBM i PTFs) has been enhanced to download new "Update access keys" as needed to prevent the access key from expiring.  This provides an automatic renewal process for the entitled customer.
• Live Partition Mobility support was added to allow the user to specify a different virtual Ethernet switch on the target server.
• PowerVM was enhanced to support an AIX Live Update where the AIX kernel is updated without rebooting the kernel.  The AIX OS level must be 7.2 or later.  Starting with AIX Version 7.2, the AIX operating system provides the AIX Live Update function which eliminates downtime associated with patching the AIX operating system. Previous releases of AIX required systems to be rebooted after an interim fix was applied to a running system. This new feature allows workloads to remain active during a Live Update operation and the operating system can use the interim fix immediately without needing to restart the entire system. In the first release of this feature, AIX Live Update will allow customers to install interim fixes (ifixes) only. For more information on AIX Live Update,  go to the IBM KnowledgeCenter link for "Live Update" 
  (https://www-01.ibm.com/support/knowledgecenter//ssw_aix_72/com.ibm.aix.install/live_update_install.htm).
  
• The management console has been enhanced to use standard FTP in its firmware update process instead of a custom implementation.  This will provide a more consistent interface for the users.
• Support for setting Power Management Tuning Parameters from the management console (Fixed Maximum Frequency (FMF), Idle Power Save, and DPS Tunables) without needing to use the Advanced System Management Interface (ASMI) on the service processor.  This allows FMF mode to be set by default without having to modify any tunable parameters using ASMI.
• Support for a Corsa PCIe adapter with accelerator FPGA for low latency connection using CAPI (Coherent Accelerator Processor Interface) attached to a FlashSystem 900 using two 8Gb optical SR Fibre Channel (FC) connections.
  Supported IBM Power Systems for this feature are the following:
  1) E880 (9119-MHE) with CAPI Activation feature #EC19 and  Corsa adapter #EJ18 Low profile on AIX.
  2) E870 (9119-MME) with CAPI Activation feature #EC18 and Corsa adapter #EJ18.Low profile on AIX.
  3) S822 (8284-22A) with CAPI  Activation feature #EC2A and Corsa adapter #EJ18.Low profile on AIX.
  4) S814 (8286-41A) with CAPI Activation feature #EC2A and Corsa adapter #EJ17 Full height on AIX.
  5) S824 (8286-42A) with CAPI Activation feature #EC2A and Corsa adapter #EJ17 Full height on AIX.
  6) S812L (8247-21L) with CAPI Activation feature #EC2A and Corsa adapter #EJ16 Low profile on Linux.
  7) S822L (8247-22L)  with CAPI Activation feature #EC2A and Corsa adapter #EJ16 Low profile on Linux.
  OS levels that support this feature are PowerVM AIX 7.2 or later and OPAL bare-metal Linux Ubuntu 15.10.
  The IBM FlashSystem 900 storage system is model 9840-AE2 (one year warranty) or 9843-AE2 (three year warranty) at the 1.4.0.0 or later firmware level with features codes #AF23, #AF24, and #AF25 supported for 1.2 TB, 2.9 TB, 5.7 TB modules, respectively.
• The Digital Power Subsystem Sweep (DPSS) FPGA, used to control P8 fan speeds and memory voltages, was enhanced to support the 840 GA level. This DPSS update is delayed to the next IPL of the CEC and adds 18 to 20 minutes to the IPL.  See the "Concurrent Firmware Updates" section above for details.
• Support for Data Center Manageability Interface (DCMI) V1.5 and Energy Star compliance.  DCMI  features were added to the Intelligent Platform Management Interface (IPMI) 2.0 implementation on the service processor.  DCMI adds platform management capability for monitoring elements such as system temperatures, power supplies, and bus errors.  It also includes automatic and manually driven recovery capabilities such as local or remote system resets, power on/off operations, logging of abnormal or  "out-of-range‟ conditions for later examination.  And It allows querying for inventory information that can help identify a failed hardware unit along with power management options for getting and setting power limits.
  Note:  A deviation from the DCMI V1.5 specification exists for 840.00 for the DCMI Configuration Parameters for DHCP Discovery.  Random back-off mode is enabled by default instead of being disabled.  The random back-off puts a random variation delay in the DHCP retry interval so that the DHCP clients are not responding at the same time. Disabling the back-off time is not required for normal operations, but if wanted, the system administrator can override the default and disable the random back-off mode by sending the “SET DCMI Configuration Parameters” for the random back-off property of the Discovery Configuration parameter.  A value of "0" for the bit means "Disabled".
• Support for PowerVM NovaLink partition management.  The NovaLink architecture enables OpenStack to work seamlessly with PowerVM by providing a direct connection to the PowerVM server rather than proxying through an HMC.  This allows for vastly improved scalability (from 30 to 200+ servers), better performance, and better alignment with the OpenStack architecture.  NovaLink is enabled via a small software package that runs within a Linux partition (Ubuntu) on a POWER8 host.  The following are the NovaLink hardware and software requirements:
      o POWER8 hardware coupled with System Firmware 840 (or later)
      o Virtual IO Server 2.2.4 (or later)
      o Ubuntu Linux 15.10 (ppc64le) (or later)
      o PowerVC 1.3 (or later)
• Support for IBM i operating system over Virtual I/O Server (VIOS) on the IBM Power System S822 (8284-22A) server.  The IBM i support requires VIOS (no native I/O) and FW840.00.  At this level, the S822 supports IBM i 7.2 or IBM i7.1 with special terms and conditions. Technology Refresh 3 or later for IBM i 7.2 or Technology Refresh 11 or later for IBM i 7.1 is required.  Multiple IBM i partitions, each up to a maximum of two cores, are supported. The Power S822 software tier is P10.
  IBM i partitions that access directly attached disk or SSD through VIOS must use 4 k byte sector drives, not 5 xx byte sector drives. The 4 k drives are required for performance reasons.
  Note:  Async or bisync adapters or crypto-cards are not supported under VIOS.  Thus IBM i applications that require use of these adapters are not a good fit for the Power S822.   IBM i 7.2 clients can connect to a LAN-attached OEM device that has downstream async connections.