VM66487: ABENDSXA004 GUEST USE OF SIMULATED AND REAL QDIO DEVICES

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• When a virtual machine is configured to exploit one or more
  simulated VSwitch NIC devices for network connectivity and
  also using a dedicated real QDIO device in a multiple vCPU
  virtual I/O configuration, there is the possibility of the
  system terminating with a SXA004 ABEND.  A real dedicated
  QDIO device can either be an OSA-Express, HiperSockets or
  a FCP device.
  A VAP002 abend is also a symptom of this problem.
  Other symptoms can be a slow or unresponsive system.  If a
  SNAPDUMP is taken many calls to HCPVAI+9D2 may exist.
  

Local fix

• N/A
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: All customers using a z/VM VSwitch for guest *
  *                 network connectivity in addition to a        *
  *                 dedicated QDIO capable device such as an OSA *
  *                 Express, HiperSockets, or a FCP Device.      *
  ****************************************************************
  * PROBLEM DESCRIPTION:                                         *
  ****************************************************************
  * RECOMMENDATION: APPLY PTF                                    *
  ****************************************************************
  z/VM may terminate with an SXA004 ABEND forcing a system IPL to
  recover, when a virtual machine is using both a VSwitch to
  provide network connectivity and real dedicated QDIO Devices
  like OSA-Express, HiperSockets or FCP.
  
  QDIO devices use a special form of I/O Interruption called an
  Adapter Interruption (AI) to present completion events for
  pending networking or SCSI operations.  The main difference
  between a tradition I/O Interruption and an AI is that
  completion events for multiple devices can be presented with
  just a single interruption.  The problem here is in the logic
  which currently merges completion events for multiple devices
  into a single I/O Interruption.  Specifically the merging of
  simulated AIs for a VSwitch device with real hardware generated
  AIs for a dedicated QDIO Device.
  
  The following virtual machine configuration is exposed to this
  issue:
  1. Configured and using multiple virtual processors (MP CONFIG)
  2. One or more virtual NICs coupled to one or multiple VSwitches
  3. One or more dedicated FCP, HiperSockets or OSA-Express
  Devices
  4. Both VSwitch and real QDIO Devices must be transferring data
  

Problem conclusion

• For a real dedicated QDIO Device, the hardware has the ability
  to reflect an Adapter Interruption (AI) directly to an
  operating system running in a virtual machine without z/VM
  involvement.  For a simulated VSwitch Device, it is z/VM's
  responsibility for generating a simulated AI to report a
  completion event.  Therefore, when running a configuration using
  both real and simulated devices, a hardware generated AI can
  inform the operating system of completion events for both real
  and simulated devices.
  
  In order to generate a simulated AI, z/VM uses a special AI
  Virtual Device Block (AIF VDEV) to serialize the presentation
  of a simulated AI to a virtual machine.  The AIF VDEV is only
  required for a simulated AI, since a real device can pass the
  AI directly to a virtual machine.  The problem at hand is
  directly related to the high frequency of z/VM's need to acquire
  the AIF VDEV Lock to determine whether it can merge a simulated
  AI with an already pending real AI.
  
  When dealing with moderate to high bandwidth data transfers on
  all QDIO devices, it's possible for a large number of
  independent completion event tasks to get queued waiting for an
  exclusive AIF VDEV Lock.  This is necessary to determine
  whether it can architecturally merge the completion event.  If
  the virtual processor z/VM selected to present the interruption
  is slow in processing for any reason while it holds the AIF VDEV
  Lock, the queue of pending tasks can get so long, it can
  exhaust all available memory in the System Execution
  Space; thus causing the system to ABEND.
  
  Additional logic is added to optimize the merging of multiple
  completion events into a single event.  This is accomplished by
  eliminating the need to acquire the AIF VDEV Lock when a
  previous completion event task is already pending to do the
  work.  This will prevent no more than a few tasks ever waiting
  for the lock at any point in time.
  

Temporary fix

• FOR RELEASE VM/ESACP/ESAR710 :
  PREREQ: VM66302 VM66426
  CO-REQ: NONE
  IF-REQ: NONE
  FOR RELEASE VM/ESA CP/ESA R720 :
  PREREQ: VM66426
  CO-REQ: NONE
  IF-REQ: NONE
  

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

  UM35807 UM35808

Modules/Macros

• HCPVAI   HCPVIS   HCPVQA
  

Fix information

• Fixed component name

  VM CP

• Fixed component ID

  568411202

Applicable component levels

• R710 PSY UM35807

     UP21/01/12 P 2101 ¢

• R720 PSY UM35808

     UP21/01/12 P 2101 ¢

Fix is available

• Select the PTF appropriate for your component level. You will be required to sign in. Distribution on physical media is not available in all countries.