VM66306: CHPIDV GUEST HANG AFTER FICON RESET NOTIFICATION

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• Linux guest hang during I/O error recovery after FICON CHPID
  reset/recovery.   Occurs only on guests defined with OPTION
  CHPIDV in User Directory.  Occurs after FICON CHPID recovery
  following a cable pull, port block, or some other condition
  which causes the CHPID to go away and then come back.
  

Local fix

• Drop into CP READ and ReIPL the guest.   This will break the
  Loop/Hang condition.
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: All z/VM Users of OPTION CHPIDV ONE in User  *
  *                 Directory.                                   *
  ****************************************************************
  * PROBLEM DESCRIPTION:                                         *
  ****************************************************************
  * RECOMMENDATION: APPLY PTF                                    *
  ****************************************************************
  This symptom is an infinite loop (with many losses of control)
  that hangs guest virtual machines, particularly Linux, when the
  guest OS goes into Reset Notification recovery.
  
  This base 6.2 SSI problem was found during device qualification
  testing on test cases that reset a FICON connection (e.g., cable
  pulls, port blocking, port toggling, etc.).  A Reset
  Notification is term of art used to describe the tap on the
  shoulder that occurs when a FICON connection becomes available
  again (after it goes away or is reset).  It tells the OS to do
  rediscovery of the connection to make sure nothing on it has
  changed.
  
  The specific bug is in VM CCW simulation code and causes the
  sense data associated with the Reset Notification to never be
  cleared after the guest reads it, specifically when the guest
  has CHPID Virtualization enabled (OPTION CHPIDV ONE in support
  of guest relocation).  As a result, the guest OS will read the
  Reset Notification sense, start doing I/O to rediscover the port
  and run right back into the same sense data again (causing the
  guest to start recovery over).  This happens forever, or until
  the guest is Re-IPLed (where as a different path finally clears
  the sense).
  

Problem conclusion

• VM CCW Simulation code in HCPIOV was modified to properly clear
  the Reset Notification sense data when a guest with OPTION
  CHPIDV ONE initially reads the sense.  I/O dispatching code in
  HCPIOS was also changed to reflect the Reset Notification error
  to a guest with CHPIDV ONE in a more timely manner.
  

Temporary fix

• FOR RELEASE VM/ESA CP/ESA R640 :
  PREREQ: NONE
  CO-REQ: NONE
  IF-REQ: NONE
  FOR RELEASE VM/ESACP/ESAR710 :
  PREREQ: NONE
  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:

  UM35524 UM35525

Modules/Macros

• HCPIOS   HCPIOV
  

Fix information

• Fixed component name

  VM CP

• Fixed component ID

  568411202

Applicable component levels

• R640 PSY UM35524

     UP19/09/12 P 2001 ¢

• R710 PSY UM35525

     UP19/09/12 P 2001 ¢

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.