VM66501: PERFORMANCE TOOLKIT DIVIDE EXCEPTION AT FCXPRC+30EC

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• Performance Toolkit can encounter a divide exception during live
  guest relocation.
  
  Between two monitor sample intervals, live guest relocation can
  cause a large, incorrect jump in the value of total emulation
  time for a processor (PFXPRBTM).  This can show up in the
  domain 0 record 2 and domain 5 record 22 monitor records.  This
  jump can result in incorrect results in performance monitor
  tools such as Performance Toolkit which might take a divide
  exception (prog 9) if the jump is large enough.
  

Local fix

• N/A
  

Problem summary

• ****************************************************************
  * USERS AFFECTED: All z/VM users of live guest relocation      *
  ****************************************************************
  * PROBLEM DESCRIPTION:                                         *
  ****************************************************************
  * RECOMMENDATION: APPLY PTF                                    *
  ****************************************************************
  The problem is caused during the trifurcation phase of a live
  guest relocation when the guest state, I/O devices, and
  penultimate memory moves happen concurrently.  VMDVTIME on the
  destination's VMDBK is migrated in the guest state move fork.
  The problem is if an inbound VMDBK is dispatched on one of the
  other two forks when this VMDVTIME update happens, then the
  calculated PFXPRBTM value will be incorrect when the VMDBK is
  undispatched.
  
  The calculation is PFXPRBTM = PFXPRBTM+(VMDVTIME-PFXVTDSP) where
  
    PFXVTDSP -- VMDVTIME at dispatch = x'7FFFFFFF00000000' which
       is the initial value set by HCPBVM when the skeleton VMDBK
       for the inbound relocation is created
    VMDVTIME -- VMDVTIME at undispatch which was just calculated
       based on the source VMDBKs value at relocation time
  
  The result of the subtraction is a negative value because
  VMDVTIME continually decreases and so does PFXPRBTM.
  Because the VMDVTIME from the source VMDBK can be very much
  lower than the initial value, the calculation leads to a large
  discrepancy in the new value of PFXPRBTM leading to a very large
  jump from the previous monitor interval.
  

Problem conclusion

• To avoid this problem, setting of the source value of VMDVTIME
  into the destination VMDBK is delayed until just before the
  guest is restarted on the destination member.  This is
  accomplished by saving the source value in another field and
  then moving that value into VMDVTIME in HCPRLGFI.  The VMDBK's
  VMAVTIME_RLO (VMDVTIME at the time of relocation) field must
  be set with that same value at this time also.
  

Temporary fix

• FOR RELEASE VM/ESA CP/ESA R640 :
  PREREQ: VM65786
  CO-REQ: NONE
  IF-REQ: NONE
  FOR RELEASE VM/ESACP/ESAR710 :
  PREREQ: VM66266 VM65786
  CO-REQ: NONE
  IF-REQ: NONE
  FOR RELEASE VM/ESA CP/ESA R720 :
  PREREQ: VM66173
  CO-REQ: NONE
  IF-REQ: NONE
  

Comments

• ×**** PE21/11/19 FIX IN ERROR. SEE APAR VM66583  FOR DESCRIPTION
  

APAR Information

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

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

  UM35839 UM35840 UM35841

Modules/Macros

• HCPRLG   HCPVMDBK
  

Fix information

• Fixed component name

  VM CP

• Fixed component ID

  568411202

Applicable component levels

• R640 PSY UM35839

     UP21/03/31 I 1000

• R710 PSY UM35840

     UP21/03/31 P 2101

• R720 PSY UM35841

     UP21/03/31 P 2201

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.