PM12402: CSL3117E SVC ABEND S0C4 MEMBER RB1MA7NB PSW=074C0000 A3DE69E4 RSN=00000004 CSLSRG20=23DE5D68 REG PROC

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• Environment: Z/OS 1.10 - IMS V10
  Shared Queue two way - 2 IMS members MS1DCTL and MS2DCTL
  (control
  region), MS1DCSLS and MS2DCLSL (sci task)
  Problem description:
  - on day 201080 - 01:41:18 - user issued the command //DIS TRAN
  IL01
    under Mainview Plex product on LPAR SECA (task MVPLEX)
  - at this time only MS1DCTL member was active
  - the //DIS TRAN command was not executed successfully
  - at 2010080 01:41:18.43 this msg was issued:
    CSL3117E SVC ABEND S0C4  MEMBER RB1MA7NB PSW=074C0000 A3DE69E4
                     RSN=00000004 CSLSRG20=23DE5D68  REG PROC
  - at 2010080 01:41:21.34 :
    CSL3117E SVC ABEND S0C4  MEMBER RB1MA7NB PSW=070C2000 814992A4
            RSN=00000004 CSLSRG20=23DE5D68   CLEANUP
    DUMPID=002 REQUESTED BY JOB (MVPLEX  )
  - after that many 0C4 abends were reported at CLS task
  (MS1DCSLS)
    DIAG=4004000184 (RETRYING)
  - on Wednesday as this task was filling spool, the customer
  brought down
    this task, but didn't get to bring it up again. The 0C4 abend
  persisted
    and the SCI didn't stay active.
  - At this night the customer brought down all IMS members in
  this Shared
    group and also did cold start in the  Shared Queue structure,
  but again
    the SCI task didn't stay up.
  - What happened in customers case is that the CMBR
  storage was freed, but then a second registration
  came along and just happened to get the same
  storage again.  The PMBR already pointed to the CMBR
  from the first registration. Now when the same  CMBR
  was chained in for the second registration, it got
  chained to itself, creating a looped CMBR chain off
  of the PMBR.
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED:   All IMS V11 SCI users.                     *
  ****************************************************************
  * PROBLEM DESCRIPTION:   ABEND0C4 in CSLSTOA0.                 *
  *                                                              *
  *                        ABEND0C4 in CSLSPLR0.                 *
  *                                                              *
  *                        CMBR chain points to storage that     *
  *                        has been released.                    *
  ****************************************************************
  * RECOMMENDATION: INSTALL CORRECTIVE SERVICE FOR APAR/PTF      *
  ****************************************************************
      SCI does not support an address space registering multiple
    times using different keys, but it does not check for this
    during registration.  When an address space tries to register
    a second time using a different key than the first
    registration, an abend is encountered in CSLSRG20.  When
    CSLSRG20 tries to clean up by deleting the storage obtained
    for the new registration, two additional ABEND0C4 abends
    are encountered due to coding errors in the cleanup code.
  
      The first abend occurs when SCI attempts to delete the input
    exit cell pool.  Before deleting the cell pool, it switches to
    the key saved in the CMBR block (which matches the key of the
    first registration).  However, the cell pool was obtained
    using the caller's key, which in this case is different.  This
    results in an ABEND0C4.
  
      The second abend occurs when SCI attempts to return a cell
    to the registration cell pool.  Before it frees the cell, the
    pointer to the cell is zeroed out (to ensure that this is not
    repeated).  This results in an ABEND0C4 when the CPOOL service
    tries to return a cell at location x'00'.
  
      Because of the way the code is structured, the second abend
    results in bypassing the code that unchains the CMBR block
    from the the CMBR blocks under a particular PMBR.  Instead of
    executing that code, it goes directly to the code that
    releases the CMBR, even though the CMBR is still on an active
    chain.
  
      The above scenario results in a CMBR chain that points to
    storage that has been released.  When this storage is reused,
    it will eventually result in multiple ABEND0C4 abends when SCI
    modules attempt to process all the CMBR blocks by follwing the
    CMBR chain.
  

Problem conclusion

• GEN:
  KEYWORDS:
  
  *** END IMS KEYWORDS ***
      A new registration error reason code is added to CSLSRG20.
    This reason code is returned when the key of a subsequent
    registration from an address space does not match the key from
    the first registration from that address space.
  
      The errors in the cleanup subroutine are fixed.
  
  
      SCI Macro Changed
      ==================
  
      CSLSRR
        A new reason code is added : SRSN_WRONGKEY.  This reason
        code is returned when the key does not match the key of
        the first registration from the address space.
  
      SCI Module Changed
      ==================
  
      CSLSRG20
        A check is added to ensure that the key matches the key
        of the first registration from this address space (if
        multiple registrations).  If the keys don't match, the
        registration is rejected with a return code of SC_ENVIRON
        and reason code of SRSN_WRONGKEY.
  
        The registration cell return is pulled into its own
        if-then-else clause.  The pointer to the cell is
        saved before the CMBR cell pointer is cleared.  The
        saved pointer is used in the CPOOL (FREE) service.
  
  
  
  
    The following publication updates are made by this APAR:
  
  
    IMS Version 11: System Programming API Reference
                    SC19-2445
  
  
      In chapter 8 (Writing a CSL SCI Client), add the following
    paragraph in the CSLSCREG section.  Add it following the
    paragraph describing the SYSEVENT DONTSWAP restriction:
  
    Restriction: One address space may register multiple times
    with SCI.  However, all registrations from one address
    space must be made while running in the same PSW key and
    state (supervisor or problem) as the first registration
    made in the address space.
  
  
      Add the following entry to table 68 (CSLSCREG return
    and reason codes):
  
  
    Return Code    Reason Code    Meaning
  
    x'01000010'    x'0000402C'    Caller key does not match key
                                   of existing registration or
                                   state (supervisor or problem)
                                   does not match state of
                                   existing registration.
  

Temporary fix

• *********
  * HIPER *
  *********
  

Comments

APAR Information

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

  PM11027

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

  UK60321

Modules/Macros

• CSLSRG20 CSLSRR
  

Fix information

• Fixed component name

  IMS V11

• Fixed component ID

  5635A0200

Applicable component levels

• R100 PSY UK60321

     UP10/09/14 P F009 Ž

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.