Conference eusact::r3_osf1_technical

410.0. "ES_SPIN_LOCK_ERROR problem" by TROOA::ras007p01.srn.dec.com::dloten (semper ubi sub ubi) Mon Feb 17 1997 21:15

We have a customer running R/3 v3.0D, Informix, DUNIX v3.2F on an Alphaserver 
4100 5/300-2 and has started encountering ES_SPIN_LOCK_ERROR problems.  He 
went to OSS and downloaded note 63381, which describes the problem and a 
solution (see below).  He has implemented the prescribed parameter change and 
restarted the system, but doesn't see the expected message in the DEV_DISP 
trace file and is still getting the ABAP error messages.  Has anyone else 
seen this problem?

-doug

===========================================================================

     R/3 note no.   63381                         17.02.1997           Page 1
     ________________________________________________________________________

     Number              0063381
     Status              Released for customer
     Set by              SAP AG on 13.12.1996

     Language            E
     Short text          EM: ES_SPINLOCK_ERROR on DEC/OSF1

     Administrator       SAP AG
     Application area    BC-OP-DIG Digital UNIX
     ________________________________________________________________________

     Long text


     Symptom
     ABAP/4 dump:
     - ES_SPIN_LOCK_ERROR
     in the trace files dev_w:
     - ERROR => Sem2Lock: 653e3d0 sem2_tryset() (timeout) [sem2xx.c ...]
     The error only occurs in this form on DEC/alpha multi-processor machines
     under OSF1 or DigitalUnix.
     The error causes the termination of an application program with an
     ABAP/4 dump, further application programs are not affected.
     Further accompanying error messages in this context:
     - ERROR => EsILock: inconsistency, time = ...
     -  RecoverAll
     -  RecoverFreeList
     - Module esxx
     - Function EsILock ...
     - Error ES_SPIN_LOCK_ERROR occurred.
     Additional key words
     ES Extended Memory management
     Assembler instructions: ldq_l stq_c
     Cause and preconditions
     Background information:
     In the Memory Manangement lock function, quick locks are used initially
     (on the basis of the assembler instructions ldq_l, stq_c, see DEC/alpha
     Architecture manual). In addition to the results "Lock set" and "Lock
     could not be set", this lock operation can also return a value "Lock
     operation was destroyed and must be repeated" on multi-processor
     machines.
     The error occurs after the 100th repetition does not return a consistent
     result.
     A consistent result does not occur if another processor is accessed at
     an adjacent address between ldq_l and stq_c (two instructions later),
     where the size of the sensitive address area depends on the address of
     the lock object, and can be as large as one page (8 KB).
     If no hardware problems exist, this error situation occurs very
     infrequently.
     The error can be avoided completely if the locking operation uses the
     slower semaphore procedure in the Memory Management. This measure is
     only recommended if the error actually occurs.
     Solution
     Change the Memory Management Locking from Spinlock to Semaphore with an
     entry in the instance profile of the form:
     em/use_sema_lock = TRUE
     The following message then occurs in the trace file dev_disp:
                                                                       Page 2

      EsILock: use semaphore instead of spinlock
     Source code corrections
     ________________________________________________________________________

     Internal system environment


     Release             30A  to  30Z

     ________________________________________________________________________

     External system environment


     OP system           DEC-UNIX
     ________________________________________________________________________