Abend Codes- Overview

An Abend (ABnormal END) is the way a mainframe program is stopped by the system when something goes wrong. This tutorial explains the main abend types and how to debug them.

What it means

• ABEND stands for ABnormal END of a task. Instead of finishing normally, the job step is terminated and the system writes a message and optionally a dump.
• There are two families of abends: System abends (Sxxx), issued by the operating system, e.g. S0C7, S0C4, S322; and User abends (Unnnn), issued by the program itself, e.g. U0001.
• Every abend comes with a completion code. The IEF450I message in the job log shows which step abended and with what code, for example: IEF450I STEP1 - ABEND=S0C7 U0000 REASON=00000007.

Common causes

• System abends are caused by program errors (bad data, bad address, wild branch) or resource problems (CPU time exceeded, space exhausted, dataset not found).
• User abends (Unnnn) are raised deliberately by the program, often to signal an expected failure like a missing input file.
• CICS uses its own codes called transaction abends, e.g. ASRA (program check), AEIN, ATNI, which work like system abends inside a CICS region.

How to find it

• Start with the IEF450I message in the job log: it tells you the failing step name, the system completion code (Sxxx) and the reason code.
• For a program failure, get a dump of storage at the time of the abend: code a SYSUDUMP, SYSABEND or SYSMDUMP DD statement in the failing step.
• SYSUDUMP gives a formatted dump of the program area (usually enough for COBOL). SYSABEND dumps the full address space (large). SYSMDUMP is used with the IPCS tool for deep analysis.
• In the dump, look at the PSW (Program Status Word) for the failing instruction address and the general registers for the data involved.

How to fix it

• Read the message first: the completion code tells you exactly which category of error to look for (data, address, loop, space).
• Use the dump to find the failing instruction offset, then match it against the compiler listing of your program to find the statement.
• Fix the root cause in the program or JCL, recompile, and rerun. Never just raise limits (like TIME or SPACE) without checking the program is healthy.

Example

• Add a dump DD statement to any failing step so you get a diagnostic dump with the abend:-
  
  //STEP1 EXEC PGM=MYPROG //SYSUDUMP DD SYSOUT=* <-- formatted dump of program area //SYSABEND DD SYSOUT=* <-- full address space dump