Experimental Physics and
Industrial Control System

"Weatherford, Jeremy W." <[email protected]> · Wed, 30 Jun 2004 10:24:28 -0500

If someone could tell me of a more appropriate place to report this, that would be wonderful.  It's not part of the EPICS base, so I didn't think it belonged in Mantis.

The bug is with both versions of cau, 20020903 and 20020802.  On receipt of any signal, cau's behavior is undefined.  It may segfault, which is the easy way out.  However, on two of our Linux boxes (RH Enterprise, kernel 2.4.21) it has reproducibly crashed destructively, allocating approx. 2G of memory and thrashing both CPUs.  When it does this, it takes about 5 minutes to die, leaving the system nearly unusable during that time.

The problem is in cau.c, at line 470 (20020802) or 473 (20020903):

   genSigInit();

genSigInit, in genSubr.c, is defined as:

   void genSigInit(void (*handler)(int))

Ooops.  Line 470/473 in cau.c should be:

   genSigInit(cauTaskSigHandler);

This allows cau to exit gracefully on receipt of a signal.  The moral of the story?  Use ANSI C prototypes!  (the offending argument-less prototype is in genDefs.h, line 236)

Jeremy Weatherford
SBC-CAT, APS

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System