Actual source code: signal.c

  1: /*
  2:       Routines to handle signals the program will receive.
  3:     Usually this will call the error handlers.
  4: */
  5: #include <petsc/private/petscimpl.h>
  6: #include <signal.h>
  7: #include <stdlib.h> /* for _Exit() */

  9: static PetscClassId SIGNAL_CLASSID = 0;

 11: struct SH {
 12:   PetscClassId classid;
 13:   PetscErrorCode (*handler)(int, void *);
 14:   void      *ctx;
 15:   struct SH *previous;
 16: };
 17: static struct SH *sh        = NULL;
 18: static PetscBool  SignalSet = PETSC_FALSE;

 20: /* Called by MPI_Abort() to suppress user-registered atexit()/on_exit() functions.
 21:    See discussion at https://gitlab.com/petsc/petsc/-/merge_requests/2745.
 22: */
 23: static void MyExit(void)
 24: {
 25:   _Exit(MPI_ERR_OTHER);
 26: }

 28: /*
 29:     PetscSignalHandler_Private - This is the signal handler called by the system. This calls
 30:              any signal handler set by PETSc or the application code.

 32:    Input Parameters: (depends on system)
 33: .    sig - integer code indicating the type of signal
 34: .    code - ??
 35: .    sigcontext - ??
 36: .    addr - ??

 38: */
 39: #if PetscDefined(HAVE_4ARG_SIGNAL_HANDLER)
 40: static void PetscSignalHandler_Private(int sig, int code, struct sigcontext *scp, char *addr)
 41: #else
 42: static void PetscSignalHandler_Private(int sig)
 43: #endif
 44: {
 45:   PetscErrorCode ierr;

 47:   if (!sh || !sh->handler) ierr = PetscSignalHandlerDefault(sig, NULL);
 48:   else {
 49:     if (sh->classid != SIGNAL_CLASSID) SETERRABORT(PETSC_COMM_WORLD, PETSC_ERR_COR, "Signal object has been corrupted");
 50:     ierr = (*sh->handler)(sig, sh->ctx);
 51:   }
 52:   if (ierr) PETSCABORT(PETSC_COMM_WORLD, PETSC_ERR_COR);
 53: }

 55: /*@
 56:   PetscSignalHandlerDefault - Default signal handler.

 58:   Not Collective

 60:   Input Parameters:
 61: + sig - signal value
 62: - ptr - unused pointer

 64:   Level: advanced

 66:   Developer Note:
 67:   This does not call `PetscError()`, it handles the entire error process, including possibly printing the traceback, directly

 69: .seealso: [](sec_errors), `PetscPushSignalHandler()`
 70: @*/
 71: PetscErrorCode PetscSignalHandlerDefault(int sig, void *ptr)
 72: {
 73:   const char *SIGNAME[64];

 75:   if (sig == SIGSEGV) PetscSignalSegvCheckPointerOrMpi();
 76:   SIGNAME[0] = "Unknown signal";
 77: #if !PetscDefined(MISSING_SIGABRT)
 78:   SIGNAME[SIGABRT] = "Abort";
 79: #endif
 80: #if !PetscDefined(MISSING_SIGALRM)
 81:   SIGNAME[SIGALRM] = "Alarm";
 82: #endif
 83: #if !PetscDefined(MISSING_SIGBUS)
 84:   SIGNAME[SIGBUS] = "BUS: Bus Error, possibly illegal memory access";
 85: #endif
 86: #if !PetscDefined(MISSING_SIGCHLD)
 87:   SIGNAME[SIGCHLD] = "CHLD";
 88: #endif
 89: #if !PetscDefined(MISSING_SIGCONT)
 90:   SIGNAME[SIGCONT] = "CONT";
 91: #endif
 92: #if !PetscDefined(MISSING_SIGFPE)
 93:   SIGNAME[SIGFPE] = "FPE: Floating Point Exception,probably divide by zero";
 94: #endif
 95: #if !PetscDefined(MISSING_SIGHUP)
 96:   SIGNAME[SIGHUP] = "Hang up: Some other process (or the batch system) has told this process to end";
 97: #endif
 98: #if !PetscDefined(MISSING_SIGILL)
 99:   SIGNAME[SIGILL] = "Illegal instruction: Likely due to memory corruption";
100: #endif
101: #if !PetscDefined(MISSING_SIGINT)
102:   SIGNAME[SIGINT] = "Interrupt";
103: #endif
104: #if !PetscDefined(MISSING_SIGKILL)
105:   SIGNAME[SIGKILL] = "Kill: Some other process (or the batch system) has told this process to end";
106: #endif
107: #if !PetscDefined(MISSING_SIGPIPE)
108:   SIGNAME[SIGPIPE] = "Broken Pipe: Likely while reading or writing to a socket";
109: #endif
110: #if !PetscDefined(MISSING_SIGQUIT)
111:   SIGNAME[SIGQUIT] = "Quit: Some other process (or the batch system) has told this process to end";
112: #endif
113: #if !PetscDefined(MISSING_SIGSEGV)
114:   SIGNAME[SIGSEGV] = "SEGV: Segmentation Violation, probably memory access out of range";
115: #endif
116: #if !PetscDefined(MISSING_SIGSYS)
117:   SIGNAME[SIGSYS] = "SYS";
118: #endif
119: #if !PetscDefined(MISSING_SIGTERM)
120:   SIGNAME[SIGTERM] = "Terminate: Some process (or the batch system) has told this process to end";
121: #endif
122: #if !PetscDefined(MISSING_SIGTRAP)
123:   SIGNAME[SIGTRAP] = "TRAP";
124: #endif
125: #if !PetscDefined(MISSING_SIGTSTP)
126:   SIGNAME[SIGTSTP] = "TSTP";
127: #endif
128: #if !PetscDefined(MISSING_SIGURG)
129:   SIGNAME[SIGURG] = "URG";
130: #endif
131: #if !PetscDefined(MISSING_SIGUSR1)
132:   SIGNAME[SIGUSR1] = "User 1";
133: #endif
134: #if !PetscDefined(MISSING_SIGUSR2)
135:   SIGNAME[SIGUSR2] = "User 2";
136: #endif

138:   signal(sig, SIG_DFL);
139:   (void)PetscSleep(PetscGlobalRank % 4); /* prevent some jumbling of error messages from different ranks */
140:   (void)(*PetscErrorPrintf)("------------------------------------------------------------------------\n");
141:   if (sig >= 0 && sig <= 20) (void)(*PetscErrorPrintf)("Caught signal number %d %s\n", sig, SIGNAME[sig]);
142:   else (void)(*PetscErrorPrintf)("Caught signal\n");

144:   (void)(*PetscErrorPrintf)("Try option -start_in_debugger or -on_error_attach_debugger\n");
145:   (void)(*PetscErrorPrintf)("or see https://petsc.org/release/faq/#valgrind and https://petsc.org/release/faq/\n");
146:   if (PetscDefined(HAVE_CUDA)) (void)(*PetscErrorPrintf)("or try https://docs.nvidia.com/compute-sanitizer/ComputeSanitizer/index.html on NVIDIA CUDA systems to find memory corruption errors\n");
147:   if (PetscDefined(USE_DEBUG)) {
148:     (void)(*PetscErrorPrintf)("---------------------  Stack Frames ------------------------------------\n");
149:     (void)PetscStackView(PETSC_STDOUT);
150:   } else {
151:     (void)(*PetscErrorPrintf)("configure using --with-debugging=yes, recompile, link, and run \n");
152:     (void)(*PetscErrorPrintf)("to get more information on the crash.\n");
153:   }
154: #if !PetscDefined(MISSING_SIGBUS)
155:   if (sig == SIGSEGV || sig == SIGBUS) {
156: #else
157:   if (sig == SIGSEGV) {
158: #endif
159:     PetscBool debug;

161:     (void)PetscMallocGetDebug(&debug, NULL, NULL);
162:     if (debug) (void)PetscMallocValidate(__LINE__, PETSC_FUNCTION_NAME, __FILE__);
163:     else (void)(*PetscErrorPrintf)("Run with -malloc_debug to check if memory corruption is causing the crash.\n");
164:   }
165:   atexit(MyExit);
166:   PETSCABORT(PETSC_COMM_WORLD, PETSC_ERR_SIG);
167:   return PETSC_SUCCESS;
168: }

170: #if !defined(PETSC_SIGNAL_CAST)
171:   #define PETSC_SIGNAL_CAST
172: #endif

174: /*@
175:   PetscPushSignalHandler - Catches the usual fatal errors and
176:   calls a user-provided routine.

178:   Not Collective, No Fortran Support

180:   Input Parameters:
181: + routine - routine to call when a signal is received
182: - ctx     - optional context needed by the routine

184:   Level: developer

186:   Note:
187:   There is no way to return to a signal handler that was set directly by the user with the UNIX signal handler API or by
188:   the loader. That information is lost with the first call to `PetscPushSignalHandler()`

190: .seealso: [](sec_errors), `PetscPopSignalHandler()`, `PetscSignalHandlerDefault()`, `PetscPushErrorHandler()`
191: @*/
192: PetscErrorCode PetscPushSignalHandler(PetscErrorCode (*routine)(int, void *), PetscCtx ctx)
193: {
194:   struct SH *newsh;

196:   PetscFunctionBegin;
197:   if (!SIGNAL_CLASSID) {
198:     /* PetscCall(PetscClassIdRegister("Signal",&SIGNAL_CLASSID)); */
199:     SIGNAL_CLASSID = 19;
200:   }
201:   if (!SignalSet && routine) {
202:     /* Do not catch ABRT, CHLD, KILL */
203: #if !PetscDefined(MISSING_SIGALRM)
204:     /* signal(SIGALRM, PETSC_SIGNAL_CAST PetscSignalHandler_Private); */
205: #endif
206: #if !PetscDefined(MISSING_SIGBUS)
207:     signal(SIGBUS, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
208: #endif
209: #if !PetscDefined(MISSING_SIGCONT)
210:     /*signal(SIGCONT, PETSC_SIGNAL_CAST PetscSignalHandler_Private);*/
211: #endif
212: #if !PetscDefined(MISSING_SIGFPE)
213:     signal(SIGFPE, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
214: #endif
215: #if !PetscDefined(MISSING_SIGHUP) && PetscDefined(HAVE_STRUCT_SIGACTION)
216:     {
217:       struct sigaction action;
218:       sigaction(SIGHUP, NULL, &action);
219:       if (action.sa_handler == SIG_IGN) {
220:         PetscCall(PetscInfo(NULL, "SIGHUP previously set to ignore, therefore not changing its signal handler\n"));
221:       } else {
222:         signal(SIGHUP, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
223:       }
224:     }
225: #endif
226: #if !PetscDefined(MISSING_SIGILL)
227:     signal(SIGILL, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
228: #endif
229: #if !PetscDefined(MISSING_SIGINT)
230:     /* signal(SIGINT, PETSC_SIGNAL_CAST PetscSignalHandler_Private); */
231: #endif
232: #if !PetscDefined(MISSING_SIGPIPE)
233:     signal(SIGPIPE, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
234: #endif
235: #if !PetscDefined(MISSING_SIGQUIT)
236:     signal(SIGQUIT, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
237: #endif
238: #if !PetscDefined(MISSING_SIGSEGV)
239:     signal(SIGSEGV, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
240: #endif
241: #if !PetscDefined(MISSING_SIGSYS)
242:     signal(SIGSYS, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
243: #endif
244: #if !PetscDefined(MISSING_SIGTERM)
245:   #if !PetscDefined(HAVE_OPENMPI)
246:     /* Open MPI may use SIGTERM to close down all its ranks; we don't want to generate many confusing PETSc error messages in that case */
247:     signal(SIGTERM, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
248:   #endif
249: #endif
250: #if !PetscDefined(MISSING_SIGTRAP)
251:     signal(SIGTRAP, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
252: #endif
253: #if !PetscDefined(MISSING_SIGTSTP)
254:     /* signal(SIGTSTP,  PETSC_SIGNAL_CAST PetscSignalHandler_Private); */
255: #endif
256: #if !PetscDefined(MISSING_SIGURG)
257:     signal(SIGURG, PETSC_SIGNAL_CAST PetscSignalHandler_Private);
258: #endif
259: #if !PetscDefined(MISSING_SIGUSR1)
260:     /* signal(SIGUSR1, PETSC_SIGNAL_CAST PetscSignalHandler_Private); */
261: #endif
262: #if !PetscDefined(MISSING_SIGUSR2)
263:     /* signal(SIGUSR2, PETSC_SIGNAL_CAST PetscSignalHandler_Private); */
264: #endif
265:     SignalSet = PETSC_TRUE;
266:   }
267:   if (!routine) {
268: #if !PetscDefined(MISSING_SIGALRM)
269:     /* signal(SIGALRM, SIG_DFL); */
270: #endif
271: #if !PetscDefined(MISSING_SIGBUS)
272:     signal(SIGBUS, SIG_DFL);
273: #endif
274: #if !PetscDefined(MISSING_SIGCONT)
275:     /* signal(SIGCONT, SIG_DFL); */
276: #endif
277: #if !PetscDefined(MISSING_SIGFPE)
278:     signal(SIGFPE, SIG_DFL);
279: #endif
280: #if !PetscDefined(MISSING_SIGHUP)
281:     signal(SIGHUP, SIG_DFL);
282: #endif
283: #if !PetscDefined(MISSING_SIGILL)
284:     signal(SIGILL, SIG_DFL);
285: #endif
286: #if !PetscDefined(MISSING_SIGINT)
287:     /* signal(SIGINT,  SIG_DFL); */
288: #endif
289: #if !PetscDefined(MISSING_SIGPIPE)
290:     signal(SIGPIPE, SIG_DFL);
291: #endif
292: #if !PetscDefined(MISSING_SIGQUIT)
293:     signal(SIGQUIT, SIG_DFL);
294: #endif
295: #if !PetscDefined(MISSING_SIGSEGV)
296:     signal(SIGSEGV, SIG_DFL);
297: #endif
298: #if !PetscDefined(MISSING_SIGSYS)
299:     signal(SIGSYS, SIG_DFL);
300: #endif
301: #if !PetscDefined(MISSING_SIGTERM)
302:     signal(SIGTERM, SIG_DFL);
303: #endif
304: #if !PetscDefined(MISSING_SIGTRAP)
305:     signal(SIGTRAP, SIG_DFL);
306: #endif
307: #if !PetscDefined(MISSING_SIGTSTP)
308:     /* signal(SIGTSTP, SIG_DFL); */
309: #endif
310: #if !PetscDefined(MISSING_SIGURG)
311:     signal(SIGURG, SIG_DFL);
312: #endif
313: #if !PetscDefined(MISSING_SIGUSR1)
314:     /* signal(SIGUSR1, SIG_DFL); */
315: #endif
316: #if !PetscDefined(MISSING_SIGUSR2)
317:     /* signal(SIGUSR2, SIG_DFL); */
318: #endif
319:     SignalSet = PETSC_FALSE;
320:   }
321:   PetscCall(PetscNew(&newsh));
322:   if (sh) {
323:     PetscCheck(sh->classid == SIGNAL_CLASSID, PETSC_COMM_SELF, PETSC_ERR_COR, "Signal object has been corrupted");
324:     newsh->previous = sh;
325:   } else newsh->previous = NULL;
326:   newsh->handler = routine;
327:   newsh->ctx     = ctx;
328:   newsh->classid = SIGNAL_CLASSID;
329:   sh             = newsh;
330:   PetscFunctionReturn(PETSC_SUCCESS);
331: }

333: /*@
334:   PetscPopSignalHandler - Removes the last signal handler that was pushed.
335:   If no signal handlers are left on the stack it will remove the PETSc signal handler.
336:   (That is PETSc will no longer catch signals).

338:   Not Collective

340:   Level: developer

342:   Note:
343:   There is no way to return to a signal handler that was set directly by the user with the UNIX signal handler API or by
344:   the loader. That information is lost with the first call to `PetscPushSignalHandler()`

346: .seealso: [](sec_errors), `PetscPushSignalHandler()`
347: @*/
348: PetscErrorCode PetscPopSignalHandler(void)
349: {
350:   struct SH *tmp;

352:   PetscFunctionBegin;
353:   if (!sh) PetscFunctionReturn(PETSC_SUCCESS);
354:   PetscCheck(sh->classid == SIGNAL_CLASSID, PETSC_COMM_SELF, PETSC_ERR_COR, "Signal object has been corrupted");

356:   tmp = sh;
357:   sh  = sh->previous;
358:   PetscCall(PetscFree(tmp));
359:   if (!sh || !sh->handler) {
360: #if !PetscDefined(MISSING_SIGALRM)
361:     /* signal(SIGALRM, SIG_DFL); */
362: #endif
363: #if !PetscDefined(MISSING_SIGBUS)
364:     signal(SIGBUS, SIG_DFL);
365: #endif
366: #if !PetscDefined(MISSING_SIGCONT)
367:     /* signal(SIGCONT, SIG_DFL); */
368: #endif
369: #if !PetscDefined(MISSING_SIGFPE)
370:     signal(SIGFPE, SIG_DFL);
371: #endif
372: #if !PetscDefined(MISSING_SIGHUP)
373:     signal(SIGHUP, SIG_DFL);
374: #endif
375: #if !PetscDefined(MISSING_SIGILL)
376:     signal(SIGILL, SIG_DFL);
377: #endif
378: #if !PetscDefined(MISSING_SIGINT)
379:     /* signal(SIGINT,  SIG_DFL); */
380: #endif
381: #if !PetscDefined(MISSING_SIGPIPE)
382:     signal(SIGPIPE, SIG_DFL);
383: #endif
384: #if !PetscDefined(MISSING_SIGQUIT)
385:     signal(SIGQUIT, SIG_DFL);
386: #endif
387: #if !PetscDefined(MISSING_SIGSEGV)
388:     signal(SIGSEGV, SIG_DFL);
389: #endif
390: #if !PetscDefined(MISSING_SIGSYS)
391:     signal(SIGSYS, SIG_DFL);
392: #endif
393: #if !PetscDefined(MISSING_SIGTERM)
394:     signal(SIGTERM, SIG_DFL);
395: #endif
396: #if !PetscDefined(MISSING_SIGTRAP)
397:     signal(SIGTRAP, SIG_DFL);
398: #endif
399: #if !PetscDefined(MISSING_SIGTSTP)
400:     /* signal(SIGTSTP, SIG_DFL); */
401: #endif
402: #if !PetscDefined(MISSING_SIGURG)
403:     signal(SIGURG, SIG_DFL);
404: #endif
405: #if !PetscDefined(MISSING_SIGUSR1)
406:     /* signal(SIGUSR1, SIG_DFL); */
407: #endif
408: #if !PetscDefined(MISSING_SIGUSR2)
409:     /* signal(SIGUSR2, SIG_DFL); */
410: #endif
411:     SignalSet = PETSC_FALSE;
412:   } else {
413:     SignalSet = PETSC_TRUE;
414:   }
415:   PetscFunctionReturn(PETSC_SUCCESS);
416: }