Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge branch 'master' of github.com:mquinson/simgrid
[simgrid.git] / src / simix / smx_process.c
1 /* Copyright (c) 2007-2014. The SimGrid Team.
2  * All rights reserved.                                                     */
3
4 /* This program is free software; you can redistribute it and/or modify it
5  * under the terms of the license (GNU LGPL) which comes with this package. */
6
7 #include "smx_private.h"
8 #include "xbt/sysdep.h"
9 #include "xbt/log.h"
10 #include "xbt/dict.h"
11 #include "mc/mc.h"
12 #include "mc/mc_replay.h"
13 #include "mc/mc_client.h"
14
15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
16                                 "Logging specific to SIMIX (process)");
17
18 unsigned long simix_process_maxpid = 0;
19
20 /**
21  * \brief Returns the current agent.
22  *
23  * This functions returns the currently running SIMIX process.
24  *
25  * \return The SIMIX process
26  */
27 XBT_INLINE smx_process_t SIMIX_process_self(void)
28 {
29   smx_context_t self_context = SIMIX_context_self();
30
31   return self_context ? SIMIX_context_get_process(self_context) : NULL;
32 }
33
34 /**
35  * \brief Returns whether a process has pending asynchronous communications.
36  * \return true if there are asynchronous communications in this process
37  */
38 int SIMIX_process_has_pending_comms(smx_process_t process) {
39
40   return xbt_fifo_size(process->comms) > 0;
41 }
42
43 /**
44  * \brief Moves a process to the list of processes to destroy.
45  */
46 void SIMIX_process_cleanup(smx_process_t process)
47 {
48   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
49       process->name, process, process->waiting_synchro);
50
51   SIMIX_process_on_exit_runall(process);
52
53   /* Unregister from the kill timer if any */
54   if (process->kill_timer != NULL)
55           SIMIX_timer_remove(process->kill_timer);
56
57   xbt_os_mutex_acquire(simix_global->mutex);
58
59   /* cancel non-blocking communications */
60   smx_synchro_t synchro;
61   while ((synchro = xbt_fifo_pop(process->comms))) {
62
63     /* make sure no one will finish the comm after this process is destroyed,
64      * because src_proc or dst_proc would be an invalid pointer */
65     SIMIX_comm_cancel(synchro);
66
67     if (synchro->comm.src_proc == process) {
68       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
69           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
70       synchro->comm.src_proc = NULL;
71
72       /* I'm not supposed to destroy a detached comm from the sender side, */
73       if (!synchro->comm.detached)
74         SIMIX_comm_destroy(synchro);
75       else
76         XBT_DEBUG("Don't destroy it since it's a detached comm");
77
78     }
79     else if (synchro->comm.dst_proc == process){
80       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
81           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
82       synchro->comm.dst_proc = NULL;
83
84       if (synchro->comm.detached && synchro->comm.refcount == 1
85           && synchro->comm.src_proc != NULL) {
86         /* the comm will be freed right now, remove it from the sender */
87         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
88       }
89       SIMIX_comm_destroy(synchro);
90     }
91     else {
92       xbt_die("Communication synchro %p is in my list but I'm not the sender "
93           "or the receiver", synchro);
94     }
95   }
96
97   XBT_DEBUG("%p should not be run anymore",process);
98   xbt_swag_remove(process, simix_global->process_list);
99   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
100   xbt_swag_insert(process, simix_global->process_to_destroy);
101   process->context->iwannadie = 0;
102
103   xbt_os_mutex_release(simix_global->mutex);
104 }
105
106 /**
107  * Garbage collection
108  *
109  * Should be called some time to time to free the memory allocated for processes
110  * that have finished (or killed).
111  */
112 void SIMIX_process_empty_trash(void)
113 {
114   smx_process_t process = NULL;
115
116   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
117     XBT_DEBUG("Getting rid of %p",process);
118
119     SIMIX_context_free(process->context);
120
121     /* Free the exception allocated at creation time */
122     free(process->running_ctx);
123     xbt_dict_free(&process->properties);
124
125     xbt_fifo_free(process->comms);
126
127     xbt_dynar_free(&process->on_exit);
128
129     xbt_free(process->name);
130     xbt_free(process);
131   }
132 }
133
134 /**
135  * \brief Creates and runs the maestro process
136  */
137 void SIMIX_create_maestro_process()
138 {
139   smx_process_t maestro = NULL;
140
141   /* Create maestro process and intilialize it */
142   maestro = xbt_new0(s_smx_process_t, 1);
143   maestro->pid = simix_process_maxpid++;
144   maestro->ppid = -1;
145   maestro->name = (char *) "";
146   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
147   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
148   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
149   maestro->simcall.issuer = maestro;
150   simix_global->maestro_process = maestro;
151   return;
152 }
153 /**
154  * \brief Stops a process.
155  *
156  * Stops the process, execute all the registered on_exit functions,
157  * register it to the list of the process to restart if needed
158  * and stops its context.
159  */
160 void SIMIX_process_stop(smx_process_t arg) {
161   /* execute the on_exit functions */
162   SIMIX_process_on_exit_runall(arg);
163   /* Add the process to the list of process to restart, only if
164    * the host is down
165    */
166   if (arg->auto_restart && !SIMIX_host_get_state(arg->smx_host)) {
167     SIMIX_host_add_auto_restart_process(arg->smx_host,arg->name,arg->code, arg->data,
168                                         sg_host_name(arg->smx_host),
169                                         SIMIX_timer_get_date(arg->kill_timer),
170                                         arg->argc,arg->argv,arg->properties,
171                                         arg->auto_restart);
172   }
173   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_name(arg->smx_host));
174   /* stop the context */
175   SIMIX_context_stop(arg->context);
176 }
177
178 /**
179  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
180  * This function frees the argument.
181  * \return the process created
182  */
183 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
184
185   smx_process_t process;
186   simix_global->create_process_function(&process,
187                                         args->name,
188                                         args->code,
189                                         args->data,
190                                         args->hostname,
191                                         args->kill_time,
192                                         args->argc,
193                                         args->argv,
194                                         args->properties,
195                                         args->auto_restart,
196                                         NULL);
197   xbt_free(args);
198   return process;
199 }
200
201
202 void simcall_HANDLER_process_create(smx_simcall_t simcall,
203                           smx_process_t *process,
204                           const char *name,
205                           xbt_main_func_t code,
206                           void *data,
207                           const char *hostname,
208                           double kill_time,
209                           int argc, char **argv,
210                           xbt_dict_t properties,
211                           int auto_restart){
212   SIMIX_process_create(process, name, code, data, hostname,
213                        kill_time, argc, argv, properties, auto_restart,
214                        simcall->issuer);
215 }
216 /**
217  * \brief Internal function to create a process.
218  *
219  * This function actually creates the process.
220  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
221  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
222  *
223  * \return the process created
224  */
225 void SIMIX_process_create(smx_process_t *process,
226                           const char *name,
227                           xbt_main_func_t code,
228                           void *data,
229                           const char *hostname,
230                           double kill_time,
231                           int argc, char **argv,
232                           xbt_dict_t properties,
233                           int auto_restart,
234                           smx_process_t parent_process)
235 {
236   *process = NULL;
237   smx_host_t host = SIMIX_host_get_by_name(hostname);
238
239   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
240
241   if (!SIMIX_host_get_state(host)) {
242     int i;
243     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
244           hostname);
245     for (i = 0; i < argc; i++)
246       xbt_free(argv[i]);
247     xbt_free(argv);
248   }
249   else {
250     *process = xbt_new0(s_smx_process_t, 1);
251
252     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
253     /* Process data */
254     (*process)->pid = simix_process_maxpid++;
255     (*process)->name = xbt_strdup(name);
256     (*process)->smx_host = host;
257     (*process)->data = data;
258     (*process)->comms = xbt_fifo_new();
259     (*process)->simcall.issuer = *process;
260
261      if (parent_process) {
262        (*process)->ppid = SIMIX_process_get_PID(parent_process);
263      } else {
264        (*process)->ppid = -1;
265      }
266
267     /* Process data for auto-restart */
268     (*process)->auto_restart = auto_restart;
269     (*process)->code = code;
270     (*process)->argc = argc;
271     (*process)->argv = argv;
272
273
274     XBT_VERB("Create context %s", (*process)->name);
275     (*process)->context = SIMIX_context_new(code, argc, argv,
276       simix_global->cleanup_process_function, *process);
277
278     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
279     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
280
281     if(MC_is_active()){
282       MC_ignore_heap((*process)->running_ctx, sizeof(*(*process)->running_ctx));
283     }
284
285     /* Add properties */
286     (*process)->properties = properties;
287
288     /* Add the process to it's host process list */
289     xbt_swag_insert(*process, SIMIX_host_priv(host)->process_list);
290
291     XBT_DEBUG("Start context '%s'", (*process)->name);
292
293     /* Now insert it in the global process list and in the process to run list */
294     xbt_swag_insert(*process, simix_global->process_list);
295     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, sg_host_name(host));
296     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
297
298     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
299       XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
300           sg_host_name((*process)->smx_host), kill_time);
301       (*process)->kill_timer = SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
302     }
303   }
304 }
305
306 /**
307  * \brief Executes the processes from simix_global->process_to_run.
308  *
309  * The processes of simix_global->process_to_run are run (in parallel if
310  * possible).  On exit, simix_global->process_to_run is empty, and
311  * simix_global->process_that_ran contains the list of processes that just ran.
312  * The two lists are swapped so, be careful when using them before and after a
313  * call to this function.
314  */
315 void SIMIX_process_runall(void)
316 {
317   SIMIX_context_runall();
318
319   xbt_dynar_t tmp = simix_global->process_that_ran;
320   simix_global->process_that_ran = simix_global->process_to_run;
321   simix_global->process_to_run = tmp;
322   xbt_dynar_reset(simix_global->process_to_run);
323 }
324
325 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
326   SIMIX_process_kill(process, simcall->issuer);
327 }
328 /**
329  * \brief Internal function to kill a SIMIX process.
330  *
331  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
332  * or directly for SIMIX internal purposes.
333  *
334  * \param process poor victim
335  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
336  */
337 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
338
339   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_name(process->smx_host));
340
341   process->context->iwannadie = 1;
342   process->blocked = 0;
343   process->suspended = 0;
344   process->doexception = 0;
345
346   /* destroy the blocking synchro if any */
347   if (process->waiting_synchro) {
348
349     switch (process->waiting_synchro->type) {
350
351     case SIMIX_SYNC_EXECUTE:
352     case SIMIX_SYNC_PARALLEL_EXECUTE:
353       SIMIX_host_execution_destroy(process->waiting_synchro);
354       break;
355
356     case SIMIX_SYNC_COMMUNICATE:
357       xbt_fifo_remove(process->comms, process->waiting_synchro);
358       SIMIX_comm_cancel(process->waiting_synchro);
359       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
360       SIMIX_comm_destroy(process->waiting_synchro);
361       break;
362
363     case SIMIX_SYNC_SLEEP:
364       SIMIX_process_sleep_destroy(process->waiting_synchro);
365       break;
366
367     case SIMIX_SYNC_JOIN:
368       SIMIX_process_sleep_destroy(process->waiting_synchro);
369       break;
370
371     case SIMIX_SYNC_SYNCHRO:
372       SIMIX_synchro_stop_waiting(process, &process->simcall);
373       SIMIX_synchro_destroy(process->waiting_synchro);
374       break;
375
376     case SIMIX_SYNC_IO:
377       SIMIX_io_destroy(process->waiting_synchro);
378       break;
379
380     }
381
382     process->waiting_synchro = NULL;
383   }
384   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
385     XBT_DEBUG("Inserting %s in the to_run list", process->name);
386     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
387   }
388
389 }
390
391 /** @brief Ask another process to raise the given exception
392  *
393  * @param cat category of exception
394  * @param value value associated to the exception
395  * @param msg string information associated to the exception
396  */
397 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
398   SMX_EXCEPTION(process, cat, value, msg);
399
400   if (process->suspended)
401     SIMIX_process_resume(process,SIMIX_process_self());
402
403   /* cancel the blocking synchro if any */
404   if (process->waiting_synchro) {
405
406     switch (process->waiting_synchro->type) {
407
408     case SIMIX_SYNC_EXECUTE:
409     case SIMIX_SYNC_PARALLEL_EXECUTE:
410       SIMIX_host_execution_cancel(process->waiting_synchro);
411       break;
412
413     case SIMIX_SYNC_COMMUNICATE:
414       xbt_fifo_remove(process->comms, process->waiting_synchro);
415       SIMIX_comm_cancel(process->waiting_synchro);
416       break;
417
418     case SIMIX_SYNC_SLEEP:
419     case SIMIX_SYNC_JOIN:
420       SIMIX_process_sleep_destroy(process->waiting_synchro);
421       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
422         XBT_DEBUG("Inserting %s in the to_run list", process->name);
423         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
424       }
425       break;
426
427     case SIMIX_SYNC_SYNCHRO:
428       SIMIX_synchro_stop_waiting(process, &process->simcall);
429       break;
430
431     case SIMIX_SYNC_IO:
432       SIMIX_io_destroy(process->waiting_synchro);
433       break;
434
435     }
436   }
437   process->waiting_synchro = NULL;
438
439 }
440
441 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
442   SIMIX_process_killall(simcall->issuer, reset_pid);
443 }
444 /**
445  * \brief Kills all running processes.
446  * \param issuer this one will not be killed
447  */
448 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
449 {
450   smx_process_t p = NULL;
451
452   while ((p = xbt_swag_extract(simix_global->process_list))) {
453     if (p != issuer) {
454       SIMIX_process_kill(p,issuer);
455     }
456   }
457
458   if (reset_pid > 0)
459     simix_process_maxpid = reset_pid;
460
461   SIMIX_context_runall();
462
463   SIMIX_process_empty_trash();
464 }
465
466 void simcall_HANDLER_process_change_host(smx_simcall_t simcall, smx_process_t process,
467                                    smx_host_t dest)
468 {
469   process->new_host = dest;
470 }
471 void SIMIX_process_change_host(smx_process_t process,
472              smx_host_t dest)
473 {
474   xbt_assert((process != NULL), "Invalid parameters");
475   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
476   process->smx_host = dest;
477   xbt_swag_insert(process, SIMIX_host_priv(dest)->process_list);
478 }
479
480
481 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
482 {
483   smx_synchro_t sync_suspend =
484       SIMIX_process_suspend(process, simcall->issuer);
485
486   if (process != simcall->issuer) {
487     SIMIX_simcall_answer(simcall);
488   } else {
489     xbt_fifo_push(sync_suspend->simcalls, simcall);
490     process->waiting_synchro = sync_suspend;
491     SIMIX_host_execution_suspend(process->waiting_synchro);
492   }
493   /* If we are suspending ourselves, then just do not finish the simcall now */
494 }
495
496 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
497 {
498   xbt_assert((process != NULL), "Invalid parameters");
499
500   if (process->suspended) {
501     XBT_DEBUG("Process '%s' is already suspended", process->name);
502     return NULL;
503   }
504
505   process->suspended = 1;
506
507   /* If we are suspending another process, and it is waiting on a sync,
508      suspend its synchronization. */
509   if (process != issuer) {
510
511     if (process->waiting_synchro) {
512
513       switch (process->waiting_synchro->type) {
514
515         case SIMIX_SYNC_EXECUTE:
516         case SIMIX_SYNC_PARALLEL_EXECUTE:
517           SIMIX_host_execution_suspend(process->waiting_synchro);
518           break;
519
520         case SIMIX_SYNC_COMMUNICATE:
521           SIMIX_comm_suspend(process->waiting_synchro);
522           break;
523
524         case SIMIX_SYNC_SLEEP:
525           SIMIX_process_sleep_suspend(process->waiting_synchro);
526           break;
527
528         case SIMIX_SYNC_SYNCHRO:
529           /* Suspension is delayed to when the process is rescheduled. */
530           break;
531
532         default:
533           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
534               (int)process->waiting_synchro->type);
535       }
536       return NULL;
537     } else {
538       /* Suspension is delayed to when the process is rescheduled. */
539       return NULL;
540     }
541   } else {
542     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
543     return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0, 0.0, 0);
544   }
545 }
546
547 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
548   SIMIX_process_resume(process, simcall->issuer);
549 }
550
551 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
552 {
553   XBT_IN("process = %p, issuer = %p", process, issuer);
554
555   if(process->context->iwannadie) {
556     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
557     return;
558   }
559
560   if(!process->suspended) return;
561   process->suspended = 0;
562
563   /* If we are resuming another process, resume the synchronization it was waiting for
564      if any. Otherwise add it to the list of process to run in the next round. */
565   if (process != issuer) {
566
567     if (process->waiting_synchro) {
568
569       switch (process->waiting_synchro->type) {
570
571         case SIMIX_SYNC_EXECUTE:
572         case SIMIX_SYNC_PARALLEL_EXECUTE:
573           SIMIX_host_execution_resume(process->waiting_synchro);
574           break;
575
576         case SIMIX_SYNC_COMMUNICATE:
577           SIMIX_comm_resume(process->waiting_synchro);
578           break;
579
580         case SIMIX_SYNC_SLEEP:
581           SIMIX_process_sleep_resume(process->waiting_synchro);
582           break;
583
584         case SIMIX_SYNC_SYNCHRO:
585           /* I cannot resume it now. This is delayed to when the process is rescheduled at
586            * the end of the synchro. */
587           break;
588
589         default:
590           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
591               (int)process->waiting_synchro->type);
592       }
593     }
594   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
595
596   XBT_OUT();
597 }
598
599 int SIMIX_process_get_maxpid(void) {
600   return simix_process_maxpid;
601 }
602
603 int SIMIX_process_count(void)
604 {
605   return xbt_swag_size(simix_global->process_list);
606 }
607
608 int SIMIX_process_get_PID(smx_process_t self){
609   if (self == NULL)
610     return 0;
611   else
612     return self->pid;
613 }
614
615 int SIMIX_process_get_PPID(smx_process_t self){
616   if (self == NULL)
617     return 0;
618   else
619     return self->ppid;
620 }
621
622 void* SIMIX_process_self_get_data(smx_process_t self)
623 {
624   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
625
626   if (!self) {
627     return NULL;
628   }
629   return SIMIX_process_get_data(self);
630 }
631
632 void SIMIX_process_self_set_data(smx_process_t self, void *data)
633 {
634   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
635
636   SIMIX_process_set_data(self, data);
637 }
638
639 void* SIMIX_process_get_data(smx_process_t process)
640 {
641   return process->data;
642 }
643
644 void SIMIX_process_set_data(smx_process_t process, void *data)
645 {
646   process->data = data;
647 }
648
649 smx_host_t SIMIX_process_get_host(smx_process_t process)
650 {
651   return process->smx_host;
652 }
653
654 /* needs to be public and without simcall because it is called
655    by exceptions and logging events */
656 const char* SIMIX_process_self_get_name(void) {
657
658   smx_process_t process = SIMIX_process_self();
659   if (process == NULL || process == simix_global->maestro_process)
660     return "";
661
662   return SIMIX_process_get_name(process);
663 }
664
665 const char* SIMIX_process_get_name(smx_process_t process)
666 {
667   return process->name;
668 }
669
670 smx_process_t SIMIX_process_get_by_name(const char* name)
671 {
672   smx_process_t proc;
673
674   xbt_swag_foreach(proc, simix_global->process_list)
675   {
676     if(!strcmp(name, proc->name))
677       return proc;
678   }
679   return NULL;
680 }
681
682 int SIMIX_process_is_suspended(smx_process_t process)
683 {
684   return process->suspended;
685 }
686
687 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
688 {
689   return process->properties;
690 }
691
692 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
693 {
694   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
695   xbt_fifo_push(sync->simcalls, simcall);
696   simcall->issuer->waiting_synchro = sync;
697 }
698
699 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
700   if (sync->sleep.surf_sleep) {
701     surf_action_cancel(sync->sleep.surf_sleep);
702
703     smx_simcall_t simcall;
704     while ((simcall = xbt_fifo_shift(sync->simcalls))) {
705       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
706       simcall->issuer->waiting_synchro = NULL;
707       if (simcall->issuer->suspended) {
708         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
709         simcall->issuer->suspended = 0;
710         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
711       } else {
712         SIMIX_simcall_answer(simcall);
713       }
714     }
715     surf_action_unref(sync->sleep.surf_sleep);
716     sync->sleep.surf_sleep = NULL;
717   }
718   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
719   return 0;
720 }
721
722 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
723 {
724   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
725   res->type = SIMIX_SYNC_JOIN;
726   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
727   return res;
728 }
729
730 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
731 {
732   if (MC_is_active() || MC_record_replay_is_active()) {
733     MC_process_clock_add(simcall->issuer, duration);
734     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
735     SIMIX_simcall_answer(simcall);
736     return;
737   }
738   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
739   xbt_fifo_push(sync->simcalls, simcall);
740   simcall->issuer->waiting_synchro = sync;
741 }
742
743 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
744 {
745   smx_synchro_t synchro;
746   smx_host_t host = process->smx_host;
747
748   /* check if the host is active */
749   if (surf_host_get_state(surf_host_resource_priv(host)) != SURF_RESOURCE_ON) {
750     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
751            sg_host_name(host));
752   }
753
754   synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
755   synchro->type = SIMIX_SYNC_SLEEP;
756   synchro->name = NULL;
757   synchro->category = NULL;
758
759   synchro->sleep.host = host;
760   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
761
762   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
763   XBT_DEBUG("Create sleep synchronization %p", synchro);
764
765   return synchro;
766 }
767
768 void SIMIX_post_process_sleep(smx_synchro_t synchro)
769 {
770   smx_simcall_t simcall;
771   e_smx_state_t state;
772   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
773
774   while ((simcall = xbt_fifo_shift(synchro->simcalls))) {
775
776     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
777       case SURF_ACTION_FAILED:
778         simcall->issuer->context->iwannadie = 1;
779         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
780         state = SIMIX_SRC_HOST_FAILURE;
781         break;
782
783       case SURF_ACTION_DONE:
784         state = SIMIX_DONE;
785         break;
786
787       default:
788         THROW_IMPOSSIBLE;
789         break;
790     }
791     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
792       simcall->issuer->context->iwannadie = 1;
793     }
794     simcall_process_sleep__set__result(simcall, state);
795     simcall->issuer->waiting_synchro = NULL;
796     if (simcall->issuer->suspended) {
797       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
798       simcall->issuer->suspended = 0;
799       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
800     } else {
801       SIMIX_simcall_answer(simcall);
802     }
803   }
804
805   SIMIX_process_sleep_destroy(synchro);
806 }
807
808 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
809 {
810   XBT_DEBUG("Destroy synchro %p", synchro);
811   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
812
813   if (synchro->sleep.surf_sleep) {
814     surf_action_unref(synchro->sleep.surf_sleep);
815     synchro->sleep.surf_sleep = NULL;
816   }
817   if (synchro->type == SIMIX_SYNC_SLEEP)
818     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
819 }
820
821 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
822 {
823   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
824   surf_action_suspend(synchro->sleep.surf_sleep);
825 }
826
827 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
828 {
829   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
830   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
831   surf_action_resume(synchro->sleep.surf_sleep);
832 }
833
834 /**
835  * \brief Calling this function makes the process to yield.
836  *
837  * Only the current process can call this function, giving back the control to
838  * maestro.
839  *
840  * \param self the current process
841  */
842 void SIMIX_process_yield(smx_process_t self)
843 {
844   XBT_DEBUG("Yield process '%s'", self->name);
845
846   /* Go into sleep and return control to maestro */
847   SIMIX_context_suspend(self->context);
848
849   /* Ok, maestro returned control to us */
850   XBT_DEBUG("Control returned to me: '%s'", self->name);
851
852   if (self->new_host) {
853     SIMIX_process_change_host(self, self->new_host);
854     self->new_host = NULL;
855   }
856
857   if (self->context->iwannadie){
858     XBT_DEBUG("I wanna die!");
859     SIMIX_process_stop(self);
860   }
861
862   if (self->suspended) {
863     XBT_DEBUG("Hey! I'm suspended.");
864     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
865     self->suspended = 0;
866     SIMIX_process_suspend(self, self);
867   }
868
869   if (self->doexception) {
870     XBT_DEBUG("Wait, maestro left me an exception");
871     self->doexception = 0;
872     SMX_THROW();
873   }
874
875 }
876
877 /* callback: context fetching */
878 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
879 {
880   return SIMIX_process_self()->running_ctx;
881 }
882
883 /* callback: termination */
884 void SIMIX_process_exception_terminate(xbt_ex_t * e)
885 {
886   xbt_ex_display(e);
887   xbt_abort();
888 }
889
890 smx_context_t SIMIX_process_get_context(smx_process_t p) {
891   return p->context;
892 }
893
894 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
895   p->context = c;
896 }
897
898 /**
899  * \brief Returns the list of processes to run.
900  */
901 xbt_dynar_t SIMIX_process_get_runnable(void)
902 {
903   return simix_global->process_to_run;
904 }
905
906 /**
907  * \brief Returns the process from PID.
908  */
909 smx_process_t SIMIX_process_from_PID(int PID)
910 {
911   smx_process_t proc;
912   xbt_swag_foreach(proc, simix_global->process_list)
913   {
914    if(proc->pid == PID)
915    return proc;
916   }
917   return NULL;
918 }
919
920 /** @brief returns a dynar containg all currently existing processes */
921 xbt_dynar_t SIMIX_processes_as_dynar(void) {
922   smx_process_t proc;
923   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
924   xbt_swag_foreach(proc, simix_global->process_list) {
925     xbt_dynar_push(res,&proc);
926   }
927   return res;
928 }
929
930
931 void SIMIX_process_on_exit_runall(smx_process_t process) {
932   s_smx_process_exit_fun_t exit_fun;
933   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
934                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
935   while (!xbt_dynar_is_empty(process->on_exit)) {
936     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
937     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
938   }
939 }
940
941 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
942   xbt_assert(process, "current process not found: are you in maestro context ?");
943
944   if (!process->on_exit) {
945     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
946   }
947
948   s_smx_process_exit_fun_t exit_fun = {fun, data};
949
950   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
951 }
952
953 /**
954  * \brief Sets the auto-restart status of the process.
955  * If set to 1, the process will be automatically restarted when its host
956  * comes back.
957  */
958 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
959   process->auto_restart = auto_restart;
960 }
961
962 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
963   return SIMIX_process_restart(process, simcall->issuer);
964 }
965 /** @brief Restart a process, starting it again from the beginning. */
966 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
967   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_name(process->smx_host));
968   //retrieve the arguments of the old process
969   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
970   s_smx_process_arg_t arg;
971   arg.code = process->code;
972   arg.hostname = sg_host_name(process->smx_host);
973   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
974   arg.argc = process->argc;
975   arg.data = process->data;
976   int i;
977   arg.argv = xbt_new(char*,process->argc + 1);
978   for (i = 0; i < arg.argc; i++) {
979     arg.argv[i] = xbt_strdup(process->argv[i]);
980   }
981   arg.argv[process->argc] = NULL;
982   arg.properties = NULL;
983   arg.auto_restart = process->auto_restart;
984   //kill the old process
985   SIMIX_process_kill(process,issuer);
986   //start the new process
987   smx_process_t new_process;
988   if (simix_global->create_process_function) {
989     simix_global->create_process_function(&new_process,
990                                           arg.argv[0],
991                                           arg.code,
992                                           arg.data,
993                                           arg.hostname,
994                                           arg.kill_time,
995                                           arg.argc,
996                                           arg.argv,
997                                           arg.properties,
998                                           arg.auto_restart,
999                                           NULL);
1000   } else {
1001     simcall_process_create(&new_process,
1002                            arg.argv[0],
1003                            arg.code,
1004                            arg.data,
1005                            arg.hostname,
1006                            arg.kill_time,
1007                            arg.argc,
1008                            arg.argv,
1009                            arg.properties,
1010                            arg.auto_restart);
1011
1012   }
1013   return new_process;
1014 }