src/simix/smx_process.c

   1 /* Copyright (c) 2007, 2008, 2009, 2010. 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 "private.h"
   8 #include "xbt/sysdep.h"
   9 #include "xbt/log.h"
  10 #include "xbt/dict.h"
  11 #include "msg/mailbox.h"
  12 #include "mc/mc.h"
  13
  14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  15                                 "Logging specific to SIMIX (process)");
  16
  17 unsigned long simix_process_maxpid = 0;
  18
  19 /**
  20  * \brief Returns the current agent.
  21  *
  22  * This functions returns the currently running SIMIX process.
  23  *
  24  * \return The SIMIX process
  25  */
  26 XBT_INLINE smx_process_t SIMIX_process_self(void)
  27 {
  28   smx_context_t self_context = SIMIX_context_self();
  29
  30   return self_context ? SIMIX_context_get_data(self_context) : NULL;
  31 }
  32
  33 /**
  34  * \brief Move a process to the list of processes to destroy.
  35  */
  36 void SIMIX_process_cleanup(smx_process_t process)
  37 {
  38   XBT_DEBUG("Cleanup process %s (%p), waiting action %p",
  39       process->name, process, process->waiting_action);
  40
  41   /* cancel non-blocking communications */
  42   smx_action_t action;
  43   while ((action = xbt_fifo_pop(process->comms))) {
  44
  45     /* make sure no one will finish the comm after this process is destroyed */
  46     SIMIX_comm_cancel(action);
  47
  48     if (action->comm.src_proc == process) {
  49       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d",
  50           action, action->comm.detached, action->state);
  51       action->comm.src_proc = NULL;
  52
  53       if (action->comm.detached) {
  54         /* the receiver was supposed to destroy the comm after completion,
  55          * but the comm will actually never finish */
  56         action->comm.refcount++;
  57       }
  58     }
  59     else if (action->comm.dst_proc == process){
  60       XBT_DEBUG("Found an unfinished recv comm %p, state %d", action, action->state);
  61       action->comm.dst_proc = NULL;
  62     }
  63     else {
  64       XBT_DEBUG("Strange, I'm not in comm %p, state = %d, src = %p, dst = %p", action,
  65           action->state, action->comm.src_proc, action->comm.dst_proc);
  66       THROW_IMPOSSIBLE;
  67     }
  68
  69     /* FIXME uncommenting this instruction crashes complex simulations
  70     SIMIX_comm_destroy(action); */
  71   }
  72
  73   /*xbt_swag_remove(process, simix_global->process_to_run);*/
  74   xbt_swag_remove(process, simix_global->process_list);
  75   xbt_swag_remove(process, process->smx_host->process_list);
  76   xbt_swag_insert(process, simix_global->process_to_destroy);
  77 }
  78
  79 /**
  80  * Garbage collection
  81  *
  82  * Should be called some time to time to free the memory allocated for processes
  83  * that have finished (or killed).
  84  */
  85 void SIMIX_process_empty_trash(void)
  86 {
  87   smx_process_t process = NULL;
  88
  89   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
  90     SIMIX_context_free(process->context);
  91
  92     /* Free the exception allocated at creation time */
  93     if (process->running_ctx)
  94       free(process->running_ctx);
  95     if (process->properties)
  96       xbt_dict_free(&process->properties);
  97
  98     xbt_fifo_free(process->comms);
  99
 100     free(process->name);
 101     process->name = NULL;
 102     free(process);
 103   }
 104 }
 105
 106 /**
 107  * \brief Creates and runs the maestro process
 108  */
 109 void SIMIX_create_maestro_process()
 110 {
 111   smx_process_t maestro = NULL;
 112
 113   /* Create maestro process and intilialize it */
 114   maestro = xbt_new0(s_smx_process_t, 1);
 115   maestro->pid = simix_process_maxpid++;
 116   maestro->name = (char *) "";
 117   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 118   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 119   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 120   maestro->request.issuer = maestro;
 121
 122   simix_global->maestro_process = maestro;
 123   return;
 124 }
 125
 126 /**
 127  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 128  * \return the process created
 129  */
 130 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 131
 132   smx_process_t process;
 133   simix_global->create_process_function(
 134       &process,
 135       args->name,
 136       args->code,
 137       args->data,
 138       args->hostname,
 139       args->argc,
 140       args->argv,
 141       args->properties);
 142
 143   return process;
 144 }
 145
 146 /**
 147  * \brief Internal function to create a process.
 148  *
 149  * This function actually creates the process.
 150  * It may be called when a REQ_PROCESS_CREATE request occurs,
 151  * or directly for SIMIX internal purposes.
 152  *
 153  * \return the process created
 154  */
 155 void SIMIX_process_create(smx_process_t *process,
 156                           const char *name,
 157                           xbt_main_func_t code,
 158                           void *data,
 159                           const char *hostname,
 160                           int argc, char **argv,
 161                           xbt_dict_t properties) {
 162
 163   *process = NULL;
 164   smx_host_t host = SIMIX_host_get_by_name(hostname);
 165
 166   XBT_DEBUG("Start process %s on host %s", name, hostname);
 167
 168   if (!SIMIX_host_get_state(host)) {
 169     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 170           hostname);
 171   }
 172   else {
 173     *process = xbt_new0(s_smx_process_t, 1);
 174
 175     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 176
 177     /* Process data */
 178     (*process)->pid = simix_process_maxpid++;
 179     (*process)->name = xbt_strdup(name);
 180     (*process)->smx_host = host;
 181     (*process)->data = data;
 182     (*process)->comms = xbt_fifo_new();
 183
 184     XBT_VERB("Create context %s", (*process)->name);
 185     (*process)->context = SIMIX_context_new(code, argc, argv,
 186         simix_global->cleanup_process_function, *process);
 187
 188     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 189     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 190
 191     /* Add properties */
 192     (*process)->properties = properties;
 193
 194     /* Add the process to it's host process list */
 195     xbt_swag_insert(*process, host->process_list);
 196
 197     XBT_DEBUG("Start context '%s'", (*process)->name);
 198
 199     /* Now insert it in the global process list and in the process to run list */
 200     xbt_swag_insert(*process, simix_global->process_list);
 201     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
 202     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 203   }
 204 }
 205
 206 /**
 207  * \brief Executes the processes from simix_global->process_to_run.
 208  *
 209  * The processes of simix_global->process_to_run are run (in parallel if
 210  * possible).  On exit, simix_global->process_to_run is empty, and
 211  * simix_global->process_that_ran contains the list of processes that just ran.
 212  * The two lists are swapped so, be careful when using them before and after a
 213  * call to this function.
 214  */
 215 void SIMIX_process_runall(void)
 216 {
 217   SIMIX_context_runall();
 218
 219   xbt_dynar_t tmp = simix_global->process_that_ran;
 220   simix_global->process_that_ran = simix_global->process_to_run;
 221   simix_global->process_to_run = tmp;
 222   xbt_dynar_reset(simix_global->process_to_run);
 223 }
 224
 225 /**
 226  * \brief Internal function to kill a SIMIX process.
 227  *
 228  * This function may be called when a REQ_PROCESS_KILL request occurs,
 229  * or directly for SIMIX internal purposes.
 230  *
 231  * \param process poor victim
 232  */
 233 void SIMIX_process_kill(smx_process_t process) {
 234
 235   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 236
 237   process->context->iwannadie = 1;
 238   process->blocked = 0;
 239   process->suspended = 0;
 240   /* FIXME: set doexception to 0 also? */
 241
 242   /* destroy the blocking action if any */
 243   if (process->waiting_action) {
 244
 245     switch (process->waiting_action->type) {
 246
 247       case SIMIX_ACTION_EXECUTE:
 248       case SIMIX_ACTION_PARALLEL_EXECUTE:
 249         SIMIX_host_execution_destroy(process->waiting_action);
 250         break;
 251
 252       case SIMIX_ACTION_COMMUNICATE:
 253         SIMIX_comm_destroy(process->waiting_action);
 254         break;
 255
 256       case SIMIX_ACTION_SLEEP:
 257         SIMIX_process_sleep_destroy(process->waiting_action);
 258         break;
 259
 260       case SIMIX_ACTION_SYNCHRO:
 261         SIMIX_synchro_stop_waiting(process, &process->request);
 262         SIMIX_synchro_destroy(process->waiting_action);
 263         break;
 264
 265       case SIMIX_ACTION_IO:
 266         THROW_UNIMPLEMENTED;
 267         break;
 268     }
 269   }
 270
 271   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 272 }
 273
 274 /**
 275  * \brief Kills all running processes.
 276  * \param issuer this one will not be killed
 277  */
 278 void SIMIX_process_killall(smx_process_t issuer)
 279 {
 280   smx_process_t p = NULL;
 281
 282   while ((p = xbt_swag_extract(simix_global->process_list))) {
 283     if (p != issuer) {
 284       SIMIX_process_kill(p);
 285     }
 286   }
 287
 288   SIMIX_context_runall(simix_global->process_to_run);
 289
 290   SIMIX_process_empty_trash();
 291 }
 292
 293 void SIMIX_process_change_host(smx_process_t process,
 294                                smx_host_t dest)
 295 {
 296   xbt_assert((process != NULL), "Invalid parameters");
 297   xbt_swag_remove(process, process->smx_host->process_list);
 298   process->smx_host = dest;
 299   xbt_swag_insert(process, dest->process_list);
 300 }
 301
 302 void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
 303 {
 304   process->new_host = dest;
 305 }
 306
 307 void SIMIX_pre_process_suspend(smx_req_t req)
 308 {
 309   smx_process_t process = req->process_suspend.process;
 310   SIMIX_process_suspend(process, req->issuer);
 311
 312   if (process != req->issuer) {
 313     SIMIX_request_answer(req);
 314   }
 315   /* If we are suspending ourselves, then just do not replay the request. */
 316 }
 317
 318 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 319 {
 320   xbt_assert((process != NULL), "Invalid parameters");
 321
 322   if (process->suspended) {
 323     XBT_DEBUG("Process '%s' is already suspended", process->name);
 324     return;
 325   }
 326
 327   process->suspended = 1;
 328
 329   /* If we are suspending another process, and it is waiting on an action,
 330      suspend its action. */
 331   if (process != issuer) {
 332
 333     if (process->waiting_action) {
 334
 335       switch (process->waiting_action->type) {
 336
 337         case SIMIX_ACTION_EXECUTE:
 338         case SIMIX_ACTION_PARALLEL_EXECUTE:
 339           SIMIX_host_execution_suspend(process->waiting_action);
 340           break;
 341
 342         case SIMIX_ACTION_COMMUNICATE:
 343           SIMIX_comm_suspend(process->waiting_action);
 344           break;
 345
 346         case SIMIX_ACTION_SLEEP:
 347           SIMIX_process_sleep_suspend(process->waiting_action);
 348           break;
 349
 350         default:
 351           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
 352               process->waiting_action->type);
 353       }
 354     }
 355   }
 356 }
 357
 358 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 359 {
 360   xbt_assert((process != NULL), "Invalid parameters");
 361
 362   if (!process->suspended) {
 363     XBT_DEBUG("Process '%s' is not suspended", process->name);
 364     return;
 365   }
 366
 367   process->suspended = 0;
 368
 369   /* If we are resuming another process, resume the action it was waiting for
 370      if any. Otherwise add it to the list of process to run in the next round. */
 371   if (process != issuer) {
 372
 373     if (process->waiting_action) {
 374
 375       switch (process->waiting_action->type) {
 376
 377         case SIMIX_ACTION_EXECUTE:
 378         case SIMIX_ACTION_PARALLEL_EXECUTE:
 379           SIMIX_host_execution_resume(process->waiting_action);
 380           break;
 381
 382         case SIMIX_ACTION_COMMUNICATE:
 383           SIMIX_comm_resume(process->waiting_action);
 384           break;
 385
 386         case SIMIX_ACTION_SLEEP:
 387           SIMIX_process_sleep_resume(process->waiting_action);
 388           break;
 389
 390         default:
 391           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
 392               process->waiting_action->type);
 393       }
 394     }
 395     else {
 396       xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 397     }
 398   }
 399 }
 400
 401 int SIMIX_process_get_maxpid(void) {
 402   return simix_process_maxpid;
 403 }
 404
 405 int SIMIX_process_count(void)
 406 {
 407   return xbt_swag_size(simix_global->process_list);
 408 }
 409
 410 void* SIMIX_process_self_get_data(void)
 411 {
 412   smx_process_t me = SIMIX_process_self();
 413   if (!me) {
 414     return NULL;
 415   }
 416   return SIMIX_process_get_data(me);
 417 }
 418
 419 void SIMIX_process_self_set_data(void *data)
 420 {
 421   SIMIX_process_set_data(SIMIX_process_self(), data);
 422 }
 423
 424 void* SIMIX_process_get_data(smx_process_t process)
 425 {
 426   return process->data;
 427 }
 428
 429 void SIMIX_process_set_data(smx_process_t process, void *data)
 430 {
 431   process->data = data;
 432 }
 433
 434 smx_host_t SIMIX_process_get_host(smx_process_t process)
 435 {
 436   return process->smx_host;
 437 }
 438
 439 /* needs to be public and without request because it is called
 440    by exceptions and logging events */
 441 const char* SIMIX_process_self_get_name(void) {
 442
 443   smx_process_t process = SIMIX_process_self();
 444   if (process == NULL || process == simix_global->maestro_process)
 445     return "";
 446
 447   return SIMIX_process_get_name(process);
 448 }
 449
 450 const char* SIMIX_process_get_name(smx_process_t process)
 451 {
 452   return process->name;
 453 }
 454
 455 smx_process_t SIMIX_process_get_by_name(const char* name)
 456 {
 457   smx_process_t proc;
 458
 459   xbt_swag_foreach(proc, simix_global->process_list)
 460   {
 461     if(!strcmp(name, proc->name))
 462       return proc;
 463   }
 464   return NULL;
 465 }
 466
 467 int SIMIX_process_is_suspended(smx_process_t process)
 468 {
 469   return process->suspended;
 470 }
 471
 472 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 473 {
 474   return process->properties;
 475 }
 476
 477 void SIMIX_pre_process_sleep(smx_req_t req)
 478 {
 479   if (MC_IS_ENABLED) {
 480     MC_process_clock_add(req->issuer, req->process_sleep.duration);
 481     req->process_sleep.result = SIMIX_DONE;
 482     SIMIX_request_answer(req);
 483     return;
 484   }
 485   smx_action_t action = SIMIX_process_sleep(req->issuer, req->process_sleep.duration);
 486   xbt_fifo_push(action->request_list, req);
 487   req->issuer->waiting_action = action;
 488 }
 489
 490 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 491 {
 492   smx_action_t action;
 493   smx_host_t host = process->smx_host;
 494
 495   /* check if the host is active */
 496   if (surf_workstation_model->extension.
 497       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 498     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 499            host->name);
 500   }
 501
 502   action = xbt_mallocator_get(simix_global->action_mallocator);
 503   action->type = SIMIX_ACTION_SLEEP;
 504   action->name = NULL;
 505 #ifdef HAVE_TRACING
 506   action->category = NULL;
 507 #endif
 508
 509   action->sleep.host = host;
 510   action->sleep.surf_sleep =
 511       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 512
 513   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 514   XBT_DEBUG("Create sleep action %p", action);
 515
 516   return action;
 517 }
 518
 519 void SIMIX_post_process_sleep(smx_action_t action)
 520 {
 521   smx_req_t req;
 522   e_smx_state_t state;
 523
 524   while ((req = xbt_fifo_shift(action->request_list))) {
 525
 526     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 527       case SURF_ACTION_FAILED:
 528         state = SIMIX_SRC_HOST_FAILURE;
 529         break;
 530
 531       case SURF_ACTION_DONE:
 532         state = SIMIX_DONE;
 533         break;
 534
 535       default:
 536         THROW_IMPOSSIBLE;
 537         break;
 538     }
 539     req->process_sleep.result = state;
 540     req->issuer->waiting_action = NULL;
 541     SIMIX_request_answer(req);
 542   }
 543   SIMIX_process_sleep_destroy(action);
 544 }
 545
 546 void SIMIX_process_sleep_destroy(smx_action_t action)
 547 {
 548   XBT_DEBUG("Destroy action %p", action);
 549   if (action->sleep.surf_sleep)
 550     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 551   xbt_mallocator_release(simix_global->action_mallocator, action);
 552 }
 553
 554 void SIMIX_process_sleep_suspend(smx_action_t action)
 555 {
 556   surf_workstation_model->suspend(action->sleep.surf_sleep);
 557 }
 558
 559 void SIMIX_process_sleep_resume(smx_action_t action)
 560 {
 561   surf_workstation_model->resume(action->sleep.surf_sleep);
 562 }
 563
 564 /**
 565  * Calling this function makes the process to yield.
 566  * Only the processes can call this function, giving back the control to maestro
 567  */
 568 void SIMIX_process_yield(void)
 569 {
 570   smx_process_t self = SIMIX_process_self();
 571
 572   XBT_DEBUG("Yield process '%s'", self->name);
 573
 574   /* Go into sleep and return control to maestro */
 575   SIMIX_context_suspend(self->context);
 576
 577   /* Ok, maestro returned control to us */
 578   XBT_DEBUG("Control returned to me: '%s'", self->name);
 579
 580   if (self->context->iwannadie){
 581     XBT_DEBUG("I wanna die!");
 582     SIMIX_context_stop(self->context);
 583   }
 584
 585   if (self->doexception) {
 586     XBT_DEBUG("Wait, maestro left me an exception");
 587     self->doexception = 0;
 588     RETHROW;
 589   }
 590
 591   if (self->new_host) {
 592     SIMIX_process_change_host(self, self->new_host);
 593     self->new_host = NULL;
 594   }
 595 }
 596
 597 /* callback: context fetching */
 598 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 599 {
 600   return SIMIX_process_self()->running_ctx;
 601 }
 602
 603 /* callback: termination */
 604 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 605 {
 606   xbt_ex_display(e);
 607   abort();
 608 }
 609
 610 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 611   return p->context;
 612 }
 613
 614 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 615   p->context = c;
 616 }