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(simix_global->process_to_run);
 218   xbt_dynar_t tmp = simix_global->process_that_ran;
 219   simix_global->process_that_ran = simix_global->process_to_run;
 220   simix_global->process_to_run = tmp;
 221   xbt_dynar_reset(simix_global->process_to_run);
 222 }
 223
 224 /**
 225  * \brief Internal function to kill a SIMIX process.
 226  *
 227  * This function may be called when a REQ_PROCESS_KILL request occurs,
 228  * or directly for SIMIX internal purposes.
 229  *
 230  * \param process poor victim
 231  */
 232 void SIMIX_process_kill(smx_process_t process) {
 233
 234   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 235
 236   process->context->iwannadie = 1;
 237   process->blocked = 0;
 238   process->suspended = 0;
 239   /* FIXME: set doexception to 0 also? */
 240
 241   /* destroy the blocking action if any */
 242   if (process->waiting_action) {
 243
 244     switch (process->waiting_action->type) {
 245
 246       case SIMIX_ACTION_EXECUTE:
 247       case SIMIX_ACTION_PARALLEL_EXECUTE:
 248         SIMIX_host_execution_destroy(process->waiting_action);
 249         break;
 250
 251       case SIMIX_ACTION_COMMUNICATE:
 252         SIMIX_comm_destroy(process->waiting_action);
 253         break;
 254
 255       case SIMIX_ACTION_SLEEP:
 256         SIMIX_process_sleep_destroy(process->waiting_action);
 257         break;
 258
 259       case SIMIX_ACTION_SYNCHRO:
 260         SIMIX_synchro_stop_waiting(process, &process->request);
 261         SIMIX_synchro_destroy(process->waiting_action);
 262         break;
 263
 264       case SIMIX_ACTION_IO:
 265         THROW_UNIMPLEMENTED;
 266         break;
 267     }
 268   }
 269
 270   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 271 }
 272
 273 /**
 274  * \brief Kills all running processes.
 275  * \param issuer this one will not be killed
 276  */
 277 void SIMIX_process_killall(smx_process_t issuer)
 278 {
 279   smx_process_t p = NULL;
 280
 281   while ((p = xbt_swag_extract(simix_global->process_list))) {
 282     if (p != issuer) {
 283       SIMIX_process_kill(p);
 284     }
 285   }
 286
 287   SIMIX_context_runall(simix_global->process_to_run);
 288
 289   SIMIX_process_empty_trash();
 290 }
 291
 292 void SIMIX_process_change_host(smx_process_t process,
 293                                smx_host_t dest)
 294 {
 295   xbt_assert((process != NULL), "Invalid parameters");
 296   xbt_swag_remove(process, process->smx_host->process_list);
 297   process->smx_host = dest;
 298   xbt_swag_insert(process, dest->process_list);
 299 }
 300
 301 void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
 302 {
 303   process->new_host = dest;
 304 }
 305
 306 void SIMIX_pre_process_suspend(smx_req_t req)
 307 {
 308   smx_process_t process = req->process_suspend.process;
 309   SIMIX_process_suspend(process, req->issuer);
 310
 311   if (process != req->issuer) {
 312     SIMIX_request_answer(req);
 313   }
 314   /* If we are suspending ourselves, then just do not replay the request. */
 315 }
 316
 317 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 318 {
 319   process->suspended = 1;
 320
 321   /* If we are suspending another process, and it is waiting on an action,
 322      suspend it's action. */
 323   if (process != issuer) {
 324
 325     if (process->waiting_action) {
 326
 327       switch (process->waiting_action->type) {
 328
 329         case SIMIX_ACTION_EXECUTE:
 330         case SIMIX_ACTION_PARALLEL_EXECUTE:
 331           SIMIX_host_execution_suspend(process->waiting_action);
 332           break;
 333
 334         case SIMIX_ACTION_COMMUNICATE:
 335           SIMIX_comm_suspend(process->waiting_action);
 336           break;
 337
 338         case SIMIX_ACTION_SLEEP:
 339           SIMIX_process_sleep_suspend(process->waiting_action);
 340           break;
 341
 342         default:
 343           THROW_IMPOSSIBLE;
 344       }
 345     }
 346   }
 347 }
 348
 349 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 350 {
 351   xbt_assert((process != NULL), "Invalid parameters");
 352
 353   process->suspended = 0;
 354
 355   /* If we are resuming another process, resume the action it was waiting for
 356      if any. Otherwise add it to the list of process to run in the next round. */
 357   if (process != issuer) {
 358
 359     if (process->waiting_action) {
 360
 361       switch (process->waiting_action->type) {
 362
 363         case SIMIX_ACTION_EXECUTE:
 364         case SIMIX_ACTION_PARALLEL_EXECUTE:
 365           SIMIX_host_execution_resume(process->waiting_action);
 366           break;
 367
 368         case SIMIX_ACTION_COMMUNICATE:
 369           SIMIX_comm_resume(process->waiting_action);
 370           break;
 371
 372         case SIMIX_ACTION_SLEEP:
 373           SIMIX_process_sleep_resume(process->waiting_action);
 374           break;
 375
 376         default:
 377           THROW_IMPOSSIBLE;
 378       }
 379     }
 380     else {
 381       xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 382     }
 383   }
 384 }
 385
 386 int SIMIX_process_get_maxpid(void) {
 387   return simix_process_maxpid;
 388 }
 389
 390 int SIMIX_process_count(void)
 391 {
 392   return xbt_swag_size(simix_global->process_list);
 393 }
 394
 395 void* SIMIX_process_self_get_data(void)
 396 {
 397   smx_process_t me = SIMIX_process_self();
 398   if (!me) {
 399     return NULL;
 400   }
 401   return SIMIX_process_get_data(me);
 402 }
 403
 404 void SIMIX_process_self_set_data(void *data)
 405 {
 406   SIMIX_process_set_data(SIMIX_process_self(), data);
 407 }
 408
 409 void* SIMIX_process_get_data(smx_process_t process)
 410 {
 411   return process->data;
 412 }
 413
 414 void SIMIX_process_set_data(smx_process_t process, void *data)
 415 {
 416   process->data = data;
 417 }
 418
 419 smx_host_t SIMIX_process_get_host(smx_process_t process)
 420 {
 421   return process->smx_host;
 422 }
 423
 424 /* needs to be public and without request because it is called
 425    by exceptions and logging events */
 426 const char* SIMIX_process_self_get_name(void) {
 427
 428   smx_process_t process = SIMIX_process_self();
 429   if (process == NULL || process == simix_global->maestro_process)
 430     return "";
 431
 432   return SIMIX_process_get_name(process);
 433 }
 434
 435 const char* SIMIX_process_get_name(smx_process_t process)
 436 {
 437   return process->name;
 438 }
 439
 440 smx_process_t SIMIX_process_get_by_name(const char* name)
 441 {
 442   smx_process_t proc;
 443
 444   xbt_swag_foreach(proc, simix_global->process_list)
 445   {
 446     if(!strcmp(name, proc->name))
 447       return proc;
 448   }
 449   return NULL;
 450 }
 451
 452 int SIMIX_process_is_suspended(smx_process_t process)
 453 {
 454   return process->suspended;
 455 }
 456
 457 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 458 {
 459   return process->properties;
 460 }
 461
 462 void SIMIX_pre_process_sleep(smx_req_t req)
 463 {
 464   if (MC_IS_ENABLED) {
 465     MC_process_clock_add(req->issuer, req->process_sleep.duration);
 466     req->process_sleep.result = SIMIX_DONE;
 467     SIMIX_request_answer(req);
 468     return;
 469   }
 470   smx_action_t action = SIMIX_process_sleep(req->issuer, req->process_sleep.duration);
 471   xbt_fifo_push(action->request_list, req);
 472   req->issuer->waiting_action = action;
 473 }
 474
 475 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 476 {
 477   smx_action_t action;
 478   smx_host_t host = process->smx_host;
 479
 480   /* check if the host is active */
 481   if (surf_workstation_model->extension.
 482       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 483     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 484            host->name);
 485   }
 486
 487   action = xbt_mallocator_get(simix_global->action_mallocator);
 488   action->type = SIMIX_ACTION_SLEEP;
 489   action->name = NULL;
 490 #ifdef HAVE_TRACING
 491   action->category = NULL;
 492 #endif
 493
 494   action->sleep.host = host;
 495   action->sleep.surf_sleep =
 496       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 497
 498   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 499   XBT_DEBUG("Create sleep action %p", action);
 500
 501   return action;
 502 }
 503
 504 void SIMIX_post_process_sleep(smx_action_t action)
 505 {
 506   smx_req_t req;
 507   e_smx_state_t state;
 508
 509   while ((req = xbt_fifo_shift(action->request_list))) {
 510
 511     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 512       case SURF_ACTION_FAILED:
 513         state = SIMIX_SRC_HOST_FAILURE;
 514         break;
 515
 516       case SURF_ACTION_DONE:
 517         state = SIMIX_DONE;
 518         break;
 519
 520       default:
 521         THROW_IMPOSSIBLE;
 522         break;
 523     }
 524     req->process_sleep.result = state;
 525     req->issuer->waiting_action = NULL;
 526     SIMIX_request_answer(req);
 527   }
 528   SIMIX_process_sleep_destroy(action);
 529 }
 530
 531 void SIMIX_process_sleep_destroy(smx_action_t action)
 532 {
 533   XBT_DEBUG("Destroy action %p", action);
 534   if (action->sleep.surf_sleep)
 535     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 536   xbt_mallocator_release(simix_global->action_mallocator, action);
 537 }
 538
 539 void SIMIX_process_sleep_suspend(smx_action_t action)
 540 {
 541   surf_workstation_model->suspend(action->sleep.surf_sleep);
 542 }
 543
 544 void SIMIX_process_sleep_resume(smx_action_t action)
 545 {
 546   surf_workstation_model->resume(action->sleep.surf_sleep);
 547 }
 548
 549 /**
 550  * Calling this function makes the process to yield.
 551  * Only the processes can call this function, giving back the control to maestro
 552  */
 553 void SIMIX_process_yield(void)
 554 {
 555   smx_process_t self = SIMIX_process_self();
 556
 557   XBT_DEBUG("Yield process '%s'", self->name);
 558
 559   /* Go into sleep and return control to maestro */
 560   SIMIX_context_suspend(self->context);
 561
 562   /* Ok, maestro returned control to us */
 563   XBT_DEBUG("Maestro returned control to me: '%s'", self->name);
 564
 565   if (self->context->iwannadie){
 566     XBT_DEBUG("I wanna die!");
 567     SIMIX_context_stop(self->context);
 568   }
 569
 570   if (self->doexception) {
 571     XBT_DEBUG("Wait, maestro left me an exception");
 572     self->doexception = 0;
 573     RETHROW;
 574   }
 575
 576   if (self->new_host) {
 577     SIMIX_process_change_host(self, self->new_host);
 578     self->new_host = NULL;
 579   }
 580 }
 581
 582 /* callback: context fetching */
 583 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 584 {
 585   return SIMIX_process_self()->running_ctx;
 586 }
 587
 588 /* callback: termination */
 589 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 590 {
 591   xbt_ex_display(e);
 592   abort();
 593 }
 594
 595 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 596   return p->context;
 597 }
 598
 599 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 600   p->context = c;
 601 }