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 "smx_private.h"
   8 #include "xbt/sysdep.h"
   9 #include "xbt/log.h"
  10 #include "xbt/dict.h"
  11 #include "mc/mc.h"
  12
  13 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  14                                 "Logging specific to SIMIX (process)");
  15
  16 unsigned long simix_process_maxpid = 0;
  17
  18 /**
  19  * \brief Returns the current agent.
  20  *
  21  * This functions returns the currently running SIMIX process.
  22  *
  23  * \return The SIMIX process
  24  */
  25 XBT_INLINE smx_process_t SIMIX_process_self(void)
  26 {
  27   smx_context_t self_context = SIMIX_context_self();
  28
  29   return self_context ? SIMIX_context_get_data(self_context) : NULL;
  30 }
  31
  32 /**
  33  * \brief Returns whether a process has pending asynchronous communications.
  34  * \return true if there are asynchronous communications in this process
  35  */
  36 int SIMIX_process_has_pending_comms(smx_process_t process) {
  37
  38   return xbt_fifo_size(process->comms) > 0;
  39 }
  40
  41 /**
  42  * \brief Moves a process to the list of processes to destroy.
  43  */
  44 void SIMIX_process_cleanup(smx_process_t process)
  45 {
  46   XBT_DEBUG("Cleanup process %s (%p), waiting action %p",
  47       process->name, process, process->waiting_action);
  48
  49   /* cancel non-blocking communications */
  50   smx_action_t action;
  51   while ((action = xbt_fifo_pop(process->comms))) {
  52
  53     /* make sure no one will finish the comm after this process is destroyed,
  54      * because src_proc or dst_proc would be an invalid pointer */
  55     SIMIX_comm_cancel(action);
  56
  57     if (action->comm.src_proc == process) {
  58       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  59           action, action->comm.detached, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
  60       action->comm.src_proc = NULL;
  61
  62       if (action->comm.detached) {
  63          if (action->comm.refcount == 0) {
  64            XBT_DEBUG("Increase the refcount before destroying it since it's detached");
  65            /* I'm not supposed to destroy a detached comm from the sender side,
  66             * unless there is no receiver matching the rdv */
  67            action->comm.refcount++;
  68            SIMIX_comm_destroy(action);
  69          }
  70          else {
  71            XBT_DEBUG("Don't destroy it since its refcount is %d", action->comm.refcount);
  72          }
  73       } else {
  74         SIMIX_comm_destroy(action);
  75       }
  76     }
  77     else if (action->comm.dst_proc == process){
  78       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  79           action, (int)action->state, action->comm.src_proc, action->comm.dst_proc);
  80       action->comm.dst_proc = NULL;
  81
  82       if (action->comm.detached && action->comm.refcount == 1
  83           && action->comm.src_proc != NULL) {
  84         /* the comm will be freed right now, remove it from the sender */
  85         xbt_fifo_remove(action->comm.src_proc->comms, action);
  86       }
  87       SIMIX_comm_destroy(action);
  88     }
  89     else {
  90       xbt_die("Communication action %p is in my list but I'm not the sender "
  91           "or the receiver", action);
  92     }
  93   }
  94
  95   /*xbt_swag_remove(process, simix_global->process_to_run);*/
  96   xbt_swag_remove(process, simix_global->process_list);
  97   xbt_swag_remove(process, process->smx_host->process_list);
  98   xbt_swag_insert(process, simix_global->process_to_destroy);
  99   process->context->iwannadie = 0;
 100 }
 101
 102 /**
 103  * Garbage collection
 104  *
 105  * Should be called some time to time to free the memory allocated for processes
 106  * that have finished (or killed).
 107  */
 108 void SIMIX_process_empty_trash(void)
 109 {
 110   smx_process_t process = NULL;
 111
 112   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
 113     SIMIX_context_free(process->context);
 114
 115     /* Free the exception allocated at creation time */
 116     free(process->running_ctx);
 117     xbt_dict_free(&process->properties);
 118
 119     xbt_fifo_free(process->comms);
 120
 121     free(process->name);
 122     free(process);
 123   }
 124 }
 125
 126 /**
 127  * \brief Creates and runs the maestro process
 128  */
 129 void SIMIX_create_maestro_process()
 130 {
 131   smx_process_t maestro = NULL;
 132
 133   /* Create maestro process and intilialize it */
 134   maestro = xbt_new0(s_smx_process_t, 1);
 135   maestro->pid = simix_process_maxpid++;
 136   maestro->name = (char *) "";
 137   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 138   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 139   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 140   maestro->simcall.issuer = maestro;
 141
 142   simix_global->maestro_process = maestro;
 143   return;
 144 }
 145
 146 /**
 147  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 148  * \return the process created
 149  */
 150 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 151
 152   smx_process_t process;
 153   simix_global->create_process_function(
 154       &process,
 155       args->name,
 156       args->code,
 157       args->data,
 158       args->hostname,
 159       args->argc,
 160       args->argv,
 161       args->properties);
 162
 163   return process;
 164 }
 165
 166 /**
 167  * \brief Internal function to create a process.
 168  *
 169  * This function actually creates the process.
 170  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 171  * or directly for SIMIX internal purposes.
 172  *
 173  * \return the process created
 174  */
 175 void SIMIX_process_create(smx_process_t *process,
 176                           const char *name,
 177                           xbt_main_func_t code,
 178                           void *data,
 179                           const char *hostname,
 180                           int argc, char **argv,
 181                           xbt_dict_t properties) {
 182
 183   *process = NULL;
 184   smx_host_t host = SIMIX_host_get_by_name(hostname);
 185
 186   XBT_DEBUG("Start process %s on host %s", name, hostname);
 187
 188   if (!SIMIX_host_get_state(host)) {
 189     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 190           hostname);
 191   }
 192   else {
 193     *process = xbt_new0(s_smx_process_t, 1);
 194
 195     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 196
 197     /* Process data */
 198     (*process)->pid = simix_process_maxpid++;
 199     (*process)->name = xbt_strdup(name);
 200     (*process)->smx_host = host;
 201     (*process)->data = data;
 202     (*process)->comms = xbt_fifo_new();
 203     (*process)->simcall.issuer = *process;
 204
 205     XBT_VERB("Create context %s", (*process)->name);
 206     (*process)->context = SIMIX_context_new(code, argc, argv,
 207         simix_global->cleanup_process_function, *process);
 208
 209     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 210     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 211
 212     /* Add properties */
 213     (*process)->properties = properties;
 214
 215     /* Add the process to it's host process list */
 216     xbt_swag_insert(*process, host->process_list);
 217
 218     XBT_DEBUG("Start context '%s'", (*process)->name);
 219
 220     /* Now insert it in the global process list and in the process to run list */
 221     xbt_swag_insert(*process, simix_global->process_list);
 222     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
 223     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 224   }
 225 }
 226
 227 /**
 228  * \brief Executes the processes from simix_global->process_to_run.
 229  *
 230  * The processes of simix_global->process_to_run are run (in parallel if
 231  * possible).  On exit, simix_global->process_to_run is empty, and
 232  * simix_global->process_that_ran contains the list of processes that just ran.
 233  * The two lists are swapped so, be careful when using them before and after a
 234  * call to this function.
 235  */
 236 void SIMIX_process_runall(void)
 237 {
 238   SIMIX_context_runall();
 239
 240   xbt_dynar_t tmp = simix_global->process_that_ran;
 241   simix_global->process_that_ran = simix_global->process_to_run;
 242   simix_global->process_to_run = tmp;
 243   xbt_dynar_reset(simix_global->process_to_run);
 244 }
 245
 246 /**
 247  * \brief Internal function to kill a SIMIX process.
 248  *
 249  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 250  * or directly for SIMIX internal purposes.
 251  *
 252  * \param process poor victim
 253  */
 254 void SIMIX_process_kill(smx_process_t process) {
 255
 256   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 257
 258   process->context->iwannadie = 1;
 259   process->blocked = 0;
 260   process->suspended = 0;
 261   /* FIXME: set doexception to 0 also? */
 262
 263   /* destroy the blocking action if any */
 264   if (process->waiting_action) {
 265
 266     switch (process->waiting_action->type) {
 267
 268       case SIMIX_ACTION_EXECUTE:
 269       case SIMIX_ACTION_PARALLEL_EXECUTE:
 270         SIMIX_host_execution_destroy(process->waiting_action);
 271         break;
 272
 273       case SIMIX_ACTION_COMMUNICATE:
 274         xbt_fifo_remove(process->comms, process->waiting_action);
 275         SIMIX_comm_destroy(process->waiting_action);
 276         break;
 277
 278       case SIMIX_ACTION_SLEEP:
 279         SIMIX_process_sleep_destroy(process->waiting_action);
 280         break;
 281
 282       case SIMIX_ACTION_SYNCHRO:
 283         SIMIX_synchro_stop_waiting(process, &process->simcall);
 284         SIMIX_synchro_destroy(process->waiting_action);
 285         break;
 286
 287       case SIMIX_ACTION_IO:
 288         SIMIX_io_destroy(process->waiting_action);
 289         break;
 290     }
 291   }
 292   if(!xbt_dynar_member(simix_global->process_to_run, &(process)))
 293     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 294 }
 295
 296 /**
 297  * \brief Kills all running processes.
 298  * \param issuer this one will not be killed
 299  */
 300 void SIMIX_process_killall(smx_process_t issuer)
 301 {
 302   smx_process_t p = NULL;
 303
 304   while ((p = xbt_swag_extract(simix_global->process_list))) {
 305     if (p != issuer) {
 306       SIMIX_process_kill(p);
 307     }
 308   }
 309
 310   SIMIX_context_runall();
 311
 312   SIMIX_process_empty_trash();
 313 }
 314
 315 void SIMIX_process_change_host(smx_process_t process,
 316                                smx_host_t dest)
 317 {
 318   xbt_assert((process != NULL), "Invalid parameters");
 319   xbt_swag_remove(process, process->smx_host->process_list);
 320   process->smx_host = dest;
 321   xbt_swag_insert(process, dest->process_list);
 322 }
 323
 324 void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
 325 {
 326   process->new_host = dest;
 327 }
 328
 329 void SIMIX_pre_process_suspend(smx_simcall_t simcall)
 330 {
 331   smx_process_t process = simcall->process_suspend.process;
 332   SIMIX_process_suspend(process, simcall->issuer);
 333
 334   if (process != simcall->issuer) {
 335     SIMIX_simcall_answer(simcall);
 336   }
 337   /* If we are suspending ourselves, then just do not finish the simcall now */
 338 }
 339
 340 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 341 {
 342   xbt_assert((process != NULL), "Invalid parameters");
 343
 344   if (process->suspended) {
 345     XBT_DEBUG("Process '%s' is already suspended", process->name);
 346     return;
 347   }
 348
 349   process->suspended = 1;
 350
 351   /* If we are suspending another process, and it is waiting on an action,
 352      suspend its action. */
 353   if (process != issuer) {
 354
 355     if (process->waiting_action) {
 356
 357       switch (process->waiting_action->type) {
 358
 359         case SIMIX_ACTION_EXECUTE:
 360         case SIMIX_ACTION_PARALLEL_EXECUTE:
 361           SIMIX_host_execution_suspend(process->waiting_action);
 362           break;
 363
 364         case SIMIX_ACTION_COMMUNICATE:
 365           SIMIX_comm_suspend(process->waiting_action);
 366           break;
 367
 368         case SIMIX_ACTION_SLEEP:
 369           SIMIX_process_sleep_suspend(process->waiting_action);
 370           break;
 371
 372         default:
 373           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
 374               (int)process->waiting_action->type);
 375       }
 376     }
 377   }
 378 }
 379
 380 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 381 {
 382   xbt_assert((process != NULL), "Invalid parameters");
 383
 384   if(process->context->iwannadie) return;
 385     process->suspended = 0;
 386
 387   /* If we are resuming another process, resume the action it was waiting for
 388      if any. Otherwise add it to the list of process to run in the next round. */
 389   if (process != issuer) {
 390
 391     if (process->waiting_action) {
 392
 393       switch (process->waiting_action->type) {
 394
 395         case SIMIX_ACTION_EXECUTE:
 396         case SIMIX_ACTION_PARALLEL_EXECUTE:
 397           SIMIX_host_execution_resume(process->waiting_action);
 398           break;
 399
 400         case SIMIX_ACTION_COMMUNICATE:
 401           SIMIX_comm_resume(process->waiting_action);
 402           break;
 403
 404         case SIMIX_ACTION_SLEEP:
 405           SIMIX_process_sleep_resume(process->waiting_action);
 406           break;
 407
 408         default:
 409           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
 410               (int)process->waiting_action->type);
 411       }
 412     }
 413     else {
 414       if(!xbt_dynar_member(simix_global->process_to_run, &(process)))
 415         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 416     }
 417   }
 418 }
 419
 420 int SIMIX_process_get_maxpid(void) {
 421   return simix_process_maxpid;
 422 }
 423
 424 int SIMIX_process_count(void)
 425 {
 426   return xbt_swag_size(simix_global->process_list);
 427 }
 428
 429 void* SIMIX_process_self_get_data(smx_process_t self)
 430 {
 431   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 432
 433   if (!self) {
 434     return NULL;
 435   }
 436   return SIMIX_process_get_data(self);
 437 }
 438
 439 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 440 {
 441   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 442
 443   SIMIX_process_set_data(self, data);
 444 }
 445
 446 void* SIMIX_process_get_data(smx_process_t process)
 447 {
 448   return process->data;
 449 }
 450
 451 void SIMIX_process_set_data(smx_process_t process, void *data)
 452 {
 453   process->data = data;
 454 }
 455
 456 smx_host_t SIMIX_process_get_host(smx_process_t process)
 457 {
 458   return process->smx_host;
 459 }
 460
 461 /* needs to be public and without simcall because it is called
 462    by exceptions and logging events */
 463 const char* SIMIX_process_self_get_name(void) {
 464
 465   smx_process_t process = SIMIX_process_self();
 466   if (process == NULL || process == simix_global->maestro_process)
 467     return "";
 468
 469   return SIMIX_process_get_name(process);
 470 }
 471
 472 const char* SIMIX_process_get_name(smx_process_t process)
 473 {
 474   return process->name;
 475 }
 476
 477 smx_process_t SIMIX_process_get_by_name(const char* name)
 478 {
 479   smx_process_t proc;
 480
 481   xbt_swag_foreach(proc, simix_global->process_list)
 482   {
 483     if(!strcmp(name, proc->name))
 484       return proc;
 485   }
 486   return NULL;
 487 }
 488
 489 int SIMIX_process_is_suspended(smx_process_t process)
 490 {
 491   return process->suspended;
 492 }
 493
 494 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 495 {
 496   return process->properties;
 497 }
 498
 499 void SIMIX_pre_process_sleep(smx_simcall_t simcall)
 500 {
 501   if (MC_IS_ENABLED) {
 502     MC_process_clock_add(simcall->issuer, simcall->process_sleep.duration);
 503     simcall->process_sleep.result = SIMIX_DONE;
 504     SIMIX_simcall_answer(simcall);
 505     return;
 506   }
 507   smx_action_t action = SIMIX_process_sleep(simcall->issuer, simcall->process_sleep.duration);
 508   xbt_fifo_push(action->simcalls, simcall);
 509   simcall->issuer->waiting_action = action;
 510 }
 511
 512 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 513 {
 514   smx_action_t action;
 515   smx_host_t host = process->smx_host;
 516
 517   /* check if the host is active */
 518   if (surf_workstation_model->extension.
 519       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 520     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 521            host->name);
 522   }
 523
 524   action = xbt_mallocator_get(simix_global->action_mallocator);
 525   action->type = SIMIX_ACTION_SLEEP;
 526   action->name = NULL;
 527 #ifdef HAVE_TRACING
 528   action->category = NULL;
 529 #endif
 530
 531   action->sleep.host = host;
 532   action->sleep.surf_sleep =
 533       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 534
 535   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 536   XBT_DEBUG("Create sleep action %p", action);
 537
 538   return action;
 539 }
 540
 541 void SIMIX_post_process_sleep(smx_action_t action)
 542 {
 543   smx_simcall_t simcall;
 544   e_smx_state_t state;
 545
 546   while ((simcall = xbt_fifo_shift(action->simcalls))) {
 547
 548     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 549       case SURF_ACTION_FAILED:
 550         state = SIMIX_SRC_HOST_FAILURE;
 551         break;
 552
 553       case SURF_ACTION_DONE:
 554         state = SIMIX_DONE;
 555         break;
 556
 557       default:
 558         THROW_IMPOSSIBLE;
 559         break;
 560     }
 561     simcall->process_sleep.result = state;
 562     simcall->issuer->waiting_action = NULL;
 563     SIMIX_simcall_answer(simcall);
 564   }
 565   SIMIX_process_sleep_destroy(action);
 566 }
 567
 568 void SIMIX_process_sleep_destroy(smx_action_t action)
 569 {
 570   XBT_DEBUG("Destroy action %p", action);
 571   if (action->sleep.surf_sleep)
 572     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 573   xbt_mallocator_release(simix_global->action_mallocator, action);
 574 }
 575
 576 void SIMIX_process_sleep_suspend(smx_action_t action)
 577 {
 578   surf_workstation_model->suspend(action->sleep.surf_sleep);
 579 }
 580
 581 void SIMIX_process_sleep_resume(smx_action_t action)
 582 {
 583   surf_workstation_model->resume(action->sleep.surf_sleep);
 584 }
 585
 586 /**
 587  * \brief Calling this function makes the process to yield.
 588  *
 589  * Only the current process can call this function, giving back the control to
 590  * maestro.
 591  *
 592  * \param self the current process
 593  */
 594 void SIMIX_process_yield(smx_process_t self)
 595 {
 596   XBT_DEBUG("Yield process '%s'", self->name);
 597
 598   /* Go into sleep and return control to maestro */
 599   SIMIX_context_suspend(self->context);
 600
 601   /* Ok, maestro returned control to us */
 602   XBT_DEBUG("Control returned to me: '%s'", self->name);
 603
 604   if (self->context->iwannadie){
 605     XBT_DEBUG("I wanna die!");
 606     SIMIX_context_stop(self->context);
 607   }
 608
 609   if (self->doexception) {
 610     XBT_DEBUG("Wait, maestro left me an exception");
 611     self->doexception = 0;
 612     SMX_THROW();
 613   }
 614
 615   if (self->new_host) {
 616     SIMIX_process_change_host(self, self->new_host);
 617     self->new_host = NULL;
 618   }
 619 }
 620
 621 /* callback: context fetching */
 622 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 623 {
 624   return SIMIX_process_self()->running_ctx;
 625 }
 626
 627 /* callback: termination */
 628 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 629 {
 630   xbt_ex_display(e);
 631   abort();
 632 }
 633
 634 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 635   return p->context;
 636 }
 637
 638 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 639   p->context = c;
 640 }
 641
 642 /**
 643  * \brief Returns the list of processes to run.
 644  */
 645 xbt_dynar_t SIMIX_process_get_runnable(void)
 646 {
 647   return simix_global->process_to_run;
 648 }
 649
 650 /**
 651  * \brief Returns the process from PID.
 652  */
 653 smx_process_t SIMIX_process_from_PID(int PID)
 654 {
 655         smx_process_t proc;
 656         xbt_swag_foreach(proc, simix_global->process_list)
 657         {
 658          if(proc->pid == PID)
 659          return proc;
 660         }
 661         return NULL;
 662 }
 663
 664 /** @brief returns a dynar containg all currently existing processes */
 665 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 666   smx_process_t proc;
 667   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 668   xbt_swag_foreach(proc, simix_global->process_list) {
 669     xbt_dynar_push(res,&proc);
 670   }
 671   return res;
 672 }