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   smx_action_t action_suspend =
 333       SIMIX_process_suspend(process, simcall->issuer);
 334
 335   if (process != simcall->issuer) {
 336     SIMIX_simcall_answer(simcall);
 337   } else {
 338     xbt_fifo_push(action_suspend->simcalls, simcall);
 339     process->waiting_action = action_suspend;
 340     SIMIX_host_execution_suspend(process->waiting_action);
 341   }
 342   /* If we are suspending ourselves, then just do not finish the simcall now */
 343 }
 344
 345 smx_action_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 346 {
 347   xbt_assert((process != NULL), "Invalid parameters");
 348
 349   if (process->suspended) {
 350     XBT_DEBUG("Process '%s' is already suspended", process->name);
 351     return NULL;
 352   }
 353
 354   process->suspended = 1;
 355
 356   /* If we are suspending another process, and it is waiting on an action,
 357      suspend its action. */
 358   if (process != issuer) {
 359
 360     if (process->waiting_action) {
 361
 362       switch (process->waiting_action->type) {
 363
 364         case SIMIX_ACTION_EXECUTE:
 365         case SIMIX_ACTION_PARALLEL_EXECUTE:
 366           SIMIX_host_execution_suspend(process->waiting_action);
 367           break;
 368
 369         case SIMIX_ACTION_COMMUNICATE:
 370           SIMIX_comm_suspend(process->waiting_action);
 371           break;
 372
 373         case SIMIX_ACTION_SLEEP:
 374           SIMIX_process_sleep_suspend(process->waiting_action);
 375           break;
 376
 377         default:
 378           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
 379               (int)process->waiting_action->type);
 380       }
 381       return NULL;
 382     } else {
 383       DIE_IMPOSSIBLE;
 384       return NULL;
 385     }
 386   } else {
 387     return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0);
 388   }
 389 }
 390
 391 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 392 {
 393   xbt_assert((process != NULL), "Invalid parameters");
 394
 395   if(process->context->iwannadie)
 396     return;
 397
 398   process->suspended = 0;
 399
 400   /* If we are resuming another process, resume the action it was waiting for
 401      if any. Otherwise add it to the list of process to run in the next round. */
 402   if (process != issuer) {
 403
 404     if (process->waiting_action) {
 405
 406       switch (process->waiting_action->type) {
 407
 408         case SIMIX_ACTION_EXECUTE:
 409         case SIMIX_ACTION_PARALLEL_EXECUTE:
 410           SIMIX_host_execution_resume(process->waiting_action);
 411           break;
 412
 413         case SIMIX_ACTION_COMMUNICATE:
 414           SIMIX_comm_resume(process->waiting_action);
 415           break;
 416
 417         case SIMIX_ACTION_SLEEP:
 418           SIMIX_process_sleep_resume(process->waiting_action);
 419           break;
 420
 421         default:
 422           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
 423               (int)process->waiting_action->type);
 424       }
 425     }
 426     else {
 427       DIE_IMPOSSIBLE;
 428 //      if(!xbt_dynar_member(simix_global->process_to_run, &(process)))
 429 //        xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 430     }
 431   }
 432 }
 433
 434 int SIMIX_process_get_maxpid(void) {
 435   return simix_process_maxpid;
 436 }
 437
 438 int SIMIX_process_count(void)
 439 {
 440   return xbt_swag_size(simix_global->process_list);
 441 }
 442
 443 void* SIMIX_process_self_get_data(smx_process_t self)
 444 {
 445   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 446
 447   if (!self) {
 448     return NULL;
 449   }
 450   return SIMIX_process_get_data(self);
 451 }
 452
 453 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 454 {
 455   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 456
 457   SIMIX_process_set_data(self, data);
 458 }
 459
 460 void* SIMIX_process_get_data(smx_process_t process)
 461 {
 462   return process->data;
 463 }
 464
 465 void SIMIX_process_set_data(smx_process_t process, void *data)
 466 {
 467   process->data = data;
 468 }
 469
 470 smx_host_t SIMIX_process_get_host(smx_process_t process)
 471 {
 472   return process->smx_host;
 473 }
 474
 475 /* needs to be public and without simcall because it is called
 476    by exceptions and logging events */
 477 const char* SIMIX_process_self_get_name(void) {
 478
 479   smx_process_t process = SIMIX_process_self();
 480   if (process == NULL || process == simix_global->maestro_process)
 481     return "";
 482
 483   return SIMIX_process_get_name(process);
 484 }
 485
 486 const char* SIMIX_process_get_name(smx_process_t process)
 487 {
 488   return process->name;
 489 }
 490
 491 smx_process_t SIMIX_process_get_by_name(const char* name)
 492 {
 493   smx_process_t proc;
 494
 495   xbt_swag_foreach(proc, simix_global->process_list)
 496   {
 497     if(!strcmp(name, proc->name))
 498       return proc;
 499   }
 500   return NULL;
 501 }
 502
 503 int SIMIX_process_is_suspended(smx_process_t process)
 504 {
 505   return process->suspended;
 506 }
 507
 508 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 509 {
 510   return process->properties;
 511 }
 512
 513 void SIMIX_pre_process_sleep(smx_simcall_t simcall)
 514 {
 515   if (MC_IS_ENABLED) {
 516     MC_process_clock_add(simcall->issuer, simcall->process_sleep.duration);
 517     simcall->process_sleep.result = SIMIX_DONE;
 518     SIMIX_simcall_answer(simcall);
 519     return;
 520   }
 521   smx_action_t action = SIMIX_process_sleep(simcall->issuer, simcall->process_sleep.duration);
 522   xbt_fifo_push(action->simcalls, simcall);
 523   simcall->issuer->waiting_action = action;
 524 }
 525
 526 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 527 {
 528   smx_action_t action;
 529   smx_host_t host = process->smx_host;
 530
 531   /* check if the host is active */
 532   if (surf_workstation_model->extension.
 533       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 534     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 535            host->name);
 536   }
 537
 538   action = xbt_mallocator_get(simix_global->action_mallocator);
 539   action->type = SIMIX_ACTION_SLEEP;
 540   action->name = NULL;
 541 #ifdef HAVE_TRACING
 542   action->category = NULL;
 543 #endif
 544
 545   action->sleep.host = host;
 546   action->sleep.surf_sleep =
 547       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 548
 549   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 550   XBT_DEBUG("Create sleep action %p", action);
 551
 552   return action;
 553 }
 554
 555 void SIMIX_post_process_sleep(smx_action_t action)
 556 {
 557   smx_simcall_t simcall;
 558   e_smx_state_t state;
 559
 560   while ((simcall = xbt_fifo_shift(action->simcalls))) {
 561
 562     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 563       case SURF_ACTION_FAILED:
 564         state = SIMIX_SRC_HOST_FAILURE;
 565         break;
 566
 567       case SURF_ACTION_DONE:
 568         state = SIMIX_DONE;
 569         break;
 570
 571       default:
 572         THROW_IMPOSSIBLE;
 573         break;
 574     }
 575     simcall->process_sleep.result = state;
 576     simcall->issuer->waiting_action = NULL;
 577     SIMIX_simcall_answer(simcall);
 578   }
 579   SIMIX_process_sleep_destroy(action);
 580 }
 581
 582 void SIMIX_process_sleep_destroy(smx_action_t action)
 583 {
 584   XBT_DEBUG("Destroy action %p", action);
 585   if (action->sleep.surf_sleep)
 586     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 587   xbt_mallocator_release(simix_global->action_mallocator, action);
 588 }
 589
 590 void SIMIX_process_sleep_suspend(smx_action_t action)
 591 {
 592   surf_workstation_model->suspend(action->sleep.surf_sleep);
 593 }
 594
 595 void SIMIX_process_sleep_resume(smx_action_t action)
 596 {
 597   surf_workstation_model->resume(action->sleep.surf_sleep);
 598 }
 599
 600 /**
 601  * \brief Calling this function makes the process to yield.
 602  *
 603  * Only the current process can call this function, giving back the control to
 604  * maestro.
 605  *
 606  * \param self the current process
 607  */
 608 void SIMIX_process_yield(smx_process_t self)
 609 {
 610   XBT_DEBUG("Yield process '%s'", self->name);
 611
 612   /* Go into sleep and return control to maestro */
 613   SIMIX_context_suspend(self->context);
 614
 615   /* Ok, maestro returned control to us */
 616   XBT_DEBUG("Control returned to me: '%s'", self->name);
 617
 618   if (self->context->iwannadie){
 619     XBT_DEBUG("I wanna die!");
 620     SIMIX_context_stop(self->context);
 621   }
 622
 623   if (self->doexception) {
 624     XBT_DEBUG("Wait, maestro left me an exception");
 625     self->doexception = 0;
 626     SMX_THROW();
 627   }
 628
 629   if (self->new_host) {
 630     SIMIX_process_change_host(self, self->new_host);
 631     self->new_host = NULL;
 632   }
 633 }
 634
 635 /* callback: context fetching */
 636 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 637 {
 638   return SIMIX_process_self()->running_ctx;
 639 }
 640
 641 /* callback: termination */
 642 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 643 {
 644   xbt_ex_display(e);
 645   abort();
 646 }
 647
 648 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 649   return p->context;
 650 }
 651
 652 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 653   p->context = c;
 654 }
 655
 656 /**
 657  * \brief Returns the list of processes to run.
 658  */
 659 xbt_dynar_t SIMIX_process_get_runnable(void)
 660 {
 661   return simix_global->process_to_run;
 662 }
 663
 664 /**
 665  * \brief Returns the process from PID.
 666  */
 667 smx_process_t SIMIX_process_from_PID(int PID)
 668 {
 669         smx_process_t proc;
 670         xbt_swag_foreach(proc, simix_global->process_list)
 671         {
 672          if(proc->pid == PID)
 673          return proc;
 674         }
 675         return NULL;
 676 }
 677
 678 /** @brief returns a dynar containg all currently existing processes */
 679 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 680   smx_process_t proc;
 681   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 682   xbt_swag_foreach(proc, simix_global->process_list) {
 683     xbt_dynar_push(res,&proc);
 684   }
 685   return res;
 686 }