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