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