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