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     if (process->running_ctx)
 104       free(process->running_ctx);
 105     if (process->properties)
 106       xbt_dict_free(&process->properties);
 107
 108     xbt_fifo_free(process->comms);
 109
 110     free(process->name);
 111     process->name = NULL;
 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
 194     XBT_VERB("Create context %s", (*process)->name);
 195     (*process)->context = SIMIX_context_new(code, argc, argv,
 196         simix_global->cleanup_process_function, *process);
 197
 198     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 199     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 200
 201     /* Add properties */
 202     (*process)->properties = properties;
 203
 204     /* Add the process to it's host process list */
 205     xbt_swag_insert(*process, host->process_list);
 206
 207     XBT_DEBUG("Start context '%s'", (*process)->name);
 208
 209     /* Now insert it in the global process list and in the process to run list */
 210     xbt_swag_insert(*process, simix_global->process_list);
 211     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
 212     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 213   }
 214 }
 215
 216 /**
 217  * \brief Executes the processes from simix_global->process_to_run.
 218  *
 219  * The processes of simix_global->process_to_run are run (in parallel if
 220  * possible).  On exit, simix_global->process_to_run is empty, and
 221  * simix_global->process_that_ran contains the list of processes that just ran.
 222  * The two lists are swapped so, be careful when using them before and after a
 223  * call to this function.
 224  */
 225 void SIMIX_process_runall(void)
 226 {
 227   SIMIX_context_runall();
 228
 229   xbt_dynar_t tmp = simix_global->process_that_ran;
 230   simix_global->process_that_ran = simix_global->process_to_run;
 231   simix_global->process_to_run = tmp;
 232   xbt_dynar_reset(simix_global->process_to_run);
 233 }
 234
 235 /**
 236  * \brief Internal function to kill a SIMIX process.
 237  *
 238  * This function may be called when a REQ_PROCESS_KILL request occurs,
 239  * or directly for SIMIX internal purposes.
 240  *
 241  * \param process poor victim
 242  */
 243 void SIMIX_process_kill(smx_process_t process) {
 244
 245   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 246
 247   process->context->iwannadie = 1;
 248   process->blocked = 0;
 249   process->suspended = 0;
 250   /* FIXME: set doexception to 0 also? */
 251
 252   /* destroy the blocking action if any */
 253   if (process->waiting_action) {
 254
 255     switch (process->waiting_action->type) {
 256
 257       case SIMIX_ACTION_EXECUTE:
 258       case SIMIX_ACTION_PARALLEL_EXECUTE:
 259         SIMIX_host_execution_destroy(process->waiting_action);
 260         break;
 261
 262       case SIMIX_ACTION_COMMUNICATE:
 263         SIMIX_comm_destroy(process->waiting_action);
 264         break;
 265
 266       case SIMIX_ACTION_SLEEP:
 267         SIMIX_process_sleep_destroy(process->waiting_action);
 268         break;
 269
 270       case SIMIX_ACTION_SYNCHRO:
 271         SIMIX_synchro_stop_waiting(process, &process->request);
 272         SIMIX_synchro_destroy(process->waiting_action);
 273         break;
 274
 275       case SIMIX_ACTION_IO:
 276         THROW_UNIMPLEMENTED;
 277         break;
 278     }
 279   }
 280
 281   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 282 }
 283
 284 /**
 285  * \brief Kills all running processes.
 286  * \param issuer this one will not be killed
 287  */
 288 void SIMIX_process_killall(smx_process_t issuer)
 289 {
 290   smx_process_t p = NULL;
 291
 292   while ((p = xbt_swag_extract(simix_global->process_list))) {
 293     if (p != issuer) {
 294       SIMIX_process_kill(p);
 295     }
 296   }
 297
 298   SIMIX_context_runall(simix_global->process_to_run);
 299
 300   SIMIX_process_empty_trash();
 301 }
 302
 303 void SIMIX_process_change_host(smx_process_t process,
 304                                smx_host_t dest)
 305 {
 306   xbt_assert((process != NULL), "Invalid parameters");
 307   xbt_swag_remove(process, process->smx_host->process_list);
 308   process->smx_host = dest;
 309   xbt_swag_insert(process, dest->process_list);
 310 }
 311
 312 void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
 313 {
 314   process->new_host = dest;
 315 }
 316
 317 void SIMIX_pre_process_suspend(smx_req_t req)
 318 {
 319   smx_process_t process = req->process_suspend.process;
 320   SIMIX_process_suspend(process, req->issuer);
 321
 322   if (process != req->issuer) {
 323     SIMIX_request_answer(req);
 324   }
 325   /* If we are suspending ourselves, then just do not replay the request. */
 326 }
 327
 328 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 329 {
 330   xbt_assert((process != NULL), "Invalid parameters");
 331
 332   if (process->suspended) {
 333     XBT_DEBUG("Process '%s' is already suspended", process->name);
 334     return;
 335   }
 336
 337   process->suspended = 1;
 338
 339   /* If we are suspending another process, and it is waiting on an action,
 340      suspend its action. */
 341   if (process != issuer) {
 342
 343     if (process->waiting_action) {
 344
 345       switch (process->waiting_action->type) {
 346
 347         case SIMIX_ACTION_EXECUTE:
 348         case SIMIX_ACTION_PARALLEL_EXECUTE:
 349           SIMIX_host_execution_suspend(process->waiting_action);
 350           break;
 351
 352         case SIMIX_ACTION_COMMUNICATE:
 353           SIMIX_comm_suspend(process->waiting_action);
 354           break;
 355
 356         case SIMIX_ACTION_SLEEP:
 357           SIMIX_process_sleep_suspend(process->waiting_action);
 358           break;
 359
 360         default:
 361           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
 362               process->waiting_action->type);
 363       }
 364     }
 365   }
 366 }
 367
 368 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 369 {
 370   xbt_assert((process != NULL), "Invalid parameters");
 371
 372   if (!process->suspended) {
 373     XBT_DEBUG("Process '%s' is not suspended", process->name);
 374     return;
 375   }
 376
 377   process->suspended = 0;
 378
 379   /* If we are resuming another process, resume the action it was waiting for
 380      if any. Otherwise add it to the list of process to run in the next round. */
 381   if (process != issuer) {
 382
 383     if (process->waiting_action) {
 384
 385       switch (process->waiting_action->type) {
 386
 387         case SIMIX_ACTION_EXECUTE:
 388         case SIMIX_ACTION_PARALLEL_EXECUTE:
 389           SIMIX_host_execution_resume(process->waiting_action);
 390           break;
 391
 392         case SIMIX_ACTION_COMMUNICATE:
 393           SIMIX_comm_resume(process->waiting_action);
 394           break;
 395
 396         case SIMIX_ACTION_SLEEP:
 397           SIMIX_process_sleep_resume(process->waiting_action);
 398           break;
 399
 400         default:
 401           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
 402               process->waiting_action->type);
 403       }
 404     }
 405     else {
 406       xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 407     }
 408   }
 409 }
 410
 411 int SIMIX_process_get_maxpid(void) {
 412   return simix_process_maxpid;
 413 }
 414
 415 int SIMIX_process_count(void)
 416 {
 417   return xbt_swag_size(simix_global->process_list);
 418 }
 419
 420 void* SIMIX_process_self_get_data(void)
 421 {
 422   smx_process_t me = SIMIX_process_self();
 423   if (!me) {
 424     return NULL;
 425   }
 426   return SIMIX_process_get_data(me);
 427 }
 428
 429 void SIMIX_process_self_set_data(void *data)
 430 {
 431   SIMIX_process_set_data(SIMIX_process_self(), data);
 432 }
 433
 434 void* SIMIX_process_get_data(smx_process_t process)
 435 {
 436   return process->data;
 437 }
 438
 439 void SIMIX_process_set_data(smx_process_t process, void *data)
 440 {
 441   process->data = data;
 442 }
 443
 444 smx_host_t SIMIX_process_get_host(smx_process_t process)
 445 {
 446   return process->smx_host;
 447 }
 448
 449 /* needs to be public and without request because it is called
 450    by exceptions and logging events */
 451 const char* SIMIX_process_self_get_name(void) {
 452
 453   smx_process_t process = SIMIX_process_self();
 454   if (process == NULL || process == simix_global->maestro_process)
 455     return "";
 456
 457   return SIMIX_process_get_name(process);
 458 }
 459
 460 const char* SIMIX_process_get_name(smx_process_t process)
 461 {
 462   return process->name;
 463 }
 464
 465 smx_process_t SIMIX_process_get_by_name(const char* name)
 466 {
 467   smx_process_t proc;
 468
 469   xbt_swag_foreach(proc, simix_global->process_list)
 470   {
 471     if(!strcmp(name, proc->name))
 472       return proc;
 473   }
 474   return NULL;
 475 }
 476
 477 int SIMIX_process_is_suspended(smx_process_t process)
 478 {
 479   return process->suspended;
 480 }
 481
 482 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 483 {
 484   return process->properties;
 485 }
 486
 487 void SIMIX_pre_process_sleep(smx_req_t req)
 488 {
 489   if (MC_IS_ENABLED) {
 490     MC_process_clock_add(req->issuer, req->process_sleep.duration);
 491     req->process_sleep.result = SIMIX_DONE;
 492     SIMIX_request_answer(req);
 493     return;
 494   }
 495   smx_action_t action = SIMIX_process_sleep(req->issuer, req->process_sleep.duration);
 496   xbt_fifo_push(action->request_list, req);
 497   req->issuer->waiting_action = action;
 498 }
 499
 500 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 501 {
 502   smx_action_t action;
 503   smx_host_t host = process->smx_host;
 504
 505   /* check if the host is active */
 506   if (surf_workstation_model->extension.
 507       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 508     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 509            host->name);
 510   }
 511
 512   action = xbt_mallocator_get(simix_global->action_mallocator);
 513   action->type = SIMIX_ACTION_SLEEP;
 514   action->name = NULL;
 515 #ifdef HAVE_TRACING
 516   action->category = NULL;
 517 #endif
 518
 519   action->sleep.host = host;
 520   action->sleep.surf_sleep =
 521       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 522
 523   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 524   XBT_DEBUG("Create sleep action %p", action);
 525
 526   return action;
 527 }
 528
 529 void SIMIX_post_process_sleep(smx_action_t action)
 530 {
 531   smx_req_t req;
 532   e_smx_state_t state;
 533
 534   while ((req = xbt_fifo_shift(action->request_list))) {
 535
 536     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 537       case SURF_ACTION_FAILED:
 538         state = SIMIX_SRC_HOST_FAILURE;
 539         break;
 540
 541       case SURF_ACTION_DONE:
 542         state = SIMIX_DONE;
 543         break;
 544
 545       default:
 546         THROW_IMPOSSIBLE;
 547         break;
 548     }
 549     req->process_sleep.result = state;
 550     req->issuer->waiting_action = NULL;
 551     SIMIX_request_answer(req);
 552   }
 553   SIMIX_process_sleep_destroy(action);
 554 }
 555
 556 void SIMIX_process_sleep_destroy(smx_action_t action)
 557 {
 558   XBT_DEBUG("Destroy action %p", action);
 559   if (action->sleep.surf_sleep)
 560     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 561   xbt_mallocator_release(simix_global->action_mallocator, action);
 562 }
 563
 564 void SIMIX_process_sleep_suspend(smx_action_t action)
 565 {
 566   surf_workstation_model->suspend(action->sleep.surf_sleep);
 567 }
 568
 569 void SIMIX_process_sleep_resume(smx_action_t action)
 570 {
 571   surf_workstation_model->resume(action->sleep.surf_sleep);
 572 }
 573
 574 /**
 575  * Calling this function makes the process to yield.
 576  * Only the processes can call this function, giving back the control to maestro
 577  */
 578 void SIMIX_process_yield(void)
 579 {
 580   smx_process_t self = SIMIX_process_self();
 581
 582   XBT_DEBUG("Yield process '%s'", self->name);
 583
 584   /* Go into sleep and return control to maestro */
 585   SIMIX_context_suspend(self->context);
 586
 587   /* Ok, maestro returned control to us */
 588   XBT_DEBUG("Control returned to me: '%s'", self->name);
 589
 590   if (self->context->iwannadie){
 591     XBT_DEBUG("I wanna die!");
 592     SIMIX_context_stop(self->context);
 593   }
 594
 595   if (self->doexception) {
 596     XBT_DEBUG("Wait, maestro left me an exception");
 597     self->doexception = 0;
 598     RETHROW;
 599   }
 600
 601   if (self->new_host) {
 602     SIMIX_process_change_host(self, self->new_host);
 603     self->new_host = NULL;
 604   }
 605 }
 606
 607 /* callback: context fetching */
 608 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 609 {
 610   return SIMIX_process_self()->running_ctx;
 611 }
 612
 613 /* callback: termination */
 614 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 615 {
 616   xbt_ex_display(e);
 617   abort();
 618 }
 619
 620 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 621   return p->context;
 622 }
 623
 624 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 625   p->context = c;
 626 }