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