src/simix/smx_process.cpp

   1 /* Copyright (c) 2007-2015. 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 "src/surf/surf_interface.hpp"
   8 #include "smx_private.h"
   9 #include "xbt/sysdep.h"
  10 #include "xbt/log.h"
  11 #include "xbt/dict.h"
  12 #include "mc/mc.h"
  13 #include "src/mc/mc_replay.h"
  14 #include "src/mc/Client.hpp"
  15 #include "src/msg/msg_private.h"
  16
  17 #include "src/simix/SynchroSleep.hpp"
  18 #include "src/simix/SynchroRaw.hpp"
  19 #include "src/simix/SynchroIo.hpp"
  20
  21 #ifdef HAVE_SMPI
  22 #include "src/smpi/private.h"
  23 #endif
  24
  25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
  26
  27 unsigned long simix_process_maxpid = 0;
  28
  29 /**
  30  * \brief Returns the current agent.
  31  *
  32  * This functions returns the currently running SIMIX process.
  33  *
  34  * \return The SIMIX process
  35  */
  36 smx_process_t SIMIX_process_self(void)
  37 {
  38   smx_context_t self_context = SIMIX_context_self();
  39
  40   return self_context ? self_context->process() : NULL;
  41 }
  42
  43 /**
  44  * \brief Returns whether a process has pending asynchronous communications.
  45  * \return true if there are asynchronous communications in this process
  46  */
  47 int SIMIX_process_has_pending_comms(smx_process_t process) {
  48
  49   return xbt_fifo_size(process->comms) > 0;
  50 }
  51
  52 /**
  53  * \brief Moves a process to the list of processes to destroy.
  54  */
  55 void SIMIX_process_cleanup(smx_process_t process)
  56 {
  57   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  58       process->name, process, process->waiting_synchro);
  59
  60   SIMIX_process_on_exit_runall(process);
  61
  62   /* Unregister from the kill timer if any */
  63   if (process->kill_timer != NULL)
  64       SIMIX_timer_remove(process->kill_timer);
  65
  66   xbt_os_mutex_acquire(simix_global->mutex);
  67
  68   /* cancel non-blocking communications */
  69   smx_synchro_t synchro;
  70   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  71     simgrid::simix::Comm *comm = static_cast<simgrid::simix::Comm*>(synchro);
  72
  73     /* make sure no one will finish the comm after this process is destroyed,
  74      * because src_proc or dst_proc would be an invalid pointer */
  75     comm->cancel();
  76
  77     if (comm->src_proc == process) {
  78       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  79           comm, comm->detached, (int)comm->state, comm->src_proc, comm->dst_proc);
  80       comm->src_proc = NULL;
  81
  82       /* I'm not supposed to destroy a detached comm from the sender side, */
  83       if (comm->detached)
  84         XBT_DEBUG("Don't destroy it since it's a detached comm and I'm the sender");
  85       else
  86         comm->unref();
  87
  88     }
  89     else if (comm->dst_proc == process){
  90       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  91           comm, (int)comm->state, comm->src_proc, comm->dst_proc);
  92       comm->dst_proc = NULL;
  93
  94       if (comm->detached && comm->refcount == 1 && comm->src_proc != NULL) {
  95         /* the comm will be freed right now, remove it from the sender */
  96         xbt_fifo_remove(comm->src_proc->comms, comm);
  97       }
  98       comm->unref();
  99     } else {
 100       xbt_die("Communication synchro %p is in my list but I'm not the sender nor the receiver", synchro);
 101     }
 102   }
 103
 104   XBT_DEBUG("%p should not be run anymore",process);
 105   xbt_swag_remove(process, simix_global->process_list);
 106   if (process->host)
 107     xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 108   xbt_swag_insert(process, simix_global->process_to_destroy);
 109   process->context->iwannadie = 0;
 110
 111   xbt_os_mutex_release(simix_global->mutex);
 112 }
 113
 114 /**
 115  * Garbage collection
 116  *
 117  * Should be called some time to time to free the memory allocated for processes
 118  * that have finished (or killed).
 119  */
 120 void SIMIX_process_empty_trash(void)
 121 {
 122   smx_process_t process = NULL;
 123
 124   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 125     XBT_DEBUG("Getting rid of %p",process);
 126
 127     delete process->context;
 128
 129     /* Free the exception allocated at creation time */
 130     free(process->running_ctx);
 131     xbt_dict_free(&process->properties);
 132
 133     xbt_fifo_free(process->comms);
 134
 135     xbt_dynar_free(&process->on_exit);
 136
 137     xbt_free(process->name);
 138     xbt_free(process);
 139   }
 140 }
 141
 142 namespace simgrid {
 143 namespace simix {
 144
 145 void create_maestro(std::function<void()> code)
 146 {
 147   smx_process_t maestro = NULL;
 148   /* Create maestro process and intilialize it */
 149   maestro = xbt_new0(s_smx_process_t, 1);
 150   maestro->pid = simix_process_maxpid++;
 151   maestro->ppid = -1;
 152   maestro->name = (char*) "";
 153   maestro->data = nullptr;
 154   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 155   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 156
 157   if (!code) {
 158     maestro->context = SIMIX_context_new(NULL, 0, nullptr, NULL, maestro);
 159   } else {
 160     if (!simix_global)
 161       xbt_die("simix is not initialized, please call MSG_init first");
 162     maestro->context =
 163       simix_global->context_factory->create_maestro(code, maestro);
 164   }
 165
 166   maestro->simcall.issuer = maestro;
 167   simix_global->maestro_process = maestro;
 168 }
 169
 170 }
 171 }
 172
 173 /**
 174  * \brief Creates and runs the maestro process
 175  */
 176 void SIMIX_maestro_create(void (*code)(void*), void* data)
 177 {
 178   simgrid::simix::create_maestro(std::bind(code, data));
 179 }
 180
 181 /**
 182  * \brief Stops a process.
 183  *
 184  * Stops the process, execute all the registered on_exit functions,
 185  * register it to the list of the process to restart if needed
 186  * and stops its context.
 187  */
 188 void SIMIX_process_stop(smx_process_t arg) {
 189   /* execute the on_exit functions */
 190   SIMIX_process_on_exit_runall(arg);
 191   /* Add the process to the list of process to restart, only if the host is down */
 192   if (arg->auto_restart && arg->host->isOff()) {
 193     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 194                                         sg_host_get_name(arg->host),
 195                                         SIMIX_timer_get_date(arg->kill_timer),
 196                                         arg->argc,arg->argv,arg->properties,
 197                                         arg->auto_restart);
 198   }
 199   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 200   arg->context->stop();
 201 }
 202
 203 /**
 204  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 205  * This function frees the argument.
 206  * \return the process created
 207  */
 208 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 209
 210   smx_process_t process = simix_global->create_process_function(
 211                                         args->name,
 212                                         args->code,
 213                                         args->data,
 214                                         args->hostname,
 215                                         args->kill_time,
 216                                         args->argc,
 217                                         args->argv,
 218                                         args->properties,
 219                                         args->auto_restart,
 220                                         NULL);
 221   xbt_free(args);
 222   return process;
 223 }
 224
 225
 226 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 227                           const char *name,
 228                           xbt_main_func_t code,
 229                           void *data,
 230                           const char *hostname,
 231                           double kill_time,
 232                           int argc, char **argv,
 233                           xbt_dict_t properties,
 234                           int auto_restart){
 235   return (void*)SIMIX_process_create(name, code, data, hostname,
 236                        kill_time, argc, argv, properties, auto_restart,
 237                        simcall->issuer);
 238 }
 239
 240 static void kill_process(void* process)
 241 {
 242   simix_global->kill_process_function((smx_process_t) process);
 243 }
 244
 245 /**
 246  * \brief Internal function to create a process.
 247  *
 248  * This function actually creates the process.
 249  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 250  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 251  *
 252  * \return the process created
 253  */
 254 smx_process_t SIMIX_process_create(
 255                           const char *name,
 256                           xbt_main_func_t code,
 257                           void *data,
 258                           const char *hostname,
 259                           double kill_time,
 260                           int argc, char **argv,
 261                           xbt_dict_t properties,
 262                           int auto_restart,
 263                           smx_process_t parent_process)
 264 {
 265   smx_process_t process = NULL;
 266   sg_host_t host = sg_host_by_name(hostname);
 267
 268   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 269
 270   if (host->isOff()) {
 271     int i;
 272     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 273           hostname);
 274     for (i = 0; i < argc; i++)
 275       xbt_free(argv[i]);
 276     xbt_free(argv);
 277   }
 278   else {
 279     process = xbt_new0(s_smx_process_t, 1);
 280
 281     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 282     /* Process data */
 283     process->pid = simix_process_maxpid++;
 284     process->name = xbt_strdup(name);
 285     process->host = host;
 286     process->data = data;
 287     process->comms = xbt_fifo_new();
 288     process->simcall.issuer = process;
 289     /* Initiliaze data segment to default value */
 290     SIMIX_segment_index_set(process, -1);
 291
 292      if (parent_process != NULL) {
 293        process->ppid = SIMIX_process_get_PID(parent_process);
 294        /* SMPI process have their own data segment and
 295           each other inherit from their father */
 296 #if HAVE_SMPI
 297        if(smpi_privatize_global_variables){
 298          if( parent_process->pid != 0){
 299            SIMIX_segment_index_set(process, parent_process->segment_index);
 300          } else {
 301            SIMIX_segment_index_set(process, process->pid - 1);
 302          }
 303        }
 304 #endif
 305      } else {
 306        process->ppid = -1;
 307      }
 308
 309     /* Process data for auto-restart */
 310     process->auto_restart = auto_restart;
 311     process->code = code;
 312     process->argc = argc;
 313     process->argv = argv;
 314
 315
 316     XBT_VERB("Create context %s", process->name);
 317     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 318
 319     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 320     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 321
 322     if(MC_is_active()){
 323       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 324     }
 325
 326     /* Add properties */
 327     process->properties = properties;
 328
 329     /* Add the process to it's host process list */
 330     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 331
 332     XBT_DEBUG("Start context '%s'", process->name);
 333
 334     /* Now insert it in the global process list and in the process to run list */
 335     xbt_swag_insert(process, simix_global->process_list);
 336     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 337     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 338
 339     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 340       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 341           sg_host_get_name(process->host), kill_time);
 342       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 343     }
 344
 345     /* Tracing the process creation */
 346     TRACE_msg_process_create(process->name, process->pid, process->host);
 347   }
 348   return process;
 349 }
 350
 351 smx_process_t SIMIX_process_attach(
 352   const char* name,
 353   void *data,
 354   const char* hostname,
 355   xbt_dict_t properties,
 356   smx_process_t parent_process)
 357 {
 358   // This is mostly a copy/paste from SIMIX_process_new(),
 359   // it'd be nice to share some code between those two functions.
 360
 361   sg_host_t host = sg_host_by_name(hostname);
 362   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 363
 364   if (host->isOff()) {
 365     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 366       name, hostname);
 367     return nullptr;
 368   }
 369
 370   smx_process_t process = xbt_new0(s_smx_process_t, 1);
 371   /* Process data */
 372   process->pid = simix_process_maxpid++;
 373   process->name = xbt_strdup(name);
 374   process->host = host;
 375   process->data = data;
 376   process->comms = xbt_fifo_new();
 377   process->simcall.issuer = process;
 378   process->ppid = -1;
 379   /* Initiliaze data segment to default value */
 380   SIMIX_segment_index_set(process, -1);
 381   if (parent_process != NULL) {
 382     process->ppid = SIMIX_process_get_PID(parent_process);
 383    /* SMPI process have their own data segment and
 384       each other inherit from their father */
 385   #if HAVE_SMPI
 386     if(smpi_privatize_global_variables){
 387       if(parent_process->pid != 0){
 388         SIMIX_segment_index_set(process, parent_process->segment_index);
 389       } else {
 390         SIMIX_segment_index_set(process, process->pid - 1);
 391       }
 392     }
 393   #endif
 394   }
 395
 396   /* Process data for auto-restart */
 397   process->auto_restart = false;
 398   process->code = nullptr;
 399   process->argc = 0;
 400   process->argv = nullptr;
 401
 402   XBT_VERB("Create context %s", process->name);
 403   if (!simix_global)
 404     xbt_die("simix is not initialized, please call MSG_init first");
 405   process->context = simix_global->context_factory->attach(
 406     simix_global->cleanup_process_function, process);
 407
 408   process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 409   XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 410
 411   if(MC_is_active()){
 412     MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 413   }
 414
 415   /* Add properties */
 416   process->properties = properties;
 417
 418   /* Add the process to it's host process list */
 419   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 420
 421   /* Now insert it in the global process list and in the process to run list */
 422   xbt_swag_insert(process, simix_global->process_list);
 423   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 424   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 425
 426   /* Tracing the process creation */
 427   TRACE_msg_process_create(process->name, process->pid, process->host);
 428
 429   auto context = dynamic_cast<simgrid::simix::AttachContext*>(process->context);
 430   if (!context)
 431     xbt_die("Not a suitable context");
 432
 433   context->attach_start();
 434   return process;
 435 }
 436
 437 void SIMIX_process_detach(void)
 438 {
 439   auto context = dynamic_cast<simgrid::simix::AttachContext*>(SIMIX_context_self());
 440   if (!context)
 441     xbt_die("Not a suitable context");
 442
 443   simix_global->cleanup_process_function(context->process());
 444
 445   // Let maestro ignore we are still alive:
 446   // xbt_swag_remove(context->process(), simix_global->process_list);
 447
 448   // TODDO, Remove from proces list:
 449   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 450
 451   context->attach_stop();
 452   // delete context;
 453 }
 454
 455 /**
 456  * \brief Executes the processes from simix_global->process_to_run.
 457  *
 458  * The processes of simix_global->process_to_run are run (in parallel if
 459  * possible).  On exit, simix_global->process_to_run is empty, and
 460  * simix_global->process_that_ran contains the list of processes that just ran.
 461  * The two lists are swapped so, be careful when using them before and after a
 462  * call to this function.
 463  */
 464 void SIMIX_process_runall(void)
 465 {
 466   SIMIX_context_runall();
 467
 468   xbt_dynar_t tmp = simix_global->process_that_ran;
 469   simix_global->process_that_ran = simix_global->process_to_run;
 470   simix_global->process_to_run = tmp;
 471   xbt_dynar_reset(simix_global->process_to_run);
 472 }
 473
 474 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 475   SIMIX_process_kill(process, simcall->issuer);
 476 }
 477 /**
 478  * \brief Internal function to kill a SIMIX process.
 479  *
 480  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 481  * or directly for SIMIX internal purposes.
 482  *
 483  * \param process poor victim
 484  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 485  */
 486 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 487
 488   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 489
 490   process->context->iwannadie = 1;
 491   process->blocked = 0;
 492   process->suspended = 0;
 493   process->doexception = 0;
 494
 495   /* destroy the blocking synchro if any */
 496   if (process->waiting_synchro) {
 497
 498     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 499     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 500     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 501     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 502     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 503
 504     if (exec != nullptr) {
 505       SIMIX_execution_destroy(process->waiting_synchro);
 506
 507     } else if (comm != nullptr) {
 508       xbt_fifo_remove(process->comms, process->waiting_synchro);
 509       comm->cancel();
 510       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 511       comm->unref();
 512
 513     } else if (sleep != nullptr) {
 514       SIMIX_process_sleep_destroy(process->waiting_synchro);
 515
 516     } else if (raw != nullptr) {
 517       SIMIX_synchro_stop_waiting(process, &process->simcall);
 518       SIMIX_synchro_destroy(process->waiting_synchro);
 519
 520     } else if (io != nullptr) {
 521       SIMIX_io_destroy(process->waiting_synchro);
 522     }
 523
 524     /*
 525     switch (process->waiting_synchro->type) {
 526     case SIMIX_SYNC_JOIN:
 527       SIMIX_process_sleep_destroy(process->waiting_synchro);
 528       break;
 529     } */
 530
 531     process->waiting_synchro = NULL;
 532   }
 533   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 534     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 535     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 536   }
 537
 538 }
 539
 540 /** @brief Ask another process to raise the given exception
 541  *
 542  * @param cat category of exception
 543  * @param value value associated to the exception
 544  * @param msg string information associated to the exception
 545  */
 546 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 547   SMX_EXCEPTION(process, cat, value, msg);
 548
 549   if (process->suspended)
 550     SIMIX_process_resume(process,SIMIX_process_self());
 551
 552   /* cancel the blocking synchro if any */
 553   if (process->waiting_synchro) {
 554
 555     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 556     if (exec != nullptr) {
 557       SIMIX_execution_cancel(process->waiting_synchro);
 558     }
 559
 560     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 561     if (comm != nullptr) {
 562       xbt_fifo_remove(process->comms, comm);
 563       comm->cancel();
 564     }
 565
 566     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 567     if (sleep != nullptr) {
 568       SIMIX_process_sleep_destroy(process->waiting_synchro);
 569       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 570         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 571         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 572       }
 573     }
 574
 575     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 576     if (raw != nullptr) {
 577       SIMIX_synchro_stop_waiting(process, &process->simcall);
 578     }
 579
 580     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 581     if (io != nullptr) {
 582       SIMIX_io_destroy(process->waiting_synchro);
 583     }
 584   }
 585   process->waiting_synchro = NULL;
 586
 587 }
 588
 589 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 590   SIMIX_process_killall(simcall->issuer, reset_pid);
 591 }
 592 /**
 593  * \brief Kills all running processes.
 594  * \param issuer this one will not be killed
 595  */
 596 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 597 {
 598   smx_process_t p = NULL;
 599
 600   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 601     if (p != issuer) {
 602       SIMIX_process_kill(p,issuer);
 603     }
 604   }
 605
 606   if (reset_pid > 0)
 607     simix_process_maxpid = reset_pid;
 608
 609   SIMIX_context_runall();
 610
 611   SIMIX_process_empty_trash();
 612 }
 613
 614 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 615 {
 616   process->new_host = dest;
 617 }
 618 void SIMIX_process_change_host(smx_process_t process,
 619              sg_host_t dest)
 620 {
 621   xbt_assert((process != NULL), "Invalid parameters");
 622   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 623   process->host = dest;
 624   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 625 }
 626
 627
 628 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 629 {
 630   smx_synchro_t sync_suspend =
 631       SIMIX_process_suspend(process, simcall->issuer);
 632
 633   if (process != simcall->issuer) {
 634     SIMIX_simcall_answer(simcall);
 635   } else {
 636     xbt_fifo_push(sync_suspend->simcalls, simcall);
 637     process->waiting_synchro = sync_suspend;
 638     SIMIX_execution_suspend(process->waiting_synchro);
 639   }
 640   /* If we are suspending ourselves, then just do not finish the simcall now */
 641 }
 642
 643 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 644 {
 645   if (process->suspended) {
 646     XBT_DEBUG("Process '%s' is already suspended", process->name);
 647     return NULL;
 648   }
 649
 650   process->suspended = 1;
 651
 652   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 653   if (process != issuer) {
 654
 655     if (process->waiting_synchro)
 656       process->waiting_synchro->suspend();
 657     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 658
 659     return NULL;
 660   } else {
 661     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 662     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 663   }
 664 }
 665
 666 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 667   SIMIX_process_resume(process, simcall->issuer);
 668 }
 669
 670 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 671 {
 672   XBT_IN("process = %p, issuer = %p", process, issuer);
 673
 674   if(process->context->iwannadie) {
 675     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 676     return;
 677   }
 678
 679   if(!process->suspended) return;
 680   process->suspended = 0;
 681
 682   /* If we are resuming another process, resume the synchronization it was waiting for
 683      if any. Otherwise add it to the list of process to run in the next round. */
 684   if (process != issuer) {
 685
 686     if (process->waiting_synchro) {
 687       process->waiting_synchro->resume();
 688     }
 689   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 690
 691   XBT_OUT();
 692 }
 693
 694 int SIMIX_process_get_maxpid(void) {
 695   return simix_process_maxpid;
 696 }
 697
 698 int SIMIX_process_count(void)
 699 {
 700   return xbt_swag_size(simix_global->process_list);
 701 }
 702
 703 int SIMIX_process_get_PID(smx_process_t self){
 704   if (self == NULL)
 705     return 0;
 706   else
 707     return self->pid;
 708 }
 709
 710 int SIMIX_process_get_PPID(smx_process_t self){
 711   if (self == NULL)
 712     return 0;
 713   else
 714     return self->ppid;
 715 }
 716
 717 void* SIMIX_process_self_get_data()
 718 {
 719   smx_process_t self = SIMIX_process_self();
 720
 721   if (!self) {
 722     return NULL;
 723   }
 724   return SIMIX_process_get_data(self);
 725 }
 726
 727 void SIMIX_process_self_set_data(void *data)
 728 {
 729   smx_process_t self = SIMIX_process_self();
 730
 731   SIMIX_process_set_data(self, data);
 732 }
 733
 734 void* SIMIX_process_get_data(smx_process_t process)
 735 {
 736   return process->data;
 737 }
 738
 739 void SIMIX_process_set_data(smx_process_t process, void *data)
 740 {
 741   process->data = data;
 742 }
 743
 744 sg_host_t SIMIX_process_get_host(smx_process_t process)
 745 {
 746   return process->host;
 747 }
 748
 749 xbt_main_func_t SIMIX_process_get_code(void){
 750   return SIMIX_process_self()->code;
 751 }
 752
 753 /* needs to be public and without simcall because it is called
 754    by exceptions and logging events */
 755 const char* SIMIX_process_self_get_name(void) {
 756
 757   smx_process_t process = SIMIX_process_self();
 758   if (process == NULL || process == simix_global->maestro_process)
 759     return "maestro";
 760
 761   return SIMIX_process_get_name(process);
 762 }
 763
 764 const char* SIMIX_process_get_name(smx_process_t process)
 765 {
 766   return process->name;
 767 }
 768
 769 smx_process_t SIMIX_process_get_by_name(const char* name)
 770 {
 771   smx_process_t proc;
 772
 773   xbt_swag_foreach(proc, simix_global->process_list)
 774   {
 775     if(!strcmp(name, proc->name))
 776       return proc;
 777   }
 778   return NULL;
 779 }
 780
 781 int SIMIX_process_is_suspended(smx_process_t process)
 782 {
 783   return process->suspended;
 784 }
 785
 786 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 787 {
 788   return process->properties;
 789 }
 790
 791 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 792 {
 793   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 794   xbt_fifo_push(sync->simcalls, simcall);
 795   simcall->issuer->waiting_synchro = sync;
 796 }
 797
 798 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t synchro){
 799   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 800
 801   if (sleep->surf_sleep) {
 802     sleep->surf_sleep->cancel();
 803
 804     smx_simcall_t simcall;
 805     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sleep->simcalls))) {
 806       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 807       simcall->issuer->waiting_synchro = NULL;
 808       if (simcall->issuer->suspended) {
 809         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 810         simcall->issuer->suspended = 0;
 811         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 812       } else {
 813         SIMIX_simcall_answer(simcall);
 814       }
 815     }
 816     sleep->surf_sleep->unref();
 817     sleep->surf_sleep = NULL;
 818   }
 819   delete sleep;
 820   return 0;
 821 }
 822
 823 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 824 {
 825   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 826   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 827   return res;
 828 }
 829
 830 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 831 {
 832   if (MC_is_active() || MC_record_replay_is_active()) {
 833     MC_process_clock_add(simcall->issuer, duration);
 834     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 835     SIMIX_simcall_answer(simcall);
 836     return;
 837   }
 838   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 839   xbt_fifo_push(sync->simcalls, simcall);
 840   simcall->issuer->waiting_synchro = sync;
 841 }
 842
 843 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 844 {
 845   sg_host_t host = process->host;
 846
 847   /* check if the host is active */
 848   if (host->isOff())
 849     THROWF(host_error, 0, "Host %s failed, you cannot call this function", sg_host_get_name(host));
 850
 851   simgrid::simix::Sleep *synchro = new simgrid::simix::Sleep();
 852   synchro->name = NULL;
 853
 854   synchro->host = host;
 855   synchro->surf_sleep = surf_host_sleep(host, duration);
 856   synchro->surf_sleep->setData(synchro);
 857   XBT_DEBUG("Create sleep synchronization %p", synchro);
 858
 859   return synchro;
 860 }
 861
 862 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 863 {
 864   smx_simcall_t simcall;
 865   e_smx_state_t state;
 866   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 867
 868   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 869
 870     switch (sleep->surf_sleep->getState()){
 871       case simgrid::surf::Action::State::failed:
 872         simcall->issuer->context->iwannadie = 1;
 873         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 874         state = SIMIX_SRC_HOST_FAILURE;
 875         break;
 876
 877       case simgrid::surf::Action::State::done:
 878         state = SIMIX_DONE;
 879         break;
 880
 881       default:
 882         THROW_IMPOSSIBLE;
 883         break;
 884     }
 885     if (simcall->issuer->host->isOff()) {
 886       simcall->issuer->context->iwannadie = 1;
 887     }
 888     simcall_process_sleep__set__result(simcall, state);
 889     simcall->issuer->waiting_synchro = NULL;
 890     if (simcall->issuer->suspended) {
 891       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 892       simcall->issuer->suspended = 0;
 893       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 894     } else {
 895       SIMIX_simcall_answer(simcall);
 896     }
 897   }
 898
 899   SIMIX_process_sleep_destroy(synchro);
 900 }
 901
 902 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 903 {
 904   XBT_DEBUG("Destroy synchro %p", synchro);
 905   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 906
 907   if (sleep->surf_sleep) {
 908     sleep->surf_sleep->unref();
 909     sleep->surf_sleep = NULL;
 910   }
 911 }
 912
 913 /**
 914  * \brief Calling this function makes the process to yield.
 915  *
 916  * Only the current process can call this function, giving back the control to
 917  * maestro.
 918  *
 919  * \param self the current process
 920  */
 921 void SIMIX_process_yield(smx_process_t self)
 922 {
 923   XBT_DEBUG("Yield process '%s'", self->name);
 924
 925   /* Go into sleep and return control to maestro */
 926   self->context->suspend();
 927
 928   /* Ok, maestro returned control to us */
 929   XBT_DEBUG("Control returned to me: '%s'", self->name);
 930
 931   if (self->new_host) {
 932     SIMIX_process_change_host(self, self->new_host);
 933     self->new_host = NULL;
 934   }
 935
 936   if (self->context->iwannadie){
 937     XBT_DEBUG("I wanna die!");
 938     SIMIX_process_stop(self);
 939   }
 940
 941   if (self->suspended) {
 942     XBT_DEBUG("Hey! I'm suspended.");
 943     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 944     self->suspended = 0;
 945     SIMIX_process_suspend(self, self);
 946   }
 947
 948   if (self->doexception) {
 949     XBT_DEBUG("Wait, maestro left me an exception");
 950     self->doexception = 0;
 951     RETHROW;
 952   }
 953
 954   if(SMPI_switch_data_segment && self->segment_index != -1){
 955     SMPI_switch_data_segment(self->segment_index);
 956   }
 957 }
 958
 959 /* callback: context fetching */
 960 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 961 {
 962   smx_process_t process = SIMIX_process_self();
 963   if (process)
 964     return process->running_ctx;
 965   else
 966     return nullptr;
 967 }
 968
 969 /* callback: termination */
 970 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 971 {
 972   xbt_ex_display(e);
 973   xbt_abort();
 974 }
 975
 976 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 977   return p->context;
 978 }
 979
 980 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 981   p->context = c;
 982 }
 983
 984 /**
 985  * \brief Returns the list of processes to run.
 986  */
 987 xbt_dynar_t SIMIX_process_get_runnable(void)
 988 {
 989   return simix_global->process_to_run;
 990 }
 991
 992 /**
 993  * \brief Returns the process from PID.
 994  */
 995 smx_process_t SIMIX_process_from_PID(int PID)
 996 {
 997   smx_process_t proc;
 998   xbt_swag_foreach(proc, simix_global->process_list) {
 999    if (proc->pid == (unsigned long) PID)
1000     return proc;
1001   }
1002   return NULL;
1003 }
1004
1005 /** @brief returns a dynar containg all currently existing processes */
1006 xbt_dynar_t SIMIX_processes_as_dynar(void) {
1007   smx_process_t proc;
1008   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
1009   xbt_swag_foreach(proc, simix_global->process_list) {
1010     xbt_dynar_push(res,&proc);
1011   }
1012   return res;
1013 }
1014
1015
1016 void SIMIX_process_on_exit_runall(smx_process_t process) {
1017   s_smx_process_exit_fun_t exit_fun;
1018   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
1019                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
1020   while (!xbt_dynar_is_empty(process->on_exit)) {
1021     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
1022     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
1023   }
1024 }
1025
1026 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
1027   xbt_assert(process, "current process not found: are you in maestro context ?");
1028
1029   if (!process->on_exit) {
1030     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
1031   }
1032
1033   s_smx_process_exit_fun_t exit_fun = {fun, data};
1034
1035   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
1036 }
1037
1038 /**
1039  * \brief Sets the auto-restart status of the process.
1040  * If set to 1, the process will be automatically restarted when its host
1041  * comes back.
1042  */
1043 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1044   process->auto_restart = auto_restart;
1045 }
1046
1047 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1048   return SIMIX_process_restart(process, simcall->issuer);
1049 }
1050 /** @brief Restart a process, starting it again from the beginning. */
1051 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1052   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1053   //retrieve the arguments of the old process
1054   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1055   s_smx_process_arg_t arg;
1056   arg.code = process->code;
1057   arg.hostname = sg_host_get_name(process->host);
1058   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1059   arg.argc = process->argc;
1060   arg.data = process->data;
1061   int i;
1062   arg.argv = xbt_new(char*,process->argc + 1);
1063   for (i = 0; i < arg.argc; i++) {
1064     arg.argv[i] = xbt_strdup(process->argv[i]);
1065   }
1066   arg.argv[process->argc] = NULL;
1067   arg.properties = NULL;
1068   arg.auto_restart = process->auto_restart;
1069   //kill the old process
1070   SIMIX_process_kill(process,issuer);
1071   //start the new process
1072   smx_process_t new_process;
1073   if (simix_global->create_process_function) {
1074     new_process = simix_global->create_process_function(
1075                                           arg.argv[0],
1076                                           arg.code,
1077                                           arg.data,
1078                                           arg.hostname,
1079                                           arg.kill_time,
1080                                           arg.argc,
1081                                           arg.argv,
1082                                           arg.properties,
1083                                           arg.auto_restart,
1084                                           NULL);
1085   } else {
1086     new_process = simcall_process_create(
1087                            arg.argv[0],
1088                            arg.code,
1089                            arg.data,
1090                            arg.hostname,
1091                            arg.kill_time,
1092                            arg.argc,
1093                            arg.argv,
1094                            arg.properties,
1095                            arg.auto_restart);
1096
1097   }
1098   return new_process;
1099 }
1100
1101 void SIMIX_segment_index_set(smx_process_t proc, int index){
1102   proc->segment_index = index;
1103 }