src/simix/ActorImpl.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 <exception>
   8 #include <functional>
   9 #include <string>
  10 #include <utility>
  11
  12 #include <boost/range/algorithm.hpp>
  13
  14 #include <xbt/functional.hpp>
  15 #include <xbt/ex.hpp>
  16 #include <xbt/sysdep.h>
  17 #include <xbt/log.h>
  18 #include <xbt/dict.h>
  19
  20 #include <simgrid/s4u/host.hpp>
  21
  22 #include <mc/mc.h>
  23
  24 #include "smx_private.h"
  25 #include "src/kernel/activity/SynchroIo.hpp"
  26 #include "src/kernel/activity/SynchroRaw.hpp"
  27 #include "src/kernel/activity/SynchroSleep.hpp"
  28 #include "src/mc/Client.hpp"
  29 #include "src/mc/mc_replay.h"
  30 #include "src/msg/msg_private.h"
  31 #include "src/surf/cpu_interface.hpp"
  32 #include "src/surf/surf_interface.hpp"
  33
  34 #ifdef HAVE_SMPI
  35 #include "src/smpi/private.h"
  36 #endif
  37
  38 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
  39
  40 unsigned long simix_process_maxpid = 0;
  41
  42 /** Increase the refcount for this process */
  43 smx_actor_t SIMIX_process_ref(smx_actor_t process)
  44 {
  45   if (process != nullptr)
  46     intrusive_ptr_add_ref(process);
  47   return process;
  48 }
  49
  50 /** Decrease the refcount for this process */
  51 void SIMIX_process_unref(smx_actor_t process)
  52 {
  53   if (process != nullptr)
  54     intrusive_ptr_release(process);
  55 }
  56
  57 /**
  58  * \brief Returns the current agent.
  59  *
  60  * This functions returns the currently running SIMIX process.
  61  *
  62  * \return The SIMIX process
  63  */
  64 smx_actor_t SIMIX_process_self()
  65 {
  66   smx_context_t self_context = SIMIX_context_self();
  67
  68   return (self_context != nullptr) ? self_context->process() : nullptr;
  69 }
  70
  71 /**
  72  * \brief Returns whether a process has pending asynchronous communications.
  73  * \return true if there are asynchronous communications in this process
  74  */
  75 int SIMIX_process_has_pending_comms(smx_actor_t process) {
  76
  77   return xbt_fifo_size(process->comms) > 0;
  78 }
  79
  80 /**
  81  * \brief Moves a process to the list of processes to destroy.
  82  */
  83 void SIMIX_process_cleanup(smx_actor_t process)
  84 {
  85   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  86       process->name.c_str(), process, process->waiting_synchro);
  87
  88   process->finished = true;
  89   SIMIX_process_on_exit_runall(process);
  90
  91   /* Unregister from the kill timer if any */
  92   if (process->kill_timer != nullptr)
  93       SIMIX_timer_remove(process->kill_timer);
  94
  95   xbt_os_mutex_acquire(simix_global->mutex);
  96
  97   /* cancel non-blocking communications */
  98   smx_activity_t synchro = static_cast<smx_activity_t>(xbt_fifo_pop(process->comms));
  99   while (synchro != nullptr) {
 100     simgrid::kernel::activity::Comm *comm = static_cast<simgrid::kernel::activity::Comm*>(synchro);
 101
 102     /* make sure no one will finish the comm after this process is destroyed,
 103      * because src_proc or dst_proc would be an invalid pointer */
 104     comm->cancel();
 105
 106     if (comm->src_proc == process) {
 107       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
 108           comm, comm->detached, (int)comm->state, comm->src_proc, comm->dst_proc);
 109       comm->src_proc = nullptr;
 110
 111       /* I'm not supposed to destroy a detached comm from the sender side, */
 112       if (comm->detached)
 113         XBT_DEBUG("Don't destroy it since it's a detached comm and I'm the sender");
 114       else
 115         comm->unref();
 116
 117     }
 118     else if (comm->dst_proc == process){
 119       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
 120           comm, (int)comm->state, comm->src_proc, comm->dst_proc);
 121       comm->dst_proc = nullptr;
 122
 123       if (comm->detached && comm->src_proc != nullptr) {
 124         /* the comm will be freed right now, remove it from the sender */
 125         xbt_fifo_remove(comm->src_proc->comms, comm);
 126       }
 127
 128       comm->unref();
 129     } else {
 130       xbt_die("Communication synchro %p is in my list but I'm not the sender nor the receiver", synchro);
 131     }
 132     synchro = static_cast<smx_activity_t>(xbt_fifo_pop(process->comms));
 133   }
 134
 135   XBT_DEBUG("%p should not be run anymore",process);
 136   xbt_swag_remove(process, simix_global->process_list);
 137   if (process->host)
 138     xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 139   xbt_swag_insert(process, simix_global->process_to_destroy);
 140   process->context->iwannadie = 0;
 141
 142   xbt_os_mutex_release(simix_global->mutex);
 143 }
 144
 145 /**
 146  * Garbage collection
 147  *
 148  * Should be called some time to time to free the memory allocated for processes
 149  * that have finished (or killed).
 150  */
 151 void SIMIX_process_empty_trash()
 152 {
 153   smx_actor_t process = nullptr;
 154
 155   while ((process = (smx_actor_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 156     XBT_DEBUG("Getting rid of %p",process);
 157     intrusive_ptr_release(process);
 158   }
 159 }
 160
 161 namespace simgrid {
 162 namespace simix {
 163
 164 ActorImpl::~ActorImpl()
 165 {
 166   delete this->context;
 167   if (this->properties)
 168     xbt_dict_free(&this->properties);
 169   if (this->comms != nullptr)
 170     xbt_fifo_free(this->comms);
 171   if (this->on_exit)
 172     xbt_dynar_free(&this->on_exit);
 173 }
 174
 175 void create_maestro(std::function<void()> code)
 176 {
 177   smx_actor_t maestro = nullptr;
 178   /* Create maestro process and intilialize it */
 179   maestro = new simgrid::simix::ActorImpl();
 180   maestro->pid = simix_process_maxpid++;
 181   maestro->ppid = -1;
 182   maestro->name = "";
 183   maestro->data = nullptr;
 184
 185   if (!code) {
 186     maestro->context = SIMIX_context_new(std::function<void()>(), nullptr, maestro);
 187   } else {
 188     if (!simix_global)
 189       xbt_die("simix is not initialized, please call MSG_init first");
 190     maestro->context =
 191       simix_global->context_factory->create_maestro(code, maestro);
 192   }
 193
 194   maestro->simcall.issuer = maestro;
 195   simix_global->maestro_process = maestro;
 196 }
 197
 198 }
 199 }
 200
 201 /** @brief Creates and runs the maestro process */
 202 void SIMIX_maestro_create(void (*code)(void*), void* data)
 203 {
 204   simgrid::simix::create_maestro(std::bind(code, data));
 205 }
 206
 207 /**
 208  * \brief Internal function to create a process.
 209  *
 210  * This function actually creates the process.
 211  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 212  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 213  *
 214  * \return the process created
 215  */
 216 smx_actor_t SIMIX_process_create(
 217                           const char *name,
 218                           std::function<void()> code,
 219                           void *data,
 220                           sg_host_t host,
 221                           double kill_time,
 222                           xbt_dict_t properties,
 223                           int auto_restart,
 224                           smx_actor_t parent_process)
 225 {
 226   smx_actor_t process = nullptr;
 227
 228   XBT_DEBUG("Start process %s on host '%s'", name, host->cname());
 229
 230   if (host->isOff()) {
 231     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name, host->cname());
 232     return nullptr;
 233   }
 234   else {
 235     process = new simgrid::simix::ActorImpl();
 236
 237     xbt_assert(code && host != nullptr, "Invalid parameters");
 238     /* Process data */
 239     process->pid = simix_process_maxpid++;
 240     process->name = simgrid::xbt::string(name);
 241     process->host = host;
 242     process->data = data;
 243     process->comms = xbt_fifo_new();
 244     process->simcall.issuer = process;
 245     /* Initiliaze data segment to default value */
 246     SIMIX_segment_index_set(process, -1);
 247
 248     if (parent_process != nullptr) {
 249       process->ppid = parent_process->pid;
 250       /* SMPI process have their own data segment and each other inherit from their father */
 251 #if HAVE_SMPI
 252        if( smpi_privatize_global_variables) {
 253          if (parent_process->pid != 0) {
 254            SIMIX_segment_index_set(process, parent_process->segment_index);
 255          } else {
 256            SIMIX_segment_index_set(process, process->pid - 1);
 257          }
 258        }
 259 #endif
 260      } else {
 261        process->ppid = -1;
 262      }
 263
 264     /* Process data for auto-restart */
 265     process->auto_restart = auto_restart;
 266     process->code = code;
 267
 268     XBT_VERB("Create context %s", process->name.c_str());
 269     process->context = SIMIX_context_new(
 270       std::move(code),
 271       simix_global->cleanup_process_function, process);
 272
 273     /* Add properties */
 274     process->properties = properties;
 275
 276     /* Add the process to it's host process list */
 277     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 278
 279     XBT_DEBUG("Start context '%s'", process->name.c_str());
 280
 281     /* Now insert it in the global process list and in the process to run list */
 282     xbt_swag_insert(process, simix_global->process_list);
 283     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 284     xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 285
 286     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 287       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->cname(), process->host->cname(), kill_time);
 288       process->kill_timer = SIMIX_timer_set(kill_time, [=]() {
 289         simix_global->kill_process_function(process);
 290       });
 291     }
 292
 293     /* Tracing the process creation */
 294     TRACE_msg_process_create(process->cname(), process->pid, process->host);
 295   }
 296   return process;
 297 }
 298
 299 smx_actor_t SIMIX_process_attach(
 300   const char* name,
 301   void *data,
 302   const char* hostname,
 303   xbt_dict_t properties,
 304   smx_actor_t parent_process)
 305 {
 306   // This is mostly a copy/paste from SIMIX_process_new(),
 307   // it'd be nice to share some code between those two functions.
 308
 309   sg_host_t host = sg_host_by_name(hostname);
 310   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 311
 312   if (host->isOff()) {
 313     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 314       name, hostname);
 315     return nullptr;
 316   }
 317
 318   smx_actor_t process = new simgrid::simix::ActorImpl();
 319   /* Process data */
 320   process->pid = simix_process_maxpid++;
 321   process->name = std::string(name);
 322   process->host = host;
 323   process->data = data;
 324   process->comms = xbt_fifo_new();
 325   process->simcall.issuer = process;
 326   process->ppid = -1;
 327   /* Initiliaze data segment to default value */
 328   SIMIX_segment_index_set(process, -1);
 329   if (parent_process != nullptr) {
 330     process->ppid = parent_process->pid;
 331     /* SMPI process have their own data segment and each other inherit from their father */
 332 #if HAVE_SMPI
 333     if (smpi_privatize_global_variables) {
 334       if (parent_process->pid != 0) {
 335         SIMIX_segment_index_set(process, parent_process->segment_index);
 336       } else {
 337         SIMIX_segment_index_set(process, process->pid - 1);
 338       }
 339     }
 340 #endif
 341   }
 342
 343   /* Process data for auto-restart */
 344   process->auto_restart = false;
 345   process->code = nullptr;
 346
 347   XBT_VERB("Create context %s", process->name.c_str());
 348   if (!simix_global)
 349     xbt_die("simix is not initialized, please call MSG_init first");
 350   process->context = simix_global->context_factory->attach(
 351     simix_global->cleanup_process_function, process);
 352
 353   /* Add properties */
 354   process->properties = properties;
 355
 356   /* Add the process to it's host process list */
 357   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 358
 359   /* Now insert it in the global process list and in the process to run list */
 360   xbt_swag_insert(process, simix_global->process_list);
 361   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 362   xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 363
 364   /* Tracing the process creation */
 365   TRACE_msg_process_create(process->cname(), process->pid, process->host);
 366
 367   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(process->context);
 368   if (!context)
 369     xbt_die("Not a suitable context");
 370
 371   context->attach_start();
 372   return process;
 373 }
 374
 375 void SIMIX_process_detach()
 376 {
 377   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(SIMIX_context_self());
 378   if (!context)
 379     xbt_die("Not a suitable context");
 380
 381   simix_global->cleanup_process_function(context->process());
 382
 383   // Let maestro ignore we are still alive:
 384   // xbt_swag_remove(context->process(), simix_global->process_list);
 385
 386   // TODO, Remove from proces list:
 387   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 388
 389   context->attach_stop();
 390   // delete context;
 391 }
 392
 393 /**
 394  * \brief Executes the processes from simix_global->process_to_run.
 395  *
 396  * The processes of simix_global->process_to_run are run (in parallel if
 397  * possible).  On exit, simix_global->process_to_run is empty, and
 398  * simix_global->process_that_ran contains the list of processes that just ran.
 399  * The two lists are swapped so, be careful when using them before and after a
 400  * call to this function.
 401  */
 402 void SIMIX_process_runall()
 403 {
 404   SIMIX_context_runall();
 405
 406   xbt_dynar_t tmp = simix_global->process_that_ran;
 407   simix_global->process_that_ran = simix_global->process_to_run;
 408   simix_global->process_to_run = tmp;
 409   xbt_dynar_reset(simix_global->process_to_run);
 410 }
 411
 412 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_actor_t process) {
 413   SIMIX_process_kill(process, simcall->issuer);
 414 }
 415 /**
 416  * \brief Internal function to kill a SIMIX process.
 417  *
 418  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 419  * or directly for SIMIX internal purposes.
 420  *
 421  * \param process poor victim
 422  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 423  */
 424 void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) {
 425
 426   XBT_DEBUG("Killing process %s@%s", process->cname(), process->host->cname());
 427
 428   process->context->iwannadie = 1;
 429   process->blocked = 0;
 430   process->suspended = 0;
 431   process->exception = nullptr;
 432
 433   /* destroy the blocking synchro if any */
 434   if (process->waiting_synchro) {
 435
 436     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 437     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 438     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 439     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 440     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 441
 442     if (exec != nullptr) {
 443       exec->unref();
 444
 445     } else if (comm != nullptr) {
 446       xbt_fifo_remove(process->comms, process->waiting_synchro);
 447       comm->cancel();
 448
 449       // Remove first occurrence of &process->simcall:
 450       auto i = boost::range::find(
 451         process->waiting_synchro->simcalls,
 452         &process->simcall);
 453       if (i != process->waiting_synchro->simcalls.end())
 454         process->waiting_synchro->simcalls.remove(&process->simcall);
 455
 456       comm->unref();
 457
 458     } else if (sleep != nullptr) {
 459       SIMIX_process_sleep_destroy(process->waiting_synchro);
 460
 461     } else if (raw != nullptr) {
 462       SIMIX_synchro_stop_waiting(process, &process->simcall);
 463       delete process->waiting_synchro;
 464
 465     } else if (io != nullptr) {
 466       SIMIX_io_destroy(process->waiting_synchro);
 467     }
 468
 469     /*
 470     switch (process->waiting_synchro->type) {
 471     case SIMIX_SYNC_JOIN:
 472       SIMIX_process_sleep_destroy(process->waiting_synchro);
 473       break;
 474     } */
 475
 476     process->waiting_synchro = nullptr;
 477   }
 478   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 479     XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 480     xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 481   }
 482 }
 483
 484 /** @brief Ask another process to raise the given exception
 485  *
 486  * @param cat category of exception
 487  * @param value value associated to the exception
 488  * @param msg string information associated to the exception
 489  */
 490 void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const char *msg) {
 491   SMX_EXCEPTION(process, cat, value, msg);
 492
 493   if (process->suspended)
 494     SIMIX_process_resume(process);
 495
 496   /* cancel the blocking synchro if any */
 497   if (process->waiting_synchro) {
 498
 499     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 500     if (exec != nullptr) {
 501       SIMIX_execution_cancel(process->waiting_synchro);
 502     }
 503
 504     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 505     if (comm != nullptr) {
 506       xbt_fifo_remove(process->comms, comm);
 507       comm->cancel();
 508     }
 509
 510     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 511     if (sleep != nullptr) {
 512       SIMIX_process_sleep_destroy(process->waiting_synchro);
 513       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 514         XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 515         xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 516       }
 517     }
 518
 519     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 520     if (raw != nullptr) {
 521       SIMIX_synchro_stop_waiting(process, &process->simcall);
 522     }
 523
 524     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 525     if (io != nullptr) {
 526       SIMIX_io_destroy(process->waiting_synchro);
 527     }
 528   }
 529   process->waiting_synchro = nullptr;
 530
 531 }
 532
 533 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 534   SIMIX_process_killall(simcall->issuer, reset_pid);
 535 }
 536 /**
 537  * \brief Kills all running processes.
 538  * \param issuer this one will not be killed
 539  */
 540 void SIMIX_process_killall(smx_actor_t issuer, int reset_pid)
 541 {
 542   smx_actor_t p = nullptr;
 543
 544   while ((p = (smx_actor_t) xbt_swag_extract(simix_global->process_list))) {
 545     if (p != issuer) {
 546       SIMIX_process_kill(p,issuer);
 547     }
 548   }
 549
 550   if (reset_pid > 0)
 551     simix_process_maxpid = reset_pid;
 552
 553   SIMIX_context_runall();
 554
 555   SIMIX_process_empty_trash();
 556 }
 557
 558 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_actor_t process, sg_host_t dest)
 559 {
 560   process->new_host = dest;
 561 }
 562 void SIMIX_process_change_host(smx_actor_t process, sg_host_t dest)
 563 {
 564   xbt_assert((process != nullptr), "Invalid parameters");
 565   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 566   process->host = dest;
 567   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 568 }
 569
 570
 571 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t process)
 572 {
 573   smx_activity_t sync_suspend = SIMIX_process_suspend(process, simcall->issuer);
 574
 575   if (process != simcall->issuer) {
 576     SIMIX_simcall_answer(simcall);
 577   } else {
 578     sync_suspend->simcalls.push_back(simcall);
 579     process->waiting_synchro = sync_suspend;
 580     process->waiting_synchro->suspend();
 581   }
 582   /* If we are suspending ourselves, then just do not finish the simcall now */
 583 }
 584
 585 smx_activity_t SIMIX_process_suspend(smx_actor_t process, smx_actor_t issuer)
 586 {
 587   if (process->suspended) {
 588     XBT_DEBUG("Process '%s' is already suspended", process->name.c_str());
 589     return nullptr;
 590   }
 591
 592   process->suspended = 1;
 593
 594   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 595   if (process != issuer) {
 596
 597     if (process->waiting_synchro)
 598       process->waiting_synchro->suspend();
 599     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 600
 601     return nullptr;
 602   } else {
 603     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0);
 604   }
 605 }
 606
 607 void SIMIX_process_resume(smx_actor_t process)
 608 {
 609   XBT_IN("process = %p", process);
 610
 611   if(process->context->iwannadie) {
 612     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 613     return;
 614   }
 615
 616   if(!process->suspended) return;
 617   process->suspended = 0;
 618
 619   /* resume the synchronization that was blocking the resumed process. */
 620   if (process->waiting_synchro)
 621     process->waiting_synchro->resume();
 622
 623   XBT_OUT();
 624 }
 625
 626 int SIMIX_process_get_maxpid() {
 627   return simix_process_maxpid;
 628 }
 629
 630 int SIMIX_process_count()
 631 {
 632   return xbt_swag_size(simix_global->process_list);
 633 }
 634
 635 int SIMIX_process_get_PID(smx_actor_t self)
 636 {
 637   if (self == nullptr)
 638     return 0;
 639   else
 640     return self->pid;
 641 }
 642
 643 void* SIMIX_process_self_get_data()
 644 {
 645   smx_actor_t self = SIMIX_process_self();
 646
 647   if (!self) {
 648     return nullptr;
 649   }
 650   return SIMIX_process_get_data(self);
 651 }
 652
 653 void SIMIX_process_self_set_data(void *data)
 654 {
 655   smx_actor_t self = SIMIX_process_self();
 656
 657   SIMIX_process_set_data(self, data);
 658 }
 659
 660 void* SIMIX_process_get_data(smx_actor_t process)
 661 {
 662   return process->data;
 663 }
 664
 665 void SIMIX_process_set_data(smx_actor_t process, void *data)
 666 {
 667   process->data = data;
 668 }
 669
 670 /* needs to be public and without simcall because it is called
 671    by exceptions and logging events */
 672 const char* SIMIX_process_self_get_name() {
 673
 674   smx_actor_t process = SIMIX_process_self();
 675   if (process == nullptr || process == simix_global->maestro_process)
 676     return "maestro";
 677
 678   return process->name.c_str();
 679 }
 680
 681 smx_actor_t SIMIX_process_get_by_name(const char* name)
 682 {
 683   smx_actor_t proc;
 684   xbt_swag_foreach(proc, simix_global->process_list) {
 685     if (proc->name == name)
 686       return proc;
 687   }
 688   return nullptr;
 689 }
 690
 691 int SIMIX_process_is_suspended(smx_actor_t process)
 692 {
 693   return process->suspended;
 694 }
 695
 696 xbt_dict_t SIMIX_process_get_properties(smx_actor_t process)
 697 {
 698   return process->properties;
 699 }
 700
 701 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
 702 {
 703   if (process->finished) {
 704     // The joined process is already finished, just wake up the issuer process right away
 705     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 706     SIMIX_simcall_answer(simcall);
 707     return;
 708   }
 709   smx_activity_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 710   sync->simcalls.push_back(simcall);
 711   simcall->issuer->waiting_synchro = sync;
 712 }
 713
 714 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_activity_t synchro){
 715   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 716
 717   if (sleep->surf_sleep) {
 718     sleep->surf_sleep->cancel();
 719
 720     while (!sleep->simcalls.empty()) {
 721       smx_simcall_t simcall = sleep->simcalls.front();
 722       sleep->simcalls.pop_front();
 723       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 724       simcall->issuer->waiting_synchro = nullptr;
 725       if (simcall->issuer->suspended) {
 726         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 727         simcall->issuer->suspended = 0;
 728         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 729       } else {
 730         SIMIX_simcall_answer(simcall);
 731       }
 732     }
 733     sleep->surf_sleep->unref();
 734     sleep->surf_sleep = nullptr;
 735   }
 736   sleep->unref();
 737   return 0;
 738 }
 739
 740 smx_activity_t SIMIX_process_join(smx_actor_t issuer, smx_actor_t process, double timeout)
 741 {
 742   smx_activity_t res = SIMIX_process_sleep(issuer, timeout);
 743   static_cast<simgrid::kernel::activity::ActivityImpl*>(res)->ref();
 744   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 745   return res;
 746 }
 747
 748 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 749 {
 750   if (MC_is_active() || MC_record_replay_is_active()) {
 751     MC_process_clock_add(simcall->issuer, duration);
 752     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 753     SIMIX_simcall_answer(simcall);
 754     return;
 755   }
 756   smx_activity_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 757   sync->simcalls.push_back(simcall);
 758   simcall->issuer->waiting_synchro = sync;
 759 }
 760
 761 smx_activity_t SIMIX_process_sleep(smx_actor_t process, double duration)
 762 {
 763   sg_host_t host = process->host;
 764
 765   if (host->isOff())
 766     THROWF(host_error, 0, "Host %s failed, you cannot sleep there.", host->cname());
 767
 768   simgrid::kernel::activity::Sleep *synchro = new simgrid::kernel::activity::Sleep();
 769   synchro->host = host;
 770   synchro->surf_sleep                       = host->pimpl_cpu->sleep(duration);
 771   synchro->surf_sleep->setData(synchro);
 772   XBT_DEBUG("Create sleep synchronization %p", synchro);
 773
 774   return synchro;
 775 }
 776
 777 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
 778 {
 779   XBT_DEBUG("Destroy synchro %p", synchro);
 780   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 781
 782   if (sleep->surf_sleep) {
 783     sleep->surf_sleep->unref();
 784     sleep->surf_sleep = nullptr;
 785     sleep->unref();
 786   }
 787 }
 788
 789 /**
 790  * \brief Calling this function makes the process to yield.
 791  *
 792  * Only the current process can call this function, giving back the control to
 793  * maestro.
 794  *
 795  * \param self the current process
 796  */
 797 void SIMIX_process_yield(smx_actor_t self)
 798 {
 799   XBT_DEBUG("Yield process '%s'", self->name.c_str());
 800
 801   /* Go into sleep and return control to maestro */
 802   self->context->suspend();
 803
 804   /* Ok, maestro returned control to us */
 805   XBT_DEBUG("Control returned to me: '%s'", self->name.c_str());
 806
 807   if (self->new_host) {
 808     SIMIX_process_change_host(self, self->new_host);
 809     self->new_host = nullptr;
 810   }
 811
 812   if (self->context->iwannadie){
 813     XBT_DEBUG("I wanna die!");
 814     self->finished = true;
 815     /* execute the on_exit functions */
 816     SIMIX_process_on_exit_runall(self);
 817     /* Add the process to the list of process to restart, only if the host is down */
 818     if (self->auto_restart && self->host->isOff()) {
 819       SIMIX_host_add_auto_restart_process(self->host, self->cname(),
 820                                           self->code, self->data,
 821                                           SIMIX_timer_get_date(self->kill_timer),
 822                                           self->properties,
 823                                           self->auto_restart);
 824     }
 825     XBT_DEBUG("Process %s@%s is dead", self->cname(), self->host->cname());
 826     self->context->stop();
 827   }
 828
 829   if (self->suspended) {
 830     XBT_DEBUG("Hey! I'm suspended.");
 831     xbt_assert(self->exception != nullptr, "Gasp! This exception may be lost by subsequent calls.");
 832     self->suspended = 0;
 833     SIMIX_process_suspend(self, self);
 834   }
 835
 836   if (self->exception != nullptr) {
 837     XBT_DEBUG("Wait, maestro left me an exception");
 838     std::exception_ptr exception = std::move(self->exception);
 839     self->exception = nullptr;
 840     std::rethrow_exception(std::move(exception));
 841   }
 842
 843   if(SMPI_switch_data_segment && self->segment_index != -1){
 844     SMPI_switch_data_segment(self->segment_index);
 845   }
 846 }
 847
 848 /* callback: termination */
 849 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 850 {
 851   xbt_ex_display(e);
 852   xbt_abort();
 853 }
 854
 855 smx_context_t SIMIX_process_get_context(smx_actor_t p) {
 856   return p->context;
 857 }
 858
 859 void SIMIX_process_set_context(smx_actor_t p,smx_context_t c) {
 860   p->context = c;
 861 }
 862
 863 /**
 864  * \brief Returns the list of processes to run.
 865  */
 866 xbt_dynar_t SIMIX_process_get_runnable()
 867 {
 868   return simix_global->process_to_run;
 869 }
 870
 871 /**
 872  * \brief Returns the process from PID.
 873  */
 874 smx_actor_t SIMIX_process_from_PID(int PID)
 875 {
 876   smx_actor_t proc;
 877   xbt_swag_foreach(proc, simix_global->process_list) {
 878    if (proc->pid == static_cast<unsigned long> (PID))
 879     return proc;
 880   }
 881   return nullptr;
 882 }
 883
 884 /** @brief returns a dynar containing all currently existing processes */
 885 xbt_dynar_t SIMIX_processes_as_dynar() {
 886   smx_actor_t proc;
 887   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_actor_t),nullptr);
 888   xbt_swag_foreach(proc, simix_global->process_list) {
 889     xbt_dynar_push(res,&proc);
 890   }
 891   return res;
 892 }
 893
 894 void SIMIX_process_on_exit_runall(smx_actor_t process) {
 895   s_smx_process_exit_fun_t exit_fun;
 896   smx_process_exit_status_t exit_status = (process->context->iwannadie) ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 897   while (!xbt_dynar_is_empty(process->on_exit)) {
 898     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 899     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 900   }
 901 }
 902
 903 void SIMIX_process_on_exit(smx_actor_t process, int_f_pvoid_pvoid_t fun, void *data) {
 904   xbt_assert(process, "current process not found: are you in maestro context ?");
 905
 906   if (!process->on_exit) {
 907     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), nullptr);
 908   }
 909
 910   s_smx_process_exit_fun_t exit_fun = {fun, data};
 911
 912   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 913 }
 914
 915 /**
 916  * \brief Sets the auto-restart status of the process.
 917  * If set to 1, the process will be automatically restarted when its host
 918  * comes back.
 919  */
 920 void SIMIX_process_auto_restart_set(smx_actor_t process, int auto_restart) {
 921   process->auto_restart = auto_restart;
 922 }
 923
 924 smx_actor_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_actor_t process) {
 925   return SIMIX_process_restart(process, simcall->issuer);
 926 }
 927 /** @brief Restart a process, starting it again from the beginning. */
 928 smx_actor_t SIMIX_process_restart(smx_actor_t process, smx_actor_t issuer) {
 929   XBT_DEBUG("Restarting process %s on %s", process->cname(), process->host->cname());
 930
 931   //retrieve the arguments of the old process
 932   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 933   simgrid::simix::ProcessArg arg;
 934   arg.name = process->name;
 935   arg.code = process->code;
 936   arg.host = process->host;
 937   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 938   arg.data = process->data;
 939   arg.properties = nullptr;
 940   arg.auto_restart = process->auto_restart;
 941
 942   //kill the old process
 943   SIMIX_process_kill(process, issuer);
 944
 945   //start the new process
 946   return simix_global->create_process_function(arg.name.c_str(), std::move(arg.code), arg.data, arg.host,
 947       arg.kill_time, arg.properties, arg.auto_restart, nullptr);
 948 }
 949
 950 void SIMIX_segment_index_set(smx_actor_t proc, int index){
 951   proc->segment_index = index;
 952 }
 953
 954 /**
 955  * \ingroup simix_process_management
 956  * \brief Creates and runs a new SIMIX process.
 957  *
 958  * The structure and the corresponding thread are created and put in the list of ready processes.
 959  *
 960  * \param name a name for the process. It is for user-level information and can be nullptr.
 961  * \param code the main function of the process
 962  * \param data a pointer to any data one may want to attach to the new object. It is for user-level information and can be nullptr.
 963  * It can be retrieved with the function \ref simcall_process_get_data.
 964  * \param hostname name of the host where the new agent is executed.
 965  * \param kill_time time when the process is killed
 966  * \param argc first argument passed to \a code
 967  * \param argv second argument passed to \a code
 968  * \param properties the properties of the process
 969  * \param auto_restart either it is autorestarting or not.
 970  */
 971 smx_actor_t simcall_process_create(const char *name,
 972                               xbt_main_func_t code,
 973                               void *data,
 974                               sg_host_t host,
 975                               double kill_time,
 976                               int argc, char **argv,
 977                               xbt_dict_t properties,
 978                               int auto_restart)
 979 {
 980   if (name == nullptr)
 981     name = "";
 982   auto wrapped_code = simgrid::xbt::wrapMain(code, argc, argv);
 983   for (int i = 0; i != argc; ++i)
 984     xbt_free(argv[i]);
 985   xbt_free(argv);
 986   smx_actor_t res = simcall_process_create(name,
 987     std::move(wrapped_code),
 988     data, host, kill_time, properties, auto_restart);
 989   return res;
 990 }
 991
 992 smx_actor_t simcall_process_create(
 993   const char *name, std::function<void()> code, void *data,
 994   sg_host_t host, double kill_time,
 995   xbt_dict_t properties, int auto_restart)
 996 {
 997   if (name == nullptr)
 998     name = "";
 999   smx_actor_t self = SIMIX_process_self();
1000   return simgrid::simix::kernelImmediate([&] {
1001     return SIMIX_process_create(name, std::move(code), data, host, kill_time, properties, auto_restart, self);
1002   });
1003 }