src/mc/checker/CommunicationDeterminismChecker.cpp

   1 /* Copyright (c) 2008-2019. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "src/mc/checker/CommunicationDeterminismChecker.hpp"
   7 #include "src/kernel/activity/MailboxImpl.hpp"
   8 #include "src/mc/Session.hpp"
   9 #include "src/mc/mc_config.hpp"
  10 #include "src/mc/mc_exit.hpp"
  11 #include "src/mc/mc_private.hpp"
  12 #include "src/mc/mc_request.hpp"
  13 #include "src/mc/mc_smx.hpp"
  14
  15 #if HAVE_SMPI
  16 #include "smpi_request.hpp"
  17 #endif
  18
  19 #include <cstdint>
  20
  21 using simgrid::mc::remote;
  22
  23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  24
  25 /********** Global variables **********/
  26
  27 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  28 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  29
  30 /********** Static functions ***********/
  31
  32 static e_mc_comm_pattern_difference_t compare_comm_pattern(simgrid::mc::PatternCommunication* comm1,
  33                                                            simgrid::mc::PatternCommunication* comm2)
  34 {
  35   if(comm1->type != comm2->type)
  36     return TYPE_DIFF;
  37   if (comm1->rdv != comm2->rdv)
  38     return RDV_DIFF;
  39   if (comm1->src_proc != comm2->src_proc)
  40     return SRC_PROC_DIFF;
  41   if (comm1->dst_proc != comm2->dst_proc)
  42     return DST_PROC_DIFF;
  43   if (comm1->tag != comm2->tag)
  44     return TAG_DIFF;
  45   if (comm1->data.size() != comm2->data.size())
  46     return DATA_SIZE_DIFF;
  47   if (comm1->data != comm2->data)
  48     return DATA_DIFF;
  49   return NONE_DIFF;
  50 }
  51
  52 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process,
  53                                       simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  54 {
  55   char* type;
  56   char* res;
  57
  58   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  59     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  60   else
  61     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  62
  63   switch(diff) {
  64   case TYPE_DIFF:
  65     res = bprintf("%s Different type for communication #%u", type, cursor);
  66     break;
  67   case RDV_DIFF:
  68     res = bprintf("%s Different rdv for communication #%u", type, cursor);
  69     break;
  70   case TAG_DIFF:
  71     res = bprintf("%s Different tag for communication #%u", type, cursor);
  72     break;
  73   case SRC_PROC_DIFF:
  74     res = bprintf("%s Different source for communication #%u", type, cursor);
  75     break;
  76   case DST_PROC_DIFF:
  77     res = bprintf("%s Different destination for communication #%u", type, cursor);
  78     break;
  79   case DATA_SIZE_DIFF:
  80     res = bprintf("%s Different data size for communication #%u", type, cursor);
  81     break;
  82   case DATA_DIFF:
  83     res = bprintf("%s Different data for communication #%u", type, cursor);
  84     break;
  85   default:
  86     res = nullptr;
  87     break;
  88   }
  89
  90   return res;
  91 }
  92
  93 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
  94                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr)
  95 {
  96   // HACK, type punning
  97   simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_comm;
  98   mc_model_checker->process().read(temp_comm, comm_addr);
  99   simgrid::kernel::activity::CommImpl* comm = temp_comm.get_buffer();
 100
 101   smx_actor_t src_proc   = mc_model_checker->process().resolve_actor(simgrid::mc::remote(comm->src_actor_.get()));
 102   smx_actor_t dst_proc   = mc_model_checker->process().resolve_actor(simgrid::mc::remote(comm->dst_actor_.get()));
 103   comm_pattern->src_proc = src_proc->get_pid();
 104   comm_pattern->dst_proc = dst_proc->get_pid();
 105   comm_pattern->src_host = MC_smx_actor_get_host_name(src_proc);
 106   comm_pattern->dst_host = MC_smx_actor_get_host_name(dst_proc);
 107   if (comm_pattern->data.size() == 0 && comm->src_buff_ != nullptr) {
 108     size_t buff_size;
 109     mc_model_checker->process().read(&buff_size, remote(comm->dst_buff_size_));
 110     comm_pattern->data.resize(buff_size);
 111     mc_model_checker->process().read_bytes(comm_pattern->data.data(), comm_pattern->data.size(),
 112                                            remote(comm->src_buff_));
 113   }
 114 }
 115
 116 namespace simgrid {
 117 namespace mc {
 118
 119 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, PatternCommunication* comm,
 120                                                                  int backtracking)
 121 {
 122   if (not backtracking) {
 123     PatternCommunicationList& list      = initial_communications_pattern[process];
 124     e_mc_comm_pattern_difference_t diff = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 125
 126     if (diff != NONE_DIFF) {
 127       if (comm->type == PatternCommunicationType::send) {
 128         this->send_deterministic = 0;
 129         if (this->send_diff != nullptr)
 130           xbt_free(this->send_diff);
 131         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 132       } else {
 133         this->recv_deterministic = 0;
 134         if (this->recv_diff != nullptr)
 135           xbt_free(this->recv_diff);
 136         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 137       }
 138       if (_sg_mc_send_determinism && not this->send_deterministic) {
 139         XBT_INFO("*********************************************************");
 140         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 141         XBT_INFO("*********************************************************");
 142         XBT_INFO("%s", this->send_diff);
 143         xbt_free(this->send_diff);
 144         this->send_diff = nullptr;
 145         mc::session->log_state();
 146         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 147       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
 148         XBT_INFO("****************************************************");
 149         XBT_INFO("***** Non-deterministic communications pattern *****");
 150         XBT_INFO("****************************************************");
 151         if (this->send_diff) {
 152           XBT_INFO("%s", this->send_diff);
 153           xbt_free(this->send_diff);
 154           this->send_diff = nullptr;
 155         }
 156         if (this->recv_diff) {
 157           XBT_INFO("%s", this->recv_diff);
 158           xbt_free(this->recv_diff);
 159           this->recv_diff = nullptr;
 160         }
 161         mc::session->log_state();
 162         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 163       }
 164     }
 165   }
 166 }
 167
 168 /********** Non Static functions ***********/
 169
 170 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, e_mc_call_type_t call_type,
 171                                                        int backtracking)
 172 {
 173   const smx_actor_t issuer = MC_smx_simcall_get_issuer(request);
 174   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 175   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 176
 177   std::unique_ptr<PatternCommunication> pattern(new PatternCommunication());
 178   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 179
 180   if (call_type == MC_CALL_TYPE_SEND) {
 181     /* Create comm pattern */
 182     pattern->type      = PatternCommunicationType::send;
 183     pattern->comm_addr = static_cast<kernel::activity::CommImpl*>(simcall_comm_isend__getraw__result(request));
 184
 185     Remote<kernel::activity::CommImpl> temp_synchro;
 186     mc_model_checker->process().read(temp_synchro,
 187                                      remote(static_cast<kernel::activity::CommImpl*>(pattern->comm_addr)));
 188     kernel::activity::CommImpl* synchro = static_cast<kernel::activity::CommImpl*>(temp_synchro.get_buffer());
 189
 190     char* remote_name = mc_model_checker->process().read<char*>(RemotePtr<char*>(
 191         (uint64_t)(synchro->get_mailbox() ? &synchro->get_mailbox()->name_ : &synchro->mbox_cpy->name_)));
 192     pattern->rdv      = mc_model_checker->process().read_string(RemotePtr<char>(remote_name));
 193     pattern->src_proc = mc_model_checker->process().resolve_actor(mc::remote(synchro->src_actor_.get()))->get_pid();
 194     pattern->src_host = MC_smx_actor_get_host_name(issuer);
 195
 196 #if HAVE_SMPI
 197     simgrid::smpi::Request mpi_request;
 198     mc_model_checker->process().read(&mpi_request,
 199                                      remote(static_cast<smpi::Request*>(simcall_comm_isend__get__data(request))));
 200     pattern->tag = mpi_request.tag();
 201 #endif
 202
 203     if (synchro->src_buff_ != nullptr) {
 204       pattern->data.resize(synchro->src_buff_size_);
 205       mc_model_checker->process().read_bytes(pattern->data.data(), pattern->data.size(), remote(synchro->src_buff_));
 206     }
 207 #if HAVE_SMPI
 208     if(mpi_request.detached()){
 209       if (this->initial_communications_pattern_done) {
 210         /* Evaluate comm determinism */
 211         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
 212         initial_communications_pattern[pattern->src_proc].index_comm++;
 213       } else {
 214         /* Store comm pattern */
 215         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
 216       }
 217       return;
 218     }
 219 #endif
 220   } else if (call_type == MC_CALL_TYPE_RECV) {
 221     pattern->type      = PatternCommunicationType::receive;
 222     pattern->comm_addr = static_cast<kernel::activity::CommImpl*>(simcall_comm_irecv__getraw__result(request));
 223
 224 #if HAVE_SMPI
 225     smpi::Request mpi_request;
 226     mc_model_checker->process().read(&mpi_request,
 227                                      remote(static_cast<smpi::Request*>(simcall_comm_irecv__get__data(request))));
 228     pattern->tag = mpi_request.tag();
 229 #endif
 230
 231     Remote<kernel::activity::CommImpl> temp_comm;
 232     mc_model_checker->process().read(temp_comm, remote(static_cast<kernel::activity::CommImpl*>(pattern->comm_addr)));
 233     kernel::activity::CommImpl* comm = temp_comm.get_buffer();
 234
 235     char* remote_name;
 236     mc_model_checker->process().read(
 237         &remote_name, remote(comm->get_mailbox() ? &xbt::string::to_string_data(comm->get_mailbox()->name_).data
 238                                                  : &xbt::string::to_string_data(comm->mbox_cpy->name_).data));
 239     pattern->rdv      = mc_model_checker->process().read_string(RemotePtr<char>(remote_name));
 240     pattern->dst_proc = mc_model_checker->process().resolve_actor(mc::remote(comm->dst_actor_.get()))->get_pid();
 241     pattern->dst_host = MC_smx_actor_get_host_name(issuer);
 242   } else
 243     xbt_die("Unexpected call_type %i", (int) call_type);
 244
 245   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
 246   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
 247 }
 248
 249 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> comm_addr,
 250                                                             unsigned int issuer, int backtracking)
 251 {
 252   /* Complete comm pattern */
 253   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
 254   auto current_comm_pattern =
 255       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 256                    [&comm_addr](PatternCommunication* comm) { return remote(comm->comm_addr) == comm_addr; });
 257   if (current_comm_pattern == std::end(incomplete_pattern))
 258     xbt_die("Corresponding communication not found!");
 259
 260   update_comm_pattern(*current_comm_pattern, comm_addr);
 261   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 262   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
 263             std::distance(begin(incomplete_pattern), current_comm_pattern));
 264   incomplete_pattern.erase(current_comm_pattern);
 265
 266   if (this->initial_communications_pattern_done) {
 267     /* Evaluate comm determinism */
 268     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
 269     initial_communications_pattern[issuer].index_comm++;
 270   } else {
 271     /* Store comm pattern */
 272     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
 273   }
 274 }
 275
 276 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s)
 277 {
 278 }
 279
 280 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
 281
 282 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
 283 {
 284   RecordTrace res;
 285   for (auto const& state : stack_)
 286     res.push_back(state->get_transition());
 287   return res;
 288 }
 289
 290 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
 291 {
 292   std::vector<std::string> trace;
 293   for (auto const& state : stack_) {
 294     smx_simcall_t req = &state->executed_req_;
 295     if (req)
 296       trace.push_back(request_to_string(req, state->transition_.argument_, RequestType::executed));
 297   }
 298   return trace;
 299 }
 300
 301 void CommunicationDeterminismChecker::log_state() // override
 302 {
 303   if (_sg_mc_comms_determinism) {
 304     if (this->send_deterministic && not this->recv_deterministic) {
 305       XBT_INFO("*******************************************************");
 306       XBT_INFO("**** Only-send-deterministic communication pattern ****");
 307       XBT_INFO("*******************************************************");
 308       XBT_INFO("%s", this->recv_diff);
 309     }
 310     if (not this->send_deterministic && this->recv_deterministic) {
 311       XBT_INFO("*******************************************************");
 312       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 313       XBT_INFO("*******************************************************");
 314       XBT_INFO("%s", this->send_diff);
 315     }
 316   }
 317   XBT_INFO("Expanded states = %lu", expanded_states_count_);
 318   XBT_INFO("Visited states = %lu", mc_model_checker->visited_states);
 319   XBT_INFO("Executed transitions = %lu", mc_model_checker->executed_transitions);
 320   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
 321   if (_sg_mc_comms_determinism)
 322     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
 323 }
 324
 325 void CommunicationDeterminismChecker::prepare()
 326 {
 327   const int maxpid = MC_smx_get_maxpid();
 328
 329   initial_communications_pattern.resize(maxpid);
 330   incomplete_communications_pattern.resize(maxpid);
 331
 332   std::unique_ptr<State> initial_state(new State(++expanded_states_count_));
 333
 334   XBT_DEBUG("********* Start communication determinism verification *********");
 335
 336   /* Get an enabled actor and insert it in the interleave set of the initial state */
 337   for (auto& actor : mc_model_checker->process().actors())
 338     if (mc::actor_is_enabled(actor.copy.get_buffer()))
 339       initial_state->add_interleaving_set(actor.copy.get_buffer());
 340
 341   stack_.push_back(std::move(initial_state));
 342 }
 343
 344 static inline bool all_communications_are_finished()
 345 {
 346   for (size_t current_actor = 1; current_actor < MC_smx_get_maxpid(); current_actor++) {
 347     if (not incomplete_communications_pattern[current_actor].empty()) {
 348       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 349       return false;
 350     }
 351   }
 352   return true;
 353 }
 354
 355 void CommunicationDeterminismChecker::restoreState()
 356 {
 357   /* Intermediate backtracking */
 358   State* last_state = stack_.back().get();
 359   if (last_state->system_state_) {
 360     last_state->system_state_->restore(&mc_model_checker->process());
 361     MC_restore_communications_pattern(last_state);
 362     return;
 363   }
 364
 365   /* Restore the initial state */
 366   mc::session->restore_initial_state();
 367
 368   unsigned n = MC_smx_get_maxpid();
 369   assert(n == incomplete_communications_pattern.size());
 370   assert(n == initial_communications_pattern.size());
 371   for (unsigned j=0; j < n ; j++) {
 372     incomplete_communications_pattern[j].clear();
 373     initial_communications_pattern[j].index_comm = 0;
 374   }
 375
 376   /* Traverse the stack from the state at position start and re-execute the transitions */
 377   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
 378     if (state == stack_.back())
 379       break;
 380
 381     int req_num             = state->transition_.argument_;
 382     smx_simcall_t saved_req = &state->executed_req_;
 383     xbt_assert(saved_req);
 384
 385     /* because we got a copy of the executed request, we have to fetch the
 386        real one, pointed by the request field of the issuer process */
 387
 388     const smx_actor_t issuer = MC_smx_simcall_get_issuer(saved_req);
 389     smx_simcall_t req = &issuer->simcall;
 390
 391     /* TODO : handle test and testany simcalls */
 392     e_mc_call_type_t call = MC_get_call_type(req);
 393     mc_model_checker->handle_simcall(state->transition_);
 394     MC_handle_comm_pattern(call, req, req_num, 1);
 395     mc_model_checker->wait_for_requests();
 396
 397     /* Update statistics */
 398     mc_model_checker->visited_states++;
 399     mc_model_checker->executed_transitions++;
 400   }
 401 }
 402
 403 void CommunicationDeterminismChecker::real_run()
 404 {
 405   std::unique_ptr<VisitedState> visited_state = nullptr;
 406   smx_simcall_t req = nullptr;
 407
 408   while (not stack_.empty()) {
 409     /* Get current state */
 410     State* cur_state = stack_.back().get();
 411
 412     XBT_DEBUG("**************************************************");
 413     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 414               cur_state->interleave_size());
 415
 416     /* Update statistics */
 417     mc_model_checker->visited_states++;
 418
 419     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 420       req = MC_state_choose_request(cur_state);
 421     else
 422       req = nullptr;
 423
 424     if (req != nullptr && visited_state == nullptr) {
 425       int req_num = cur_state->transition_.argument_;
 426
 427       XBT_DEBUG("Execute: %s", request_to_string(req, req_num, RequestType::simix).c_str());
 428
 429       std::string req_str;
 430       if (dot_output != nullptr)
 431         req_str = request_get_dot_output(req, req_num);
 432
 433       mc_model_checker->executed_transitions++;
 434
 435       /* TODO : handle test and testany simcalls */
 436       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 437       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 438         call = MC_get_call_type(req);
 439
 440       /* Answer the request */
 441       mc_model_checker->handle_simcall(cur_state->transition_);
 442       /* After this call req is no longer useful */
 443
 444       MC_handle_comm_pattern(call, req, req_num, 0);
 445
 446       /* Wait for requests (schedules processes) */
 447       mc_model_checker->wait_for_requests();
 448
 449       /* Create the new expanded state */
 450       std::unique_ptr<State> next_state(new State(++expanded_states_count_));
 451
 452       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 453        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 454        * with the initial pattern. */
 455       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 456
 457       if (_sg_mc_max_visited_states != 0)
 458         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
 459       else
 460         visited_state = nullptr;
 461
 462       if (visited_state == nullptr) {
 463         /* Get enabled actors and insert them in the interleave set of the next state */
 464         for (auto& actor : mc_model_checker->process().actors())
 465           if (simgrid::mc::actor_is_enabled(actor.copy.get_buffer()))
 466             next_state->add_interleaving_set(actor.copy.get_buffer());
 467
 468         if (dot_output != nullptr)
 469           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 470
 471       } else if (dot_output != nullptr)
 472         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
 473                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 474
 475       stack_.push_back(std::move(next_state));
 476     } else {
 477       if (stack_.size() > (std::size_t) _sg_mc_max_depth)
 478         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 479       else if (visited_state != nullptr)
 480         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
 481             visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 482       else
 483         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 484
 485       this->initial_communications_pattern_done = true;
 486
 487       /* Trash the current state, no longer needed */
 488       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
 489       stack_.pop_back();
 490
 491       visited_state = nullptr;
 492
 493       /* Check for deadlocks */
 494       if (mc_model_checker->checkDeadlock()) {
 495         MC_show_deadlock();
 496         throw simgrid::mc::DeadlockError();
 497       }
 498
 499       while (not stack_.empty()) {
 500         std::unique_ptr<State> state(std::move(stack_.back()));
 501         stack_.pop_back();
 502         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 503           /* We found a back-tracking point, let's loop */
 504           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
 505           stack_.push_back(std::move(state));
 506
 507           this->restoreState();
 508
 509           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
 510
 511           break;
 512         } else {
 513           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
 514         }
 515       }
 516     }
 517   }
 518
 519   mc::session->log_state();
 520 }
 521
 522 void CommunicationDeterminismChecker::run()
 523 {
 524   XBT_INFO("Check communication determinism");
 525   mc::session->initialize();
 526
 527   this->prepare();
 528
 529   this->real_run();
 530 }
 531
 532 Checker* createCommunicationDeterminismChecker(Session& s)
 533 {
 534   return new CommunicationDeterminismChecker(s);
 535 }
 536
 537 } // namespace mc
 538 } // namespace simgrid