src/mc/checker/CommunicationDeterminismChecker.cpp

   1 /* Copyright (c) 2008-2022. 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
  13 #include <cstdint>
  14
  15 using api = simgrid::mc::Api;
  16
  17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  18
  19 /********** Global variables **********/
  20
  21 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  22 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  23
  24 /********** Static functions ***********/
  25
  26 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
  27                                                                const simgrid::mc::PatternCommunication* comm2)
  28 {
  29   using simgrid::mc::CommPatternDifference;
  30   if (comm1->type != comm2->type)
  31     return CommPatternDifference::TYPE;
  32   if (comm1->rdv != comm2->rdv)
  33     return CommPatternDifference::RDV;
  34   if (comm1->src_proc != comm2->src_proc)
  35     return CommPatternDifference::SRC_PROC;
  36   if (comm1->dst_proc != comm2->dst_proc)
  37     return CommPatternDifference::DST_PROC;
  38   if (comm1->tag != comm2->tag)
  39     return CommPatternDifference::TAG;
  40   if (comm1->data.size() != comm2->data.size())
  41     return CommPatternDifference::DATA_SIZE;
  42   if (comm1->data != comm2->data)
  43     return CommPatternDifference::DATA;
  44   return CommPatternDifference::NONE;
  45 }
  46
  47 static void patterns_copy(std::vector<simgrid::mc::PatternCommunication*>& dest,
  48                              std::vector<simgrid::mc::PatternCommunication> const& source)
  49 {
  50   dest.clear();
  51   for (simgrid::mc::PatternCommunication const& comm : source) {
  52     auto* copy_comm = new simgrid::mc::PatternCommunication(comm.dup());
  53     dest.push_back(copy_comm);
  54   }
  55 }
  56
  57 static void restore_communications_pattern(simgrid::mc::State* state)
  58 {
  59   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
  60     initial_communications_pattern[i].index_comm = state->communication_indices_[i];
  61
  62   const unsigned long maxpid = api::get().get_maxpid();
  63   for (unsigned long i = 0; i < maxpid; i++)
  64     patterns_copy(incomplete_communications_pattern[i], state->incomplete_comm_pattern_[i]);
  65 }
  66
  67 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t process,
  68                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  69 {
  70   char* type;
  71   char* res;
  72
  73   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  74     type = bprintf("The send communications pattern of the process %ld is different!", process - 1);
  75   else
  76     type = bprintf("The recv communications pattern of the process %ld is different!", process - 1);
  77
  78   using simgrid::mc::CommPatternDifference;
  79   switch (diff) {
  80     case CommPatternDifference::TYPE:
  81       res = bprintf("%s Different type for communication #%u", type, cursor);
  82       break;
  83     case CommPatternDifference::RDV:
  84       res = bprintf("%s Different rdv for communication #%u", type, cursor);
  85       break;
  86     case CommPatternDifference::TAG:
  87       res = bprintf("%s Different tag for communication #%u", type, cursor);
  88       break;
  89     case CommPatternDifference::SRC_PROC:
  90       res = bprintf("%s Different source for communication #%u", type, cursor);
  91       break;
  92     case CommPatternDifference::DST_PROC:
  93       res = bprintf("%s Different destination for communication #%u", type, cursor);
  94       break;
  95     case CommPatternDifference::DATA_SIZE:
  96       res = bprintf("%s Different data size for communication #%u", type, cursor);
  97       break;
  98     case CommPatternDifference::DATA:
  99       res = bprintf("%s Different data for communication #%u", type, cursor);
 100       break;
 101     default:
 102       res = nullptr;
 103       break;
 104   }
 105
 106   return res;
 107 }
 108
 109 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
 110                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> const& comm_addr)
 111 {
 112   auto src_proc = api::get().get_src_actor(comm_addr);
 113   auto dst_proc = api::get().get_dst_actor(comm_addr);
 114   comm_pattern->src_proc = src_proc->get_pid();
 115   comm_pattern->dst_proc = dst_proc->get_pid();
 116
 117   if (comm_pattern->data.empty()) {
 118     comm_pattern->data = api::get().get_pattern_comm_data(comm_addr);
 119   }
 120 }
 121
 122 namespace simgrid {
 123 namespace mc {
 124
 125 void CommunicationDeterminismChecker::deterministic_comm_pattern(aid_t process, const PatternCommunication* comm,
 126                                                                  bool backtracking)
 127 {
 128   if (not backtracking) {
 129     PatternCommunicationList& list = initial_communications_pattern[process];
 130     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 131
 132     if (diff != CommPatternDifference::NONE) {
 133       if (comm->type == PatternCommunicationType::send) {
 134         this->send_deterministic = false;
 135         if (this->send_diff != nullptr)
 136           xbt_free(this->send_diff);
 137         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 138       } else {
 139         this->recv_deterministic = false;
 140         if (this->recv_diff != nullptr)
 141           xbt_free(this->recv_diff);
 142         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 143       }
 144       if (_sg_mc_send_determinism && not this->send_deterministic) {
 145         XBT_INFO("*********************************************************");
 146         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 147         XBT_INFO("*********************************************************");
 148         XBT_INFO("%s", this->send_diff);
 149         xbt_free(this->send_diff);
 150         this->send_diff = nullptr;
 151         api::get().log_state();
 152         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 153       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
 154         XBT_INFO("****************************************************");
 155         XBT_INFO("***** Non-deterministic communications pattern *****");
 156         XBT_INFO("****************************************************");
 157         if (this->send_diff) {
 158           XBT_INFO("%s", this->send_diff);
 159           xbt_free(this->send_diff);
 160           this->send_diff = nullptr;
 161         }
 162         if (this->recv_diff) {
 163           XBT_INFO("%s", this->recv_diff);
 164           xbt_free(this->recv_diff);
 165           this->recv_diff = nullptr;
 166         }
 167         api::get().log_state();
 168         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 169       }
 170     }
 171   }
 172 }
 173
 174 /********** Non Static functions ***********/
 175
 176 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, bool backtracking)
 177 {
 178   const smx_actor_t issuer                                     = api::get().simcall_get_issuer(request);
 179   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 180   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 181
 182   auto pattern   = std::make_unique<PatternCommunication>();
 183   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 184
 185   if (call_type == CallType::SEND) {
 186     /* Create comm pattern */
 187     pattern->type      = PatternCommunicationType::send;
 188     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 189     pattern->rdv       = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 190     pattern->src_proc  = api::get().get_pattern_comm_src_proc(pattern->comm_addr);
 191
 192 #if HAVE_SMPI
 193     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
 194 #endif
 195     pattern->data = api::get().get_pattern_comm_data(pattern->comm_addr);
 196
 197 #if HAVE_SMPI
 198     auto send_detached = api::get().check_send_request_detached(request);
 199     if (send_detached) {
 200       if (this->initial_communications_pattern_done) {
 201         /* Evaluate comm determinism */
 202         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
 203         initial_communications_pattern[pattern->src_proc].index_comm++;
 204       } else {
 205         /* Store comm pattern */
 206         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
 207       }
 208       return;
 209     }
 210 #endif
 211   } else if (call_type == CallType::RECV) {
 212     pattern->type = PatternCommunicationType::receive;
 213     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 214
 215 #if HAVE_SMPI
 216     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
 217 #endif
 218     pattern->rdv = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 219     pattern->dst_proc = api::get().get_pattern_comm_dst_proc(pattern->comm_addr);
 220   } else
 221     xbt_die("Unexpected call_type %i", (int)call_type);
 222
 223   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
 224   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
 225 }
 226
 227 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> const& comm_addr,
 228                                                             aid_t issuer, bool backtracking)
 229 {
 230   /* Complete comm pattern */
 231   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
 232   auto current_comm_pattern =
 233       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 234                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
 235   xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
 236
 237   update_comm_pattern(*current_comm_pattern, comm_addr);
 238   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 239   XBT_DEBUG("Remove incomplete comm pattern for process %ld at cursor %zd", issuer,
 240             std::distance(begin(incomplete_pattern), current_comm_pattern));
 241   incomplete_pattern.erase(current_comm_pattern);
 242
 243   if (this->initial_communications_pattern_done) {
 244     /* Evaluate comm determinism */
 245     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
 246     initial_communications_pattern[issuer].index_comm++;
 247   } else {
 248     /* Store comm pattern */
 249     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
 250   }
 251 }
 252
 253 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session* session) : Checker(session) {}
 254
 255 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
 256
 257 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
 258 {
 259   RecordTrace res;
 260   for (auto const& state : stack_)
 261     res.push_back(state->get_transition());
 262   return res;
 263 }
 264
 265 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
 266 {
 267   std::vector<std::string> trace;
 268   for (auto const& state : stack_)
 269     trace.push_back(state->get_transition()->to_string());
 270   return trace;
 271 }
 272
 273 void CommunicationDeterminismChecker::log_state() // override
 274 {
 275   if (_sg_mc_comms_determinism) {
 276     if (this->send_deterministic && not this->recv_deterministic) {
 277       XBT_INFO("*******************************************************");
 278       XBT_INFO("**** Only-send-deterministic communication pattern ****");
 279       XBT_INFO("*******************************************************");
 280       XBT_INFO("%s", this->recv_diff);
 281     }
 282     if (not this->send_deterministic && this->recv_deterministic) {
 283       XBT_INFO("*******************************************************");
 284       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 285       XBT_INFO("*******************************************************");
 286       XBT_INFO("%s", this->send_diff);
 287     }
 288   }
 289   XBT_INFO("Expanded states = %ld", State::get_expanded_states());
 290   XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
 291   XBT_INFO("Executed transitions = %lu", Transition::get_executed_transitions());
 292   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
 293   if (_sg_mc_comms_determinism)
 294     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
 295 }
 296
 297 void CommunicationDeterminismChecker::prepare()
 298 {
 299   const unsigned long maxpid = api::get().get_maxpid();
 300
 301   initial_communications_pattern.resize(maxpid);
 302   incomplete_communications_pattern.resize(maxpid);
 303
 304   auto initial_state = std::make_unique<State>();
 305
 306   XBT_DEBUG("********* Start communication determinism verification *********");
 307
 308   /* Add all enabled actors to the interleave set of the initial state */
 309   for (auto& act : api::get().get_actors()) {
 310     auto actor = act.copy.get_buffer();
 311     if (get_session().actor_is_enabled(actor->get_pid()))
 312       initial_state->mark_todo(actor->get_pid());
 313   }
 314
 315   stack_.push_back(std::move(initial_state));
 316 }
 317
 318 static inline bool all_communications_are_finished()
 319 {
 320   const unsigned long maxpid = api::get().get_maxpid();
 321   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
 322     if (not incomplete_communications_pattern[current_actor].empty()) {
 323       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 324       return false;
 325     }
 326   }
 327   return true;
 328 }
 329
 330 void CommunicationDeterminismChecker::restoreState()
 331 {
 332   /* Intermediate backtracking */
 333   State* last_state = stack_.back().get();
 334   if (last_state->system_state_) {
 335     api::get().restore_state(last_state->system_state_);
 336     restore_communications_pattern(last_state);
 337     return;
 338   }
 339
 340   get_session().restore_initial_state();
 341
 342   const unsigned long maxpid = api::get().get_maxpid();
 343   assert(maxpid == incomplete_communications_pattern.size());
 344   assert(maxpid == initial_communications_pattern.size());
 345   for (unsigned long j = 0; j < maxpid; j++) {
 346     incomplete_communications_pattern[j].clear();
 347     initial_communications_pattern[j].index_comm = 0;
 348   }
 349
 350   /* Traverse the stack from the state at position start and re-execute the transitions */
 351   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
 352     if (state == stack_.back())
 353       break;
 354
 355     int req_num                    = state->get_transition()->times_considered_;
 356     const s_smx_simcall* saved_req = &state->executed_req_;
 357     xbt_assert(saved_req);
 358
 359     /* because we got a copy of the executed request, we have to fetch the
 360        real one, pointed by the request field of the issuer process */
 361
 362     const smx_actor_t issuer = api::get().simcall_get_issuer(saved_req);
 363     smx_simcall_t req        = &issuer->simcall_;
 364
 365     /* TODO : handle test and testany simcalls */
 366     CallType call = MC_get_call_type(req);
 367     state->get_transition()->replay();
 368     handle_comm_pattern(call, req, req_num, 1);
 369
 370     /* Update statistics */
 371     api::get().mc_inc_visited_states();
 372   }
 373 }
 374
 375 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value,
 376                                                           bool backtracking)
 377 {
 378   using simgrid::mc::CallType;
 379   switch(call_type) {
 380     case CallType::NONE:
 381       break;
 382     case CallType::SEND:
 383     case CallType::RECV:
 384       get_comm_pattern(req, call_type, backtracking);
 385       break;
 386     case CallType::WAIT:
 387     case CallType::WAITANY: {
 388       RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr;
 389       if (call_type == CallType::WAIT)
 390         comm_addr = remote(simcall_comm_wait__getraw__comm(req));
 391       else
 392         comm_addr = api::get().get_comm_waitany_raw_addr(req, value);
 393       auto simcall_issuer = api::get().simcall_get_issuer(req);
 394       complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
 395     } break;
 396   default:
 397     xbt_die("Unexpected call type %i", (int)call_type);
 398   }
 399 }
 400
 401 void CommunicationDeterminismChecker::real_run()
 402 {
 403   std::unique_ptr<VisitedState> visited_state = nullptr;
 404
 405   while (not stack_.empty()) {
 406     /* Get current state */
 407     State* cur_state = stack_.back().get();
 408
 409     XBT_DEBUG("**************************************************");
 410     XBT_DEBUG("Exploration depth = %zu (state = %ld, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 411               cur_state->count_todo());
 412
 413     /* Update statistics */
 414     api::get().mc_inc_visited_states();
 415
 416     int next_transition = -1;
 417     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 418       next_transition = cur_state->next_transition();
 419
 420     if (next_transition >= 0 && visited_state == nullptr) {
 421       cur_state->execute_next(next_transition);
 422
 423       int req_num       = cur_state->get_transition()->times_considered_;
 424       smx_simcall_t req = &cur_state->executed_req_;
 425
 426       XBT_DEBUG("Execute: %s", cur_state->get_transition()->to_string().c_str());
 427
 428       std::string req_str;
 429       if (dot_output != nullptr)
 430         req_str = api::get().request_get_dot_output(cur_state->get_transition());
 431
 432       /* TODO : handle test and testany simcalls */
 433       CallType call = CallType::NONE;
 434       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 435         call = MC_get_call_type(req);
 436
 437       handle_comm_pattern(call, req, req_num, 0);
 438
 439       /* Create the new expanded state */
 440       auto next_state = std::make_unique<State>();
 441
 442       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 443        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 444        * with the initial pattern. */
 445       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 446
 447       if (_sg_mc_max_visited_states != 0)
 448         visited_state = visited_states_.addVisitedState(next_state->num_, next_state.get(), compare_snapshots);
 449       else
 450         visited_state = nullptr;
 451
 452       if (visited_state == nullptr) {
 453         /* Add all enabled actors to the interleave set of the next state */
 454         for (auto& act : api::get().get_actors()) {
 455           auto actor = act.copy.get_buffer();
 456           if (get_session().actor_is_enabled(actor->get_pid()))
 457             next_state->mark_todo(actor->get_pid());
 458         }
 459
 460         if (dot_output != nullptr)
 461           fprintf(dot_output, "\"%ld\" -> \"%ld\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 462
 463       } else if (dot_output != nullptr)
 464         fprintf(dot_output, "\"%ld\" -> \"%ld\" [%s];\n", cur_state->num_,
 465                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 466
 467       stack_.push_back(std::move(next_state));
 468     } else {
 469       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
 470         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 471       else if (visited_state != nullptr)
 472         XBT_DEBUG("State already visited (equal to state %ld), exploration stopped on this path.",
 473                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 474       else
 475         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 476
 477       this->initial_communications_pattern_done = true;
 478
 479       /* Trash the current state, no longer needed */
 480       XBT_DEBUG("Delete state %ld at depth %zu", cur_state->num_, stack_.size());
 481       stack_.pop_back();
 482
 483       visited_state = nullptr;
 484
 485       api::get().mc_check_deadlock();
 486
 487       while (not stack_.empty()) {
 488         std::unique_ptr<State> state(std::move(stack_.back()));
 489         stack_.pop_back();
 490         if (state->count_todo() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 491           /* We found a back-tracking point, let's loop */
 492           XBT_DEBUG("Back-tracking to state %ld at depth %zu", state->num_, stack_.size() + 1);
 493           stack_.push_back(std::move(state));
 494
 495           this->restoreState();
 496
 497           XBT_DEBUG("Back-tracking to state %ld at depth %zu done", stack_.back()->num_, stack_.size());
 498
 499           break;
 500         } else {
 501           XBT_DEBUG("Delete state %ld at depth %zu", state->num_, stack_.size() + 1);
 502         }
 503       }
 504     }
 505   }
 506
 507   api::get().log_state();
 508 }
 509
 510 void CommunicationDeterminismChecker::run()
 511 {
 512   XBT_INFO("Check communication determinism");
 513   get_session().take_initial_snapshot();
 514
 515   this->prepare();
 516   this->real_run();
 517 }
 518
 519 Checker* create_communication_determinism_checker(Session* session)
 520 {
 521   return new CommunicationDeterminismChecker(session);
 522 }
 523
 524 } // namespace mc
 525 } // namespace simgrid