src/mc/checker/CommunicationDeterminismChecker.cpp

   1 /* Copyright (c) 2008-2021. 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/mc_config.hpp"
   9 #include "src/mc/mc_exit.hpp"
  10 #include "src/mc/mc_private.hpp"
  11 #include "src/mc/mc_request.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   for (unsigned long i = 0; i < api::get().get_maxpid(); i++)
  63     patterns_copy(incomplete_communications_pattern[i], state->incomplete_comm_pattern_[i]);
  64 }
  65
  66 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t process,
  67                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  68 {
  69   char* type;
  70   char* res;
  71
  72   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  73     type = bprintf("The send communications pattern of the process %ld is different!", process - 1);
  74   else
  75     type = bprintf("The recv communications pattern of the process %ld is different!", process - 1);
  76
  77   using simgrid::mc::CommPatternDifference;
  78   switch (diff) {
  79     case CommPatternDifference::TYPE:
  80       res = bprintf("%s Different type for communication #%u", type, cursor);
  81       break;
  82     case CommPatternDifference::RDV:
  83       res = bprintf("%s Different rdv for communication #%u", type, cursor);
  84       break;
  85     case CommPatternDifference::TAG:
  86       res = bprintf("%s Different tag for communication #%u", type, cursor);
  87       break;
  88     case CommPatternDifference::SRC_PROC:
  89       res = bprintf("%s Different source for communication #%u", type, cursor);
  90       break;
  91     case CommPatternDifference::DST_PROC:
  92       res = bprintf("%s Different destination for communication #%u", type, cursor);
  93       break;
  94     case CommPatternDifference::DATA_SIZE:
  95       res = bprintf("%s Different data size for communication #%u", type, cursor);
  96       break;
  97     case CommPatternDifference::DATA:
  98       res = bprintf("%s Different data for communication #%u", type, cursor);
  99       break;
 100     default:
 101       res = nullptr;
 102       break;
 103   }
 104
 105   return res;
 106 }
 107
 108 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
 109                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> const& comm_addr)
 110 {
 111   auto src_proc = api::get().get_src_actor(comm_addr);
 112   auto dst_proc = api::get().get_dst_actor(comm_addr);
 113   comm_pattern->src_proc = src_proc->get_pid();
 114   comm_pattern->dst_proc = dst_proc->get_pid();
 115   comm_pattern->src_host = &api::get().get_actor_host_name(src_proc);
 116   comm_pattern->dst_host = &api::get().get_actor_host_name(dst_proc);
 117
 118   if (comm_pattern->data.empty()) {
 119     comm_pattern->data = api::get().get_pattern_comm_data(comm_addr);
 120   }
 121 }
 122
 123 namespace simgrid {
 124 namespace mc {
 125
 126 void CommunicationDeterminismChecker::deterministic_comm_pattern(aid_t process, const PatternCommunication* comm,
 127                                                                  int backtracking)
 128 {
 129   if (not backtracking) {
 130     PatternCommunicationList& list = initial_communications_pattern[process];
 131     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 132
 133     if (diff != CommPatternDifference::NONE) {
 134       if (comm->type == PatternCommunicationType::send) {
 135         this->send_deterministic = false;
 136         if (this->send_diff != nullptr)
 137           xbt_free(this->send_diff);
 138         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 139       } else {
 140         this->recv_deterministic = false;
 141         if (this->recv_diff != nullptr)
 142           xbt_free(this->recv_diff);
 143         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 144       }
 145       if (_sg_mc_send_determinism && not this->send_deterministic) {
 146         XBT_INFO("*********************************************************");
 147         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 148         XBT_INFO("*********************************************************");
 149         XBT_INFO("%s", this->send_diff);
 150         xbt_free(this->send_diff);
 151         this->send_diff = nullptr;
 152         api::get().log_state();
 153         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 154       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
 155         XBT_INFO("****************************************************");
 156         XBT_INFO("***** Non-deterministic communications pattern *****");
 157         XBT_INFO("****************************************************");
 158         if (this->send_diff) {
 159           XBT_INFO("%s", this->send_diff);
 160           xbt_free(this->send_diff);
 161           this->send_diff = nullptr;
 162         }
 163         if (this->recv_diff) {
 164           XBT_INFO("%s", this->recv_diff);
 165           xbt_free(this->recv_diff);
 166           this->recv_diff = nullptr;
 167         }
 168         api::get().log_state();
 169         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 170       }
 171     }
 172   }
 173 }
 174
 175 /********** Non Static functions ***********/
 176
 177 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, int backtracking)
 178 {
 179   const smx_actor_t issuer                                     = api::get().simcall_get_issuer(request);
 180   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 181   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 182
 183   auto pattern   = std::make_unique<PatternCommunication>();
 184   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 185
 186   if (call_type == CallType::SEND) {
 187     /* Create comm pattern */
 188     pattern->type      = PatternCommunicationType::send;
 189     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 190     pattern->rdv       = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 191     pattern->src_proc  = api::get().get_pattern_comm_src_proc(pattern->comm_addr);
 192     pattern->src_host  = &api::get().get_actor_host_name(issuer);
 193
 194 #if HAVE_SMPI
 195     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
 196 #endif
 197     pattern->data = api::get().get_pattern_comm_data(pattern->comm_addr);
 198
 199 #if HAVE_SMPI
 200     auto send_detached = api::get().check_send_request_detached(request);
 201     if (send_detached) {
 202       if (this->initial_communications_pattern_done) {
 203         /* Evaluate comm determinism */
 204         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
 205         initial_communications_pattern[pattern->src_proc].index_comm++;
 206       } else {
 207         /* Store comm pattern */
 208         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
 209       }
 210       return;
 211     }
 212 #endif
 213   } else if (call_type == CallType::RECV) {
 214     pattern->type = PatternCommunicationType::receive;
 215     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 216
 217 #if HAVE_SMPI
 218     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
 219 #endif
 220     pattern->rdv = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 221     pattern->dst_proc = api::get().get_pattern_comm_dst_proc(pattern->comm_addr);
 222     pattern->dst_host = &api::get().get_actor_host_name(issuer);
 223   } else
 224     xbt_die("Unexpected call_type %i", (int)call_type);
 225
 226   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
 227   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
 228 }
 229
 230 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> const& comm_addr, aid_t issuer,
 231                                                             int backtracking)
 232 {
 233   /* Complete comm pattern */
 234   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
 235   auto current_comm_pattern =
 236       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 237                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
 238   if (current_comm_pattern == std::end(incomplete_pattern))
 239     xbt_die("Corresponding communication not found!");
 240
 241   update_comm_pattern(*current_comm_pattern, comm_addr);
 242   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 243   XBT_DEBUG("Remove incomplete comm pattern for process %ld at cursor %zd", issuer,
 244             std::distance(begin(incomplete_pattern), current_comm_pattern));
 245   incomplete_pattern.erase(current_comm_pattern);
 246
 247   if (this->initial_communications_pattern_done) {
 248     /* Evaluate comm determinism */
 249     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
 250     initial_communications_pattern[issuer].index_comm++;
 251   } else {
 252     /* Store comm pattern */
 253     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
 254   }
 255 }
 256
 257 CommunicationDeterminismChecker::CommunicationDeterminismChecker() : Checker() {}
 258
 259 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
 260
 261 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
 262 {
 263   RecordTrace res;
 264   for (auto const& state : stack_)
 265     res.push_back(state->get_transition());
 266   return res;
 267 }
 268
 269 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
 270 {
 271   std::vector<std::string> trace;
 272   for (auto const& state : stack_) {
 273     smx_simcall_t req = &state->executed_req_;
 274     trace.push_back(api::get().request_to_string(req, state->transition_.times_considered_, RequestType::executed));
 275   }
 276   return trace;
 277 }
 278
 279 void CommunicationDeterminismChecker::log_state() // override
 280 {
 281   if (_sg_mc_comms_determinism) {
 282     if (this->send_deterministic && not this->recv_deterministic) {
 283       XBT_INFO("*******************************************************");
 284       XBT_INFO("**** Only-send-deterministic communication pattern ****");
 285       XBT_INFO("*******************************************************");
 286       XBT_INFO("%s", this->recv_diff);
 287     }
 288     if (not this->send_deterministic && this->recv_deterministic) {
 289       XBT_INFO("*******************************************************");
 290       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 291       XBT_INFO("*******************************************************");
 292       XBT_INFO("%s", this->send_diff);
 293     }
 294   }
 295   XBT_INFO("Expanded states = %lu", expanded_states_count_);
 296   XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
 297   XBT_INFO("Executed transitions = %lu", api::get().mc_get_executed_trans());
 298   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
 299   if (_sg_mc_comms_determinism)
 300     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
 301 }
 302
 303 void CommunicationDeterminismChecker::prepare()
 304 {
 305   const auto maxpid = api::get().get_maxpid();
 306
 307   initial_communications_pattern.resize(maxpid);
 308   incomplete_communications_pattern.resize(maxpid);
 309
 310   ++expanded_states_count_;
 311   auto initial_state = std::make_unique<State>(expanded_states_count_);
 312
 313   XBT_DEBUG("********* Start communication determinism verification *********");
 314
 315   /* Get an enabled actor and insert it in the interleave set of the initial state */
 316   auto actors = api::get().get_actors();
 317   for (auto& actor : actors)
 318     if (api::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 319       initial_state->mark_todo(actor.copy.get_buffer());
 320
 321   stack_.push_back(std::move(initial_state));
 322 }
 323
 324 static inline bool all_communications_are_finished()
 325 {
 326   auto maxpid = api::get().get_maxpid();
 327   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
 328     if (not incomplete_communications_pattern[current_actor].empty()) {
 329       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 330       return false;
 331     }
 332   }
 333   return true;
 334 }
 335
 336 void CommunicationDeterminismChecker::restoreState()
 337 {
 338   /* Intermediate backtracking */
 339   State* last_state = stack_.back().get();
 340   if (last_state->system_state_) {
 341     api::get().restore_state(last_state->system_state_);
 342     restore_communications_pattern(last_state);
 343     return;
 344   }
 345
 346   /* Restore the initial state */
 347   api::get().restore_initial_state();
 348
 349   unsigned long n = api::get().get_maxpid();
 350   assert(n == incomplete_communications_pattern.size());
 351   assert(n == initial_communications_pattern.size());
 352   for (unsigned long j = 0; j < n; j++) {
 353     incomplete_communications_pattern[j].clear();
 354     initial_communications_pattern[j].index_comm = 0;
 355   }
 356
 357   /* Traverse the stack from the state at position start and re-execute the transitions */
 358   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
 359     if (state == stack_.back())
 360       break;
 361
 362     int req_num                    = state->transition_.times_considered_;
 363     const s_smx_simcall* saved_req = &state->executed_req_;
 364     xbt_assert(saved_req);
 365
 366     /* because we got a copy of the executed request, we have to fetch the
 367        real one, pointed by the request field of the issuer process */
 368
 369     const smx_actor_t issuer = api::get().simcall_get_issuer(saved_req);
 370     smx_simcall_t req        = &issuer->simcall_;
 371
 372     /* TODO : handle test and testany simcalls */
 373     CallType call = MC_get_call_type(req);
 374     api::get().handle_simcall(state->transition_);
 375     handle_comm_pattern(call, req, req_num, 1);
 376     api::get().mc_wait_for_requests();
 377
 378     /* Update statistics */
 379     api::get().mc_inc_visited_states();
 380     api::get().mc_inc_executed_trans();
 381   }
 382 }
 383
 384 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value, int backtracking)
 385 {
 386   using simgrid::mc::CallType;
 387   switch(call_type) {
 388     case CallType::NONE:
 389       break;
 390     case CallType::SEND:
 391     case CallType::RECV:
 392       get_comm_pattern(req, call_type, backtracking);
 393       break;
 394     case CallType::WAIT:
 395     case CallType::WAITANY: {
 396       RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr;
 397       if (call_type == CallType::WAIT)
 398         comm_addr = api::get().get_comm_wait_raw_addr(req);
 399       else
 400         comm_addr = api::get().get_comm_waitany_raw_addr(req, value);
 401       auto simcall_issuer = api::get().simcall_get_issuer(req);
 402       complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
 403     } break;
 404   default:
 405     xbt_die("Unexpected call type %i", (int)call_type);
 406   }
 407 }
 408
 409 void CommunicationDeterminismChecker::real_run()
 410 {
 411   std::unique_ptr<VisitedState> visited_state = nullptr;
 412   smx_simcall_t req                           = nullptr;
 413
 414   while (not stack_.empty()) {
 415     /* Get current state */
 416     State* cur_state = stack_.back().get();
 417
 418     XBT_DEBUG("**************************************************");
 419     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 420               cur_state->count_todo());
 421
 422     /* Update statistics */
 423     api::get().mc_inc_visited_states();
 424
 425     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 426       req = api::get().mc_state_choose_request(cur_state);
 427     else
 428       req = nullptr;
 429
 430     if (req != nullptr && visited_state == nullptr) {
 431       int req_num = cur_state->transition_.times_considered_;
 432
 433       XBT_DEBUG("Execute: %s", api::get().request_to_string(req, req_num, RequestType::simix).c_str());
 434
 435       std::string req_str;
 436       if (dot_output != nullptr)
 437         req_str = api::get().request_get_dot_output(req, req_num);
 438
 439       api::get().mc_inc_executed_trans();
 440
 441       /* TODO : handle test and testany simcalls */
 442       CallType call = CallType::NONE;
 443       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 444         call = MC_get_call_type(req);
 445
 446       /* Answer the request */
 447       api::get().handle_simcall(cur_state->transition_);
 448       /* After this call req is no longer useful */
 449
 450       handle_comm_pattern(call, req, req_num, 0);
 451
 452       /* Wait for requests (schedules processes) */
 453       api::get().mc_wait_for_requests();
 454
 455       /* Create the new expanded state */
 456       ++expanded_states_count_;
 457       auto next_state = std::make_unique<State>(expanded_states_count_);
 458
 459       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 460        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 461        * with the initial pattern. */
 462       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 463
 464       if (_sg_mc_max_visited_states != 0)
 465         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
 466       else
 467         visited_state = nullptr;
 468
 469       if (visited_state == nullptr) {
 470         /* Get enabled actors and insert them in the interleave set of the next state */
 471         auto actors = api::get().get_actors();
 472         for (auto& actor : actors)
 473           if (api::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 474             next_state->mark_todo(actor.copy.get_buffer());
 475
 476         if (dot_output != nullptr)
 477           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 478
 479       } else if (dot_output != nullptr)
 480         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
 481                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 482
 483       stack_.push_back(std::move(next_state));
 484     } else {
 485       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
 486         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 487       else if (visited_state != nullptr)
 488         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
 489                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 490       else
 491         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 492
 493       this->initial_communications_pattern_done = true;
 494
 495       /* Trash the current state, no longer needed */
 496       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
 497       stack_.pop_back();
 498
 499       visited_state = nullptr;
 500
 501       api::get().mc_check_deadlock();
 502
 503       while (not stack_.empty()) {
 504         std::unique_ptr<State> state(std::move(stack_.back()));
 505         stack_.pop_back();
 506         if (state->count_todo() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 507           /* We found a back-tracking point, let's loop */
 508           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
 509           stack_.push_back(std::move(state));
 510
 511           this->restoreState();
 512
 513           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
 514
 515           break;
 516         } else {
 517           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
 518         }
 519       }
 520     }
 521   }
 522
 523   api::get().log_state();
 524 }
 525
 526 void CommunicationDeterminismChecker::run()
 527 {
 528   XBT_INFO("Check communication determinism");
 529   api::get().session_initialize();
 530
 531   this->prepare();
 532   this->real_run();
 533 }
 534
 535 Checker* createCommunicationDeterminismChecker()
 536 {
 537   return new CommunicationDeterminismChecker();
 538 }
 539
 540 } // namespace mc
 541 } // namespace simgrid