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 #include "src/mc/mc_smx.hpp"
  13
  14 #if HAVE_SMPI
  15 #include "smpi_request.hpp"
  16 #endif
  17
  18 #include <cstdint>
  19
  20 using api = simgrid::mc::Api;
  21
  22 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  23
  24 /********** Global variables **********/
  25
  26 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  27 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  28
  29 /********** Static functions ***********/
  30
  31 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
  32                                                                const simgrid::mc::PatternCommunication* comm2)
  33 {
  34   using simgrid::mc::CommPatternDifference;
  35   if (comm1->type != comm2->type)
  36     return CommPatternDifference::TYPE;
  37   if (comm1->rdv != comm2->rdv)
  38     return CommPatternDifference::RDV;
  39   if (comm1->src_proc != comm2->src_proc)
  40     return CommPatternDifference::SRC_PROC;
  41   if (comm1->dst_proc != comm2->dst_proc)
  42     return CommPatternDifference::DST_PROC;
  43   if (comm1->tag != comm2->tag)
  44     return CommPatternDifference::TAG;
  45   if (comm1->data.size() != comm2->data.size())
  46     return CommPatternDifference::DATA_SIZE;
  47   if (comm1->data != comm2->data)
  48     return CommPatternDifference::DATA;
  49   return CommPatternDifference::NONE;
  50 }
  51
  52 static void patterns_copy(std::vector<simgrid::mc::PatternCommunication*>& dest,
  53                              std::vector<simgrid::mc::PatternCommunication> const& source)
  54 {
  55   dest.clear();
  56   for (simgrid::mc::PatternCommunication const& comm : source) {
  57     auto* copy_comm = new simgrid::mc::PatternCommunication(comm.dup());
  58     dest.push_back(copy_comm);
  59   }
  60 }
  61
  62 static void restore_communications_pattern(simgrid::mc::State* state)
  63 {
  64   for (size_t i = 0; i < initial_communications_pattern.size(); i++)
  65     initial_communications_pattern[i].index_comm = state->communication_indices_[i];
  66
  67   for (unsigned long i = 0; i < api::get().get_maxpid(); i++)
  68     patterns_copy(incomplete_communications_pattern[i], state->incomplete_comm_pattern_[i]);
  69 }
  70
  71 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t process,
  72                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  73 {
  74   char* type;
  75   char* res;
  76
  77   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  78     type = bprintf("The send communications pattern of the process %ld is different!", process - 1);
  79   else
  80     type = bprintf("The recv communications pattern of the process %ld is different!", process - 1);
  81
  82   using simgrid::mc::CommPatternDifference;
  83   switch (diff) {
  84     case CommPatternDifference::TYPE:
  85       res = bprintf("%s Different type for communication #%u", type, cursor);
  86       break;
  87     case CommPatternDifference::RDV:
  88       res = bprintf("%s Different rdv for communication #%u", type, cursor);
  89       break;
  90     case CommPatternDifference::TAG:
  91       res = bprintf("%s Different tag for communication #%u", type, cursor);
  92       break;
  93     case CommPatternDifference::SRC_PROC:
  94       res = bprintf("%s Different source for communication #%u", type, cursor);
  95       break;
  96     case CommPatternDifference::DST_PROC:
  97       res = bprintf("%s Different destination for communication #%u", type, cursor);
  98       break;
  99     case CommPatternDifference::DATA_SIZE:
 100       res = bprintf("%s Different data size for communication #%u", type, cursor);
 101       break;
 102     case CommPatternDifference::DATA:
 103       res = bprintf("%s Different data for communication #%u", type, cursor);
 104       break;
 105     default:
 106       res = nullptr;
 107       break;
 108   }
 109
 110   return res;
 111 }
 112
 113 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
 114                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> const& comm_addr)
 115 {
 116   auto src_proc = api::get().get_src_actor(comm_addr);
 117   auto dst_proc = api::get().get_dst_actor(comm_addr);
 118   comm_pattern->src_proc = src_proc->get_pid();
 119   comm_pattern->dst_proc = dst_proc->get_pid();
 120   comm_pattern->src_host = &api::get().get_actor_host_name(src_proc);
 121   comm_pattern->dst_host = &api::get().get_actor_host_name(dst_proc);
 122
 123   if (comm_pattern->data.empty()) {
 124     comm_pattern->data = api::get().get_pattern_comm_data(comm_addr);
 125   }
 126 }
 127
 128 namespace simgrid {
 129 namespace mc {
 130
 131 void CommunicationDeterminismChecker::deterministic_comm_pattern(aid_t process, const PatternCommunication* comm,
 132                                                                  int backtracking)
 133 {
 134   if (not backtracking) {
 135     PatternCommunicationList& list = initial_communications_pattern[process];
 136     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 137
 138     if (diff != CommPatternDifference::NONE) {
 139       if (comm->type == PatternCommunicationType::send) {
 140         this->send_deterministic = false;
 141         if (this->send_diff != nullptr)
 142           xbt_free(this->send_diff);
 143         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 144       } else {
 145         this->recv_deterministic = false;
 146         if (this->recv_diff != nullptr)
 147           xbt_free(this->recv_diff);
 148         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 149       }
 150       if (_sg_mc_send_determinism && not this->send_deterministic) {
 151         XBT_INFO("*********************************************************");
 152         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 153         XBT_INFO("*********************************************************");
 154         XBT_INFO("%s", this->send_diff);
 155         xbt_free(this->send_diff);
 156         this->send_diff = nullptr;
 157         api::get().log_state();
 158         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 159       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
 160         XBT_INFO("****************************************************");
 161         XBT_INFO("***** Non-deterministic communications pattern *****");
 162         XBT_INFO("****************************************************");
 163         if (this->send_diff) {
 164           XBT_INFO("%s", this->send_diff);
 165           xbt_free(this->send_diff);
 166           this->send_diff = nullptr;
 167         }
 168         if (this->recv_diff) {
 169           XBT_INFO("%s", this->recv_diff);
 170           xbt_free(this->recv_diff);
 171           this->recv_diff = nullptr;
 172         }
 173         api::get().log_state();
 174         api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 175       }
 176     }
 177   }
 178 }
 179
 180 /********** Non Static functions ***********/
 181
 182 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, int backtracking)
 183 {
 184   const smx_actor_t issuer                                     = api::get().simcall_get_issuer(request);
 185   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 186   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 187
 188   auto pattern   = std::make_unique<PatternCommunication>();
 189   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 190
 191   if (call_type == CallType::SEND) {
 192     /* Create comm pattern */
 193     pattern->type      = PatternCommunicationType::send;
 194     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 195     pattern->rdv       = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 196     pattern->src_proc  = api::get().get_pattern_comm_src_proc(pattern->comm_addr);
 197     pattern->src_host  = &api::get().get_actor_host_name(issuer);
 198
 199 #if HAVE_SMPI
 200     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
 201 #endif
 202     pattern->data = api::get().get_pattern_comm_data(pattern->comm_addr);
 203
 204 #if HAVE_SMPI
 205     auto send_detached = api::get().check_send_request_detached(request);
 206     if (send_detached) {
 207       if (this->initial_communications_pattern_done) {
 208         /* Evaluate comm determinism */
 209         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
 210         initial_communications_pattern[pattern->src_proc].index_comm++;
 211       } else {
 212         /* Store comm pattern */
 213         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
 214       }
 215       return;
 216     }
 217 #endif
 218   } else if (call_type == CallType::RECV) {
 219     pattern->type = PatternCommunicationType::receive;
 220     pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
 221
 222 #if HAVE_SMPI
 223     pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
 224 #endif
 225     pattern->rdv = api::get().get_pattern_comm_rdv(pattern->comm_addr);
 226     pattern->dst_proc = api::get().get_pattern_comm_dst_proc(pattern->comm_addr);
 227     pattern->dst_host = &api::get().get_actor_host_name(issuer);
 228   } else
 229     xbt_die("Unexpected call_type %i", (int)call_type);
 230
 231   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
 232   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
 233 }
 234
 235 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> const& comm_addr, aid_t issuer,
 236                                                             int backtracking)
 237 {
 238   /* Complete comm pattern */
 239   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
 240   auto current_comm_pattern =
 241       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 242                    [&comm_addr](const PatternCommunication* comm) { return (comm->comm_addr == comm_addr); });
 243   if (current_comm_pattern == std::end(incomplete_pattern))
 244     xbt_die("Corresponding communication not found!");
 245
 246   update_comm_pattern(*current_comm_pattern, comm_addr);
 247   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 248   XBT_DEBUG("Remove incomplete comm pattern for process %ld at cursor %zd", issuer,
 249             std::distance(begin(incomplete_pattern), current_comm_pattern));
 250   incomplete_pattern.erase(current_comm_pattern);
 251
 252   if (this->initial_communications_pattern_done) {
 253     /* Evaluate comm determinism */
 254     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
 255     initial_communications_pattern[issuer].index_comm++;
 256   } else {
 257     /* Store comm pattern */
 258     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
 259   }
 260 }
 261
 262 CommunicationDeterminismChecker::CommunicationDeterminismChecker() : Checker() {}
 263
 264 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
 265
 266 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
 267 {
 268   RecordTrace res;
 269   for (auto const& state : stack_)
 270     res.push_back(state->get_transition());
 271   return res;
 272 }
 273
 274 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
 275 {
 276   std::vector<std::string> trace;
 277   for (auto const& state : stack_) {
 278     smx_simcall_t req = &state->executed_req_;
 279     trace.push_back(api::get().request_to_string(req, state->transition_.times_considered_, RequestType::executed));
 280   }
 281   return trace;
 282 }
 283
 284 void CommunicationDeterminismChecker::log_state() // override
 285 {
 286   if (_sg_mc_comms_determinism) {
 287     if (this->send_deterministic && not this->recv_deterministic) {
 288       XBT_INFO("*******************************************************");
 289       XBT_INFO("**** Only-send-deterministic communication pattern ****");
 290       XBT_INFO("*******************************************************");
 291       XBT_INFO("%s", this->recv_diff);
 292     }
 293     if (not this->send_deterministic && this->recv_deterministic) {
 294       XBT_INFO("*******************************************************");
 295       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 296       XBT_INFO("*******************************************************");
 297       XBT_INFO("%s", this->send_diff);
 298     }
 299   }
 300   XBT_INFO("Expanded states = %lu", expanded_states_count_);
 301   XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
 302   XBT_INFO("Executed transitions = %lu", api::get().mc_get_executed_trans());
 303   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
 304   if (_sg_mc_comms_determinism)
 305     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
 306 }
 307
 308 void CommunicationDeterminismChecker::prepare()
 309 {
 310   const auto maxpid = api::get().get_maxpid();
 311
 312   initial_communications_pattern.resize(maxpid);
 313   incomplete_communications_pattern.resize(maxpid);
 314
 315   ++expanded_states_count_;
 316   auto initial_state = std::make_unique<State>(expanded_states_count_);
 317
 318   XBT_DEBUG("********* Start communication determinism verification *********");
 319
 320   /* Get an enabled actor and insert it in the interleave set of the initial state */
 321   auto actors = api::get().get_actors();
 322   for (auto& actor : actors)
 323     if (api::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 324       initial_state->mark_todo(actor.copy.get_buffer());
 325
 326   stack_.push_back(std::move(initial_state));
 327 }
 328
 329 static inline bool all_communications_are_finished()
 330 {
 331   auto maxpid = api::get().get_maxpid();
 332   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
 333     if (not incomplete_communications_pattern[current_actor].empty()) {
 334       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 335       return false;
 336     }
 337   }
 338   return true;
 339 }
 340
 341 void CommunicationDeterminismChecker::restoreState()
 342 {
 343   /* Intermediate backtracking */
 344   State* last_state = stack_.back().get();
 345   if (last_state->system_state_) {
 346     api::get().restore_state(last_state->system_state_);
 347     restore_communications_pattern(last_state);
 348     return;
 349   }
 350
 351   /* Restore the initial state */
 352   api::get().restore_initial_state();
 353
 354   unsigned long n = api::get().get_maxpid();
 355   assert(n == incomplete_communications_pattern.size());
 356   assert(n == initial_communications_pattern.size());
 357   for (unsigned long j = 0; j < n; j++) {
 358     incomplete_communications_pattern[j].clear();
 359     initial_communications_pattern[j].index_comm = 0;
 360   }
 361
 362   /* Traverse the stack from the state at position start and re-execute the transitions */
 363   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
 364     if (state == stack_.back())
 365       break;
 366
 367     int req_num                    = state->transition_.times_considered_;
 368     const s_smx_simcall* saved_req = &state->executed_req_;
 369     xbt_assert(saved_req);
 370
 371     /* because we got a copy of the executed request, we have to fetch the
 372        real one, pointed by the request field of the issuer process */
 373
 374     const smx_actor_t issuer = api::get().simcall_get_issuer(saved_req);
 375     smx_simcall_t req        = &issuer->simcall_;
 376
 377     /* TODO : handle test and testany simcalls */
 378     CallType call = MC_get_call_type(req);
 379     api::get().handle_simcall(state->transition_);
 380     handle_comm_pattern(call, req, req_num, 1);
 381     api::get().mc_wait_for_requests();
 382
 383     /* Update statistics */
 384     api::get().mc_inc_visited_states();
 385     api::get().mc_inc_executed_trans();
 386   }
 387 }
 388
 389 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value, int backtracking)
 390 {
 391   using simgrid::mc::CallType;
 392   switch(call_type) {
 393     case CallType::NONE:
 394       break;
 395     case CallType::SEND:
 396     case CallType::RECV:
 397       get_comm_pattern(req, call_type, backtracking);
 398       break;
 399     case CallType::WAIT:
 400     case CallType::WAITANY: {
 401       RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr;
 402       if (call_type == CallType::WAIT)
 403         comm_addr = api::get().get_comm_wait_raw_addr(req);
 404       else
 405         comm_addr = api::get().get_comm_waitany_raw_addr(req, value);
 406       auto simcall_issuer = api::get().simcall_get_issuer(req);
 407       complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
 408     } break;
 409   default:
 410     xbt_die("Unexpected call type %i", (int)call_type);
 411   }
 412 }
 413
 414 void CommunicationDeterminismChecker::real_run()
 415 {
 416   std::unique_ptr<VisitedState> visited_state = nullptr;
 417   smx_simcall_t req                           = nullptr;
 418
 419   while (not stack_.empty()) {
 420     /* Get current state */
 421     State* cur_state = stack_.back().get();
 422
 423     XBT_DEBUG("**************************************************");
 424     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 425               cur_state->count_todo());
 426
 427     /* Update statistics */
 428     api::get().mc_inc_visited_states();
 429
 430     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 431       req = api::get().mc_state_choose_request(cur_state);
 432     else
 433       req = nullptr;
 434
 435     if (req != nullptr && visited_state == nullptr) {
 436       int req_num = cur_state->transition_.times_considered_;
 437
 438       XBT_DEBUG("Execute: %s", api::get().request_to_string(req, req_num, RequestType::simix).c_str());
 439
 440       std::string req_str;
 441       if (dot_output != nullptr)
 442         req_str = api::get().request_get_dot_output(req, req_num);
 443
 444       api::get().mc_inc_executed_trans();
 445
 446       /* TODO : handle test and testany simcalls */
 447       CallType call = CallType::NONE;
 448       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 449         call = MC_get_call_type(req);
 450
 451       /* Answer the request */
 452       api::get().handle_simcall(cur_state->transition_);
 453       /* After this call req is no longer useful */
 454
 455       handle_comm_pattern(call, req, req_num, 0);
 456
 457       /* Wait for requests (schedules processes) */
 458       api::get().mc_wait_for_requests();
 459
 460       /* Create the new expanded state */
 461       ++expanded_states_count_;
 462       auto next_state = std::make_unique<State>(expanded_states_count_);
 463
 464       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 465        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 466        * with the initial pattern. */
 467       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 468
 469       if (_sg_mc_max_visited_states != 0)
 470         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
 471       else
 472         visited_state = nullptr;
 473
 474       if (visited_state == nullptr) {
 475         /* Get enabled actors and insert them in the interleave set of the next state */
 476         auto actors = api::get().get_actors();
 477         for (auto& actor : actors)
 478           if (api::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 479             next_state->mark_todo(actor.copy.get_buffer());
 480
 481         if (dot_output != nullptr)
 482           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 483
 484       } else if (dot_output != nullptr)
 485         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
 486                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 487
 488       stack_.push_back(std::move(next_state));
 489     } else {
 490       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
 491         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 492       else if (visited_state != nullptr)
 493         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
 494                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 495       else
 496         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 497
 498       this->initial_communications_pattern_done = true;
 499
 500       /* Trash the current state, no longer needed */
 501       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
 502       stack_.pop_back();
 503
 504       visited_state = nullptr;
 505
 506       api::get().mc_check_deadlock();
 507
 508       while (not stack_.empty()) {
 509         std::unique_ptr<State> state(std::move(stack_.back()));
 510         stack_.pop_back();
 511         if (state->count_todo() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 512           /* We found a back-tracking point, let's loop */
 513           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
 514           stack_.push_back(std::move(state));
 515
 516           this->restoreState();
 517
 518           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
 519
 520           break;
 521         } else {
 522           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
 523         }
 524       }
 525     }
 526   }
 527
 528   api::get().log_state();
 529 }
 530
 531 void CommunicationDeterminismChecker::run()
 532 {
 533   XBT_INFO("Check communication determinism");
 534   api::get().session_initialize();
 535
 536   this->prepare();
 537   this->real_run();
 538 }
 539
 540 Checker* createCommunicationDeterminismChecker()
 541 {
 542   return new CommunicationDeterminismChecker();
 543 }
 544
 545 } // namespace mc
 546 } // namespace simgrid