src/mc/checker/CommunicationDeterminismChecker.cpp

   1 /* Copyright (c) 2008-2020. 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_api.hpp"
  10 #include "src/mc/mc_config.hpp"
  11 #include "src/mc/mc_exit.hpp"
  12 #include "src/mc/mc_private.hpp"
  13 #include "src/mc/mc_request.hpp"
  14 #include "src/mc/mc_smx.hpp"
  15
  16 #if HAVE_SMPI
  17 #include "smpi_request.hpp"
  18 #endif
  19
  20 #include <cstdint>
  21
  22 using mcapi = simgrid::mc::mc_api;
  23
  24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  25
  26 /********** Global variables **********/
  27
  28 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  29 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  30
  31 /********** Static functions ***********/
  32
  33 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
  34                                                                const simgrid::mc::PatternCommunication* comm2)
  35 {
  36   using simgrid::mc::CommPatternDifference;
  37   if (comm1->type != comm2->type)
  38     return CommPatternDifference::TYPE;
  39   if (comm1->rdv != comm2->rdv)
  40     return CommPatternDifference::RDV;
  41   if (comm1->src_proc != comm2->src_proc)
  42     return CommPatternDifference::SRC_PROC;
  43   if (comm1->dst_proc != comm2->dst_proc)
  44     return CommPatternDifference::DST_PROC;
  45   if (comm1->tag != comm2->tag)
  46     return CommPatternDifference::TAG;
  47   if (comm1->data.size() != comm2->data.size())
  48     return CommPatternDifference::DATA_SIZE;
  49   if (comm1->data != comm2->data)
  50     return CommPatternDifference::DATA;
  51   return CommPatternDifference::NONE;
  52 }
  53
  54 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, int process,
  55                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  56 {
  57   char* type;
  58   char* res;
  59
  60   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  61     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  62   else
  63     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  64
  65   using simgrid::mc::CommPatternDifference;
  66   switch (diff) {
  67     case CommPatternDifference::TYPE:
  68       res = bprintf("%s Different type for communication #%u", type, cursor);
  69       break;
  70     case CommPatternDifference::RDV:
  71       res = bprintf("%s Different rdv for communication #%u", type, cursor);
  72       break;
  73     case CommPatternDifference::TAG:
  74       res = bprintf("%s Different tag for communication #%u", type, cursor);
  75       break;
  76     case CommPatternDifference::SRC_PROC:
  77       res = bprintf("%s Different source for communication #%u", type, cursor);
  78       break;
  79     case CommPatternDifference::DST_PROC:
  80       res = bprintf("%s Different destination for communication #%u", type, cursor);
  81       break;
  82     case CommPatternDifference::DATA_SIZE:
  83       res = bprintf("%s Different data size for communication #%u", type, cursor);
  84       break;
  85     case CommPatternDifference::DATA:
  86       res = bprintf("%s Different data for communication #%u", type, cursor);
  87       break;
  88     default:
  89       res = nullptr;
  90       break;
  91   }
  92
  93   return res;
  94 }
  95
  96 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
  97                                 const simgrid::kernel::activity::CommImpl* comm_addr)
  98 {
  99   auto src_proc = mcapi::get().get_src_actor(comm_addr);
 100   auto dst_proc = mcapi::get().get_dst_actor(comm_addr);
 101   comm_pattern->src_proc = src_proc->get_pid();
 102   comm_pattern->dst_proc = dst_proc->get_pid();
 103   comm_pattern->src_host = mcapi::get().get_actor_host_name(src_proc);
 104   comm_pattern->dst_host = mcapi::get().get_actor_host_name(dst_proc);
 105
 106   if (comm_pattern->data.empty()) {
 107     auto pattern_data = mcapi::get().get_pattern_comm_data(comm_addr);
 108     if (pattern_data.data() != nullptr) {
 109       auto data_size = pattern_data.size();
 110       comm_pattern->data.resize(data_size);
 111       memcpy(comm_pattern->data.data(), pattern_data.data(), data_size);
 112     }
 113    }
 114 }
 115
 116 namespace simgrid {
 117 namespace mc {
 118
 119 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, const PatternCommunication* comm,
 120                                                                  int backtracking)
 121 {
 122   if (not backtracking) {
 123     PatternCommunicationList& list = initial_communications_pattern[process];
 124     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 125
 126     if (diff != CommPatternDifference::NONE) {
 127       if (comm->type == PatternCommunicationType::send) {
 128         this->send_deterministic = false;
 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 = false;
 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         mcapi::get().log_state();
 146         mcapi::get().mc_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         mcapi::get().log_state();
 162         mcapi::get().mc_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, CallType call_type, int backtracking)
 171 {
 172   const smx_actor_t issuer                                     = mcapi::get().simcall_get_issuer(request);
 173   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 174   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 175
 176   auto pattern   = std::make_unique<PatternCommunication>();
 177   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 178
 179   if (call_type == CallType::SEND) {
 180     /* Create comm pattern */
 181     pattern->type      = PatternCommunicationType::send;
 182     pattern->comm_addr = mcapi::get().get_comm_isend_raw_addr(request);
 183     pattern->rdv      = mcapi::get().get_pattern_comm_rdv(pattern->comm_addr);
 184     pattern->src_proc = mcapi::get().get_pattern_comm_src_proc(pattern->comm_addr);
 185     pattern->src_host = mc_api::get().get_actor_host_name(issuer);
 186
 187 #if HAVE_SMPI
 188     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
 189 #endif
 190     auto pattern_data = mcapi::get().get_pattern_comm_data(pattern->comm_addr);
 191     if (pattern_data.data() != nullptr) {
 192       auto data_size = pattern_data.size();
 193       pattern->data.resize(data_size);
 194       memcpy(pattern->data.data(), pattern_data.data(), data_size);
 195     }
 196
 197 #if HAVE_SMPI
 198     auto send_detached = mcapi::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 = mcapi::get().get_comm_isend_raw_addr(request);
 214
 215 #if HAVE_SMPI
 216     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
 217 #endif
 218     auto comm_addr = pattern->comm_addr;
 219     pattern->rdv = mcapi::get().get_pattern_comm_rdv(comm_addr);
 220     pattern->dst_proc = mcapi::get().get_pattern_comm_dst_proc(comm_addr);
 221     pattern->dst_host = mcapi::get().get_actor_host_name(issuer);
 222   } else
 223     xbt_die("Unexpected call_type %i", (int)call_type);
 224
 225   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
 226   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
 227 }
 228
 229 void CommunicationDeterminismChecker::complete_comm_pattern(const kernel::activity::CommImpl* comm_addr,
 230                                                             unsigned int issuer, int backtracking)
 231 {
 232   /* Complete comm pattern */
 233   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
 234   auto current_comm_pattern =
 235       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
 236                    [&comm_addr](const PatternCommunication* comm) { return mcapi::get().comm_addr_equal(comm->comm_addr, comm_addr); });
 237   if (current_comm_pattern == std::end(incomplete_pattern))
 238     xbt_die("Corresponding communication not found!");
 239
 240   update_comm_pattern(*current_comm_pattern, comm_addr);
 241   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
 242   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
 243             std::distance(begin(incomplete_pattern), current_comm_pattern));
 244   incomplete_pattern.erase(current_comm_pattern);
 245
 246   if (this->initial_communications_pattern_done) {
 247     /* Evaluate comm determinism */
 248     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
 249     initial_communications_pattern[issuer].index_comm++;
 250   } else {
 251     /* Store comm pattern */
 252     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
 253   }
 254 }
 255
 256 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s) {}
 257
 258 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
 259
 260 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
 261 {
 262   RecordTrace res;
 263   for (auto const& state : stack_)
 264     res.push_back(state->get_transition());
 265   return res;
 266 }
 267
 268 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
 269 {
 270   std::vector<std::string> trace;
 271   for (auto const& state : stack_) {
 272     smx_simcall_t req = &state->executed_req_;
 273     trace.push_back(mcapi::get().request_to_string(req, state->transition_.argument_, RequestType::executed));
 274   }
 275   return trace;
 276 }
 277
 278 void CommunicationDeterminismChecker::log_state() // override
 279 {
 280   if (_sg_mc_comms_determinism) {
 281     if (this->send_deterministic && not this->recv_deterministic) {
 282       XBT_INFO("*******************************************************");
 283       XBT_INFO("**** Only-send-deterministic communication pattern ****");
 284       XBT_INFO("*******************************************************");
 285       XBT_INFO("%s", this->recv_diff);
 286     }
 287     if (not this->send_deterministic && this->recv_deterministic) {
 288       XBT_INFO("*******************************************************");
 289       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
 290       XBT_INFO("*******************************************************");
 291       XBT_INFO("%s", this->send_diff);
 292     }
 293   }
 294   XBT_INFO("Expanded states = %lu", expanded_states_count_);
 295   XBT_INFO("Visited states = %lu", mcapi::get().mc_get_visited_states());
 296   XBT_INFO("Executed transitions = %lu", mcapi::get().mc_get_executed_trans());
 297   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
 298   if (_sg_mc_comms_determinism)
 299     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
 300 }
 301
 302 void CommunicationDeterminismChecker::prepare()
 303 {
 304   const int maxpid = mcapi::get().get_maxpid();
 305
 306   initial_communications_pattern.resize(maxpid);
 307   incomplete_communications_pattern.resize(maxpid);
 308
 309   ++expanded_states_count_;
 310   auto initial_state = std::make_unique<State>(expanded_states_count_);
 311
 312   XBT_DEBUG("********* Start communication determinism verification *********");
 313
 314   /* Get an enabled actor and insert it in the interleave set of the initial state */
 315   auto actors = mcapi::get().get_actors();
 316   for (auto& actor : actors)
 317     if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 318       initial_state->add_interleaving_set(actor.copy.get_buffer());
 319
 320   stack_.push_back(std::move(initial_state));
 321 }
 322
 323 static inline bool all_communications_are_finished()
 324 {
 325   auto maxpid = mcapi::get().get_maxpid();
 326   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
 327     if (not incomplete_communications_pattern[current_actor].empty()) {
 328       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
 329       return false;
 330     }
 331   }
 332   return true;
 333 }
 334
 335 void CommunicationDeterminismChecker::restoreState()
 336 {
 337   /* Intermediate backtracking */
 338   State* last_state = stack_.back().get();
 339   if (last_state->system_state_) {
 340     mc_api::get().restore_state(last_state->system_state_);
 341     MC_restore_communications_pattern(last_state);
 342     return;
 343   }
 344
 345   /* Restore the initial state */
 346   mcapi::get().restore_initial_state();
 347
 348   unsigned n = mcapi::get().get_maxpid();
 349   assert(n == incomplete_communications_pattern.size());
 350   assert(n == initial_communications_pattern.size());
 351   for (unsigned j = 0; j < n; j++) {
 352     incomplete_communications_pattern[j].clear();
 353     initial_communications_pattern[j].index_comm = 0;
 354   }
 355
 356   /* Traverse the stack from the state at position start and re-execute the transitions */
 357   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
 358     if (state == stack_.back())
 359       break;
 360
 361     int req_num                    = state->transition_.argument_;
 362     const s_smx_simcall* saved_req = &state->executed_req_;
 363     xbt_assert(saved_req);
 364
 365     /* because we got a copy of the executed request, we have to fetch the
 366        real one, pointed by the request field of the issuer process */
 367
 368     const smx_actor_t issuer = mcapi::get().simcall_get_issuer(saved_req);
 369     smx_simcall_t req        = &issuer->simcall_;
 370
 371     /* TODO : handle test and testany simcalls */
 372     CallType call = MC_get_call_type(req);
 373     mcapi::get().handle_simcall(state->transition_);
 374     handle_comm_pattern(call, req, req_num, 1);
 375     mcapi::get().mc_wait_for_requests();
 376
 377     /* Update statistics */
 378     mcapi::get().mc_inc_visited_states();
 379     mcapi::get().mc_inc_executed_trans();
 380   }
 381 }
 382
 383 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value, int backtracking)
 384 {
 385   using simgrid::mc::CallType;
 386   switch(call_type) {
 387     case CallType::NONE:
 388       break;
 389     case CallType::SEND:
 390     case CallType::RECV:
 391       get_comm_pattern(req, call_type, backtracking);
 392       break;
 393     case CallType::WAIT:
 394     case CallType::WAITANY: {
 395       simgrid::kernel::activity::CommImpl* comm_addr = nullptr;
 396       if (call_type == CallType::WAIT)
 397         comm_addr = mcapi::get().get_comm_wait_raw_addr(req);
 398       else
 399         comm_addr = mcapi::get().get_comm_waitany_raw_addr(req, value);
 400       auto simcall_issuer = mcapi::get().simcall_get_issuer(req);
 401       complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
 402     } break;
 403   default:
 404     xbt_die("Unexpected call type %i", (int)call_type);
 405   }
 406 }
 407
 408 void CommunicationDeterminismChecker::real_run()
 409 {
 410   std::unique_ptr<VisitedState> visited_state = nullptr;
 411   smx_simcall_t req                           = nullptr;
 412
 413   while (not stack_.empty()) {
 414     /* Get current state */
 415     State* cur_state = stack_.back().get();
 416
 417     XBT_DEBUG("**************************************************");
 418     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 419               cur_state->interleave_size());
 420
 421     /* Update statistics */
 422     mcapi::get().mc_inc_visited_states();
 423
 424     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 425       req = mcapi::get().mc_state_choose_request(cur_state);
 426     else
 427       req = nullptr;
 428
 429     if (req != nullptr && visited_state == nullptr) {
 430       int req_num = cur_state->transition_.argument_;
 431
 432       XBT_DEBUG("Execute: %s", mcapi::get().request_to_string(req, req_num, RequestType::simix).c_str());
 433
 434       std::string req_str;
 435       if (dot_output != nullptr)
 436         req_str = mcapi::get().request_get_dot_output(req, req_num);
 437
 438       mcapi::get().mc_inc_executed_trans();
 439
 440       /* TODO : handle test and testany simcalls */
 441       CallType call = CallType::NONE;
 442       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 443         call = MC_get_call_type(req);
 444
 445       /* Answer the request */
 446       mcapi::get().handle_simcall(cur_state->transition_);
 447       /* After this call req is no longer useful */
 448
 449       handle_comm_pattern(call, req, req_num, 0);
 450
 451       /* Wait for requests (schedules processes) */
 452       mcapi::get().mc_wait_for_requests();
 453
 454       /* Create the new expanded state */
 455       ++expanded_states_count_;
 456       auto next_state = std::make_unique<State>(expanded_states_count_);
 457
 458       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 459        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 460        * with the initial pattern. */
 461       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 462
 463       if (_sg_mc_max_visited_states != 0)
 464         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
 465       else
 466         visited_state = nullptr;
 467
 468       if (visited_state == nullptr) {
 469         /* Get enabled actors and insert them in the interleave set of the next state */
 470         auto actors = mcapi::get().get_actors();
 471         for (auto& actor : actors)
 472           if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 473             next_state->add_interleaving_set(actor.copy.get_buffer());
 474
 475         if (dot_output != nullptr)
 476           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 477
 478       } else if (dot_output != nullptr)
 479         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
 480                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 481
 482       stack_.push_back(std::move(next_state));
 483     } else {
 484       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
 485         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 486       else if (visited_state != nullptr)
 487         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
 488                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 489       else
 490         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 491
 492       this->initial_communications_pattern_done = true;
 493
 494       /* Trash the current state, no longer needed */
 495       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
 496       stack_.pop_back();
 497
 498       visited_state = nullptr;
 499
 500       /* Check for deadlocks */
 501       if (mcapi::get().mc_check_deadlock()) {
 502         mcapi::get().mc_show_deadlock();
 503         throw simgrid::mc::DeadlockError();
 504       }
 505
 506       while (not stack_.empty()) {
 507         std::unique_ptr<State> state(std::move(stack_.back()));
 508         stack_.pop_back();
 509         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 510           /* We found a back-tracking point, let's loop */
 511           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
 512           stack_.push_back(std::move(state));
 513
 514           this->restoreState();
 515
 516           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
 517
 518           break;
 519         } else {
 520           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
 521         }
 522       }
 523     }
 524   }
 525
 526   mcapi::get().log_state();
 527 }
 528
 529 void CommunicationDeterminismChecker::run()
 530 {
 531   XBT_INFO("Check communication determinism");
 532   mcapi::get().s_initialize();
 533
 534   this->prepare();
 535   this->real_run();
 536 }
 537
 538 Checker* createCommunicationDeterminismChecker(Session& s)
 539 {
 540   return new CommunicationDeterminismChecker(s);
 541 }
 542
 543 } // namespace mc
 544 } // namespace simgrid