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