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 simgrid::mc::remote;
  23 using mcapi = simgrid::mc::mc_api;
  24
  25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
  26
  27 /********** Global variables **********/
  28
  29 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
  30 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
  31
  32 /********** Static functions ***********/
  33
  34 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
  35                                                                const simgrid::mc::PatternCommunication* comm2)
  36 {
  37   using simgrid::mc::CommPatternDifference;
  38   if (comm1->type != comm2->type)
  39     return CommPatternDifference::TYPE;
  40   if (comm1->rdv != comm2->rdv)
  41     return CommPatternDifference::RDV;
  42   if (comm1->src_proc != comm2->src_proc)
  43     return CommPatternDifference::SRC_PROC;
  44   if (comm1->dst_proc != comm2->dst_proc)
  45     return CommPatternDifference::DST_PROC;
  46   if (comm1->tag != comm2->tag)
  47     return CommPatternDifference::TAG;
  48   if (comm1->data.size() != comm2->data.size())
  49     return CommPatternDifference::DATA_SIZE;
  50   if (comm1->data != comm2->data)
  51     return CommPatternDifference::DATA;
  52   return CommPatternDifference::NONE;
  53 }
  54
  55 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, int process,
  56                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
  57 {
  58   char* type;
  59   char* res;
  60
  61   if (comm->type == simgrid::mc::PatternCommunicationType::send)
  62     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  63   else
  64     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  65
  66   using simgrid::mc::CommPatternDifference;
  67   switch (diff) {
  68     case CommPatternDifference::TYPE:
  69       res = bprintf("%s Different type for communication #%u", type, cursor);
  70       break;
  71     case CommPatternDifference::RDV:
  72       res = bprintf("%s Different rdv for communication #%u", type, cursor);
  73       break;
  74     case CommPatternDifference::TAG:
  75       res = bprintf("%s Different tag for communication #%u", type, cursor);
  76       break;
  77     case CommPatternDifference::SRC_PROC:
  78       res = bprintf("%s Different source for communication #%u", type, cursor);
  79       break;
  80     case CommPatternDifference::DST_PROC:
  81       res = bprintf("%s Different destination for communication #%u", type, cursor);
  82       break;
  83     case CommPatternDifference::DATA_SIZE:
  84       res = bprintf("%s Different data size for communication #%u", type, cursor);
  85       break;
  86     case CommPatternDifference::DATA:
  87       res = bprintf("%s Different data for communication #%u", type, cursor);
  88       break;
  89     default:
  90       res = nullptr;
  91       break;
  92   }
  93
  94   return res;
  95 }
  96
  97 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
  98                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr)
  99 {
 100   auto src_proc = mcapi::get().get_src_actor(comm_addr);
 101   auto dst_proc = mcapi::get().get_dst_actor(comm_addr);
 102   comm_pattern->src_proc = src_proc->get_pid();
 103   comm_pattern->dst_proc = dst_proc->get_pid();
 104   comm_pattern->src_host = mcapi::get().get_actor_host_name(src_proc);
 105   comm_pattern->dst_host = mcapi::get().get_actor_host_name(dst_proc);
 106
 107   if (comm_pattern->data.empty()) {
 108     auto pattern_data = mcapi::get().get_pattern_comm_data(comm_addr);
 109     if (pattern_data.data() != nullptr) {
 110       auto data_size = pattern_data.size();
 111       comm_pattern->data.resize(data_size);
 112       memcpy(comm_pattern->data.data(), pattern_data.data(), data_size);
 113     }
 114    }
 115 }
 116
 117 namespace simgrid {
 118 namespace mc {
 119
 120 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, const PatternCommunication* comm,
 121                                                                  int backtracking)
 122 {
 123   if (not backtracking) {
 124     PatternCommunicationList& list = initial_communications_pattern[process];
 125     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
 126
 127     if (diff != CommPatternDifference::NONE) {
 128       if (comm->type == PatternCommunicationType::send) {
 129         this->send_deterministic = false;
 130         if (this->send_diff != nullptr)
 131           xbt_free(this->send_diff);
 132         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 133       } else {
 134         this->recv_deterministic = false;
 135         if (this->recv_diff != nullptr)
 136           xbt_free(this->recv_diff);
 137         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
 138       }
 139       if (_sg_mc_send_determinism && not this->send_deterministic) {
 140         XBT_INFO("*********************************************************");
 141         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 142         XBT_INFO("*********************************************************");
 143         XBT_INFO("%s", this->send_diff);
 144         xbt_free(this->send_diff);
 145         this->send_diff = nullptr;
 146         mcapi::get().log_state();
 147         mcapi::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 148       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
 149         XBT_INFO("****************************************************");
 150         XBT_INFO("***** Non-deterministic communications pattern *****");
 151         XBT_INFO("****************************************************");
 152         if (this->send_diff) {
 153           XBT_INFO("%s", this->send_diff);
 154           xbt_free(this->send_diff);
 155           this->send_diff = nullptr;
 156         }
 157         if (this->recv_diff) {
 158           XBT_INFO("%s", this->recv_diff);
 159           xbt_free(this->recv_diff);
 160           this->recv_diff = nullptr;
 161         }
 162         mcapi::get().log_state();
 163         mcapi::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
 164       }
 165     }
 166   }
 167 }
 168
 169 /********** Non Static functions ***********/
 170
 171 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, int backtracking)
 172 {
 173   const smx_actor_t issuer                                     = mcapi::get().mc_smx_simcall_get_issuer(request);
 174   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
 175   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
 176
 177   auto pattern   = std::make_unique<PatternCommunication>();
 178   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
 179
 180   if (call_type == CallType::SEND) {
 181     /* Create comm pattern */
 182     pattern->type      = PatternCommunicationType::send;
 183     pattern->comm_addr = mcapi::get().get_pattern_comm_addr(request);
 184     pattern->rdv      = mcapi::get().get_pattern_comm_rdv(pattern->comm_addr);
 185     pattern->src_proc = mcapi::get().get_pattern_comm_src_proc(pattern->comm_addr);
 186     pattern->src_host = mc_api::get().get_actor_host_name(issuer);
 187
 188 #if HAVE_SMPI
 189     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
 190 #endif
 191     auto pattern_data = mcapi::get().get_pattern_comm_data(pattern->comm_addr);
 192     if (pattern_data.data() != nullptr) {
 193       auto data_size = pattern_data.size();
 194       pattern->data.resize(data_size);
 195       memcpy(pattern->data.data(), pattern_data.data(), data_size);
 196     }
 197
 198 #if HAVE_SMPI
 199     auto send_detached = mcapi::get().check_send_request_detached(request);
 200     if (send_detached) {
 201       if (this->initial_communications_pattern_done) {
 202         /* Evaluate comm determinism */
 203         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
 204         initial_communications_pattern[pattern->src_proc].index_comm++;
 205       } else {
 206         /* Store comm pattern */
 207         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
 208       }
 209       return;
 210     }
 211 #endif
 212   } else if (call_type == CallType::RECV) {
 213     pattern->type = PatternCommunicationType::receive;
 214     pattern->comm_addr = mcapi::get().get_pattern_comm_addr(request);
 215
 216 #if HAVE_SMPI
 217     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
 218 #endif
 219     auto comm_addr = pattern->comm_addr;
 220     pattern->rdv = mcapi::get().get_pattern_comm_rdv(comm_addr);
 221     pattern->dst_proc = mcapi::get().get_pattern_comm_dst_proc(comm_addr);
 222     pattern->dst_host = mcapi::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> comm_addr,
 231                                                             unsigned int issuer, 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 remote(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 %u 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(Session& s) : Checker(s) {}
 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(mcapi::get().request_to_string(req, state->transition_.argument_, 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", mcapi::get().mc_get_visited_states());
 297   XBT_INFO("Executed transitions = %lu", mcapi::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 int maxpid = mcapi::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 = mcapi::get().get_actors();
 317   for (auto& actor : actors)
 318     if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 319       initial_state->add_interleaving_set(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 = mcapi::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     mc_api::get().restore_state(last_state->system_state_);
 342     MC_restore_communications_pattern(last_state);
 343     return;
 344   }
 345
 346   /* Restore the initial state */
 347   mcapi::get().s_restore_initial_state();
 348
 349   unsigned n = mcapi::get().get_maxpid();
 350   assert(n == incomplete_communications_pattern.size());
 351   assert(n == initial_communications_pattern.size());
 352   for (unsigned 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_.argument_;
 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 = mcapi::get().mc_smx_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     mcapi::get().handle_simcall(state->transition_);
 375     MC_handle_comm_pattern(call, req, req_num, 1);
 376     mcapi::get().mc_wait_for_requests();
 377
 378     /* Update statistics */
 379     mcapi::get().mc_inc_visited_states();
 380     mcapi::get().mc_inc_executed_trans();
 381   }
 382 }
 383
 384 void CommunicationDeterminismChecker::real_run()
 385 {
 386   std::unique_ptr<VisitedState> visited_state = nullptr;
 387   smx_simcall_t req                           = nullptr;
 388
 389   while (not stack_.empty()) {
 390     /* Get current state */
 391     State* cur_state = stack_.back().get();
 392
 393     XBT_DEBUG("**************************************************");
 394     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
 395               cur_state->interleave_size());
 396
 397     /* Update statistics */
 398     mcapi::get().mc_inc_visited_states();
 399
 400     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
 401       req = mcapi::get().mc_state_choose_request(cur_state);
 402     else
 403       req = nullptr;
 404
 405     if (req != nullptr && visited_state == nullptr) {
 406       int req_num = cur_state->transition_.argument_;
 407
 408       XBT_DEBUG("Execute: %s", mcapi::get().request_to_string(req, req_num, RequestType::simix).c_str());
 409
 410       std::string req_str;
 411       if (dot_output != nullptr)
 412         req_str = mcapi::get().request_get_dot_output(req, req_num);
 413
 414       mcapi::get().mc_inc_executed_trans();
 415
 416       /* TODO : handle test and testany simcalls */
 417       CallType call = CallType::NONE;
 418       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
 419         call = MC_get_call_type(req);
 420
 421       /* Answer the request */
 422       mcapi::get().handle_simcall(cur_state->transition_);
 423       /* After this call req is no longer useful */
 424
 425       MC_handle_comm_pattern(call, req, req_num, 0);
 426
 427       /* Wait for requests (schedules processes) */
 428       mcapi::get().mc_wait_for_requests();
 429
 430       /* Create the new expanded state */
 431       ++expanded_states_count_;
 432       auto next_state = std::make_unique<State>(expanded_states_count_);
 433
 434       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
 435        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
 436        * with the initial pattern. */
 437       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
 438
 439       if (_sg_mc_max_visited_states != 0)
 440         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
 441       else
 442         visited_state = nullptr;
 443
 444       if (visited_state == nullptr) {
 445         /* Get enabled actors and insert them in the interleave set of the next state */
 446         auto actors = mcapi::get().get_actors();
 447         for (auto& actor : actors)
 448           if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
 449             next_state->add_interleaving_set(actor.copy.get_buffer());
 450
 451         if (dot_output != nullptr)
 452           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
 453
 454       } else if (dot_output != nullptr)
 455         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
 456                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
 457
 458       stack_.push_back(std::move(next_state));
 459     } else {
 460       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
 461         XBT_WARN("/!\\ Max depth reached! /!\\ ");
 462       else if (visited_state != nullptr)
 463         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
 464                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
 465       else
 466         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
 467
 468       this->initial_communications_pattern_done = true;
 469
 470       /* Trash the current state, no longer needed */
 471       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
 472       stack_.pop_back();
 473
 474       visited_state = nullptr;
 475
 476       /* Check for deadlocks */
 477       if (mcapi::get().mc_check_deadlock()) {
 478         mcapi::get().mc_show_deadlock();
 479         throw simgrid::mc::DeadlockError();
 480       }
 481
 482       while (not stack_.empty()) {
 483         std::unique_ptr<State> state(std::move(stack_.back()));
 484         stack_.pop_back();
 485         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
 486           /* We found a back-tracking point, let's loop */
 487           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
 488           stack_.push_back(std::move(state));
 489
 490           this->restoreState();
 491
 492           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
 493
 494           break;
 495         } else {
 496           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
 497         }
 498       }
 499     }
 500   }
 501
 502   mcapi::get().log_state();
 503 }
 504
 505 void CommunicationDeterminismChecker::run()
 506 {
 507   XBT_INFO("Check communication determinism");
 508   mcapi::get().s_initialize();
 509
 510   this->prepare();
 511   this->real_run();
 512 }
 513
 514 Checker* createCommunicationDeterminismChecker(Session& s)
 515 {
 516   return new CommunicationDeterminismChecker(s);
 517 }
 518
 519 } // namespace mc
 520 } // namespace simgrid