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