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