1 /* Copyright (c) 2008-2022. The SimGrid Team. All rights reserved. */
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. */
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_config.hpp"
10 #include "src/mc/mc_exit.hpp"
11 #include "src/mc/mc_private.hpp"
15 using api = simgrid::mc::Api;
17 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
19 /********** Global variables **********/
21 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
22 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
24 /********** Static functions ***********/
26 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
27 const simgrid::mc::PatternCommunication* comm2)
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;
47 static void patterns_copy(std::vector<simgrid::mc::PatternCommunication*>& dest,
48 std::vector<simgrid::mc::PatternCommunication> const& source)
51 for (simgrid::mc::PatternCommunication const& comm : source) {
52 auto* copy_comm = new simgrid::mc::PatternCommunication(comm.dup());
53 dest.push_back(copy_comm);
57 static void restore_communications_pattern(simgrid::mc::State* state)
59 for (size_t i = 0; i < initial_communications_pattern.size(); i++)
60 initial_communications_pattern[i].index_comm = state->communication_indices_[i];
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]);
67 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, aid_t process,
68 const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
73 if (comm->type == simgrid::mc::PatternCommunicationType::send)
74 type = bprintf("The send communications pattern of the process %ld is different!", process - 1);
76 type = bprintf("The recv communications pattern of the process %ld is different!", process - 1);
78 using simgrid::mc::CommPatternDifference;
80 case CommPatternDifference::TYPE:
81 res = bprintf("%s Different type for communication #%u", type, cursor);
83 case CommPatternDifference::RDV:
84 res = bprintf("%s Different rdv for communication #%u", type, cursor);
86 case CommPatternDifference::TAG:
87 res = bprintf("%s Different tag for communication #%u", type, cursor);
89 case CommPatternDifference::SRC_PROC:
90 res = bprintf("%s Different source for communication #%u", type, cursor);
92 case CommPatternDifference::DST_PROC:
93 res = bprintf("%s Different destination for communication #%u", type, cursor);
95 case CommPatternDifference::DATA_SIZE:
96 res = bprintf("%s Different data size for communication #%u", type, cursor);
98 case CommPatternDifference::DATA:
99 res = bprintf("%s Different data for communication #%u", type, cursor);
109 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
110 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> const& comm_addr)
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);
119 if (comm_pattern->data.empty()) {
120 comm_pattern->data = api::get().get_pattern_comm_data(comm_addr);
127 void CommunicationDeterminismChecker::deterministic_comm_pattern(aid_t process, const PatternCommunication* comm,
130 if (not backtracking) {
131 PatternCommunicationList& list = initial_communications_pattern[process];
132 CommPatternDifference diff = compare_comm_pattern(list.list[list.index_comm].get(), comm);
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);
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);
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;
164 if (this->recv_diff) {
165 XBT_INFO("%s", this->recv_diff);
166 xbt_free(this->recv_diff);
167 this->recv_diff = nullptr;
169 api::get().log_state();
170 api::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
176 /********** Non Static functions ***********/
178 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, bool backtracking)
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()];
184 auto pattern = std::make_unique<PatternCommunication>();
185 pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
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);
196 pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
198 pattern->data = api::get().get_pattern_comm_data(pattern->comm_addr);
201 auto send_detached = api::get().check_send_request_detached(request);
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++;
208 /* Store comm pattern */
209 initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
214 } else if (call_type == CallType::RECV) {
215 pattern->type = PatternCommunicationType::receive;
216 pattern->comm_addr = api::get().get_comm_isend_raw_addr(request);
219 pattern->tag = api::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
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);
225 xbt_die("Unexpected call_type %i", (int)call_type);
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());
231 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> const& comm_addr,
232 aid_t issuer, bool backtracking)
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 xbt_assert(current_comm_pattern != std::end(incomplete_pattern), "Corresponding communication not found!");
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 %ld at cursor %zd", issuer,
244 std::distance(begin(incomplete_pattern), current_comm_pattern));
245 incomplete_pattern.erase(current_comm_pattern);
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++;
252 /* Store comm pattern */
253 initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
257 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session* session) : Checker(session) {}
259 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
261 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
264 for (auto const& state : stack_)
265 res.push_back(*state->get_transition());
269 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
271 std::vector<std::string> trace;
272 for (auto const& state : stack_)
273 trace.push_back(state->get_transition()->to_string());
277 void CommunicationDeterminismChecker::log_state() // override
279 if (_sg_mc_comms_determinism) {
280 if (this->send_deterministic && not this->recv_deterministic) {
281 XBT_INFO("*******************************************************");
282 XBT_INFO("**** Only-send-deterministic communication pattern ****");
283 XBT_INFO("*******************************************************");
284 XBT_INFO("%s", this->recv_diff);
286 if (not this->send_deterministic && this->recv_deterministic) {
287 XBT_INFO("*******************************************************");
288 XBT_INFO("**** Only-recv-deterministic communication pattern ****");
289 XBT_INFO("*******************************************************");
290 XBT_INFO("%s", this->send_diff);
293 XBT_INFO("Expanded states = %lu", expanded_states_count_);
294 XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
295 XBT_INFO("Executed transitions = %lu", Transition::get_executed_transitions());
296 XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
297 if (_sg_mc_comms_determinism)
298 XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
301 void CommunicationDeterminismChecker::prepare()
303 const unsigned long maxpid = api::get().get_maxpid();
305 initial_communications_pattern.resize(maxpid);
306 incomplete_communications_pattern.resize(maxpid);
308 ++expanded_states_count_;
309 auto initial_state = std::make_unique<State>(expanded_states_count_);
311 XBT_DEBUG("********* Start communication determinism verification *********");
313 /* Add all enabled actors to the interleave set of the initial state */
314 for (auto& act : api::get().get_actors()) {
315 auto actor = act.copy.get_buffer();
316 if (get_session().actor_is_enabled(actor->get_pid()))
317 initial_state->mark_todo(actor->get_pid());
320 stack_.push_back(std::move(initial_state));
323 static inline bool all_communications_are_finished()
325 const unsigned long maxpid = api::get().get_maxpid();
326 for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
327 if (not incomplete_communications_pattern[current_actor].empty()) {
328 XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
335 void CommunicationDeterminismChecker::restoreState()
337 /* Intermediate backtracking */
338 State* last_state = stack_.back().get();
339 if (last_state->system_state_) {
340 api::get().restore_state(last_state->system_state_);
341 restore_communications_pattern(last_state);
345 get_session().restore_initial_state();
347 const unsigned long maxpid = api::get().get_maxpid();
348 assert(maxpid == incomplete_communications_pattern.size());
349 assert(maxpid == initial_communications_pattern.size());
350 for (unsigned long j = 0; j < maxpid; j++) {
351 incomplete_communications_pattern[j].clear();
352 initial_communications_pattern[j].index_comm = 0;
355 /* Traverse the stack from the state at position start and re-execute the transitions */
356 for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
357 if (state == stack_.back())
360 int req_num = state->get_transition()->times_considered_;
361 const s_smx_simcall* saved_req = &state->executed_req_;
362 xbt_assert(saved_req);
364 /* because we got a copy of the executed request, we have to fetch the
365 real one, pointed by the request field of the issuer process */
367 const smx_actor_t issuer = api::get().simcall_get_issuer(saved_req);
368 smx_simcall_t req = &issuer->simcall_;
370 /* TODO : handle test and testany simcalls */
371 CallType call = MC_get_call_type(req);
372 state->get_transition()->replay();
373 handle_comm_pattern(call, req, req_num, 1);
375 /* Update statistics */
376 api::get().mc_inc_visited_states();
380 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value,
383 using simgrid::mc::CallType;
389 get_comm_pattern(req, call_type, backtracking);
392 case CallType::WAITANY: {
393 RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr;
394 if (call_type == CallType::WAIT)
395 comm_addr = remote(simcall_comm_wait__getraw__comm(req));
397 comm_addr = api::get().get_comm_waitany_raw_addr(req, value);
398 auto simcall_issuer = api::get().simcall_get_issuer(req);
399 complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
402 xbt_die("Unexpected call type %i", (int)call_type);
406 void CommunicationDeterminismChecker::real_run()
408 std::unique_ptr<VisitedState> visited_state = nullptr;
410 while (not stack_.empty()) {
411 /* Get current state */
412 State* cur_state = stack_.back().get();
414 XBT_DEBUG("**************************************************");
415 XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
416 cur_state->count_todo());
418 /* Update statistics */
419 api::get().mc_inc_visited_states();
421 int next_transition = -1;
422 if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
423 next_transition = cur_state->next_transition();
425 if (next_transition >= 0 && visited_state == nullptr) {
426 cur_state->execute_next(next_transition);
428 aid_t aid = cur_state->get_transition()->aid_;
429 int req_num = cur_state->get_transition()->times_considered_;
430 smx_simcall_t req = &cur_state->executed_req_;
432 XBT_DEBUG("Execute: %s", cur_state->get_transition()->to_string().c_str());
435 if (dot_output != nullptr)
436 req_str = api::get().request_get_dot_output(aid, req_num);
438 /* TODO : handle test and testany simcalls */
439 CallType call = CallType::NONE;
440 if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
441 call = MC_get_call_type(req);
443 handle_comm_pattern(call, req, req_num, 0);
445 /* Create the new expanded state */
446 ++expanded_states_count_;
447 auto next_state = std::make_unique<State>(expanded_states_count_);
449 /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
450 * least, data are transferred). These communications are incomplete and they cannot be analyzed and compared
451 * with the initial pattern. */
452 bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
454 if (_sg_mc_max_visited_states != 0)
455 visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
457 visited_state = nullptr;
459 if (visited_state == nullptr) {
460 /* Add all enabled actors to the interleave set of the next state */
461 for (auto& act : api::get().get_actors()) {
462 auto actor = act.copy.get_buffer();
463 if (get_session().actor_is_enabled(actor->get_pid()))
464 next_state->mark_todo(actor->get_pid());
467 if (dot_output != nullptr)
468 fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
470 } else if (dot_output != nullptr)
471 fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
472 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
474 stack_.push_back(std::move(next_state));
476 if (stack_.size() > (std::size_t)_sg_mc_max_depth)
477 XBT_WARN("/!\\ Max depth reached! /!\\ ");
478 else if (visited_state != nullptr)
479 XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
480 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
482 XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
484 this->initial_communications_pattern_done = true;
486 /* Trash the current state, no longer needed */
487 XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
490 visited_state = nullptr;
492 api::get().mc_check_deadlock();
494 while (not stack_.empty()) {
495 std::unique_ptr<State> state(std::move(stack_.back()));
497 if (state->count_todo() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
498 /* We found a back-tracking point, let's loop */
499 XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
500 stack_.push_back(std::move(state));
502 this->restoreState();
504 XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
508 XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
514 api::get().log_state();
517 void CommunicationDeterminismChecker::run()
519 XBT_INFO("Check communication determinism");
520 get_session().take_initial_snapshot();
526 Checker* create_communication_determinism_checker(Session* session)
528 return new CommunicationDeterminismChecker(session);
532 } // namespace simgrid