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, int 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, aid_t issuer,
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 smx_simcall_t req = &state->executed_req_;
274 trace.push_back(api::get().request_to_string(req, state->transition_.times_considered_));
279 void CommunicationDeterminismChecker::log_state() // override
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);
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);
295 XBT_INFO("Expanded states = %lu", expanded_states_count_);
296 XBT_INFO("Visited states = %lu", api::get().mc_get_visited_states());
297 XBT_INFO("Executed transitions = %lu", api::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");
303 void CommunicationDeterminismChecker::prepare()
305 const unsigned long maxpid = api::get().get_maxpid();
307 initial_communications_pattern.resize(maxpid);
308 incomplete_communications_pattern.resize(maxpid);
310 ++expanded_states_count_;
311 auto initial_state = std::make_unique<State>(expanded_states_count_);
313 XBT_DEBUG("********* Start communication determinism verification *********");
315 /* Add all enabled actors to the interleave set of the initial state */
316 for (auto& act : api::get().get_actors()) {
317 auto actor = act.copy.get_buffer();
318 if (get_session().actor_is_enabled(actor->get_pid()))
319 initial_state->mark_todo(actor);
322 stack_.push_back(std::move(initial_state));
325 static inline bool all_communications_are_finished()
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.");
337 void CommunicationDeterminismChecker::restoreState()
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);
347 get_session().restore_initial_state();
349 const unsigned long maxpid = api::get().get_maxpid();
350 assert(maxpid == incomplete_communications_pattern.size());
351 assert(maxpid == initial_communications_pattern.size());
352 for (unsigned long j = 0; j < maxpid; j++) {
353 incomplete_communications_pattern[j].clear();
354 initial_communications_pattern[j].index_comm = 0;
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())
362 int req_num = state->transition_.times_considered_;
363 const s_smx_simcall* saved_req = &state->executed_req_;
364 xbt_assert(saved_req);
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 */
369 const smx_actor_t issuer = api::get().simcall_get_issuer(saved_req);
370 smx_simcall_t req = &issuer->simcall_;
372 /* TODO : handle test and testany simcalls */
373 CallType call = MC_get_call_type(req);
374 api::get().handle_simcall(state->transition_);
375 handle_comm_pattern(call, req, req_num, 1);
376 api::get().mc_wait_for_requests();
378 /* Update statistics */
379 api::get().mc_inc_visited_states();
380 api::get().mc_inc_executed_trans();
384 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value, int backtracking)
386 using simgrid::mc::CallType;
392 get_comm_pattern(req, call_type, backtracking);
395 case CallType::WAITANY: {
396 RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr;
397 if (call_type == CallType::WAIT)
398 comm_addr = remote(simcall_comm_wait__getraw__comm(req));
400 comm_addr = api::get().get_comm_waitany_raw_addr(req, value);
401 auto simcall_issuer = api::get().simcall_get_issuer(req);
402 complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
405 xbt_die("Unexpected call type %i", (int)call_type);
409 void CommunicationDeterminismChecker::real_run()
411 std::unique_ptr<VisitedState> visited_state = nullptr;
412 smx_simcall_t req = nullptr;
414 while (not stack_.empty()) {
415 /* Get current state */
416 State* cur_state = stack_.back().get();
418 XBT_DEBUG("**************************************************");
419 XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
420 cur_state->count_todo());
422 /* Update statistics */
423 api::get().mc_inc_visited_states();
425 if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
426 req = api::get().mc_state_choose_request(cur_state);
430 if (req != nullptr && visited_state == nullptr) {
431 int req_num = cur_state->transition_.times_considered_;
433 XBT_DEBUG("Execute: %s", api::get().request_to_string(req, req_num).c_str());
436 if (dot_output != nullptr)
437 req_str = api::get().request_get_dot_output(req, req_num);
439 api::get().mc_inc_executed_trans();
441 /* TODO : handle test and testany simcalls */
442 CallType call = CallType::NONE;
443 if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
444 call = MC_get_call_type(req);
446 /* Answer the request */
447 api::get().handle_simcall(cur_state->transition_);
448 /* After this call req is no longer useful */
450 handle_comm_pattern(call, req, req_num, 0);
452 /* Wait for requests (schedules processes) */
453 api::get().mc_wait_for_requests();
455 /* Create the new expanded state */
456 ++expanded_states_count_;
457 auto next_state = std::make_unique<State>(expanded_states_count_);
459 /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
460 * least, data are transferred). These communications are incomplete and they cannot be analyzed and compared
461 * with the initial pattern. */
462 bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
464 if (_sg_mc_max_visited_states != 0)
465 visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
467 visited_state = nullptr;
469 if (visited_state == nullptr) {
470 /* Add all enabled actors to the interleave set of the next state */
471 for (auto& act : api::get().get_actors()) {
472 auto actor = act.copy.get_buffer();
473 if (get_session().actor_is_enabled(actor->get_pid()))
474 next_state->mark_todo(actor);
477 if (dot_output != nullptr)
478 fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
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());
484 stack_.push_back(std::move(next_state));
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);
492 XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
494 this->initial_communications_pattern_done = true;
496 /* Trash the current state, no longer needed */
497 XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
500 visited_state = nullptr;
502 api::get().mc_check_deadlock();
504 while (not stack_.empty()) {
505 std::unique_ptr<State> state(std::move(stack_.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));
512 this->restoreState();
514 XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
518 XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
524 api::get().log_state();
527 void CommunicationDeterminismChecker::run()
529 XBT_INFO("Check communication determinism");
530 get_session().take_initial_snapshot();
536 Checker* create_communication_determinism_checker(Session* session)
538 return new CommunicationDeterminismChecker(session);
542 } // namespace simgrid