1 /* Copyright (c) 2008-2017. 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. */
9 #include <xbt/dynar.hpp>
11 #include <xbt/sysdep.h>
13 #include "src/mc/Transition.hpp"
14 #include "src/mc/VisitedState.hpp"
15 #include "src/mc/checker/CommunicationDeterminismChecker.hpp"
16 #include "src/mc/mc_exit.hpp"
17 #include "src/mc/mc_private.hpp"
18 #include "src/mc/mc_record.hpp"
19 #include "src/mc/mc_request.hpp"
20 #include "src/mc/mc_smx.hpp"
21 #include "src/mc/mc_state.hpp"
22 #include "src/mc/remote/Client.hpp"
24 #include "smpi_request.hpp"
26 using simgrid::mc::remote;
28 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
30 /********** Global variables **********/
32 xbt_dynar_t initial_communications_pattern;
33 xbt_dynar_t incomplete_communications_pattern;
35 /********** Static functions ***********/
37 static e_mc_comm_pattern_difference_t compare_comm_pattern(simgrid::mc::PatternCommunication* comm1,
38 simgrid::mc::PatternCommunication* comm2)
40 if(comm1->type != comm2->type)
42 if (comm1->rdv != comm2->rdv)
44 if (comm1->src_proc != comm2->src_proc)
46 if (comm1->dst_proc != comm2->dst_proc)
48 if (comm1->tag != comm2->tag)
50 if (comm1->data.size() != comm2->data.size())
51 return DATA_SIZE_DIFF;
52 if (comm1->data != comm2->data)
57 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process,
58 simgrid::mc::PatternCommunication* comm, unsigned int cursor)
63 if (comm->type == simgrid::mc::PatternCommunicationType::send)
64 type = bprintf("The send communications pattern of the process %d is different!", process - 1);
66 type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
70 res = bprintf("%s Different type for communication #%u", type, cursor);
73 res = bprintf("%s Different rdv for communication #%u", type, cursor);
76 res = bprintf("%s Different tag for communication #%u", type, cursor);
79 res = bprintf("%s Different source for communication #%u", type, cursor);
82 res = bprintf("%s Different destination for communication #%u", type, cursor);
85 res = bprintf("%s\n Different data size for communication #%u", type, cursor);
88 res = bprintf("%s\n Different data for communication #%u", type, cursor);
98 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
99 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr)
101 // HACK, type punning
102 simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_comm;
103 mc_model_checker->process().read(temp_comm, comm_addr);
104 simgrid::kernel::activity::CommImpl* comm = temp_comm.getBuffer();
106 smx_actor_t src_proc = mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->src_proc));
107 smx_actor_t dst_proc = mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->dst_proc));
108 comm_pattern->src_proc = src_proc->pid;
109 comm_pattern->dst_proc = dst_proc->pid;
110 comm_pattern->src_host = MC_smx_actor_get_host_name(src_proc);
111 comm_pattern->dst_host = MC_smx_actor_get_host_name(dst_proc);
112 if (comm_pattern->data.size() == 0 && comm->src_buff != nullptr) {
114 mc_model_checker->process().read(&buff_size, remote(comm->dst_buff_size));
115 comm_pattern->data.resize(buff_size);
116 mc_model_checker->process().read_bytes(comm_pattern->data.data(), comm_pattern->data.size(),
117 remote(comm->src_buff));
124 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, simgrid::mc::PatternCommunication* comm,
127 simgrid::mc::PatternCommunicationList* list =
128 xbt_dynar_get_as(initial_communications_pattern, process, simgrid::mc::PatternCommunicationList*);
130 if (not backtracking) {
131 e_mc_comm_pattern_difference_t diff = compare_comm_pattern(list->list[list->index_comm].get(), comm);
133 if (diff != NONE_DIFF) {
134 if (comm->type == simgrid::mc::PatternCommunicationType::send) {
135 this->send_deterministic = 0;
136 if (this->send_diff != nullptr)
137 xbt_free(this->send_diff);
138 this->send_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
140 this->recv_deterministic = 0;
141 if (this->recv_diff != nullptr)
142 xbt_free(this->recv_diff);
143 this->recv_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
145 if (_sg_mc_send_determinism && not this->send_deterministic) {
146 XBT_INFO("*********************************************************");
147 XBT_INFO("***** Non-send-deterministic communications pattern *****");
148 XBT_INFO("*********************************************************");
149 XBT_INFO("%s", this->send_diff);
150 xbt_free(this->send_diff);
151 this->send_diff = nullptr;
152 simgrid::mc::session->logState();
153 mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
154 } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
155 XBT_INFO("****************************************************");
156 XBT_INFO("***** Non-deterministic communications pattern *****");
157 XBT_INFO("****************************************************");
158 XBT_INFO("%s", this->send_diff);
159 XBT_INFO("%s", this->recv_diff);
160 xbt_free(this->send_diff);
161 this->send_diff = nullptr;
162 xbt_free(this->recv_diff);
163 this->recv_diff = nullptr;
164 simgrid::mc::session->logState();
165 mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
171 /********** Non Static functions ***********/
173 void CommunicationDeterminismChecker::get_comm_pattern(xbt_dynar_t list, smx_simcall_t request,
174 e_mc_call_type_t call_type, int backtracking)
176 const smx_actor_t issuer = MC_smx_simcall_get_issuer(request);
177 simgrid::mc::PatternCommunicationList* initial_pattern =
178 xbt_dynar_get_as(initial_communications_pattern, issuer->pid, simgrid::mc::PatternCommunicationList*);
179 xbt_dynar_t incomplete_pattern = xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
181 std::unique_ptr<simgrid::mc::PatternCommunication> pattern =
182 std::unique_ptr<simgrid::mc::PatternCommunication>(new simgrid::mc::PatternCommunication());
183 pattern->index = initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
185 if (call_type == MC_CALL_TYPE_SEND) {
186 /* Create comm pattern */
187 pattern->type = simgrid::mc::PatternCommunicationType::send;
188 pattern->comm_addr = static_cast<simgrid::kernel::activity::CommImpl*>(simcall_comm_isend__getraw__result(request));
190 simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_synchro;
191 mc_model_checker->process().read(temp_synchro,
192 remote(static_cast<simgrid::kernel::activity::CommImpl*>(pattern->comm_addr)));
193 simgrid::kernel::activity::CommImpl* synchro =
194 static_cast<simgrid::kernel::activity::CommImpl*>(temp_synchro.getBuffer());
196 char* remote_name = mc_model_checker->process().read<char*>(
197 (std::uint64_t)(synchro->mbox ? &synchro->mbox->name_ : &synchro->mbox_cpy->name_));
198 pattern->rdv = mc_model_checker->process().read_string(remote_name);
199 pattern->src_proc = mc_model_checker->process().resolveActor(simgrid::mc::remote(synchro->src_proc))->pid;
200 pattern->src_host = MC_smx_actor_get_host_name(issuer);
202 simgrid::smpi::Request mpi_request =
203 mc_model_checker->process().read<simgrid::smpi::Request>((std::uint64_t)simcall_comm_isend__get__data(request));
204 pattern->tag = mpi_request.tag();
206 if (synchro->src_buff != nullptr) {
207 pattern->data.resize(synchro->src_buff_size);
208 mc_model_checker->process().read_bytes(pattern->data.data(), pattern->data.size(), remote(synchro->src_buff));
210 if(mpi_request.detached()){
211 if (not this->initial_communications_pattern_done) {
212 /* Store comm pattern */
213 simgrid::mc::PatternCommunicationList* list =
214 xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, simgrid::mc::PatternCommunicationList*);
215 list->list.push_back(std::move(pattern));
217 /* Evaluate comm determinism */
218 this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
219 xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, simgrid::mc::PatternCommunicationList*)
224 } else if (call_type == MC_CALL_TYPE_RECV) {
225 pattern->type = simgrid::mc::PatternCommunicationType::receive;
226 pattern->comm_addr = static_cast<simgrid::kernel::activity::CommImpl*>(simcall_comm_irecv__getraw__result(request));
228 simgrid::smpi::Request mpi_request;
229 mc_model_checker->process().read(&mpi_request,
230 remote((simgrid::smpi::Request*)simcall_comm_irecv__get__data(request)));
231 pattern->tag = mpi_request.tag();
233 simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_comm;
234 mc_model_checker->process().read(temp_comm,
235 remote(static_cast<simgrid::kernel::activity::CommImpl*>(pattern->comm_addr)));
236 simgrid::kernel::activity::CommImpl* comm = temp_comm.getBuffer();
239 mc_model_checker->process().read(
240 &remote_name, remote(comm->mbox ? &simgrid::xbt::string::to_string_data(comm->mbox->name_).data
241 : &simgrid::xbt::string::to_string_data(comm->mbox_cpy->name_).data));
242 pattern->rdv = mc_model_checker->process().read_string(remote_name);
243 pattern->dst_proc = mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->dst_proc))->pid;
244 pattern->dst_host = MC_smx_actor_get_host_name(issuer);
246 xbt_die("Unexpected call_type %i", (int) call_type);
248 XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern.get(), issuer->pid);
249 xbt_dynar_t dynar = xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
250 simgrid::mc::PatternCommunication* pattern2 = pattern.release();
251 xbt_dynar_push(dynar, &pattern2);
254 void CommunicationDeterminismChecker::complete_comm_pattern(
255 xbt_dynar_t list, simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr, unsigned int issuer,
258 simgrid::mc::PatternCommunication* current_comm_pattern;
259 unsigned int cursor = 0;
260 std::unique_ptr<simgrid::mc::PatternCommunication> comm_pattern;
263 /* Complete comm pattern */
264 xbt_dynar_foreach(xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern)
265 if (remote(current_comm_pattern->comm_addr) == comm_addr) {
266 update_comm_pattern(current_comm_pattern, comm_addr);
268 simgrid::mc::PatternCommunication* temp;
269 xbt_dynar_remove_at(xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &temp);
270 comm_pattern = std::unique_ptr<simgrid::mc::PatternCommunication>(temp);
271 XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
276 xbt_die("Corresponding communication not found!");
278 simgrid::mc::PatternCommunicationList* pattern =
279 xbt_dynar_get_as(initial_communications_pattern, issuer, simgrid::mc::PatternCommunicationList*);
281 if (not this->initial_communications_pattern_done)
282 /* Store comm pattern */
283 pattern->list.push_back(std::move(comm_pattern));
285 /* Evaluate comm determinism */
286 this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
287 pattern->index_comm++;
291 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& session) : Checker(session)
295 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
297 RecordTrace CommunicationDeterminismChecker::getRecordTrace() // override
300 for (auto const& state : stack_)
301 res.push_back(state->getTransition());
305 std::vector<std::string> CommunicationDeterminismChecker::getTextualTrace() // override
307 std::vector<std::string> trace;
308 for (auto const& state : stack_) {
309 smx_simcall_t req = &state->executed_req;
312 simgrid::mc::request_to_string(req, state->transition.argument, simgrid::mc::RequestType::executed));
317 void CommunicationDeterminismChecker::logState() // override
319 if (_sg_mc_comms_determinism && not this->recv_deterministic && this->send_deterministic) {
320 XBT_INFO("******************************************************");
321 XBT_INFO("**** Only-send-deterministic communication pattern ****");
322 XBT_INFO("******************************************************");
323 XBT_INFO("%s", this->recv_diff);
324 } else if (_sg_mc_comms_determinism && not this->send_deterministic && this->recv_deterministic) {
325 XBT_INFO("******************************************************");
326 XBT_INFO("**** Only-recv-deterministic communication pattern ****");
327 XBT_INFO("******************************************************");
328 XBT_INFO("%s", this->send_diff);
330 XBT_INFO("Expanded states = %lu", expandedStatesCount_);
331 XBT_INFO("Visited states = %lu", mc_model_checker->visited_states);
332 XBT_INFO("Executed transitions = %lu", mc_model_checker->executed_transitions);
333 XBT_INFO("Send-deterministic : %s", not this->send_deterministic ? "No" : "Yes");
334 if (_sg_mc_comms_determinism)
335 XBT_INFO("Recv-deterministic : %s", not this->recv_deterministic ? "No" : "Yes");
338 void CommunicationDeterminismChecker::prepare()
340 const int maxpid = MC_smx_get_maxpid();
342 // Create initial_communications_pattern elements:
343 initial_communications_pattern = simgrid::xbt::newDeleteDynar<simgrid::mc::PatternCommunicationList*>();
344 for (int i = 0; i < maxpid; i++) {
345 simgrid::mc::PatternCommunicationList* process_list_pattern = new simgrid::mc::PatternCommunicationList();
346 xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
349 // Create incomplete_communications_pattern elements:
350 incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
351 for (int i = 0; i < maxpid; i++) {
352 xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(simgrid::mc::PatternCommunication*), nullptr);
353 xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
356 std::unique_ptr<simgrid::mc::State> initial_state =
357 std::unique_ptr<simgrid::mc::State>(new simgrid::mc::State(++expandedStatesCount_));
359 XBT_DEBUG("********* Start communication determinism verification *********");
361 /* Get an enabled actor and insert it in the interleave set of the initial state */
362 for (auto& actor : mc_model_checker->process().actors())
363 if (simgrid::mc::actor_is_enabled(actor.copy.getBuffer()))
364 initial_state->addInterleavingSet(actor.copy.getBuffer());
366 stack_.push_back(std::move(initial_state));
369 static inline bool all_communications_are_finished()
371 for (size_t current_actor = 1; current_actor < MC_smx_get_maxpid(); current_actor++) {
372 xbt_dynar_t pattern = xbt_dynar_get_as(incomplete_communications_pattern, current_actor, xbt_dynar_t);
373 if (not xbt_dynar_is_empty(pattern)) {
374 XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
381 void CommunicationDeterminismChecker::restoreState()
383 /* Intermediate backtracking */
384 simgrid::mc::State* state = stack_.back().get();
385 if (state->system_state) {
386 simgrid::mc::restore_snapshot(state->system_state);
387 MC_restore_communications_pattern(state);
391 /* Restore the initial state */
392 simgrid::mc::session->restoreInitialState();
394 unsigned n = MC_smx_get_maxpid();
395 assert(n == xbt_dynar_length(incomplete_communications_pattern));
396 assert(n == xbt_dynar_length(initial_communications_pattern));
397 for (unsigned j=0; j < n ; j++) {
398 xbt_dynar_reset((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, j, xbt_dynar_t));
399 xbt_dynar_get_as(initial_communications_pattern, j, simgrid::mc::PatternCommunicationList*)->index_comm = 0;
402 /* Traverse the stack from the state at position start and re-execute the transitions */
403 for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
404 if (state == stack_.back())
407 int req_num = state->transition.argument;
408 smx_simcall_t saved_req = &state->executed_req;
409 xbt_assert(saved_req);
411 /* because we got a copy of the executed request, we have to fetch the
412 real one, pointed by the request field of the issuer process */
414 const smx_actor_t issuer = MC_smx_simcall_get_issuer(saved_req);
415 smx_simcall_t req = &issuer->simcall;
417 /* TODO : handle test and testany simcalls */
418 e_mc_call_type_t call = MC_get_call_type(req);
419 mc_model_checker->handle_simcall(state->transition);
420 MC_handle_comm_pattern(call, req, req_num, nullptr, 1);
421 mc_model_checker->wait_for_requests();
423 /* Update statistics */
424 mc_model_checker->visited_states++;
425 mc_model_checker->executed_transitions++;
429 void CommunicationDeterminismChecker::main()
431 std::unique_ptr<simgrid::mc::VisitedState> visited_state = nullptr;
432 smx_simcall_t req = nullptr;
434 while (not stack_.empty()) {
435 /* Get current state */
436 simgrid::mc::State* state = stack_.back().get();
438 XBT_DEBUG("**************************************************");
439 XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), state->num,
440 state->interleaveSize());
442 /* Update statistics */
443 mc_model_checker->visited_states++;
445 if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
446 req = MC_state_get_request(state);
450 if (req != nullptr && visited_state == nullptr) {
452 int req_num = state->transition.argument;
454 XBT_DEBUG("Execute: %s", simgrid::mc::request_to_string(req, req_num, simgrid::mc::RequestType::simix).c_str());
457 if (dot_output != nullptr)
458 req_str = simgrid::mc::request_get_dot_output(req, req_num);
460 mc_model_checker->executed_transitions++;
462 /* TODO : handle test and testany simcalls */
463 e_mc_call_type_t call = MC_CALL_TYPE_NONE;
464 if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
465 call = MC_get_call_type(req);
467 /* Answer the request */
468 mc_model_checker->handle_simcall(state->transition);
469 /* After this call req is no longer useful */
471 if (not this->initial_communications_pattern_done)
472 MC_handle_comm_pattern(call, req, req_num, initial_communications_pattern, 0);
474 MC_handle_comm_pattern(call, req, req_num, nullptr, 0);
476 /* Wait for requests (schedules processes) */
477 mc_model_checker->wait_for_requests();
479 /* Create the new expanded state */
480 std::unique_ptr<simgrid::mc::State> next_state =
481 std::unique_ptr<simgrid::mc::State>(new simgrid::mc::State(++expandedStatesCount_));
483 /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
484 * least, data are transferred). These communications are incomplete and they cannot be analyzed and compared
485 * with the initial pattern. */
486 bool compare_snapshots = all_communications_are_finished() && this->initial_communications_pattern_done;
488 if (_sg_mc_max_visited_states != 0)
489 visited_state = visitedStates_.addVisitedState(expandedStatesCount_, next_state.get(), compare_snapshots);
491 visited_state = nullptr;
493 if (visited_state == nullptr) {
495 /* Get enabled actors and insert them in the interleave set of the next state */
496 for (auto& actor : mc_model_checker->process().actors())
497 if (simgrid::mc::actor_is_enabled(actor.copy.getBuffer()))
498 next_state->addInterleavingSet(actor.copy.getBuffer());
500 if (dot_output != nullptr)
501 fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n",
502 state->num, next_state->num, req_str.c_str());
504 } else if (dot_output != nullptr)
505 fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
506 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
508 stack_.push_back(std::move(next_state));
512 if (stack_.size() > (std::size_t) _sg_mc_max_depth)
513 XBT_WARN("/!\\ Max depth reached! /!\\ ");
514 else if (visited_state != nullptr)
515 XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
516 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
518 XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
520 if (not this->initial_communications_pattern_done)
521 this->initial_communications_pattern_done = 1;
523 /* Trash the current state, no longer needed */
524 XBT_DEBUG("Delete state %d at depth %zu", state->num, stack_.size());
527 visited_state = nullptr;
529 /* Check for deadlocks */
530 if (mc_model_checker->checkDeadlock()) {
532 throw simgrid::mc::DeadlockError();
535 while (not stack_.empty()) {
536 std::unique_ptr<simgrid::mc::State> state = std::move(stack_.back());
538 if (state->interleaveSize() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
539 /* We found a back-tracking point, let's loop */
540 XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num, stack_.size() + 1);
541 stack_.push_back(std::move(state));
543 this->restoreState();
545 XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num, stack_.size());
549 XBT_DEBUG("Delete state %d at depth %zu", state->num, stack_.size() + 1);
555 simgrid::mc::session->logState();
558 void CommunicationDeterminismChecker::run()
560 XBT_INFO("Check communication determinism");
561 simgrid::mc::session->initialize();
568 Checker* createCommunicationDeterminismChecker(Session& session)
570 return new CommunicationDeterminismChecker(session);