Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge remote-tracking branch 'upstream/master' into wifi_energy
[simgrid.git] / 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_config.hpp"
10 #include "src/mc/mc_exit.hpp"
11 #include "src/mc/mc_private.hpp"
12 #include "src/mc/mc_request.hpp"
13 #include "src/mc/mc_smx.hpp"
14
15 #if HAVE_SMPI
16 #include "smpi_request.hpp"
17 #endif
18
19 #include <cstdint>
20
21 using simgrid::mc::remote;
22
23 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
24
25 /********** Global variables **********/
26
27 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
28 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
29
30 /********** Static functions ***********/
31
32 static e_mc_comm_pattern_difference_t compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
33                                                            const simgrid::mc::PatternCommunication* comm2)
34 {
35   if(comm1->type != comm2->type)
36     return TYPE_DIFF;
37   if (comm1->rdv != comm2->rdv)
38     return RDV_DIFF;
39   if (comm1->src_proc != comm2->src_proc)
40     return SRC_PROC_DIFF;
41   if (comm1->dst_proc != comm2->dst_proc)
42     return DST_PROC_DIFF;
43   if (comm1->tag != comm2->tag)
44     return TAG_DIFF;
45   if (comm1->data.size() != comm2->data.size())
46     return DATA_SIZE_DIFF;
47   if (comm1->data != comm2->data)
48     return DATA_DIFF;
49   return NONE_DIFF;
50 }
51
52 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process,
53                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
54 {
55   char* type;
56   char* res;
57
58   if (comm->type == simgrid::mc::PatternCommunicationType::send)
59     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
60   else
61     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
62
63   switch(diff) {
64   case TYPE_DIFF:
65     res = bprintf("%s Different type for communication #%u", type, cursor);
66     break;
67   case RDV_DIFF:
68     res = bprintf("%s Different rdv for communication #%u", type, cursor);
69     break;
70   case TAG_DIFF:
71     res = bprintf("%s Different tag for communication #%u", type, cursor);
72     break;
73   case SRC_PROC_DIFF:
74     res = bprintf("%s Different source for communication #%u", type, cursor);
75     break;
76   case DST_PROC_DIFF:
77     res = bprintf("%s Different destination for communication #%u", type, cursor);
78     break;
79   case DATA_SIZE_DIFF:
80     res = bprintf("%s Different data size for communication #%u", type, cursor);
81     break;
82   case DATA_DIFF:
83     res = bprintf("%s Different data for communication #%u", type, cursor);
84     break;
85   default:
86     res = nullptr;
87     break;
88   }
89
90   return res;
91 }
92
93 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
94                                 simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr)
95 {
96   // HACK, type punning
97   simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_comm;
98   mc_model_checker->get_remote_simulation().read(temp_comm, comm_addr);
99   const simgrid::kernel::activity::CommImpl* comm = temp_comm.get_buffer();
100
101   smx_actor_t src_proc =
102       mc_model_checker->get_remote_simulation().resolve_actor(simgrid::mc::remote(comm->src_actor_.get()));
103   smx_actor_t dst_proc =
104       mc_model_checker->get_remote_simulation().resolve_actor(simgrid::mc::remote(comm->dst_actor_.get()));
105   comm_pattern->src_proc = src_proc->get_pid();
106   comm_pattern->dst_proc = dst_proc->get_pid();
107   comm_pattern->src_host = MC_smx_actor_get_host_name(src_proc);
108   comm_pattern->dst_host = MC_smx_actor_get_host_name(dst_proc);
109   if (comm_pattern->data.size() == 0 && comm->src_buff_ != nullptr) {
110     size_t buff_size;
111     mc_model_checker->get_remote_simulation().read(&buff_size, remote(comm->dst_buff_size_));
112     comm_pattern->data.resize(buff_size);
113     mc_model_checker->get_remote_simulation().read_bytes(comm_pattern->data.data(), comm_pattern->data.size(),
114                                                          remote(comm->src_buff_));
115   }
116 }
117
118 namespace simgrid {
119 namespace mc {
120
121 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, const PatternCommunication* comm,
122                                                                  int backtracking)
123 {
124   if (not backtracking) {
125     PatternCommunicationList& list      = initial_communications_pattern[process];
126     e_mc_comm_pattern_difference_t diff = compare_comm_pattern(list.list[list.index_comm].get(), comm);
127
128     if (diff != NONE_DIFF) {
129       if (comm->type == PatternCommunicationType::send) {
130         this->send_deterministic = false;
131         if (this->send_diff != nullptr)
132           xbt_free(this->send_diff);
133         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
134       } else {
135         this->recv_deterministic = false;
136         if (this->recv_diff != nullptr)
137           xbt_free(this->recv_diff);
138         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
139       }
140       if (_sg_mc_send_determinism && not this->send_deterministic) {
141         XBT_INFO("*********************************************************");
142         XBT_INFO("***** Non-send-deterministic communications pattern *****");
143         XBT_INFO("*********************************************************");
144         XBT_INFO("%s", this->send_diff);
145         xbt_free(this->send_diff);
146         this->send_diff = nullptr;
147         mc::session->log_state();
148         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
149       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
150         XBT_INFO("****************************************************");
151         XBT_INFO("***** Non-deterministic communications pattern *****");
152         XBT_INFO("****************************************************");
153         if (this->send_diff) {
154           XBT_INFO("%s", this->send_diff);
155           xbt_free(this->send_diff);
156           this->send_diff = nullptr;
157         }
158         if (this->recv_diff) {
159           XBT_INFO("%s", this->recv_diff);
160           xbt_free(this->recv_diff);
161           this->recv_diff = nullptr;
162         }
163         mc::session->log_state();
164         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
165       }
166     }
167   }
168 }
169
170 /********** Non Static functions ***********/
171
172 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, e_mc_call_type_t call_type,
173                                                        int backtracking)
174 {
175   const smx_actor_t issuer = MC_smx_simcall_get_issuer(request);
176   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
177   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
178
179   std::unique_ptr<PatternCommunication> pattern(new PatternCommunication());
180   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
181
182   if (call_type == MC_CALL_TYPE_SEND) {
183     /* Create comm pattern */
184     pattern->type      = PatternCommunicationType::send;
185     pattern->comm_addr = static_cast<kernel::activity::CommImpl*>(simcall_comm_isend__getraw__result(request));
186
187     Remote<kernel::activity::CommImpl> temp_synchro;
188     mc_model_checker->get_remote_simulation().read(temp_synchro, remote(pattern->comm_addr));
189     const kernel::activity::CommImpl* synchro = temp_synchro.get_buffer();
190
191     char* remote_name = mc_model_checker->get_remote_simulation().read<char*>(RemotePtr<char*>(
192         (uint64_t)(synchro->get_mailbox() ? &synchro->get_mailbox()->name_ : &synchro->mbox_cpy->name_)));
193     pattern->rdv      = mc_model_checker->get_remote_simulation().read_string(RemotePtr<char>(remote_name));
194     pattern->src_proc =
195         mc_model_checker->get_remote_simulation().resolve_actor(mc::remote(synchro->src_actor_.get()))->get_pid();
196     pattern->src_host = MC_smx_actor_get_host_name(issuer);
197
198 #if HAVE_SMPI
199     simgrid::smpi::Request mpi_request;
200     mc_model_checker->get_remote_simulation().read(
201         &mpi_request, remote(static_cast<smpi::Request*>(simcall_comm_isend__get__data(request))));
202     pattern->tag = mpi_request.tag();
203 #endif
204
205     if (synchro->src_buff_ != nullptr) {
206       pattern->data.resize(synchro->src_buff_size_);
207       mc_model_checker->get_remote_simulation().read_bytes(pattern->data.data(), pattern->data.size(),
208                                                            remote(synchro->src_buff_));
209     }
210 #if HAVE_SMPI
211     if(mpi_request.detached()){
212       if (this->initial_communications_pattern_done) {
213         /* Evaluate comm determinism */
214         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
215         initial_communications_pattern[pattern->src_proc].index_comm++;
216       } else {
217         /* Store comm pattern */
218         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
219       }
220       return;
221     }
222 #endif
223   } else if (call_type == MC_CALL_TYPE_RECV) {
224     pattern->type      = PatternCommunicationType::receive;
225     pattern->comm_addr = static_cast<kernel::activity::CommImpl*>(simcall_comm_irecv__getraw__result(request));
226
227 #if HAVE_SMPI
228     smpi::Request mpi_request;
229     mc_model_checker->get_remote_simulation().read(
230         &mpi_request, remote(static_cast<smpi::Request*>(simcall_comm_irecv__get__data(request))));
231     pattern->tag = mpi_request.tag();
232 #endif
233
234     Remote<kernel::activity::CommImpl> temp_comm;
235     mc_model_checker->get_remote_simulation().read(temp_comm, remote(pattern->comm_addr));
236     const kernel::activity::CommImpl* comm = temp_comm.get_buffer();
237
238     char* remote_name;
239     mc_model_checker->get_remote_simulation().read(
240         &remote_name, remote(comm->get_mailbox() ? &xbt::string::to_string_data(comm->get_mailbox()->name_).data
241                                                  : &xbt::string::to_string_data(comm->mbox_cpy->name_).data));
242     pattern->rdv = mc_model_checker->get_remote_simulation().read_string(RemotePtr<char>(remote_name));
243     pattern->dst_proc =
244         mc_model_checker->get_remote_simulation().resolve_actor(mc::remote(comm->dst_actor_.get()))->get_pid();
245     pattern->dst_host = MC_smx_actor_get_host_name(issuer);
246   } else
247     xbt_die("Unexpected call_type %i", (int) call_type);
248
249   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
250   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
251 }
252
253 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> comm_addr,
254                                                             unsigned int issuer, int backtracking)
255 {
256   /* Complete comm pattern */
257   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
258   auto current_comm_pattern =
259       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
260                    [&comm_addr](const PatternCommunication* comm) { return remote(comm->comm_addr) == comm_addr; });
261   if (current_comm_pattern == std::end(incomplete_pattern))
262     xbt_die("Corresponding communication not found!");
263
264   update_comm_pattern(*current_comm_pattern, comm_addr);
265   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
266   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
267             std::distance(begin(incomplete_pattern), current_comm_pattern));
268   incomplete_pattern.erase(current_comm_pattern);
269
270   if (this->initial_communications_pattern_done) {
271     /* Evaluate comm determinism */
272     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
273     initial_communications_pattern[issuer].index_comm++;
274   } else {
275     /* Store comm pattern */
276     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
277   }
278 }
279
280 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s)
281 {
282 }
283
284 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
285
286 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
287 {
288   RecordTrace res;
289   for (auto const& state : stack_)
290     res.push_back(state->get_transition());
291   return res;
292 }
293
294 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
295 {
296   std::vector<std::string> trace;
297   for (auto const& state : stack_) {
298     smx_simcall_t req = &state->executed_req_;
299     if (req)
300       trace.push_back(request_to_string(req, state->transition_.argument_, RequestType::executed));
301   }
302   return trace;
303 }
304
305 void CommunicationDeterminismChecker::log_state() // override
306 {
307   if (_sg_mc_comms_determinism) {
308     if (this->send_deterministic && not this->recv_deterministic) {
309       XBT_INFO("*******************************************************");
310       XBT_INFO("**** Only-send-deterministic communication pattern ****");
311       XBT_INFO("*******************************************************");
312       XBT_INFO("%s", this->recv_diff);
313     }
314     if (not this->send_deterministic && this->recv_deterministic) {
315       XBT_INFO("*******************************************************");
316       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
317       XBT_INFO("*******************************************************");
318       XBT_INFO("%s", this->send_diff);
319     }
320   }
321   XBT_INFO("Expanded states = %lu", expanded_states_count_);
322   XBT_INFO("Visited states = %lu", mc_model_checker->visited_states);
323   XBT_INFO("Executed transitions = %lu", mc_model_checker->executed_transitions);
324   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
325   if (_sg_mc_comms_determinism)
326     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
327 }
328
329 void CommunicationDeterminismChecker::prepare()
330 {
331   const int maxpid = MC_smx_get_maxpid();
332
333   initial_communications_pattern.resize(maxpid);
334   incomplete_communications_pattern.resize(maxpid);
335
336   std::unique_ptr<State> initial_state(new State(++expanded_states_count_));
337
338   XBT_DEBUG("********* Start communication determinism verification *********");
339
340   /* Get an enabled actor and insert it in the interleave set of the initial state */
341   for (auto& actor : mc_model_checker->get_remote_simulation().actors())
342     if (mc::actor_is_enabled(actor.copy.get_buffer()))
343       initial_state->add_interleaving_set(actor.copy.get_buffer());
344
345   stack_.push_back(std::move(initial_state));
346 }
347
348 static inline bool all_communications_are_finished()
349 {
350   for (size_t current_actor = 1; current_actor < MC_smx_get_maxpid(); current_actor++) {
351     if (not incomplete_communications_pattern[current_actor].empty()) {
352       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
353       return false;
354     }
355   }
356   return true;
357 }
358
359 void CommunicationDeterminismChecker::restoreState()
360 {
361   /* Intermediate backtracking */
362   State* last_state = stack_.back().get();
363   if (last_state->system_state_) {
364     last_state->system_state_->restore(&mc_model_checker->get_remote_simulation());
365     MC_restore_communications_pattern(last_state);
366     return;
367   }
368
369   /* Restore the initial state */
370   mc::session->restore_initial_state();
371
372   unsigned n = MC_smx_get_maxpid();
373   assert(n == incomplete_communications_pattern.size());
374   assert(n == initial_communications_pattern.size());
375   for (unsigned j=0; j < n ; j++) {
376     incomplete_communications_pattern[j].clear();
377     initial_communications_pattern[j].index_comm = 0;
378   }
379
380   /* Traverse the stack from the state at position start and re-execute the transitions */
381   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
382     if (state == stack_.back())
383       break;
384
385     int req_num             = state->transition_.argument_;
386     const s_smx_simcall* saved_req = &state->executed_req_;
387     xbt_assert(saved_req);
388
389     /* because we got a copy of the executed request, we have to fetch the
390        real one, pointed by the request field of the issuer process */
391
392     const smx_actor_t issuer = MC_smx_simcall_get_issuer(saved_req);
393     smx_simcall_t req        = &issuer->simcall_;
394
395     /* TODO : handle test and testany simcalls */
396     e_mc_call_type_t call = MC_get_call_type(req);
397     mc_model_checker->handle_simcall(state->transition_);
398     MC_handle_comm_pattern(call, req, req_num, 1);
399     mc_model_checker->wait_for_requests();
400
401     /* Update statistics */
402     mc_model_checker->visited_states++;
403     mc_model_checker->executed_transitions++;
404   }
405 }
406
407 void CommunicationDeterminismChecker::real_run()
408 {
409   std::unique_ptr<VisitedState> visited_state = nullptr;
410   smx_simcall_t req = nullptr;
411
412   while (not stack_.empty()) {
413     /* Get current state */
414     State* cur_state = stack_.back().get();
415
416     XBT_DEBUG("**************************************************");
417     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
418               cur_state->interleave_size());
419
420     /* Update statistics */
421     mc_model_checker->visited_states++;
422
423     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
424       req = MC_state_choose_request(cur_state);
425     else
426       req = nullptr;
427
428     if (req != nullptr && visited_state == nullptr) {
429       int req_num = cur_state->transition_.argument_;
430
431       XBT_DEBUG("Execute: %s", request_to_string(req, req_num, RequestType::simix).c_str());
432
433       std::string req_str;
434       if (dot_output != nullptr)
435         req_str = request_get_dot_output(req, req_num);
436
437       mc_model_checker->executed_transitions++;
438
439       /* TODO : handle test and testany simcalls */
440       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
441       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
442         call = MC_get_call_type(req);
443
444       /* Answer the request */
445       mc_model_checker->handle_simcall(cur_state->transition_);
446       /* After this call req is no longer useful */
447
448       MC_handle_comm_pattern(call, req, req_num, 0);
449
450       /* Wait for requests (schedules processes) */
451       mc_model_checker->wait_for_requests();
452
453       /* Create the new expanded state */
454       std::unique_ptr<State> next_state(new State(++expanded_states_count_));
455
456       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
457        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
458        * with the initial pattern. */
459       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
460
461       if (_sg_mc_max_visited_states != 0)
462         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
463       else
464         visited_state = nullptr;
465
466       if (visited_state == nullptr) {
467         /* Get enabled actors and insert them in the interleave set of the next state */
468         for (auto& actor : mc_model_checker->get_remote_simulation().actors())
469           if (simgrid::mc::actor_is_enabled(actor.copy.get_buffer()))
470             next_state->add_interleaving_set(actor.copy.get_buffer());
471
472         if (dot_output != nullptr)
473           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
474
475       } else if (dot_output != nullptr)
476         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
477                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
478
479       stack_.push_back(std::move(next_state));
480     } else {
481       if (stack_.size() > (std::size_t) _sg_mc_max_depth)
482         XBT_WARN("/!\\ Max depth reached! /!\\ ");
483       else if (visited_state != nullptr)
484         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
485             visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
486       else
487         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
488
489       this->initial_communications_pattern_done = true;
490
491       /* Trash the current state, no longer needed */
492       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
493       stack_.pop_back();
494
495       visited_state = nullptr;
496
497       /* Check for deadlocks */
498       if (mc_model_checker->checkDeadlock()) {
499         MC_show_deadlock();
500         throw simgrid::mc::DeadlockError();
501       }
502
503       while (not stack_.empty()) {
504         std::unique_ptr<State> state(std::move(stack_.back()));
505         stack_.pop_back();
506         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
507           /* We found a back-tracking point, let's loop */
508           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
509           stack_.push_back(std::move(state));
510
511           this->restoreState();
512
513           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
514
515           break;
516         } else {
517           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
518         }
519       }
520     }
521   }
522
523   mc::session->log_state();
524 }
525
526 void CommunicationDeterminismChecker::run()
527 {
528   XBT_INFO("Check communication determinism");
529   mc::session->initialize();
530
531   this->prepare();
532   this->real_run();
533 }
534
535 Checker* createCommunicationDeterminismChecker(Session& s)
536 {
537   return new CommunicationDeterminismChecker(s);
538 }
539
540 } // namespace mc
541 } // namespace simgrid