Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
Merge branch 'master' of github.com:simgrid/simgrid into dev_11
[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 #include "src/mc/mc_api.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
190     Remote<kernel::activity::CommImpl> temp_synchro;
191     mc_model_checker->get_remote_simulation().read(temp_synchro, remote(pattern->comm_addr));
192     const kernel::activity::CommImpl* synchro = temp_synchro.get_buffer();
193
194     // char* remote_name = mc_model_checker->get_remote_simulation().read<char*>(RemotePtr<char*>(
195     //     (uint64_t)(synchro->get_mailbox() ? &synchro->get_mailbox()->name_ : &synchro->mbox_cpy->name_)));
196     pattern->rdv      = mcapi::get().get_pattern_comm_rdv(pattern->comm_addr);
197     pattern->src_proc = mcapi::get().get_pattern_comm_src_proc(pattern->comm_addr);
198     pattern->src_host = mc_api::get().get_actor_host_name(issuer);
199
200 #if HAVE_SMPI
201     simgrid::smpi::Request mpi_request;
202     mc_model_checker->get_remote_simulation().read(
203         &mpi_request, remote(static_cast<smpi::Request*>(simcall_comm_isend__get__data(request))));
204     pattern->tag = mpi_request.tag();
205 #endif
206
207     // if (synchro->src_buff_ != nullptr) {
208     //   pattern->data.resize(synchro->src_buff_size_);
209     //   mc_model_checker->get_remote_simulation().read_bytes(pattern->data.data(), pattern->data.size(),
210     //                                                        remote(synchro->src_buff_));
211     // }
212
213     auto pattern_data = mcapi::get().get_pattern_comm_data(pattern->comm_addr);
214     if(pattern_data.data() != nullptr) {
215       auto data_size = pattern_data.size();
216       pattern->data.resize(data_size);
217       memcpy(pattern->data.data(), pattern_data.data(), data_size);
218     }
219 #if HAVE_SMPI
220     if(mpi_request.detached()){
221       if (this->initial_communications_pattern_done) {
222         /* Evaluate comm determinism */
223         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
224         initial_communications_pattern[pattern->src_proc].index_comm++;
225       } else {
226         /* Store comm pattern */
227         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
228       }
229       return;
230     }
231 #endif
232   } else if (call_type == CallType::RECV) {
233     pattern->type      = PatternCommunicationType::receive;
234     pattern->comm_addr = static_cast<kernel::activity::CommImpl*>(simcall_comm_irecv__getraw__result(request));
235
236 #if HAVE_SMPI
237     smpi::Request mpi_request;
238     mc_model_checker->get_remote_simulation().read(
239         &mpi_request, remote(static_cast<smpi::Request*>(simcall_comm_irecv__get__data(request))));
240     pattern->tag = mpi_request.tag();
241 #endif
242
243     Remote<kernel::activity::CommImpl> temp_comm;
244     mc_model_checker->get_remote_simulation().read(temp_comm, remote(pattern->comm_addr));
245     const kernel::activity::CommImpl* comm = temp_comm.get_buffer();
246
247     char* remote_name;
248     mc_model_checker->get_remote_simulation().read(
249         &remote_name, remote(comm->get_mailbox() ? &xbt::string::to_string_data(comm->get_mailbox()->name_).data
250                                                  : &xbt::string::to_string_data(comm->mbox_cpy->name_).data));
251     pattern->rdv = mc_model_checker->get_remote_simulation().read_string(RemotePtr<char>(remote_name));
252     pattern->dst_proc =
253         mc_model_checker->get_remote_simulation().resolve_actor(mc::remote(comm->dst_actor_.get()))->get_pid();
254     pattern->dst_host = MC_smx_actor_get_host_name(issuer);
255   } else
256     xbt_die("Unexpected call_type %i", (int) call_type);
257
258   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
259   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
260 }
261
262 void CommunicationDeterminismChecker::complete_comm_pattern(RemotePtr<kernel::activity::CommImpl> comm_addr,
263                                                             unsigned int issuer, int backtracking)
264 {
265   /* Complete comm pattern */
266   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
267   auto current_comm_pattern =
268       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
269                    [&comm_addr](const PatternCommunication* comm) { return remote(comm->comm_addr) == comm_addr; });
270   if (current_comm_pattern == std::end(incomplete_pattern))
271     xbt_die("Corresponding communication not found!");
272
273   update_comm_pattern(*current_comm_pattern, comm_addr);
274   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
275   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
276             std::distance(begin(incomplete_pattern), current_comm_pattern));
277   incomplete_pattern.erase(current_comm_pattern);
278
279   if (this->initial_communications_pattern_done) {
280     /* Evaluate comm determinism */
281     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
282     initial_communications_pattern[issuer].index_comm++;
283   } else {
284     /* Store comm pattern */
285     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
286   }
287 }
288
289 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s)
290 {
291 }
292
293 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
294
295 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
296 {
297   RecordTrace res;
298   for (auto const& state : stack_)
299     res.push_back(state->get_transition());
300   return res;
301 }
302
303 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
304 {
305   std::vector<std::string> trace;
306   for (auto const& state : stack_) {
307     smx_simcall_t req = &state->executed_req_;
308     trace.push_back(mcapi::get().request_to_string(req, state->transition_.argument_, RequestType::executed));
309   }
310   return trace;
311 }
312
313 void CommunicationDeterminismChecker::log_state() // override
314 {
315   if (_sg_mc_comms_determinism) {
316     if (this->send_deterministic && not this->recv_deterministic) {
317       XBT_INFO("*******************************************************");
318       XBT_INFO("**** Only-send-deterministic communication pattern ****");
319       XBT_INFO("*******************************************************");
320       XBT_INFO("%s", this->recv_diff);
321     }
322     if (not this->send_deterministic && this->recv_deterministic) {
323       XBT_INFO("*******************************************************");
324       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
325       XBT_INFO("*******************************************************");
326       XBT_INFO("%s", this->send_diff);
327     }
328   }
329   XBT_INFO("Expanded states = %lu", expanded_states_count_);
330   XBT_INFO("Visited states = %lu", mcapi::get().mc_get_visited_states());
331   XBT_INFO("Executed transitions = %lu", mcapi::get().mc_get_executed_trans());
332   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
333   if (_sg_mc_comms_determinism)
334     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
335 }
336
337 void CommunicationDeterminismChecker::prepare()
338 {
339   const int maxpid = mcapi::get().get_maxpid();
340
341   initial_communications_pattern.resize(maxpid);
342   incomplete_communications_pattern.resize(maxpid);
343
344   ++expanded_states_count_;
345   auto initial_state = std::make_unique<State>(expanded_states_count_);
346
347   XBT_DEBUG("********* Start communication determinism verification *********");
348
349   /* Get an enabled actor and insert it in the interleave set of the initial state */
350   auto actors = mcapi::get().get_actors();
351   for (auto& actor : actors)
352     if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
353       initial_state->add_interleaving_set(actor.copy.get_buffer());
354
355   stack_.push_back(std::move(initial_state));
356 }
357
358 static inline bool all_communications_are_finished()
359 {
360   auto maxpid = mcapi::get().get_maxpid();
361   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
362     if (not incomplete_communications_pattern[current_actor].empty()) {
363       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
364       return false;
365     }
366   }
367   return true;
368 }
369
370 void CommunicationDeterminismChecker::restoreState()
371 {
372   /* Intermediate backtracking */
373   State* last_state = stack_.back().get();
374   if (last_state->system_state_) {
375     last_state->system_state_->restore(&mcapi::get().mc_get_remote_simulation());
376     MC_restore_communications_pattern(last_state);
377     return;
378   }
379
380   /* Restore the initial state */
381   mcapi::get().s_restore_initial_state();
382
383   unsigned n = mcapi::get().get_maxpid();
384   assert(n == incomplete_communications_pattern.size());
385   assert(n == initial_communications_pattern.size());
386   for (unsigned j=0; j < n ; j++) {
387     incomplete_communications_pattern[j].clear();
388     initial_communications_pattern[j].index_comm = 0;
389   }
390
391   /* Traverse the stack from the state at position start and re-execute the transitions */
392   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
393     if (state == stack_.back())
394       break;
395
396     int req_num             = state->transition_.argument_;
397     const s_smx_simcall* saved_req = &state->executed_req_;
398     xbt_assert(saved_req);
399
400     /* because we got a copy of the executed request, we have to fetch the
401        real one, pointed by the request field of the issuer process */
402
403     const smx_actor_t issuer = mcapi::get().mc_smx_simcall_get_issuer(saved_req);
404     smx_simcall_t req        = &issuer->simcall_;
405
406     /* TODO : handle test and testany simcalls */
407     CallType call = MC_get_call_type(req);
408     mcapi::get().handle_simcall(state->transition_);
409     MC_handle_comm_pattern(call, req, req_num, 1);
410     mcapi::get().mc_wait_for_requests();
411
412     /* Update statistics */
413     mcapi::get().mc_inc_visited_states();
414     mcapi::get().mc_inc_executed_trans();
415   }
416 }
417
418 void CommunicationDeterminismChecker::real_run()
419 {
420   std::unique_ptr<VisitedState> visited_state = nullptr;
421   smx_simcall_t req = nullptr;
422
423   while (not stack_.empty()) {
424     /* Get current state */
425     State* cur_state = stack_.back().get();
426
427     XBT_DEBUG("**************************************************");
428     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
429               cur_state->interleave_size());
430
431     /* Update statistics */
432     mcapi::get().mc_inc_visited_states();
433
434     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
435       req = mcapi::get().mc_state_choose_request(cur_state);
436     else
437       req = nullptr;
438
439     if (req != nullptr && visited_state == nullptr) {
440       int req_num = cur_state->transition_.argument_;
441
442       XBT_DEBUG("Execute: %s", mcapi::get().request_to_string(req, req_num, RequestType::simix).c_str());
443
444       std::string req_str;
445       if (dot_output != nullptr)
446         req_str = mcapi::get().request_get_dot_output(req, req_num);
447
448       mcapi::get().mc_inc_executed_trans();
449
450       /* TODO : handle test and testany simcalls */
451       CallType call = CallType::NONE;
452       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
453         call = MC_get_call_type(req);
454
455       /* Answer the request */
456       mcapi::get().handle_simcall(cur_state->transition_);
457       /* After this call req is no longer useful */
458
459       MC_handle_comm_pattern(call, req, req_num, 0);
460
461
462       /* Wait for requests (schedules processes) */
463       mcapi::get().mc_wait_for_requests();
464
465       /* Create the new expanded state */
466       ++expanded_states_count_;
467       auto next_state = std::make_unique<State>(expanded_states_count_);
468
469       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
470        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
471        * with the initial pattern. */
472       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
473
474       if (_sg_mc_max_visited_states != 0)
475         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
476       else
477         visited_state = nullptr;
478
479       if (visited_state == nullptr) {
480         /* Get enabled actors and insert them in the interleave set of the next state */
481         auto actors = mcapi::get().mc_get_remote_simulation().actors();
482         for (auto& actor : actors)
483           if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
484             next_state->add_interleaving_set(actor.copy.get_buffer());
485
486         if (dot_output != nullptr)
487           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
488
489       } else if (dot_output != nullptr)
490         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
491                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
492
493       stack_.push_back(std::move(next_state));
494     } else {
495       if (stack_.size() > (std::size_t) _sg_mc_max_depth)
496         XBT_WARN("/!\\ Max depth reached! /!\\ ");
497       else if (visited_state != nullptr)
498         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
499             visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
500       else
501         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
502
503       this->initial_communications_pattern_done = true;
504
505       /* Trash the current state, no longer needed */
506       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
507       stack_.pop_back();
508
509       visited_state = nullptr;
510
511       /* Check for deadlocks */
512       if (mcapi::get().mc_check_deadlock()) {
513         mcapi::get().mc_show_deadlock();
514         throw simgrid::mc::DeadlockError();
515       }
516
517       while (not stack_.empty()) {
518         std::unique_ptr<State> state(std::move(stack_.back()));
519         stack_.pop_back();
520         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
521           /* We found a back-tracking point, let's loop */
522           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
523           stack_.push_back(std::move(state));
524
525           this->restoreState();
526
527           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
528
529           break;
530         } else {
531           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
532         }
533       }
534     }
535   }
536
537   mcapi::get().s_log_state();
538 }
539
540 void CommunicationDeterminismChecker::run()
541 {
542   XBT_INFO("Check communication determinism");
543   mcapi::get().s_initialize();
544
545   this->prepare();
546   this->real_run();
547 }
548
549 Checker* createCommunicationDeterminismChecker(Session& s)
550 {
551   return new CommunicationDeterminismChecker(s);
552 }
553
554 } // namespace mc
555 } // namespace simgrid