Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
31d3df5dc2380dbfe7ec8f4993c272fe76cb490b
[simgrid.git] / src / mc / checker / CommunicationDeterminismChecker.cpp
1 /* Copyright (c) 2008-2019. 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(simgrid::mc::PatternCommunication* comm1,
33                                                            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                                       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\n Different data size for communication #%u", type, cursor);
81     break;
82   case DATA_DIFF:
83     res = bprintf("%s\n 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->process().read(temp_comm, comm_addr);
99   simgrid::kernel::activity::CommImpl* comm = temp_comm.getBuffer();
100
101   smx_actor_t src_proc   = mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->src_actor_.get()));
102   smx_actor_t dst_proc   = mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->dst_actor_.get()));
103   comm_pattern->src_proc = src_proc->get_pid();
104   comm_pattern->dst_proc = dst_proc->get_pid();
105   comm_pattern->src_host = MC_smx_actor_get_host_name(src_proc);
106   comm_pattern->dst_host = MC_smx_actor_get_host_name(dst_proc);
107   if (comm_pattern->data.size() == 0 && comm->src_buff_ != nullptr) {
108     size_t buff_size;
109     mc_model_checker->process().read(&buff_size, remote(comm->dst_buff_size_));
110     comm_pattern->data.resize(buff_size);
111     mc_model_checker->process().read_bytes(comm_pattern->data.data(), comm_pattern->data.size(),
112                                            remote(comm->src_buff_));
113   }
114 }
115
116 namespace simgrid {
117 namespace mc {
118
119 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, simgrid::mc::PatternCommunication* comm,
120                                                                  int backtracking)
121 {
122   simgrid::mc::PatternCommunicationList& list = initial_communications_pattern[process];
123
124   if (not backtracking) {
125     e_mc_comm_pattern_difference_t diff = compare_comm_pattern(list.list[list.index_comm].get(), comm);
126
127     if (diff != NONE_DIFF) {
128       if (comm->type == simgrid::mc::PatternCommunicationType::send) {
129         this->send_deterministic = 0;
130         if (this->send_diff != nullptr)
131           xbt_free(this->send_diff);
132         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
133       } else {
134         this->recv_deterministic = 0;
135         if (this->recv_diff != nullptr)
136           xbt_free(this->recv_diff);
137         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
138       }
139       if (_sg_mc_send_determinism && not this->send_deterministic) {
140         XBT_INFO("*********************************************************");
141         XBT_INFO("***** Non-send-deterministic communications pattern *****");
142         XBT_INFO("*********************************************************");
143         XBT_INFO("%s", this->send_diff);
144         xbt_free(this->send_diff);
145         this->send_diff = nullptr;
146         simgrid::mc::session->log_state();
147         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
148       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
149         XBT_INFO("****************************************************");
150         XBT_INFO("***** Non-deterministic communications pattern *****");
151         XBT_INFO("****************************************************");
152         XBT_INFO("%s", this->send_diff);
153         XBT_INFO("%s", this->recv_diff);
154         xbt_free(this->send_diff);
155         this->send_diff = nullptr;
156         xbt_free(this->recv_diff);
157         this->recv_diff = nullptr;
158         simgrid::mc::session->log_state();
159         mc_model_checker->exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
160       }
161     }
162   }
163 }
164
165 /********** Non Static functions ***********/
166
167 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, e_mc_call_type_t call_type,
168                                                        int backtracking)
169 {
170   const smx_actor_t issuer = MC_smx_simcall_get_issuer(request);
171   const simgrid::mc::PatternCommunicationList& initial_pattern = initial_communications_pattern[issuer->get_pid()];
172   const std::vector<simgrid::mc::PatternCommunication*>& incomplete_pattern =
173       incomplete_communications_pattern[issuer->get_pid()];
174
175   std::unique_ptr<simgrid::mc::PatternCommunication> pattern(new simgrid::mc::PatternCommunication());
176   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
177
178   if (call_type == MC_CALL_TYPE_SEND) {
179     /* Create comm pattern */
180     pattern->type = simgrid::mc::PatternCommunicationType::send;
181     pattern->comm_addr = static_cast<simgrid::kernel::activity::CommImpl*>(simcall_comm_isend__getraw__result(request));
182
183     simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_synchro;
184     mc_model_checker->process().read(temp_synchro,
185                                      remote(static_cast<simgrid::kernel::activity::CommImpl*>(pattern->comm_addr)));
186     simgrid::kernel::activity::CommImpl* synchro =
187         static_cast<simgrid::kernel::activity::CommImpl*>(temp_synchro.getBuffer());
188
189     char* remote_name = mc_model_checker->process().read<char*>(RemotePtr<char*>(
190         (uint64_t)(synchro->get_mailbox() ? &synchro->get_mailbox()->name_ : &synchro->mbox_cpy->name_)));
191     pattern->rdv      = mc_model_checker->process().read_string(RemotePtr<char>(remote_name));
192     pattern->src_proc =
193         mc_model_checker->process().resolveActor(simgrid::mc::remote(synchro->src_actor_.get()))->get_pid();
194     pattern->src_host = MC_smx_actor_get_host_name(issuer);
195
196 #if HAVE_SMPI
197     simgrid::smpi::Request mpi_request;
198     mc_model_checker->process().read(
199         &mpi_request, remote(static_cast<simgrid::smpi::Request*>(simcall_comm_isend__get__data(request))));
200     pattern->tag = mpi_request.tag();
201 #endif
202
203     if (synchro->src_buff_ != nullptr) {
204       pattern->data.resize(synchro->src_buff_size_);
205       mc_model_checker->process().read_bytes(pattern->data.data(), pattern->data.size(), remote(synchro->src_buff_));
206     }
207 #if HAVE_SMPI
208     if(mpi_request.detached()){
209       if (not this->initial_communications_pattern_done) {
210         /* Store comm pattern */
211         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
212       } else {
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       }
217       return;
218     }
219 #endif
220   } else if (call_type == MC_CALL_TYPE_RECV) {
221     pattern->type = simgrid::mc::PatternCommunicationType::receive;
222     pattern->comm_addr = static_cast<simgrid::kernel::activity::CommImpl*>(simcall_comm_irecv__getraw__result(request));
223
224 #if HAVE_SMPI
225     simgrid::smpi::Request mpi_request;
226     mc_model_checker->process().read(
227         &mpi_request, remote(static_cast<simgrid::smpi::Request*>(simcall_comm_irecv__get__data(request))));
228     pattern->tag = mpi_request.tag();
229 #endif
230
231     simgrid::mc::Remote<simgrid::kernel::activity::CommImpl> temp_comm;
232     mc_model_checker->process().read(temp_comm,
233                                      remote(static_cast<simgrid::kernel::activity::CommImpl*>(pattern->comm_addr)));
234     simgrid::kernel::activity::CommImpl* comm = temp_comm.getBuffer();
235
236     char* remote_name;
237     mc_model_checker->process().read(&remote_name,
238                                      remote(comm->get_mailbox()
239                                                 ? &simgrid::xbt::string::to_string_data(comm->get_mailbox()->name_).data
240                                                 : &simgrid::xbt::string::to_string_data(comm->mbox_cpy->name_).data));
241     pattern->rdv      = mc_model_checker->process().read_string(RemotePtr<char>(remote_name));
242     pattern->dst_proc =
243         mc_model_checker->process().resolveActor(simgrid::mc::remote(comm->dst_actor_.get()))->get_pid();
244     pattern->dst_host = MC_smx_actor_get_host_name(issuer);
245   } else
246     xbt_die("Unexpected call_type %i", (int) call_type);
247
248   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
249   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
250 }
251
252 void CommunicationDeterminismChecker::complete_comm_pattern(
253     simgrid::mc::RemotePtr<simgrid::kernel::activity::CommImpl> comm_addr, unsigned int issuer, int backtracking)
254 {
255   /* Complete comm pattern */
256   std::vector<simgrid::mc::PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
257   auto current_comm_pattern = std::find_if(
258       begin(incomplete_pattern), end(incomplete_pattern),
259       [&comm_addr](simgrid::mc::PatternCommunication* comm) { return remote(comm->comm_addr) == comm_addr; });
260   if (current_comm_pattern == std::end(incomplete_pattern))
261     xbt_die("Corresponding communication not found!");
262
263   update_comm_pattern(*current_comm_pattern, comm_addr);
264   std::unique_ptr<simgrid::mc::PatternCommunication> comm_pattern(*current_comm_pattern);
265   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
266             std::distance(begin(incomplete_pattern), current_comm_pattern));
267   incomplete_pattern.erase(current_comm_pattern);
268
269   if (not this->initial_communications_pattern_done)
270     /* Store comm pattern */
271     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
272   else {
273     /* Evaluate comm determinism */
274     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
275     initial_communications_pattern[issuer].index_comm++;
276   }
277 }
278
279 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s)
280 {
281 }
282
283 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
284
285 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
286 {
287   RecordTrace res;
288   for (auto const& state : stack_)
289     res.push_back(state->getTransition());
290   return res;
291 }
292
293 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
294 {
295   std::vector<std::string> trace;
296   for (auto const& state : stack_) {
297     smx_simcall_t req = &state->executed_req;
298     if (req)
299       trace.push_back(
300           simgrid::mc::request_to_string(req, state->transition.argument, simgrid::mc::RequestType::executed));
301   }
302   return trace;
303 }
304
305 void CommunicationDeterminismChecker::log_state() // override
306 {
307   if (_sg_mc_comms_determinism && not this->recv_deterministic && this->send_deterministic) {
308     XBT_INFO("******************************************************");
309     XBT_INFO("**** Only-send-deterministic communication pattern ****");
310     XBT_INFO("******************************************************");
311     XBT_INFO("%s", this->recv_diff);
312   } else if (_sg_mc_comms_determinism && not this->send_deterministic && this->recv_deterministic) {
313     XBT_INFO("******************************************************");
314     XBT_INFO("**** Only-recv-deterministic communication pattern ****");
315     XBT_INFO("******************************************************");
316     XBT_INFO("%s", this->send_diff);
317   }
318   XBT_INFO("Expanded states = %lu", expanded_states_count_);
319   XBT_INFO("Visited states = %lu", mc_model_checker->visited_states);
320   XBT_INFO("Executed transitions = %lu", mc_model_checker->executed_transitions);
321   XBT_INFO("Send-deterministic : %s", not this->send_deterministic ? "No" : "Yes");
322   if (_sg_mc_comms_determinism)
323     XBT_INFO("Recv-deterministic : %s", not this->recv_deterministic ? "No" : "Yes");
324 }
325
326 void CommunicationDeterminismChecker::prepare()
327 {
328   const int maxpid = MC_smx_get_maxpid();
329
330   initial_communications_pattern.resize(maxpid);
331   incomplete_communications_pattern.resize(maxpid);
332
333   std::unique_ptr<simgrid::mc::State> initial_state =
334       std::unique_ptr<simgrid::mc::State>(new simgrid::mc::State(++expanded_states_count_));
335
336   XBT_DEBUG("********* Start communication determinism verification *********");
337
338   /* Get an enabled actor and insert it in the interleave set of the initial state */
339   for (auto& actor : mc_model_checker->process().actors())
340     if (simgrid::mc::actor_is_enabled(actor.copy.getBuffer()))
341       initial_state->addInterleavingSet(actor.copy.getBuffer());
342
343   stack_.push_back(std::move(initial_state));
344 }
345
346 static inline bool all_communications_are_finished()
347 {
348   for (size_t current_actor = 1; current_actor < MC_smx_get_maxpid(); current_actor++) {
349     if (not incomplete_communications_pattern[current_actor].empty()) {
350       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
351       return false;
352     }
353   }
354   return true;
355 }
356
357 void CommunicationDeterminismChecker::restoreState()
358 {
359   /* Intermediate backtracking */
360   simgrid::mc::State* last_state = stack_.back().get();
361   if (last_state->system_state) {
362     last_state->system_state->restore(&mc_model_checker->process());
363     MC_restore_communications_pattern(last_state);
364     return;
365   }
366
367   /* Restore the initial state */
368   simgrid::mc::session->restore_initial_state();
369
370   unsigned n = MC_smx_get_maxpid();
371   assert(n == incomplete_communications_pattern.size());
372   assert(n == initial_communications_pattern.size());
373   for (unsigned j=0; j < n ; j++) {
374     incomplete_communications_pattern[j].clear();
375     initial_communications_pattern[j].index_comm = 0;
376   }
377
378   /* Traverse the stack from the state at position start and re-execute the transitions */
379   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
380     if (state == stack_.back())
381       break;
382
383     int req_num = state->transition.argument;
384     smx_simcall_t saved_req = &state->executed_req;
385     xbt_assert(saved_req);
386
387     /* because we got a copy of the executed request, we have to fetch the
388        real one, pointed by the request field of the issuer process */
389
390     const smx_actor_t issuer = MC_smx_simcall_get_issuer(saved_req);
391     smx_simcall_t req = &issuer->simcall;
392
393     /* TODO : handle test and testany simcalls */
394     e_mc_call_type_t call = MC_get_call_type(req);
395     mc_model_checker->handle_simcall(state->transition);
396     MC_handle_comm_pattern(call, req, req_num, 1);
397     mc_model_checker->wait_for_requests();
398
399     /* Update statistics */
400     mc_model_checker->visited_states++;
401     mc_model_checker->executed_transitions++;
402   }
403 }
404
405 void CommunicationDeterminismChecker::real_run()
406 {
407   std::unique_ptr<simgrid::mc::VisitedState> visited_state = nullptr;
408   smx_simcall_t req = nullptr;
409
410   while (not stack_.empty()) {
411     /* Get current state */
412     simgrid::mc::State* cur_state = stack_.back().get();
413
414     XBT_DEBUG("**************************************************");
415     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num,
416               cur_state->interleaveSize());
417
418     /* Update statistics */
419     mc_model_checker->visited_states++;
420
421     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
422       req = MC_state_get_request(cur_state);
423     else
424       req = nullptr;
425
426     if (req != nullptr && visited_state == nullptr) {
427
428       int req_num = cur_state->transition.argument;
429
430       XBT_DEBUG("Execute: %s", simgrid::mc::request_to_string(req, req_num, simgrid::mc::RequestType::simix).c_str());
431
432       std::string req_str;
433       if (dot_output != nullptr)
434         req_str = simgrid::mc::request_get_dot_output(req, req_num);
435
436       mc_model_checker->executed_transitions++;
437
438       /* TODO : handle test and testany simcalls */
439       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
440       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
441         call = MC_get_call_type(req);
442
443       /* Answer the request */
444       mc_model_checker->handle_simcall(cur_state->transition);
445       /* After this call req is no longer useful */
446
447       MC_handle_comm_pattern(call, req, req_num, 0);
448
449       /* Wait for requests (schedules processes) */
450       mc_model_checker->wait_for_requests();
451
452       /* Create the new expanded state */
453       std::unique_ptr<simgrid::mc::State> next_state =
454           std::unique_ptr<simgrid::mc::State>(new simgrid::mc::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 = all_communications_are_finished() && this->initial_communications_pattern_done;
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
468         /* Get enabled actors and insert them in the interleave set of the next state */
469         for (auto& actor : mc_model_checker->process().actors())
470           if (simgrid::mc::actor_is_enabled(actor.copy.getBuffer()))
471             next_state->addInterleavingSet(actor.copy.getBuffer());
472
473         if (dot_output != nullptr)
474           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num, next_state->num, req_str.c_str());
475
476       } else if (dot_output != nullptr)
477         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num,
478                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
479
480       stack_.push_back(std::move(next_state));
481
482     } else {
483
484       if (stack_.size() > (std::size_t) _sg_mc_max_depth)
485         XBT_WARN("/!\\ Max depth reached! /!\\ ");
486       else if (visited_state != nullptr)
487         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
488             visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
489       else
490         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
491
492       if (not this->initial_communications_pattern_done)
493         this->initial_communications_pattern_done = 1;
494
495       /* Trash the current state, no longer needed */
496       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num, stack_.size());
497       stack_.pop_back();
498
499       visited_state = nullptr;
500
501       /* Check for deadlocks */
502       if (mc_model_checker->checkDeadlock()) {
503         MC_show_deadlock();
504         throw simgrid::mc::DeadlockError();
505       }
506
507       while (not stack_.empty()) {
508         std::unique_ptr<simgrid::mc::State> state = std::move(stack_.back());
509         stack_.pop_back();
510         if (state->interleaveSize() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
511           /* We found a back-tracking point, let's loop */
512           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num, stack_.size() + 1);
513           stack_.push_back(std::move(state));
514
515           this->restoreState();
516
517           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num, stack_.size());
518
519           break;
520         } else {
521           XBT_DEBUG("Delete state %d at depth %zu", state->num, stack_.size() + 1);
522         }
523       }
524     }
525   }
526
527   simgrid::mc::session->log_state();
528 }
529
530 void CommunicationDeterminismChecker::run()
531 {
532   XBT_INFO("Check communication determinism");
533   simgrid::mc::session->initialize();
534
535   this->prepare();
536
537   this->real_run();
538 }
539
540 Checker* createCommunicationDeterminismChecker(Session& s)
541 {
542   return new CommunicationDeterminismChecker(s);
543 }
544
545 }
546 }