Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
static function
[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_api.hpp"
10 #include "src/mc/mc_config.hpp"
11 #include "src/mc/mc_exit.hpp"
12 #include "src/mc/mc_private.hpp"
13 #include "src/mc/mc_request.hpp"
14 #include "src/mc/mc_smx.hpp"
15
16 #if HAVE_SMPI
17 #include "smpi_request.hpp"
18 #endif
19
20 #include <cstdint>
21
22 using mcapi = simgrid::mc::mc_api;
23
24 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc, "Logging specific to MC communication determinism detection");
25
26 /********** Global variables **********/
27
28 std::vector<simgrid::mc::PatternCommunicationList> initial_communications_pattern;
29 std::vector<std::vector<simgrid::mc::PatternCommunication*>> incomplete_communications_pattern;
30
31 /********** Static functions ***********/
32
33 static simgrid::mc::CommPatternDifference compare_comm_pattern(const simgrid::mc::PatternCommunication* comm1,
34                                                                const simgrid::mc::PatternCommunication* comm2)
35 {
36   using simgrid::mc::CommPatternDifference;
37   if (comm1->type != comm2->type)
38     return CommPatternDifference::TYPE;
39   if (comm1->rdv != comm2->rdv)
40     return CommPatternDifference::RDV;
41   if (comm1->src_proc != comm2->src_proc)
42     return CommPatternDifference::SRC_PROC;
43   if (comm1->dst_proc != comm2->dst_proc)
44     return CommPatternDifference::DST_PROC;
45   if (comm1->tag != comm2->tag)
46     return CommPatternDifference::TAG;
47   if (comm1->data.size() != comm2->data.size())
48     return CommPatternDifference::DATA_SIZE;
49   if (comm1->data != comm2->data)
50     return CommPatternDifference::DATA;
51   return CommPatternDifference::NONE;
52 }
53
54 static void patterns_copy(std::vector<simgrid::mc::PatternCommunication*>& dest,
55                              std::vector<simgrid::mc::PatternCommunication> const& source)
56 {
57   dest.clear();
58   for (simgrid::mc::PatternCommunication const& comm : source) {
59     auto* copy_comm = new simgrid::mc::PatternCommunication(comm.dup());
60     dest.push_back(copy_comm);
61   }
62 }
63
64 static void restore_communications_pattern(simgrid::mc::State* state)
65 {
66   for (unsigned i = 0; i < initial_communications_pattern.size(); i++)
67     initial_communications_pattern[i].index_comm = state->communication_indices_[i];
68
69   for (unsigned i = 0; i < mcapi::get().get_maxpid(); i++)
70     patterns_copy(incomplete_communications_pattern[i], state->incomplete_comm_pattern_[i]);
71 }
72
73 static char* print_determinism_result(simgrid::mc::CommPatternDifference diff, int process,
74                                       const simgrid::mc::PatternCommunication* comm, unsigned int cursor)
75 {
76   char* type;
77   char* res;
78
79   if (comm->type == simgrid::mc::PatternCommunicationType::send)
80     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
81   else
82     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
83
84   using simgrid::mc::CommPatternDifference;
85   switch (diff) {
86     case CommPatternDifference::TYPE:
87       res = bprintf("%s Different type for communication #%u", type, cursor);
88       break;
89     case CommPatternDifference::RDV:
90       res = bprintf("%s Different rdv for communication #%u", type, cursor);
91       break;
92     case CommPatternDifference::TAG:
93       res = bprintf("%s Different tag for communication #%u", type, cursor);
94       break;
95     case CommPatternDifference::SRC_PROC:
96       res = bprintf("%s Different source for communication #%u", type, cursor);
97       break;
98     case CommPatternDifference::DST_PROC:
99       res = bprintf("%s Different destination for communication #%u", type, cursor);
100       break;
101     case CommPatternDifference::DATA_SIZE:
102       res = bprintf("%s Different data size for communication #%u", type, cursor);
103       break;
104     case CommPatternDifference::DATA:
105       res = bprintf("%s Different data for communication #%u", type, cursor);
106       break;
107     default:
108       res = nullptr;
109       break;
110   }
111
112   return res;
113 }
114
115 static void update_comm_pattern(simgrid::mc::PatternCommunication* comm_pattern,
116                                 const simgrid::kernel::activity::CommImpl* comm_addr)
117 {
118   auto src_proc = mcapi::get().get_src_actor(comm_addr);
119   auto dst_proc = mcapi::get().get_dst_actor(comm_addr);
120   comm_pattern->src_proc = src_proc->get_pid();
121   comm_pattern->dst_proc = dst_proc->get_pid();
122   comm_pattern->src_host = mcapi::get().get_actor_host_name(src_proc);
123   comm_pattern->dst_host = mcapi::get().get_actor_host_name(dst_proc);
124
125   if (comm_pattern->data.empty()) {
126     auto pattern_data = mcapi::get().get_pattern_comm_data(comm_addr);
127     if (pattern_data.data() != nullptr) {
128       auto data_size = pattern_data.size();
129       comm_pattern->data.resize(data_size);
130       memcpy(comm_pattern->data.data(), pattern_data.data(), data_size);
131     }
132    }
133 }
134
135 namespace simgrid {
136 namespace mc {
137
138 void CommunicationDeterminismChecker::deterministic_comm_pattern(int process, const PatternCommunication* comm,
139                                                                  int backtracking)
140 {
141   if (not backtracking) {
142     PatternCommunicationList& list = initial_communications_pattern[process];
143     CommPatternDifference diff     = compare_comm_pattern(list.list[list.index_comm].get(), comm);
144
145     if (diff != CommPatternDifference::NONE) {
146       if (comm->type == PatternCommunicationType::send) {
147         this->send_deterministic = false;
148         if (this->send_diff != nullptr)
149           xbt_free(this->send_diff);
150         this->send_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
151       } else {
152         this->recv_deterministic = false;
153         if (this->recv_diff != nullptr)
154           xbt_free(this->recv_diff);
155         this->recv_diff = print_determinism_result(diff, process, comm, list.index_comm + 1);
156       }
157       if (_sg_mc_send_determinism && not this->send_deterministic) {
158         XBT_INFO("*********************************************************");
159         XBT_INFO("***** Non-send-deterministic communications pattern *****");
160         XBT_INFO("*********************************************************");
161         XBT_INFO("%s", this->send_diff);
162         xbt_free(this->send_diff);
163         this->send_diff = nullptr;
164         mcapi::get().log_state();
165         mcapi::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
166       } else if (_sg_mc_comms_determinism && (not this->send_deterministic && not this->recv_deterministic)) {
167         XBT_INFO("****************************************************");
168         XBT_INFO("***** Non-deterministic communications pattern *****");
169         XBT_INFO("****************************************************");
170         if (this->send_diff) {
171           XBT_INFO("%s", this->send_diff);
172           xbt_free(this->send_diff);
173           this->send_diff = nullptr;
174         }
175         if (this->recv_diff) {
176           XBT_INFO("%s", this->recv_diff);
177           xbt_free(this->recv_diff);
178           this->recv_diff = nullptr;
179         }
180         mcapi::get().log_state();
181         mcapi::get().mc_exit(SIMGRID_MC_EXIT_NON_DETERMINISM);
182       }
183     }
184   }
185 }
186
187 /********** Non Static functions ***********/
188
189 void CommunicationDeterminismChecker::get_comm_pattern(smx_simcall_t request, CallType call_type, int backtracking)
190 {
191   const smx_actor_t issuer                                     = mcapi::get().simcall_get_issuer(request);
192   const mc::PatternCommunicationList& initial_pattern          = initial_communications_pattern[issuer->get_pid()];
193   const std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer->get_pid()];
194
195   auto pattern   = std::make_unique<PatternCommunication>();
196   pattern->index = initial_pattern.index_comm + incomplete_pattern.size();
197
198   if (call_type == CallType::SEND) {
199     /* Create comm pattern */
200     pattern->type      = PatternCommunicationType::send;
201     pattern->comm_addr = mcapi::get().get_comm_isend_raw_addr(request);
202     pattern->rdv      = mcapi::get().get_pattern_comm_rdv(pattern->comm_addr);
203     pattern->src_proc = mcapi::get().get_pattern_comm_src_proc(pattern->comm_addr);
204     pattern->src_host = mc_api::get().get_actor_host_name(issuer);
205
206 #if HAVE_SMPI
207     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_ISEND);
208 #endif
209     auto pattern_data = mcapi::get().get_pattern_comm_data(pattern->comm_addr);
210     if (pattern_data.data() != nullptr) {
211       auto data_size = pattern_data.size();
212       pattern->data.resize(data_size);
213       memcpy(pattern->data.data(), pattern_data.data(), data_size);
214     }
215
216 #if HAVE_SMPI
217     auto send_detached = mcapi::get().check_send_request_detached(request);
218     if (send_detached) {
219       if (this->initial_communications_pattern_done) {
220         /* Evaluate comm determinism */
221         this->deterministic_comm_pattern(pattern->src_proc, pattern.get(), backtracking);
222         initial_communications_pattern[pattern->src_proc].index_comm++;
223       } else {
224         /* Store comm pattern */
225         initial_communications_pattern[pattern->src_proc].list.push_back(std::move(pattern));
226       }
227       return;
228     }
229 #endif
230   } else if (call_type == CallType::RECV) {
231     pattern->type = PatternCommunicationType::receive;
232     pattern->comm_addr = mcapi::get().get_comm_isend_raw_addr(request);
233
234 #if HAVE_SMPI
235     pattern->tag = mcapi::get().get_smpi_request_tag(request, simgrid::simix::Simcall::COMM_IRECV);
236 #endif
237     auto comm_addr = pattern->comm_addr;
238     pattern->rdv = mcapi::get().get_pattern_comm_rdv(comm_addr);
239     pattern->dst_proc = mcapi::get().get_pattern_comm_dst_proc(comm_addr);
240     pattern->dst_host = mcapi::get().get_actor_host_name(issuer);
241   } else
242     xbt_die("Unexpected call_type %i", (int)call_type);
243
244   XBT_DEBUG("Insert incomplete comm pattern %p for process %ld", pattern.get(), issuer->get_pid());
245   incomplete_communications_pattern[issuer->get_pid()].push_back(pattern.release());
246 }
247
248 void CommunicationDeterminismChecker::complete_comm_pattern(const kernel::activity::CommImpl* comm_addr,
249                                                             unsigned int issuer, int backtracking)
250 {
251   /* Complete comm pattern */
252   std::vector<PatternCommunication*>& incomplete_pattern = incomplete_communications_pattern[issuer];
253   auto current_comm_pattern =
254       std::find_if(begin(incomplete_pattern), end(incomplete_pattern),
255                    [&comm_addr](const PatternCommunication* comm) { return mcapi::get().comm_addr_equal(comm->comm_addr, comm_addr); });
256   if (current_comm_pattern == std::end(incomplete_pattern))
257     xbt_die("Corresponding communication not found!");
258
259   update_comm_pattern(*current_comm_pattern, comm_addr);
260   std::unique_ptr<PatternCommunication> comm_pattern(*current_comm_pattern);
261   XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %zd", issuer,
262             std::distance(begin(incomplete_pattern), current_comm_pattern));
263   incomplete_pattern.erase(current_comm_pattern);
264
265   if (this->initial_communications_pattern_done) {
266     /* Evaluate comm determinism */
267     this->deterministic_comm_pattern(issuer, comm_pattern.get(), backtracking);
268     initial_communications_pattern[issuer].index_comm++;
269   } else {
270     /* Store comm pattern */
271     initial_communications_pattern[issuer].list.push_back(std::move(comm_pattern));
272   }
273 }
274
275 CommunicationDeterminismChecker::CommunicationDeterminismChecker(Session& s) : Checker(s) {}
276
277 CommunicationDeterminismChecker::~CommunicationDeterminismChecker() = default;
278
279 RecordTrace CommunicationDeterminismChecker::get_record_trace() // override
280 {
281   RecordTrace res;
282   for (auto const& state : stack_)
283     res.push_back(state->get_transition());
284   return res;
285 }
286
287 std::vector<std::string> CommunicationDeterminismChecker::get_textual_trace() // override
288 {
289   std::vector<std::string> trace;
290   for (auto const& state : stack_) {
291     smx_simcall_t req = &state->executed_req_;
292     trace.push_back(mcapi::get().request_to_string(req, state->transition_.argument_, RequestType::executed));
293   }
294   return trace;
295 }
296
297 void CommunicationDeterminismChecker::log_state() // override
298 {
299   if (_sg_mc_comms_determinism) {
300     if (this->send_deterministic && not this->recv_deterministic) {
301       XBT_INFO("*******************************************************");
302       XBT_INFO("**** Only-send-deterministic communication pattern ****");
303       XBT_INFO("*******************************************************");
304       XBT_INFO("%s", this->recv_diff);
305     }
306     if (not this->send_deterministic && this->recv_deterministic) {
307       XBT_INFO("*******************************************************");
308       XBT_INFO("**** Only-recv-deterministic communication pattern ****");
309       XBT_INFO("*******************************************************");
310       XBT_INFO("%s", this->send_diff);
311     }
312   }
313   XBT_INFO("Expanded states = %lu", expanded_states_count_);
314   XBT_INFO("Visited states = %lu", mcapi::get().mc_get_visited_states());
315   XBT_INFO("Executed transitions = %lu", mcapi::get().mc_get_executed_trans());
316   XBT_INFO("Send-deterministic : %s", this->send_deterministic ? "Yes" : "No");
317   if (_sg_mc_comms_determinism)
318     XBT_INFO("Recv-deterministic : %s", this->recv_deterministic ? "Yes" : "No");
319 }
320
321 void CommunicationDeterminismChecker::prepare()
322 {
323   const int maxpid = mcapi::get().get_maxpid();
324
325   initial_communications_pattern.resize(maxpid);
326   incomplete_communications_pattern.resize(maxpid);
327
328   ++expanded_states_count_;
329   auto initial_state = std::make_unique<State>(expanded_states_count_);
330
331   XBT_DEBUG("********* Start communication determinism verification *********");
332
333   /* Get an enabled actor and insert it in the interleave set of the initial state */
334   auto actors = mcapi::get().get_actors();
335   for (auto& actor : actors)
336     if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
337       initial_state->add_interleaving_set(actor.copy.get_buffer());
338
339   stack_.push_back(std::move(initial_state));
340 }
341
342 static inline bool all_communications_are_finished()
343 {
344   auto maxpid = mcapi::get().get_maxpid();
345   for (size_t current_actor = 1; current_actor < maxpid; current_actor++) {
346     if (not incomplete_communications_pattern[current_actor].empty()) {
347       XBT_DEBUG("Some communications are not finished, cannot stop the exploration! State not visited.");
348       return false;
349     }
350   }
351   return true;
352 }
353
354 void CommunicationDeterminismChecker::restoreState()
355 {
356   /* Intermediate backtracking */
357   State* last_state = stack_.back().get();
358   if (last_state->system_state_) {
359     mc_api::get().restore_state(last_state->system_state_);
360     restore_communications_pattern(last_state);
361     return;
362   }
363
364   /* Restore the initial state */
365   mcapi::get().restore_initial_state();
366
367   unsigned n = mcapi::get().get_maxpid();
368   assert(n == incomplete_communications_pattern.size());
369   assert(n == initial_communications_pattern.size());
370   for (unsigned j = 0; j < n; j++) {
371     incomplete_communications_pattern[j].clear();
372     initial_communications_pattern[j].index_comm = 0;
373   }
374
375   /* Traverse the stack from the state at position start and re-execute the transitions */
376   for (std::unique_ptr<simgrid::mc::State> const& state : stack_) {
377     if (state == stack_.back())
378       break;
379
380     int req_num                    = state->transition_.argument_;
381     const s_smx_simcall* saved_req = &state->executed_req_;
382     xbt_assert(saved_req);
383
384     /* because we got a copy of the executed request, we have to fetch the
385        real one, pointed by the request field of the issuer process */
386
387     const smx_actor_t issuer = mcapi::get().simcall_get_issuer(saved_req);
388     smx_simcall_t req        = &issuer->simcall_;
389
390     /* TODO : handle test and testany simcalls */
391     CallType call = MC_get_call_type(req);
392     mcapi::get().handle_simcall(state->transition_);
393     handle_comm_pattern(call, req, req_num, 1);
394     mcapi::get().mc_wait_for_requests();
395
396     /* Update statistics */
397     mcapi::get().mc_inc_visited_states();
398     mcapi::get().mc_inc_executed_trans();
399   }
400 }
401
402 void CommunicationDeterminismChecker::handle_comm_pattern(simgrid::mc::CallType call_type, smx_simcall_t req, int value, int backtracking)
403 {
404   using simgrid::mc::CallType;
405   switch(call_type) {
406     case CallType::NONE:
407       break;
408     case CallType::SEND:
409     case CallType::RECV:
410       get_comm_pattern(req, call_type, backtracking);
411       break;
412     case CallType::WAIT:
413     case CallType::WAITANY: {
414       simgrid::kernel::activity::CommImpl* comm_addr = nullptr;
415       if (call_type == CallType::WAIT)
416         comm_addr = mcapi::get().get_comm_wait_raw_addr(req);
417       else
418         comm_addr = mcapi::get().get_comm_waitany_raw_addr(req, value);
419       auto simcall_issuer = mcapi::get().simcall_get_issuer(req);
420       complete_comm_pattern(comm_addr, simcall_issuer->get_pid(), backtracking);
421     } break;
422   default:
423     xbt_die("Unexpected call type %i", (int)call_type);
424   }
425 }
426
427 void CommunicationDeterminismChecker::real_run()
428 {
429   std::unique_ptr<VisitedState> visited_state = nullptr;
430   smx_simcall_t req                           = nullptr;
431
432   while (not stack_.empty()) {
433     /* Get current state */
434     State* cur_state = stack_.back().get();
435
436     XBT_DEBUG("**************************************************");
437     XBT_DEBUG("Exploration depth = %zu (state = %d, interleaved processes = %zu)", stack_.size(), cur_state->num_,
438               cur_state->interleave_size());
439
440     /* Update statistics */
441     mcapi::get().mc_inc_visited_states();
442
443     if (stack_.size() <= (std::size_t)_sg_mc_max_depth)
444       req = mcapi::get().mc_state_choose_request(cur_state);
445     else
446       req = nullptr;
447
448     if (req != nullptr && visited_state == nullptr) {
449       int req_num = cur_state->transition_.argument_;
450
451       XBT_DEBUG("Execute: %s", mcapi::get().request_to_string(req, req_num, RequestType::simix).c_str());
452
453       std::string req_str;
454       if (dot_output != nullptr)
455         req_str = mcapi::get().request_get_dot_output(req, req_num);
456
457       mcapi::get().mc_inc_executed_trans();
458
459       /* TODO : handle test and testany simcalls */
460       CallType call = CallType::NONE;
461       if (_sg_mc_comms_determinism || _sg_mc_send_determinism)
462         call = MC_get_call_type(req);
463
464       /* Answer the request */
465       mcapi::get().handle_simcall(cur_state->transition_);
466       /* After this call req is no longer useful */
467
468       handle_comm_pattern(call, req, req_num, 0);
469
470       /* Wait for requests (schedules processes) */
471       mcapi::get().mc_wait_for_requests();
472
473       /* Create the new expanded state */
474       ++expanded_states_count_;
475       auto next_state = std::make_unique<State>(expanded_states_count_);
476
477       /* If comm determinism verification, we cannot stop the exploration if some communications are not finished (at
478        * least, data are transferred). These communications  are incomplete and they cannot be analyzed and compared
479        * with the initial pattern. */
480       bool compare_snapshots = this->initial_communications_pattern_done && all_communications_are_finished();
481
482       if (_sg_mc_max_visited_states != 0)
483         visited_state = visited_states_.addVisitedState(expanded_states_count_, next_state.get(), compare_snapshots);
484       else
485         visited_state = nullptr;
486
487       if (visited_state == nullptr) {
488         /* Get enabled actors and insert them in the interleave set of the next state */
489         auto actors = mcapi::get().get_actors();
490         for (auto& actor : actors)
491           if (mcapi::get().actor_is_enabled(actor.copy.get_buffer()->get_pid()))
492             next_state->add_interleaving_set(actor.copy.get_buffer());
493
494         if (dot_output != nullptr)
495           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_, next_state->num_, req_str.c_str());
496
497       } else if (dot_output != nullptr)
498         fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", cur_state->num_,
499                 visited_state->original_num == -1 ? visited_state->num : visited_state->original_num, req_str.c_str());
500
501       stack_.push_back(std::move(next_state));
502     } else {
503       if (stack_.size() > (std::size_t)_sg_mc_max_depth)
504         XBT_WARN("/!\\ Max depth reached! /!\\ ");
505       else if (visited_state != nullptr)
506         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.",
507                   visited_state->original_num == -1 ? visited_state->num : visited_state->original_num);
508       else
509         XBT_DEBUG("There are no more processes to interleave. (depth %zu)", stack_.size());
510
511       this->initial_communications_pattern_done = true;
512
513       /* Trash the current state, no longer needed */
514       XBT_DEBUG("Delete state %d at depth %zu", cur_state->num_, stack_.size());
515       stack_.pop_back();
516
517       visited_state = nullptr;
518
519       /* Check for deadlocks */
520       if (mcapi::get().mc_check_deadlock()) {
521         mcapi::get().mc_show_deadlock();
522         throw simgrid::mc::DeadlockError();
523       }
524
525       while (not stack_.empty()) {
526         std::unique_ptr<State> state(std::move(stack_.back()));
527         stack_.pop_back();
528         if (state->interleave_size() && stack_.size() < (std::size_t)_sg_mc_max_depth) {
529           /* We found a back-tracking point, let's loop */
530           XBT_DEBUG("Back-tracking to state %d at depth %zu", state->num_, stack_.size() + 1);
531           stack_.push_back(std::move(state));
532
533           this->restoreState();
534
535           XBT_DEBUG("Back-tracking to state %d at depth %zu done", stack_.back()->num_, stack_.size());
536
537           break;
538         } else {
539           XBT_DEBUG("Delete state %d at depth %zu", state->num_, stack_.size() + 1);
540         }
541       }
542     }
543   }
544
545   mcapi::get().log_state();
546 }
547
548 void CommunicationDeterminismChecker::run()
549 {
550   XBT_INFO("Check communication determinism");
551   mcapi::get().s_initialize();
552
553   this->prepare();
554   this->real_run();
555 }
556
557 Checker* createCommunicationDeterminismChecker(Session& s)
558 {
559   return new CommunicationDeterminismChecker(s);
560 }
561
562 } // namespace mc
563 } // namespace simgrid