8 #include "snapshot-interface.h"
10 #include "clockvector.h"
11 #include "cyclegraph.h"
17 #define INITIAL_THREAD_ID 0
21 /** @brief Constructor */
22 ModelChecker::ModelChecker(struct model_params params) :
23 /* Initialize default scheduler */
25 scheduler(new Scheduler()),
27 num_feasible_executions(0),
29 earliest_diverge(NULL),
30 action_trace(new action_list_t()),
31 thread_map(new HashTable<int, Thread *, int>()),
32 obj_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
33 lock_waiters_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
34 obj_thrd_map(new HashTable<void *, std::vector<action_list_t>, uintptr_t, 4 >()),
35 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
36 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
37 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
38 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
39 node_stack(new NodeStack()),
40 mo_graph(new CycleGraph()),
41 failed_promise(false),
42 too_many_reads(false),
44 bad_synchronization(false)
46 /* Allocate this "size" on the snapshotting heap */
47 priv = (struct model_snapshot_members *)calloc(1, sizeof(*priv));
48 /* First thread created will have id INITIAL_THREAD_ID */
49 priv->next_thread_id = INITIAL_THREAD_ID;
51 /* Initialize a model-checker thread, for special ModelActions */
52 model_thread = new Thread(get_next_id());
53 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
56 /** @brief Destructor */
57 ModelChecker::~ModelChecker()
59 for (unsigned int i = 0; i < get_num_threads(); i++)
60 delete thread_map->get(i);
65 delete lock_waiters_map;
68 for (unsigned int i = 0; i < promises->size(); i++)
69 delete (*promises)[i];
72 delete pending_rel_seqs;
74 delete thrd_last_action;
81 * Restores user program to initial state and resets all model-checker data
84 void ModelChecker::reset_to_initial_state()
86 DEBUG("+++ Resetting to initial state +++\n");
87 node_stack->reset_execution();
88 failed_promise = false;
89 too_many_reads = false;
90 bad_synchronization = false;
92 snapshotObject->backTrackBeforeStep(0);
95 /** @return a thread ID for a new Thread */
96 thread_id_t ModelChecker::get_next_id()
98 return priv->next_thread_id++;
101 /** @return the number of user threads created during this execution */
102 unsigned int ModelChecker::get_num_threads()
104 return priv->next_thread_id;
107 /** @return The currently executing Thread. */
108 Thread * ModelChecker::get_current_thread()
110 return scheduler->get_current_thread();
113 /** @return a sequence number for a new ModelAction */
114 modelclock_t ModelChecker::get_next_seq_num()
116 return ++priv->used_sequence_numbers;
120 * @brief Choose the next thread to execute.
122 * This function chooses the next thread that should execute. It can force the
123 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
124 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
125 * The model-checker may have no preference regarding the next thread (i.e.,
126 * when exploring a new execution ordering), in which case this will return
128 * @param curr The current ModelAction. This action might guide the choice of
130 * @return The next thread to run. If the model-checker has no preference, NULL.
132 Thread * ModelChecker::get_next_thread(ModelAction *curr)
137 /* Do not split atomic actions. */
139 return thread_current();
140 /* The THREAD_CREATE action points to the created Thread */
141 else if (curr->get_type() == THREAD_CREATE)
142 return (Thread *)curr->get_location();
145 /* Have we completed exploring the preselected path? */
149 /* Else, we are trying to replay an execution */
150 ModelAction *next = node_stack->get_next()->get_action();
152 if (next == diverge) {
153 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
154 earliest_diverge=diverge;
156 Node *nextnode = next->get_node();
157 Node *prevnode = nextnode->get_parent();
158 scheduler->update_sleep_set(prevnode);
160 /* Reached divergence point */
161 if (nextnode->increment_promise()) {
162 /* The next node will try to satisfy a different set of promises. */
163 tid = next->get_tid();
164 node_stack->pop_restofstack(2);
165 } else if (nextnode->increment_read_from()) {
166 /* The next node will read from a different value. */
167 tid = next->get_tid();
168 node_stack->pop_restofstack(2);
169 } else if (nextnode->increment_future_value()) {
170 /* The next node will try to read from a different future value. */
171 tid = next->get_tid();
172 node_stack->pop_restofstack(2);
173 } else if (nextnode->increment_relseq_break()) {
174 /* The next node will try to resolve a release sequence differently */
175 tid = next->get_tid();
176 node_stack->pop_restofstack(2);
178 /* Make a different thread execute for next step */
179 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
180 tid = prevnode->get_next_backtrack();
181 /* Make sure the backtracked thread isn't sleeping. */
182 node_stack->pop_restofstack(1);
183 if (diverge==earliest_diverge) {
184 earliest_diverge=prevnode->get_action();
187 /* The correct sleep set is in the parent node. */
190 DEBUG("*** Divergence point ***\n");
194 tid = next->get_tid();
196 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
197 ASSERT(tid != THREAD_ID_T_NONE);
198 return thread_map->get(id_to_int(tid));
202 * We need to know what the next actions of all threads in the sleep
203 * set will be. This method computes them and stores the actions at
204 * the corresponding thread object's pending action.
207 void ModelChecker::execute_sleep_set() {
208 for(unsigned int i=0;i<get_num_threads();i++) {
209 thread_id_t tid=int_to_id(i);
210 Thread *thr=get_thread(tid);
211 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
212 thr->set_state(THREAD_RUNNING);
213 scheduler->next_thread(thr);
214 Thread::swap(&system_context, thr);
215 priv->current_action->set_sleep_flag();
216 thr->set_pending(priv->current_action);
219 priv->current_action = NULL;
222 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
223 for(unsigned int i=0;i<get_num_threads();i++) {
224 thread_id_t tid=int_to_id(i);
225 Thread *thr=get_thread(tid);
226 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
227 ModelAction *pending_act=thr->get_pending();
228 if (pending_act->could_synchronize_with(curr)) {
229 //Remove this thread from sleep set
230 scheduler->remove_sleep(thr);
237 * Queries the model-checker for more executions to explore and, if one
238 * exists, resets the model-checker state to execute a new execution.
240 * @return If there are more executions to explore, return true. Otherwise,
243 bool ModelChecker::next_execution()
249 if (isfinalfeasible()) {
250 printf("Earliest divergence point since last feasible execution:\n");
251 if (earliest_diverge)
252 earliest_diverge->print();
254 printf("(Not set)\n");
256 earliest_diverge = NULL;
257 num_feasible_executions++;
260 DEBUG("Number of acquires waiting on pending release sequences: %zu\n",
261 pending_rel_seqs->size());
263 if (isfinalfeasible() || DBG_ENABLED())
266 if ((diverge = get_next_backtrack()) == NULL)
270 printf("Next execution will diverge at:\n");
274 reset_to_initial_state();
278 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
280 switch (act->get_type()) {
284 /* linear search: from most recent to oldest */
285 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
286 action_list_t::reverse_iterator rit;
287 for (rit = list->rbegin(); rit != list->rend(); rit++) {
288 ModelAction *prev = *rit;
289 if (prev->could_synchronize_with(act))
295 case ATOMIC_TRYLOCK: {
296 /* linear search: from most recent to oldest */
297 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
298 action_list_t::reverse_iterator rit;
299 for (rit = list->rbegin(); rit != list->rend(); rit++) {
300 ModelAction *prev = *rit;
301 if (act->is_conflicting_lock(prev))
306 case ATOMIC_UNLOCK: {
307 /* linear search: from most recent to oldest */
308 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
309 action_list_t::reverse_iterator rit;
310 for (rit = list->rbegin(); rit != list->rend(); rit++) {
311 ModelAction *prev = *rit;
312 if (!act->same_thread(prev)&&prev->is_failed_trylock())
323 /** This method finds backtracking points where we should try to
324 * reorder the parameter ModelAction against.
326 * @param the ModelAction to find backtracking points for.
328 void ModelChecker::set_backtracking(ModelAction *act)
330 Thread *t = get_thread(act);
331 ModelAction * prev = get_last_conflict(act);
335 Node * node = prev->get_node()->get_parent();
337 int low_tid, high_tid;
338 if (node->is_enabled(t)) {
339 low_tid = id_to_int(act->get_tid());
340 high_tid = low_tid+1;
343 high_tid = get_num_threads();
346 for(int i = low_tid; i < high_tid; i++) {
347 thread_id_t tid = int_to_id(i);
349 /* Don't backtrack into a point where the thread is disabled or sleeping. */
350 if (node->get_enabled_array()[i]!=THREAD_ENABLED)
353 /* Check if this has been explored already */
354 if (node->has_been_explored(tid))
357 /* See if fairness allows */
358 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
360 for(int t=0;t<node->get_num_threads();t++) {
361 thread_id_t tother=int_to_id(t);
362 if (node->is_enabled(tother) && node->has_priority(tother)) {
370 /* Cache the latest backtracking point */
371 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
372 priv->next_backtrack = prev;
374 /* If this is a new backtracking point, mark the tree */
375 if (!node->set_backtrack(tid))
377 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
378 id_to_int(prev->get_tid()),
379 id_to_int(t->get_id()));
388 * Returns last backtracking point. The model checker will explore a different
389 * path for this point in the next execution.
390 * @return The ModelAction at which the next execution should diverge.
392 ModelAction * ModelChecker::get_next_backtrack()
394 ModelAction *next = priv->next_backtrack;
395 priv->next_backtrack = NULL;
400 * Processes a read or rmw model action.
401 * @param curr is the read model action to process.
402 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
403 * @return True if processing this read updates the mo_graph.
405 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
408 bool updated = false;
410 const ModelAction *reads_from = curr->get_node()->get_read_from();
411 if (reads_from != NULL) {
412 mo_graph->startChanges();
414 value = reads_from->get_value();
415 bool r_status = false;
417 if (!second_part_of_rmw) {
418 check_recency(curr, reads_from);
419 r_status = r_modification_order(curr, reads_from);
423 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
424 mo_graph->rollbackChanges();
425 too_many_reads = false;
429 curr->read_from(reads_from);
430 mo_graph->commitChanges();
431 mo_check_promises(curr->get_tid(), reads_from);
434 } else if (!second_part_of_rmw) {
435 /* Read from future value */
436 value = curr->get_node()->get_future_value();
437 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
438 curr->read_from(NULL);
439 Promise *valuepromise = new Promise(curr, value, expiration);
440 promises->push_back(valuepromise);
442 get_thread(curr)->set_return_value(value);
448 * Processes a lock, trylock, or unlock model action. @param curr is
449 * the read model action to process.
451 * The try lock operation checks whether the lock is taken. If not,
452 * it falls to the normal lock operation case. If so, it returns
455 * The lock operation has already been checked that it is enabled, so
456 * it just grabs the lock and synchronizes with the previous unlock.
458 * The unlock operation has to re-enable all of the threads that are
459 * waiting on the lock.
461 * @return True if synchronization was updated; false otherwise
463 bool ModelChecker::process_mutex(ModelAction *curr) {
464 std::mutex *mutex = (std::mutex *)curr->get_location();
465 struct std::mutex_state *state = mutex->get_state();
466 switch (curr->get_type()) {
467 case ATOMIC_TRYLOCK: {
468 bool success = !state->islocked;
469 curr->set_try_lock(success);
471 get_thread(curr)->set_return_value(0);
474 get_thread(curr)->set_return_value(1);
476 //otherwise fall into the lock case
478 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock) {
479 printf("Lock access before initialization\n");
482 state->islocked = true;
483 ModelAction *unlock = get_last_unlock(curr);
484 //synchronize with the previous unlock statement
485 if (unlock != NULL) {
486 curr->synchronize_with(unlock);
491 case ATOMIC_UNLOCK: {
493 state->islocked = false;
494 //wake up the other threads
495 action_list_t *waiters = lock_waiters_map->get_safe_ptr(curr->get_location());
496 //activate all the waiting threads
497 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
498 scheduler->wake(get_thread(*rit));
510 * Process a write ModelAction
511 * @param curr The ModelAction to process
512 * @return True if the mo_graph was updated or promises were resolved
514 bool ModelChecker::process_write(ModelAction *curr)
516 bool updated_mod_order = w_modification_order(curr);
517 bool updated_promises = resolve_promises(curr);
519 if (promises->size() == 0) {
520 for (unsigned int i = 0; i < futurevalues->size(); i++) {
521 struct PendingFutureValue pfv = (*futurevalues)[i];
522 //Do more ambitious checks now that mo is more complete
523 if (mo_may_allow(pfv.writer, pfv.act)&&
524 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
525 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
526 priv->next_backtrack = pfv.act;
528 futurevalues->resize(0);
531 mo_graph->commitChanges();
532 mo_check_promises(curr->get_tid(), curr);
534 get_thread(curr)->set_return_value(VALUE_NONE);
535 return updated_mod_order || updated_promises;
539 * @brief Process the current action for thread-related activity
541 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
542 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
543 * synchronization, etc. This function is a no-op for non-THREAD actions
544 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
546 * @param curr The current action
547 * @return True if synchronization was updated or a thread completed
549 bool ModelChecker::process_thread_action(ModelAction *curr)
551 bool updated = false;
553 switch (curr->get_type()) {
554 case THREAD_CREATE: {
555 Thread *th = (Thread *)curr->get_location();
556 th->set_creation(curr);
560 Thread *waiting, *blocking;
561 waiting = get_thread(curr);
562 blocking = (Thread *)curr->get_location();
563 if (!blocking->is_complete()) {
564 blocking->push_wait_list(curr);
565 scheduler->sleep(waiting);
567 do_complete_join(curr);
568 updated = true; /* trigger rel-seq checks */
572 case THREAD_FINISH: {
573 Thread *th = get_thread(curr);
574 while (!th->wait_list_empty()) {
575 ModelAction *act = th->pop_wait_list();
576 Thread *wake = get_thread(act);
577 scheduler->wake(wake);
578 do_complete_join(act);
579 updated = true; /* trigger rel-seq checks */
582 updated = true; /* trigger rel-seq checks */
586 check_promises(curr->get_tid(), NULL, curr->get_cv());
597 * @brief Process the current action for release sequence fixup activity
599 * Performs model-checker release sequence fixups for the current action,
600 * forcing a single pending release sequence to break (with a given, potential
601 * "loose" write) or to complete (i.e., synchronize). If a pending release
602 * sequence forms a complete release sequence, then we must perform the fixup
603 * synchronization, mo_graph additions, etc.
605 * @param curr The current action; must be a release sequence fixup action
606 * @param work_queue The work queue to which to add work items as they are
609 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
611 const ModelAction *write = curr->get_node()->get_relseq_break();
612 struct release_seq *sequence = pending_rel_seqs->back();
613 pending_rel_seqs->pop_back();
615 ModelAction *acquire = sequence->acquire;
616 const ModelAction *rf = sequence->rf;
617 const ModelAction *release = sequence->release;
621 ASSERT(release->same_thread(rf));
625 * @todo Forcing a synchronization requires that we set
626 * modification order constraints. For instance, we can't allow
627 * a fixup sequence in which two separate read-acquire
628 * operations read from the same sequence, where the first one
629 * synchronizes and the other doesn't. Essentially, we can't
630 * allow any writes to insert themselves between 'release' and
634 /* Must synchronize */
635 if (!acquire->synchronize_with(release)) {
636 set_bad_synchronization();
639 /* Re-check all pending release sequences */
640 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
641 /* Re-check act for mo_graph edges */
642 work_queue->push_back(MOEdgeWorkEntry(acquire));
644 /* propagate synchronization to later actions */
645 action_list_t::reverse_iterator rit = action_trace->rbegin();
646 for (; (*rit) != acquire; rit++) {
647 ModelAction *propagate = *rit;
648 if (acquire->happens_before(propagate)) {
649 propagate->synchronize_with(acquire);
650 /* Re-check 'propagate' for mo_graph edges */
651 work_queue->push_back(MOEdgeWorkEntry(propagate));
655 /* Break release sequence with new edges:
656 * release --mo--> write --mo--> rf */
657 mo_graph->addEdge(release, write);
658 mo_graph->addEdge(write, rf);
661 /* See if we have realized a data race */
662 if (checkDataRaces())
667 * Initialize the current action by performing one or more of the following
668 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
669 * in the NodeStack, manipulating backtracking sets, allocating and
670 * initializing clock vectors, and computing the promises to fulfill.
672 * @param curr The current action, as passed from the user context; may be
673 * freed/invalidated after the execution of this function
674 * @return The current action, as processed by the ModelChecker. Is only the
675 * same as the parameter @a curr if this is a newly-explored action.
677 ModelAction * ModelChecker::initialize_curr_action(ModelAction *curr)
679 ModelAction *newcurr;
681 if (curr->is_rmwc() || curr->is_rmw()) {
682 newcurr = process_rmw(curr);
685 if (newcurr->is_rmw())
686 compute_promises(newcurr);
690 curr->set_seq_number(get_next_seq_num());
692 newcurr = node_stack->explore_action(curr, scheduler->get_enabled());
694 /* First restore type and order in case of RMW operation */
696 newcurr->copy_typeandorder(curr);
698 ASSERT(curr->get_location() == newcurr->get_location());
699 newcurr->copy_from_new(curr);
701 /* Discard duplicate ModelAction; use action from NodeStack */
704 /* Always compute new clock vector */
705 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
709 /* Always compute new clock vector */
710 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
712 * Perform one-time actions when pushing new ModelAction onto
715 if (newcurr->is_write())
716 compute_promises(newcurr);
717 else if (newcurr->is_relseq_fixup())
718 compute_relseq_breakwrites(newcurr);
724 * This method checks whether a model action is enabled at the given point.
725 * At this point, it checks whether a lock operation would be successful at this point.
726 * If not, it puts the thread in a waiter list.
727 * @param curr is the ModelAction to check whether it is enabled.
728 * @return a bool that indicates whether the action is enabled.
730 bool ModelChecker::check_action_enabled(ModelAction *curr) {
731 if (curr->is_lock()) {
732 std::mutex * lock = (std::mutex *)curr->get_location();
733 struct std::mutex_state * state = lock->get_state();
734 if (state->islocked) {
735 //Stick the action in the appropriate waiting queue
736 lock_waiters_map->get_safe_ptr(curr->get_location())->push_back(curr);
745 * This is the heart of the model checker routine. It performs model-checking
746 * actions corresponding to a given "current action." Among other processes, it
747 * calculates reads-from relationships, updates synchronization clock vectors,
748 * forms a memory_order constraints graph, and handles replay/backtrack
749 * execution when running permutations of previously-observed executions.
751 * @param curr The current action to process
752 * @return The next Thread that must be executed. May be NULL if ModelChecker
753 * makes no choice (e.g., according to replay execution, combining RMW actions,
756 Thread * ModelChecker::check_current_action(ModelAction *curr)
759 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
761 if (!check_action_enabled(curr)) {
762 /* Make the execution look like we chose to run this action
763 * much later, when a lock is actually available to release */
764 get_current_thread()->set_pending(curr);
765 scheduler->sleep(get_current_thread());
766 return get_next_thread(NULL);
769 wake_up_sleeping_actions(curr);
771 ModelAction *newcurr = initialize_curr_action(curr);
774 /* Add the action to lists before any other model-checking tasks */
775 if (!second_part_of_rmw)
776 add_action_to_lists(newcurr);
778 /* Build may_read_from set for newly-created actions */
779 if (curr == newcurr && curr->is_read())
780 build_reads_from_past(curr);
783 /* Initialize work_queue with the "current action" work */
784 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
785 while (!work_queue.empty()) {
786 WorkQueueEntry work = work_queue.front();
787 work_queue.pop_front();
790 case WORK_CHECK_CURR_ACTION: {
791 ModelAction *act = work.action;
792 bool update = false; /* update this location's release seq's */
793 bool update_all = false; /* update all release seq's */
795 if (process_thread_action(curr))
798 if (act->is_read() && process_read(act, second_part_of_rmw))
801 if (act->is_write() && process_write(act))
804 if (act->is_mutex_op() && process_mutex(act))
807 if (act->is_relseq_fixup())
808 process_relseq_fixup(curr, &work_queue);
811 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
813 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
816 case WORK_CHECK_RELEASE_SEQ:
817 resolve_release_sequences(work.location, &work_queue);
819 case WORK_CHECK_MO_EDGES: {
820 /** @todo Complete verification of work_queue */
821 ModelAction *act = work.action;
822 bool updated = false;
824 if (act->is_read()) {
825 const ModelAction *rf = act->get_reads_from();
826 if (rf != NULL && r_modification_order(act, rf))
829 if (act->is_write()) {
830 if (w_modification_order(act))
833 mo_graph->commitChanges();
836 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
845 check_curr_backtracking(curr);
846 set_backtracking(curr);
847 return get_next_thread(curr);
851 * Complete a THREAD_JOIN operation, by synchronizing with the THREAD_FINISH
852 * operation from the Thread it is joining with. Must be called after the
853 * completion of the Thread in question.
854 * @param join The THREAD_JOIN action
856 void ModelChecker::do_complete_join(ModelAction *join)
858 Thread *blocking = (Thread *)join->get_location();
859 ModelAction *act = get_last_action(blocking->get_id());
860 join->synchronize_with(act);
863 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
864 Node *currnode = curr->get_node();
865 Node *parnode = currnode->get_parent();
867 if ((!parnode->backtrack_empty() ||
868 !currnode->read_from_empty() ||
869 !currnode->future_value_empty() ||
870 !currnode->promise_empty() ||
871 !currnode->relseq_break_empty())
872 && (!priv->next_backtrack ||
873 *curr > *priv->next_backtrack)) {
874 priv->next_backtrack = curr;
878 bool ModelChecker::promises_expired() {
879 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
880 Promise *promise = (*promises)[promise_index];
881 if (promise->get_expiration()<priv->used_sequence_numbers) {
888 /** @return whether the current partial trace must be a prefix of a
890 bool ModelChecker::isfeasibleprefix() {
891 return promises->size() == 0 && pending_rel_seqs->size() == 0;
894 /** @return whether the current partial trace is feasible. */
895 bool ModelChecker::isfeasible() {
896 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
897 DEBUG("Infeasible: RMW violation\n");
899 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
902 /** @return whether the current partial trace is feasible other than
903 * multiple RMW reading from the same store. */
904 bool ModelChecker::isfeasibleotherthanRMW() {
906 if (mo_graph->checkForCycles())
907 DEBUG("Infeasible: modification order cycles\n");
909 DEBUG("Infeasible: failed promise\n");
911 DEBUG("Infeasible: too many reads\n");
912 if (bad_synchronization)
913 DEBUG("Infeasible: bad synchronization ordering\n");
914 if (promises_expired())
915 DEBUG("Infeasible: promises expired\n");
917 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
920 /** Returns whether the current completed trace is feasible. */
921 bool ModelChecker::isfinalfeasible() {
922 if (DBG_ENABLED() && promises->size() != 0)
923 DEBUG("Infeasible: unrevolved promises\n");
925 return isfeasible() && promises->size() == 0;
928 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
929 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
930 ModelAction *lastread = get_last_action(act->get_tid());
931 lastread->process_rmw(act);
932 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
933 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
934 mo_graph->commitChanges();
940 * Checks whether a thread has read from the same write for too many times
941 * without seeing the effects of a later write.
944 * 1) there must a different write that we could read from that would satisfy the modification order,
945 * 2) we must have read from the same value in excess of maxreads times, and
946 * 3) that other write must have been in the reads_from set for maxreads times.
948 * If so, we decide that the execution is no longer feasible.
950 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
951 if (params.maxreads != 0) {
953 if (curr->get_node()->get_read_from_size() <= 1)
955 //Must make sure that execution is currently feasible... We could
956 //accidentally clear by rolling back
959 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
960 int tid = id_to_int(curr->get_tid());
963 if ((int)thrd_lists->size() <= tid)
965 action_list_t *list = &(*thrd_lists)[tid];
967 action_list_t::reverse_iterator rit = list->rbegin();
969 for (; (*rit) != curr; rit++)
971 /* go past curr now */
974 action_list_t::reverse_iterator ritcopy = rit;
975 //See if we have enough reads from the same value
977 for (; count < params.maxreads; rit++,count++) {
978 if (rit==list->rend())
980 ModelAction *act = *rit;
984 if (act->get_reads_from() != rf)
986 if (act->get_node()->get_read_from_size() <= 1)
989 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
991 const ModelAction * write = curr->get_node()->get_read_from_at(i);
993 //Need a different write
997 /* Test to see whether this is a feasible write to read from*/
998 mo_graph->startChanges();
999 r_modification_order(curr, write);
1000 bool feasiblereadfrom = isfeasible();
1001 mo_graph->rollbackChanges();
1003 if (!feasiblereadfrom)
1007 bool feasiblewrite = true;
1008 //new we need to see if this write works for everyone
1010 for (int loop = count; loop>0; loop--,rit++) {
1011 ModelAction *act=*rit;
1012 bool foundvalue = false;
1013 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1014 if (act->get_node()->get_read_from_at(i)==write) {
1020 feasiblewrite = false;
1024 if (feasiblewrite) {
1025 too_many_reads = true;
1033 * Updates the mo_graph with the constraints imposed from the current
1036 * Basic idea is the following: Go through each other thread and find
1037 * the lastest action that happened before our read. Two cases:
1039 * (1) The action is a write => that write must either occur before
1040 * the write we read from or be the write we read from.
1042 * (2) The action is a read => the write that that action read from
1043 * must occur before the write we read from or be the same write.
1045 * @param curr The current action. Must be a read.
1046 * @param rf The action that curr reads from. Must be a write.
1047 * @return True if modification order edges were added; false otherwise
1049 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1051 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1054 ASSERT(curr->is_read());
1056 /* Iterate over all threads */
1057 for (i = 0; i < thrd_lists->size(); i++) {
1058 /* Iterate over actions in thread, starting from most recent */
1059 action_list_t *list = &(*thrd_lists)[i];
1060 action_list_t::reverse_iterator rit;
1061 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1062 ModelAction *act = *rit;
1065 * Include at most one act per-thread that "happens
1066 * before" curr. Don't consider reflexively.
1068 if (act->happens_before(curr) && act != curr) {
1069 if (act->is_write()) {
1071 mo_graph->addEdge(act, rf);
1075 const ModelAction *prevreadfrom = act->get_reads_from();
1076 //if the previous read is unresolved, keep going...
1077 if (prevreadfrom == NULL)
1080 if (rf != prevreadfrom) {
1081 mo_graph->addEdge(prevreadfrom, rf);
1093 /** This method fixes up the modification order when we resolve a
1094 * promises. The basic problem is that actions that occur after the
1095 * read curr could not property add items to the modification order
1098 * So for each thread, we find the earliest item that happens after
1099 * the read curr. This is the item we have to fix up with additional
1100 * constraints. If that action is write, we add a MO edge between
1101 * the Action rf and that action. If the action is a read, we add a
1102 * MO edge between the Action rf, and whatever the read accessed.
1104 * @param curr is the read ModelAction that we are fixing up MO edges for.
1105 * @param rf is the write ModelAction that curr reads from.
1108 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1110 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1112 ASSERT(curr->is_read());
1114 /* Iterate over all threads */
1115 for (i = 0; i < thrd_lists->size(); i++) {
1116 /* Iterate over actions in thread, starting from most recent */
1117 action_list_t *list = &(*thrd_lists)[i];
1118 action_list_t::reverse_iterator rit;
1119 ModelAction *lastact = NULL;
1121 /* Find last action that happens after curr that is either not curr or a rmw */
1122 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1123 ModelAction *act = *rit;
1124 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1130 /* Include at most one act per-thread that "happens before" curr */
1131 if (lastact != NULL) {
1132 if (lastact==curr) {
1133 //Case 1: The resolved read is a RMW, and we need to make sure
1134 //that the write portion of the RMW mod order after rf
1136 mo_graph->addEdge(rf, lastact);
1137 } else if (lastact->is_read()) {
1138 //Case 2: The resolved read is a normal read and the next
1139 //operation is a read, and we need to make sure the value read
1140 //is mod ordered after rf
1142 const ModelAction *postreadfrom = lastact->get_reads_from();
1143 if (postreadfrom != NULL&&rf != postreadfrom)
1144 mo_graph->addEdge(rf, postreadfrom);
1146 //Case 3: The resolved read is a normal read and the next
1147 //operation is a write, and we need to make sure that the
1148 //write is mod ordered after rf
1150 mo_graph->addEdge(rf, lastact);
1158 * Updates the mo_graph with the constraints imposed from the current write.
1160 * Basic idea is the following: Go through each other thread and find
1161 * the lastest action that happened before our write. Two cases:
1163 * (1) The action is a write => that write must occur before
1166 * (2) The action is a read => the write that that action read from
1167 * must occur before the current write.
1169 * This method also handles two other issues:
1171 * (I) Sequential Consistency: Making sure that if the current write is
1172 * seq_cst, that it occurs after the previous seq_cst write.
1174 * (II) Sending the write back to non-synchronizing reads.
1176 * @param curr The current action. Must be a write.
1177 * @return True if modification order edges were added; false otherwise
1179 bool ModelChecker::w_modification_order(ModelAction *curr)
1181 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1184 ASSERT(curr->is_write());
1186 if (curr->is_seqcst()) {
1187 /* We have to at least see the last sequentially consistent write,
1188 so we are initialized. */
1189 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1190 if (last_seq_cst != NULL) {
1191 mo_graph->addEdge(last_seq_cst, curr);
1196 /* Iterate over all threads */
1197 for (i = 0; i < thrd_lists->size(); i++) {
1198 /* Iterate over actions in thread, starting from most recent */
1199 action_list_t *list = &(*thrd_lists)[i];
1200 action_list_t::reverse_iterator rit;
1201 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1202 ModelAction *act = *rit;
1205 * 1) If RMW and it actually read from something, then we
1206 * already have all relevant edges, so just skip to next
1209 * 2) If RMW and it didn't read from anything, we should
1210 * whatever edge we can get to speed up convergence.
1212 * 3) If normal write, we need to look at earlier actions, so
1213 * continue processing list.
1215 if (curr->is_rmw()) {
1216 if (curr->get_reads_from()!=NULL)
1225 * Include at most one act per-thread that "happens
1228 if (act->happens_before(curr)) {
1230 * Note: if act is RMW, just add edge:
1232 * The following edge should be handled elsewhere:
1233 * readfrom(act) --mo--> act
1235 if (act->is_write())
1236 mo_graph->addEdge(act, curr);
1237 else if (act->is_read()) {
1238 //if previous read accessed a null, just keep going
1239 if (act->get_reads_from() == NULL)
1241 mo_graph->addEdge(act->get_reads_from(), curr);
1245 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1246 !act->same_thread(curr)) {
1247 /* We have an action that:
1248 (1) did not happen before us
1249 (2) is a read and we are a write
1250 (3) cannot synchronize with us
1251 (4) is in a different thread
1253 that read could potentially read from our write. Note that
1254 these checks are overly conservative at this point, we'll
1255 do more checks before actually removing the
1259 if (thin_air_constraint_may_allow(curr, act)) {
1261 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1262 struct PendingFutureValue pfv = {curr,act};
1263 futurevalues->push_back(pfv);
1273 /** Arbitrary reads from the future are not allowed. Section 29.3
1274 * part 9 places some constraints. This method checks one result of constraint
1275 * constraint. Others require compiler support. */
1276 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1277 if (!writer->is_rmw())
1280 if (!reader->is_rmw())
1283 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1284 if (search == reader)
1286 if (search->get_tid() == reader->get_tid() &&
1287 search->happens_before(reader))
1294 /** Arbitrary reads from the future are not allowed. Section 29.3
1295 * part 9 places some constraints. This method checks one result of constraint
1296 * constraint. Others require compiler support. */
1297 bool ModelChecker::mo_may_allow(const ModelAction * writer, const ModelAction *reader) {
1298 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(reader->get_location());
1300 //Get write that follows reader action
1301 action_list_t *list = &(*thrd_lists)[id_to_int(reader->get_tid())];
1302 action_list_t::reverse_iterator rit;
1303 ModelAction *first_write_after_read=NULL;
1305 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1306 ModelAction *act = *rit;
1309 if (act->is_write())
1310 first_write_after_read=act;
1313 if (first_write_after_read==NULL)
1317 //return !mo_graph->checkReachable(first_write_after_read, writer);
1323 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1324 * The ModelAction under consideration is expected to be taking part in
1325 * release/acquire synchronization as an object of the "reads from" relation.
1326 * Note that this can only provide release sequence support for RMW chains
1327 * which do not read from the future, as those actions cannot be traced until
1328 * their "promise" is fulfilled. Similarly, we may not even establish the
1329 * presence of a release sequence with certainty, as some modification order
1330 * constraints may be decided further in the future. Thus, this function
1331 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1332 * and a boolean representing certainty.
1334 * @param rf The action that might be part of a release sequence. Must be a
1336 * @param release_heads A pass-by-reference style return parameter. After
1337 * execution of this function, release_heads will contain the heads of all the
1338 * relevant release sequences, if any exists with certainty
1339 * @param pending A pass-by-reference style return parameter which is only used
1340 * when returning false (i.e., uncertain). Returns most information regarding
1341 * an uncertain release sequence, including any write operations that might
1342 * break the sequence.
1343 * @return true, if the ModelChecker is certain that release_heads is complete;
1346 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1347 rel_heads_list_t *release_heads,
1348 struct release_seq *pending) const
1350 /* Only check for release sequences if there are no cycles */
1351 if (mo_graph->checkForCycles())
1355 ASSERT(rf->is_write());
1357 if (rf->is_release())
1358 release_heads->push_back(rf);
1360 break; /* End of RMW chain */
1362 /** @todo Need to be smarter here... In the linux lock
1363 * example, this will run to the beginning of the program for
1365 /** @todo The way to be smarter here is to keep going until 1
1366 * thread has a release preceded by an acquire and you've seen
1369 /* acq_rel RMW is a sufficient stopping condition */
1370 if (rf->is_acquire() && rf->is_release())
1371 return true; /* complete */
1373 rf = rf->get_reads_from();
1376 /* read from future: need to settle this later */
1378 return false; /* incomplete */
1381 if (rf->is_release())
1382 return true; /* complete */
1384 /* else relaxed write; check modification order for contiguous subsequence
1385 * -> rf must be same thread as release */
1386 int tid = id_to_int(rf->get_tid());
1387 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(rf->get_location());
1388 action_list_t *list = &(*thrd_lists)[tid];
1389 action_list_t::const_reverse_iterator rit;
1391 /* Find rf in the thread list */
1392 rit = std::find(list->rbegin(), list->rend(), rf);
1393 ASSERT(rit != list->rend());
1395 /* Find the last write/release */
1396 for (; rit != list->rend(); rit++)
1397 if ((*rit)->is_release())
1399 if (rit == list->rend()) {
1400 /* No write-release in this thread */
1401 return true; /* complete */
1403 ModelAction *release = *rit;
1405 ASSERT(rf->same_thread(release));
1407 pending->writes.clear();
1409 bool certain = true;
1410 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1411 if (id_to_int(rf->get_tid()) == (int)i)
1413 list = &(*thrd_lists)[i];
1415 /* Can we ensure no future writes from this thread may break
1416 * the release seq? */
1417 bool future_ordered = false;
1419 ModelAction *last = get_last_action(int_to_id(i));
1420 Thread *th = get_thread(int_to_id(i));
1421 if ((last && rf->happens_before(last)) ||
1422 !scheduler->is_enabled(th) ||
1424 future_ordered = true;
1426 ASSERT(!th->is_model_thread() || future_ordered);
1428 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1429 const ModelAction *act = *rit;
1430 /* Reach synchronization -> this thread is complete */
1431 if (act->happens_before(release))
1433 if (rf->happens_before(act)) {
1434 future_ordered = true;
1438 /* Only writes can break release sequences */
1439 if (!act->is_write())
1442 /* Check modification order */
1443 if (mo_graph->checkReachable(rf, act)) {
1444 /* rf --mo--> act */
1445 future_ordered = true;
1448 if (mo_graph->checkReachable(act, release))
1449 /* act --mo--> release */
1451 if (mo_graph->checkReachable(release, act) &&
1452 mo_graph->checkReachable(act, rf)) {
1453 /* release --mo-> act --mo--> rf */
1454 return true; /* complete */
1456 /* act may break release sequence */
1457 pending->writes.push_back(act);
1460 if (!future_ordered)
1461 certain = false; /* This thread is uncertain */
1465 release_heads->push_back(release);
1466 pending->writes.clear();
1468 pending->release = release;
1475 * A public interface for getting the release sequence head(s) with which a
1476 * given ModelAction must synchronize. This function only returns a non-empty
1477 * result when it can locate a release sequence head with certainty. Otherwise,
1478 * it may mark the internal state of the ModelChecker so that it will handle
1479 * the release sequence at a later time, causing @a act to update its
1480 * synchronization at some later point in execution.
1481 * @param act The 'acquire' action that may read from a release sequence
1482 * @param release_heads A pass-by-reference return parameter. Will be filled
1483 * with the head(s) of the release sequence(s), if they exists with certainty.
1484 * @see ModelChecker::release_seq_heads
1486 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1488 const ModelAction *rf = act->get_reads_from();
1489 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1490 sequence->acquire = act;
1492 if (!release_seq_heads(rf, release_heads, sequence)) {
1493 /* add act to 'lazy checking' list */
1494 pending_rel_seqs->push_back(sequence);
1496 snapshot_free(sequence);
1501 * Attempt to resolve all stashed operations that might synchronize with a
1502 * release sequence for a given location. This implements the "lazy" portion of
1503 * determining whether or not a release sequence was contiguous, since not all
1504 * modification order information is present at the time an action occurs.
1506 * @param location The location/object that should be checked for release
1507 * sequence resolutions. A NULL value means to check all locations.
1508 * @param work_queue The work queue to which to add work items as they are
1510 * @return True if any updates occurred (new synchronization, new mo_graph
1513 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1515 bool updated = false;
1516 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1517 while (it != pending_rel_seqs->end()) {
1518 struct release_seq *pending = *it;
1519 ModelAction *act = pending->acquire;
1521 /* Only resolve sequences on the given location, if provided */
1522 if (location && act->get_location() != location) {
1527 const ModelAction *rf = act->get_reads_from();
1528 rel_heads_list_t release_heads;
1530 complete = release_seq_heads(rf, &release_heads, pending);
1531 for (unsigned int i = 0; i < release_heads.size(); i++) {
1532 if (!act->has_synchronized_with(release_heads[i])) {
1533 if (act->synchronize_with(release_heads[i]))
1536 set_bad_synchronization();
1541 /* Re-check all pending release sequences */
1542 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1543 /* Re-check act for mo_graph edges */
1544 work_queue->push_back(MOEdgeWorkEntry(act));
1546 /* propagate synchronization to later actions */
1547 action_list_t::reverse_iterator rit = action_trace->rbegin();
1548 for (; (*rit) != act; rit++) {
1549 ModelAction *propagate = *rit;
1550 if (act->happens_before(propagate)) {
1551 propagate->synchronize_with(act);
1552 /* Re-check 'propagate' for mo_graph edges */
1553 work_queue->push_back(MOEdgeWorkEntry(propagate));
1558 it = pending_rel_seqs->erase(it);
1559 snapshot_free(pending);
1565 // If we resolved promises or data races, see if we have realized a data race.
1566 if (checkDataRaces()) {
1574 * Performs various bookkeeping operations for the current ModelAction. For
1575 * instance, adds action to the per-object, per-thread action vector and to the
1576 * action trace list of all thread actions.
1578 * @param act is the ModelAction to add.
1580 void ModelChecker::add_action_to_lists(ModelAction *act)
1582 int tid = id_to_int(act->get_tid());
1583 action_trace->push_back(act);
1585 obj_map->get_safe_ptr(act->get_location())->push_back(act);
1587 std::vector<action_list_t> *vec = obj_thrd_map->get_safe_ptr(act->get_location());
1588 if (tid >= (int)vec->size())
1589 vec->resize(priv->next_thread_id);
1590 (*vec)[tid].push_back(act);
1592 if ((int)thrd_last_action->size() <= tid)
1593 thrd_last_action->resize(get_num_threads());
1594 (*thrd_last_action)[tid] = act;
1598 * @brief Get the last action performed by a particular Thread
1599 * @param tid The thread ID of the Thread in question
1600 * @return The last action in the thread
1602 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1604 int threadid = id_to_int(tid);
1605 if (threadid < (int)thrd_last_action->size())
1606 return (*thrd_last_action)[id_to_int(tid)];
1612 * Gets the last memory_order_seq_cst write (in the total global sequence)
1613 * performed on a particular object (i.e., memory location), not including the
1615 * @param curr The current ModelAction; also denotes the object location to
1617 * @return The last seq_cst write
1619 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1621 void *location = curr->get_location();
1622 action_list_t *list = obj_map->get_safe_ptr(location);
1623 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1624 action_list_t::reverse_iterator rit;
1625 for (rit = list->rbegin(); rit != list->rend(); rit++)
1626 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1632 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1633 * location). This function identifies the mutex according to the current
1634 * action, which is presumed to perform on the same mutex.
1635 * @param curr The current ModelAction; also denotes the object location to
1637 * @return The last unlock operation
1639 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1641 void *location = curr->get_location();
1642 action_list_t *list = obj_map->get_safe_ptr(location);
1643 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1644 action_list_t::reverse_iterator rit;
1645 for (rit = list->rbegin(); rit != list->rend(); rit++)
1646 if ((*rit)->is_unlock())
1651 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1653 ModelAction *parent = get_last_action(tid);
1655 parent = get_thread(tid)->get_creation();
1660 * Returns the clock vector for a given thread.
1661 * @param tid The thread whose clock vector we want
1662 * @return Desired clock vector
1664 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1666 return get_parent_action(tid)->get_cv();
1670 * Resolve a set of Promises with a current write. The set is provided in the
1671 * Node corresponding to @a write.
1672 * @param write The ModelAction that is fulfilling Promises
1673 * @return True if promises were resolved; false otherwise
1675 bool ModelChecker::resolve_promises(ModelAction *write)
1677 bool resolved = false;
1678 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1680 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1681 Promise *promise = (*promises)[promise_index];
1682 if (write->get_node()->get_promise(i)) {
1683 ModelAction *read = promise->get_action();
1684 if (read->is_rmw()) {
1685 mo_graph->addRMWEdge(write, read);
1687 read->read_from(write);
1688 //First fix up the modification order for actions that happened
1690 r_modification_order(read, write);
1691 //Next fix up the modification order for actions that happened
1693 post_r_modification_order(read, write);
1694 //Make sure the promise's value matches the write's value
1695 ASSERT(promise->get_value() == write->get_value());
1698 promises->erase(promises->begin() + promise_index);
1699 threads_to_check.push_back(read->get_tid());
1706 //Check whether reading these writes has made threads unable to
1709 for(unsigned int i=0;i<threads_to_check.size();i++)
1710 mo_check_promises(threads_to_check[i], write);
1716 * Compute the set of promises that could potentially be satisfied by this
1717 * action. Note that the set computation actually appears in the Node, not in
1719 * @param curr The ModelAction that may satisfy promises
1721 void ModelChecker::compute_promises(ModelAction *curr)
1723 for (unsigned int i = 0; i < promises->size(); i++) {
1724 Promise *promise = (*promises)[i];
1725 const ModelAction *act = promise->get_action();
1726 if (!act->happens_before(curr) &&
1728 !act->could_synchronize_with(curr) &&
1729 !act->same_thread(curr) &&
1730 promise->get_value() == curr->get_value()) {
1731 curr->get_node()->set_promise(i);
1736 /** Checks promises in response to change in ClockVector Threads. */
1737 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
1739 for (unsigned int i = 0; i < promises->size(); i++) {
1740 Promise *promise = (*promises)[i];
1741 const ModelAction *act = promise->get_action();
1742 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
1743 merge_cv->synchronized_since(act)) {
1744 if (promise->increment_threads(tid)) {
1745 //Promise has failed
1746 failed_promise = true;
1753 /** Checks promises in response to addition to modification order for threads.
1755 * pthread is the thread that performed the read that created the promise
1757 * pread is the read that created the promise
1759 * pwrite is either the first write to same location as pread by
1760 * pthread that is sequenced after pread or the value read by the
1761 * first read to the same lcoation as pread by pthread that is
1762 * sequenced after pread..
1764 * 1. If tid=pthread, then we check what other threads are reachable
1765 * through the mode order starting with pwrite. Those threads cannot
1766 * perform a write that will resolve the promise due to modification
1767 * order constraints.
1769 * 2. If the tid is not pthread, we check whether pwrite can reach the
1770 * action write through the modification order. If so, that thread
1771 * cannot perform a future write that will resolve the promise due to
1772 * modificatin order constraints.
1774 * @parem tid The thread that either read from the model action
1775 * write, or actually did the model action write.
1777 * @parem write The ModelAction representing the relevant write.
1780 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
1781 void * location = write->get_location();
1782 for (unsigned int i = 0; i < promises->size(); i++) {
1783 Promise *promise = (*promises)[i];
1784 const ModelAction *act = promise->get_action();
1786 //Is this promise on the same location?
1787 if ( act->get_location() != location )
1790 //same thread as the promise
1791 if ( act->get_tid()==tid ) {
1793 //do we have a pwrite for the promise, if not, set it
1794 if (promise->get_write() == NULL ) {
1795 promise->set_write(write);
1797 if (mo_graph->checkPromise(write, promise)) {
1798 failed_promise = true;
1803 //Don't do any lookups twice for the same thread
1804 if (promise->has_sync_thread(tid))
1807 if (mo_graph->checkReachable(promise->get_write(), write)) {
1808 if (promise->increment_threads(tid)) {
1809 failed_promise = true;
1817 * Compute the set of writes that may break the current pending release
1818 * sequence. This information is extracted from previou release sequence
1821 * @param curr The current ModelAction. Must be a release sequence fixup
1824 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
1826 if (pending_rel_seqs->empty())
1829 struct release_seq *pending = pending_rel_seqs->back();
1830 for (unsigned int i = 0; i < pending->writes.size(); i++) {
1831 const ModelAction *write = pending->writes[i];
1832 curr->get_node()->add_relseq_break(write);
1835 /* NULL means don't break the sequence; just synchronize */
1836 curr->get_node()->add_relseq_break(NULL);
1840 * Build up an initial set of all past writes that this 'read' action may read
1841 * from. This set is determined by the clock vector's "happens before"
1843 * @param curr is the current ModelAction that we are exploring; it must be a
1846 void ModelChecker::build_reads_from_past(ModelAction *curr)
1848 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1850 ASSERT(curr->is_read());
1852 ModelAction *last_seq_cst = NULL;
1854 /* Track whether this object has been initialized */
1855 bool initialized = false;
1857 if (curr->is_seqcst()) {
1858 last_seq_cst = get_last_seq_cst(curr);
1859 /* We have to at least see the last sequentially consistent write,
1860 so we are initialized. */
1861 if (last_seq_cst != NULL)
1865 /* Iterate over all threads */
1866 for (i = 0; i < thrd_lists->size(); i++) {
1867 /* Iterate over actions in thread, starting from most recent */
1868 action_list_t *list = &(*thrd_lists)[i];
1869 action_list_t::reverse_iterator rit;
1870 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1871 ModelAction *act = *rit;
1873 /* Only consider 'write' actions */
1874 if (!act->is_write() || act == curr)
1877 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
1878 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
1879 DEBUG("Adding action to may_read_from:\n");
1880 if (DBG_ENABLED()) {
1885 if (curr->get_sleep_flag()) {
1886 if (sleep_can_read_from(curr, act))
1887 curr->get_node()->add_read_from(act);
1889 curr->get_node()->add_read_from(act);
1892 /* Include at most one act per-thread that "happens before" curr */
1893 if (act->happens_before(curr)) {
1901 /** @todo Need a more informative way of reporting errors. */
1902 printf("ERROR: may read from uninitialized atomic\n");
1905 if (DBG_ENABLED() || !initialized) {
1906 printf("Reached read action:\n");
1908 printf("Printing may_read_from\n");
1909 curr->get_node()->print_may_read_from();
1910 printf("End printing may_read_from\n");
1913 ASSERT(initialized);
1916 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
1918 Node *prevnode=write->get_node()->get_parent();
1919 bool thread_sleep=prevnode->get_enabled_array()[id_to_int(curr->get_tid())]==THREAD_SLEEP_SET;
1920 if (write->is_release()&&thread_sleep)
1922 if (!write->is_rmw()) {
1925 if (write->get_reads_from()==NULL)
1927 write=write->get_reads_from();
1931 static void print_list(action_list_t *list)
1933 action_list_t::iterator it;
1935 printf("---------------------------------------------------------------------\n");
1937 unsigned int hash=0;
1939 for (it = list->begin(); it != list->end(); it++) {
1941 hash=hash^(hash<<3)^((*it)->hash());
1943 printf("HASH %u\n", hash);
1944 printf("---------------------------------------------------------------------\n");
1947 #if SUPPORT_MOD_ORDER_DUMP
1948 void ModelChecker::dumpGraph(char *filename) {
1950 sprintf(buffer, "%s.dot",filename);
1951 FILE *file=fopen(buffer, "w");
1952 fprintf(file, "digraph %s {\n",filename);
1953 mo_graph->dumpNodes(file);
1954 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
1956 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
1957 ModelAction *action=*it;
1958 if (action->is_read()) {
1959 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
1960 if (action->get_reads_from()!=NULL)
1961 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
1963 if (thread_array[action->get_tid()] != NULL) {
1964 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
1967 thread_array[action->get_tid()]=action;
1969 fprintf(file,"}\n");
1970 model_free(thread_array);
1975 void ModelChecker::print_summary()
1978 printf("Number of executions: %d\n", num_executions);
1979 printf("Number of feasible executions: %d\n", num_feasible_executions);
1980 printf("Total nodes created: %d\n", node_stack->get_total_nodes());
1982 #if SUPPORT_MOD_ORDER_DUMP
1984 char buffername[100];
1985 sprintf(buffername, "exec%04u", num_executions);
1986 mo_graph->dumpGraphToFile(buffername);
1987 sprintf(buffername, "graph%04u", num_executions);
1988 dumpGraph(buffername);
1991 if (!isfinalfeasible())
1992 printf("INFEASIBLE EXECUTION!\n");
1993 print_list(action_trace);
1998 * Add a Thread to the system for the first time. Should only be called once
2000 * @param t The Thread to add
2002 void ModelChecker::add_thread(Thread *t)
2004 thread_map->put(id_to_int(t->get_id()), t);
2005 scheduler->add_thread(t);
2009 * Removes a thread from the scheduler.
2010 * @param the thread to remove.
2012 void ModelChecker::remove_thread(Thread *t)
2014 scheduler->remove_thread(t);
2018 * @brief Get a Thread reference by its ID
2019 * @param tid The Thread's ID
2020 * @return A Thread reference
2022 Thread * ModelChecker::get_thread(thread_id_t tid) const
2024 return thread_map->get(id_to_int(tid));
2028 * @brief Get a reference to the Thread in which a ModelAction was executed
2029 * @param act The ModelAction
2030 * @return A Thread reference
2032 Thread * ModelChecker::get_thread(ModelAction *act) const
2034 return get_thread(act->get_tid());
2038 * Switch from a user-context to the "master thread" context (a.k.a. system
2039 * context). This switch is made with the intention of exploring a particular
2040 * model-checking action (described by a ModelAction object). Must be called
2041 * from a user-thread context.
2043 * @param act The current action that will be explored. May be NULL only if
2044 * trace is exiting via an assertion (see ModelChecker::set_assert and
2045 * ModelChecker::has_asserted).
2046 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2048 int ModelChecker::switch_to_master(ModelAction *act)
2051 Thread *old = thread_current();
2052 set_current_action(act);
2053 old->set_state(THREAD_READY);
2054 return Thread::swap(old, &system_context);
2058 * Takes the next step in the execution, if possible.
2059 * @return Returns true (success) if a step was taken and false otherwise.
2061 bool ModelChecker::take_step() {
2065 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2067 if (curr->get_state() == THREAD_READY) {
2068 ASSERT(priv->current_action);
2070 priv->nextThread = check_current_action(priv->current_action);
2071 priv->current_action = NULL;
2073 if (curr->is_blocked() || curr->is_complete())
2074 scheduler->remove_thread(curr);
2079 Thread *next = scheduler->next_thread(priv->nextThread);
2081 /* Infeasible -> don't take any more steps */
2085 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2086 next ? id_to_int(next->get_id()) : -1);
2089 * Launch end-of-execution release sequence fixups only when there are:
2091 * (1) no more user threads to run (or when execution replay chooses
2092 * the 'model_thread')
2093 * (2) pending release sequences
2094 * (3) pending assertions (i.e., data races)
2095 * (4) no pending promises
2097 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2098 isfinalfeasible() && !unrealizedraces.empty()) {
2099 printf("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2100 pending_rel_seqs->size());
2101 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2102 std::memory_order_seq_cst, NULL, VALUE_NONE,
2104 set_current_action(fixup);
2108 /* next == NULL -> don't take any more steps */
2112 next->set_state(THREAD_RUNNING);
2114 if (next->get_pending() != NULL) {
2115 /* restart a pending action */
2116 set_current_action(next->get_pending());
2117 next->set_pending(NULL);
2118 next->set_state(THREAD_READY);
2122 /* Return false only if swap fails with an error */
2123 return (Thread::swap(&system_context, next) == 0);
2126 /** Runs the current execution until threre are no more steps to take. */
2127 void ModelChecker::finish_execution() {
2130 while (take_step());