9 #include "snapshot-interface.h"
11 #include "clockvector.h"
12 #include "cyclegraph.h"
15 #include "threads-model.h"
18 #define INITIAL_THREAD_ID 0
23 bug_message(const char *str) {
24 const char *fmt = " [BUG] %s\n";
25 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
26 sprintf(msg, fmt, str);
28 ~bug_message() { if (msg) snapshot_free(msg); }
31 void print() { model_print("%s", msg); }
37 * Structure for holding small ModelChecker members that should be snapshotted
39 struct model_snapshot_members {
40 model_snapshot_members() :
42 /* First thread created will have id INITIAL_THREAD_ID */
43 next_thread_id(INITIAL_THREAD_ID),
44 used_sequence_numbers(0),
49 failed_promise(false),
50 too_many_reads(false),
51 bad_synchronization(false),
55 ~model_snapshot_members() {
56 for (unsigned int i = 0; i < bugs.size(); i++)
61 ModelAction *current_action;
62 unsigned int next_thread_id;
63 modelclock_t used_sequence_numbers;
65 ModelAction *next_backtrack;
66 std::vector< bug_message *, SnapshotAlloc<bug_message *> > bugs;
67 struct execution_stats stats;
70 /** @brief Incorrectly-ordered synchronization was made */
71 bool bad_synchronization;
77 /** @brief Constructor */
78 ModelChecker::ModelChecker(struct model_params params) :
79 /* Initialize default scheduler */
81 scheduler(new Scheduler()),
83 earliest_diverge(NULL),
84 action_trace(new action_list_t()),
85 thread_map(new HashTable<int, Thread *, int>()),
86 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
87 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
88 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
89 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
90 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
91 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
92 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
93 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
94 thrd_last_fence_release(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >()),
95 node_stack(new NodeStack()),
96 priv(new struct model_snapshot_members()),
97 mo_graph(new CycleGraph())
99 /* Initialize a model-checker thread, for special ModelActions */
100 model_thread = new Thread(get_next_id());
101 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
104 /** @brief Destructor */
105 ModelChecker::~ModelChecker()
107 for (unsigned int i = 0; i < get_num_threads(); i++)
108 delete thread_map->get(i);
113 delete lock_waiters_map;
114 delete condvar_waiters_map;
117 for (unsigned int i = 0; i < promises->size(); i++)
118 delete (*promises)[i];
121 delete pending_rel_seqs;
123 delete thrd_last_action;
124 delete thrd_last_fence_release;
131 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr) {
132 action_list_t * tmp=hash->get(ptr);
134 tmp=new action_list_t();
140 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr) {
141 std::vector<action_list_t> * tmp=hash->get(ptr);
143 tmp=new std::vector<action_list_t>();
150 * Restores user program to initial state and resets all model-checker data
153 void ModelChecker::reset_to_initial_state()
155 DEBUG("+++ Resetting to initial state +++\n");
156 node_stack->reset_execution();
158 /* Print all model-checker output before rollback */
161 snapshotObject->backTrackBeforeStep(0);
164 /** @return a thread ID for a new Thread */
165 thread_id_t ModelChecker::get_next_id()
167 return priv->next_thread_id++;
170 /** @return the number of user threads created during this execution */
171 unsigned int ModelChecker::get_num_threads() const
173 return priv->next_thread_id;
176 /** @return The currently executing Thread. */
177 Thread * ModelChecker::get_current_thread() const
179 return scheduler->get_current_thread();
182 /** @return a sequence number for a new ModelAction */
183 modelclock_t ModelChecker::get_next_seq_num()
185 return ++priv->used_sequence_numbers;
188 Node * ModelChecker::get_curr_node() const
190 return node_stack->get_head();
194 * @brief Choose the next thread to execute.
196 * This function chooses the next thread that should execute. It can force the
197 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
198 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
199 * The model-checker may have no preference regarding the next thread (i.e.,
200 * when exploring a new execution ordering), in which case this will return
202 * @param curr The current ModelAction. This action might guide the choice of
204 * @return The next thread to run. If the model-checker has no preference, NULL.
206 Thread * ModelChecker::get_next_thread(ModelAction *curr)
211 /* Do not split atomic actions. */
213 return thread_current();
214 /* The THREAD_CREATE action points to the created Thread */
215 else if (curr->get_type() == THREAD_CREATE)
216 return (Thread *)curr->get_location();
219 /* Have we completed exploring the preselected path? */
223 /* Else, we are trying to replay an execution */
224 ModelAction *next = node_stack->get_next()->get_action();
226 if (next == diverge) {
227 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
228 earliest_diverge=diverge;
230 Node *nextnode = next->get_node();
231 Node *prevnode = nextnode->get_parent();
232 scheduler->update_sleep_set(prevnode);
234 /* Reached divergence point */
235 if (nextnode->increment_misc()) {
236 /* The next node will try to satisfy a different misc_index values. */
237 tid = next->get_tid();
238 node_stack->pop_restofstack(2);
239 } else if (nextnode->increment_promise()) {
240 /* The next node will try to satisfy a different set of promises. */
241 tid = next->get_tid();
242 node_stack->pop_restofstack(2);
243 } else if (nextnode->increment_read_from()) {
244 /* The next node will read from a different value. */
245 tid = next->get_tid();
246 node_stack->pop_restofstack(2);
247 } else if (nextnode->increment_future_value()) {
248 /* The next node will try to read from a different future value. */
249 tid = next->get_tid();
250 node_stack->pop_restofstack(2);
251 } else if (nextnode->increment_relseq_break()) {
252 /* The next node will try to resolve a release sequence differently */
253 tid = next->get_tid();
254 node_stack->pop_restofstack(2);
256 /* Make a different thread execute for next step */
257 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
258 tid = prevnode->get_next_backtrack();
259 /* Make sure the backtracked thread isn't sleeping. */
260 node_stack->pop_restofstack(1);
261 if (diverge==earliest_diverge) {
262 earliest_diverge=prevnode->get_action();
265 /* The correct sleep set is in the parent node. */
268 DEBUG("*** Divergence point ***\n");
272 tid = next->get_tid();
274 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
275 ASSERT(tid != THREAD_ID_T_NONE);
276 return thread_map->get(id_to_int(tid));
280 * We need to know what the next actions of all threads in the sleep
281 * set will be. This method computes them and stores the actions at
282 * the corresponding thread object's pending action.
285 void ModelChecker::execute_sleep_set() {
286 for(unsigned int i=0;i<get_num_threads();i++) {
287 thread_id_t tid=int_to_id(i);
288 Thread *thr=get_thread(tid);
289 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET &&
290 thr->get_pending() == NULL ) {
291 thr->set_state(THREAD_RUNNING);
292 scheduler->next_thread(thr);
293 Thread::swap(&system_context, thr);
294 priv->current_action->set_sleep_flag();
295 thr->set_pending(priv->current_action);
298 priv->current_action = NULL;
301 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
302 for(unsigned int i=0;i<get_num_threads();i++) {
303 thread_id_t tid=int_to_id(i);
304 Thread *thr=get_thread(tid);
305 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
306 ModelAction *pending_act=thr->get_pending();
307 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
308 //Remove this thread from sleep set
309 scheduler->remove_sleep(thr);
315 /** @brief Alert the model-checker that an incorrectly-ordered
316 * synchronization was made */
317 void ModelChecker::set_bad_synchronization()
319 priv->bad_synchronization = true;
322 bool ModelChecker::has_asserted() const
324 return priv->asserted;
327 void ModelChecker::set_assert()
329 priv->asserted = true;
333 * Check if we are in a deadlock. Should only be called at the end of an
334 * execution, although it should not give false positives in the middle of an
335 * execution (there should be some ENABLED thread).
337 * @return True if program is in a deadlock; false otherwise
339 bool ModelChecker::is_deadlocked() const
341 bool blocking_threads = false;
342 for (unsigned int i = 0; i < get_num_threads(); i++) {
343 thread_id_t tid = int_to_id(i);
346 Thread *t = get_thread(tid);
347 if (!t->is_model_thread() && t->get_pending())
348 blocking_threads = true;
350 return blocking_threads;
354 * Check if this is a complete execution. That is, have all thread completed
355 * execution (rather than exiting because sleep sets have forced a redundant
358 * @return True if the execution is complete.
360 bool ModelChecker::is_complete_execution() const
362 for (unsigned int i = 0; i < get_num_threads(); i++)
363 if (is_enabled(int_to_id(i)))
369 * @brief Assert a bug in the executing program.
371 * Use this function to assert any sort of bug in the user program. If the
372 * current trace is feasible (actually, a prefix of some feasible execution),
373 * then this execution will be aborted, printing the appropriate message. If
374 * the current trace is not yet feasible, the error message will be stashed and
375 * printed if the execution ever becomes feasible.
377 * @param msg Descriptive message for the bug (do not include newline char)
378 * @return True if bug is immediately-feasible
380 bool ModelChecker::assert_bug(const char *msg)
382 priv->bugs.push_back(new bug_message(msg));
384 if (isfeasibleprefix()) {
392 * @brief Assert a bug in the executing program, asserted by a user thread
393 * @see ModelChecker::assert_bug
394 * @param msg Descriptive message for the bug (do not include newline char)
396 void ModelChecker::assert_user_bug(const char *msg)
398 /* If feasible bug, bail out now */
400 switch_to_master(NULL);
403 /** @return True, if any bugs have been reported for this execution */
404 bool ModelChecker::have_bug_reports() const
406 return priv->bugs.size() != 0;
409 /** @brief Print bug report listing for this execution (if any bugs exist) */
410 void ModelChecker::print_bugs() const
412 if (have_bug_reports()) {
413 model_print("Bug report: %zu bug%s detected\n",
415 priv->bugs.size() > 1 ? "s" : "");
416 for (unsigned int i = 0; i < priv->bugs.size(); i++)
417 priv->bugs[i]->print();
422 * @brief Record end-of-execution stats
424 * Must be run when exiting an execution. Records various stats.
425 * @see struct execution_stats
427 void ModelChecker::record_stats()
430 if (!isfeasibleprefix())
431 stats.num_infeasible++;
432 else if (have_bug_reports())
433 stats.num_buggy_executions++;
434 else if (is_complete_execution())
435 stats.num_complete++;
437 stats.num_redundant++;
440 /** @brief Print execution stats */
441 void ModelChecker::print_stats() const
443 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
444 model_print("Number of redundant executions: %d\n", stats.num_redundant);
445 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
446 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
447 model_print("Total executions: %d\n", stats.num_total);
448 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
452 * @brief End-of-exeuction print
453 * @param printbugs Should any existing bugs be printed?
455 void ModelChecker::print_execution(bool printbugs) const
457 print_program_output();
459 if (DBG_ENABLED() || params.verbose) {
460 model_print("Earliest divergence point since last feasible execution:\n");
461 if (earliest_diverge)
462 earliest_diverge->print();
464 model_print("(Not set)\n");
470 /* Don't print invalid bugs */
479 * Queries the model-checker for more executions to explore and, if one
480 * exists, resets the model-checker state to execute a new execution.
482 * @return If there are more executions to explore, return true. Otherwise,
485 bool ModelChecker::next_execution()
488 /* Is this execution a feasible execution that's worth bug-checking? */
489 bool complete = isfeasibleprefix() && (is_complete_execution() ||
492 /* End-of-execution bug checks */
495 assert_bug("Deadlock detected");
503 if (DBG_ENABLED() || params.verbose || have_bug_reports())
504 print_execution(complete);
506 clear_program_output();
509 earliest_diverge = NULL;
511 if ((diverge = get_next_backtrack()) == NULL)
515 model_print("Next execution will diverge at:\n");
519 reset_to_initial_state();
523 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
525 switch (act->get_type()) {
530 /* Optimization: relaxed operations don't need backtracking */
531 if (act->is_relaxed())
533 /* linear search: from most recent to oldest */
534 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
535 action_list_t::reverse_iterator rit;
536 for (rit = list->rbegin(); rit != list->rend(); rit++) {
537 ModelAction *prev = *rit;
538 if (prev->could_synchronize_with(act))
544 case ATOMIC_TRYLOCK: {
545 /* linear search: from most recent to oldest */
546 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
547 action_list_t::reverse_iterator rit;
548 for (rit = list->rbegin(); rit != list->rend(); rit++) {
549 ModelAction *prev = *rit;
550 if (act->is_conflicting_lock(prev))
555 case ATOMIC_UNLOCK: {
556 /* linear search: from most recent to oldest */
557 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
558 action_list_t::reverse_iterator rit;
559 for (rit = list->rbegin(); rit != list->rend(); rit++) {
560 ModelAction *prev = *rit;
561 if (!act->same_thread(prev)&&prev->is_failed_trylock())
567 /* linear search: from most recent to oldest */
568 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
569 action_list_t::reverse_iterator rit;
570 for (rit = list->rbegin(); rit != list->rend(); rit++) {
571 ModelAction *prev = *rit;
572 if (!act->same_thread(prev)&&prev->is_failed_trylock())
574 if (!act->same_thread(prev)&&prev->is_notify())
580 case ATOMIC_NOTIFY_ALL:
581 case ATOMIC_NOTIFY_ONE: {
582 /* linear search: from most recent to oldest */
583 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
584 action_list_t::reverse_iterator rit;
585 for (rit = list->rbegin(); rit != list->rend(); rit++) {
586 ModelAction *prev = *rit;
587 if (!act->same_thread(prev)&&prev->is_wait())
598 /** This method finds backtracking points where we should try to
599 * reorder the parameter ModelAction against.
601 * @param the ModelAction to find backtracking points for.
603 void ModelChecker::set_backtracking(ModelAction *act)
605 Thread *t = get_thread(act);
606 ModelAction * prev = get_last_conflict(act);
610 Node * node = prev->get_node()->get_parent();
612 int low_tid, high_tid;
613 if (node->is_enabled(t)) {
614 low_tid = id_to_int(act->get_tid());
615 high_tid = low_tid+1;
618 high_tid = get_num_threads();
621 for(int i = low_tid; i < high_tid; i++) {
622 thread_id_t tid = int_to_id(i);
624 /* Make sure this thread can be enabled here. */
625 if (i >= node->get_num_threads())
628 /* Don't backtrack into a point where the thread is disabled or sleeping. */
629 if (node->enabled_status(tid)!=THREAD_ENABLED)
632 /* Check if this has been explored already */
633 if (node->has_been_explored(tid))
636 /* See if fairness allows */
637 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
639 for(int t=0;t<node->get_num_threads();t++) {
640 thread_id_t tother=int_to_id(t);
641 if (node->is_enabled(tother) && node->has_priority(tother)) {
649 /* Cache the latest backtracking point */
650 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
651 priv->next_backtrack = prev;
653 /* If this is a new backtracking point, mark the tree */
654 if (!node->set_backtrack(tid))
656 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
657 id_to_int(prev->get_tid()),
658 id_to_int(t->get_id()));
667 * Returns last backtracking point. The model checker will explore a different
668 * path for this point in the next execution.
669 * @return The ModelAction at which the next execution should diverge.
671 ModelAction * ModelChecker::get_next_backtrack()
673 ModelAction *next = priv->next_backtrack;
674 priv->next_backtrack = NULL;
679 * Processes a read or rmw model action.
680 * @param curr is the read model action to process.
681 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
682 * @return True if processing this read updates the mo_graph.
684 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
686 uint64_t value = VALUE_NONE;
687 bool updated = false;
689 const ModelAction *reads_from = curr->get_node()->get_read_from();
690 if (reads_from != NULL) {
691 mo_graph->startChanges();
693 value = reads_from->get_value();
694 bool r_status = false;
696 if (!second_part_of_rmw) {
697 check_recency(curr, reads_from);
698 r_status = r_modification_order(curr, reads_from);
702 if (!second_part_of_rmw&&is_infeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
703 mo_graph->rollbackChanges();
704 priv->too_many_reads = false;
708 read_from(curr, reads_from);
709 mo_graph->commitChanges();
710 mo_check_promises(curr->get_tid(), reads_from);
713 } else if (!second_part_of_rmw) {
714 /* Read from future value */
715 value = curr->get_node()->get_future_value();
716 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
717 read_from(curr, NULL);
718 Promise *valuepromise = new Promise(curr, value, expiration);
719 promises->push_back(valuepromise);
721 get_thread(curr)->set_return_value(value);
727 * Processes a lock, trylock, or unlock model action. @param curr is
728 * the read model action to process.
730 * The try lock operation checks whether the lock is taken. If not,
731 * it falls to the normal lock operation case. If so, it returns
734 * The lock operation has already been checked that it is enabled, so
735 * it just grabs the lock and synchronizes with the previous unlock.
737 * The unlock operation has to re-enable all of the threads that are
738 * waiting on the lock.
740 * @return True if synchronization was updated; false otherwise
742 bool ModelChecker::process_mutex(ModelAction *curr) {
743 std::mutex *mutex=NULL;
744 struct std::mutex_state *state=NULL;
746 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
747 mutex = (std::mutex *)curr->get_location();
748 state = mutex->get_state();
749 } else if(curr->is_wait()) {
750 mutex = (std::mutex *)curr->get_value();
751 state = mutex->get_state();
754 switch (curr->get_type()) {
755 case ATOMIC_TRYLOCK: {
756 bool success = !state->islocked;
757 curr->set_try_lock(success);
759 get_thread(curr)->set_return_value(0);
762 get_thread(curr)->set_return_value(1);
764 //otherwise fall into the lock case
766 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
767 assert_bug("Lock access before initialization");
768 state->islocked = true;
769 ModelAction *unlock = get_last_unlock(curr);
770 //synchronize with the previous unlock statement
771 if (unlock != NULL) {
772 curr->synchronize_with(unlock);
777 case ATOMIC_UNLOCK: {
779 state->islocked = false;
780 //wake up the other threads
781 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
782 //activate all the waiting threads
783 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
784 scheduler->wake(get_thread(*rit));
791 state->islocked = false;
792 //wake up the other threads
793 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
794 //activate all the waiting threads
795 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
796 scheduler->wake(get_thread(*rit));
799 //check whether we should go to sleep or not...simulate spurious failures
800 if (curr->get_node()->get_misc()==0) {
801 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
803 scheduler->sleep(get_current_thread());
807 case ATOMIC_NOTIFY_ALL: {
808 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
809 //activate all the waiting threads
810 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
811 scheduler->wake(get_thread(*rit));
816 case ATOMIC_NOTIFY_ONE: {
817 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
818 int wakeupthread=curr->get_node()->get_misc();
819 action_list_t::iterator it = waiters->begin();
820 advance(it, wakeupthread);
821 scheduler->wake(get_thread(*it));
833 * Process a write ModelAction
834 * @param curr The ModelAction to process
835 * @return True if the mo_graph was updated or promises were resolved
837 bool ModelChecker::process_write(ModelAction *curr)
839 bool updated_mod_order = w_modification_order(curr);
840 bool updated_promises = resolve_promises(curr);
842 if (promises->size() == 0) {
843 for (unsigned int i = 0; i < futurevalues->size(); i++) {
844 struct PendingFutureValue pfv = (*futurevalues)[i];
845 //Do more ambitious checks now that mo is more complete
846 if (mo_may_allow(pfv.writer, pfv.act)&&
847 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
848 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
849 priv->next_backtrack = pfv.act;
851 futurevalues->resize(0);
854 mo_graph->commitChanges();
855 mo_check_promises(curr->get_tid(), curr);
857 get_thread(curr)->set_return_value(VALUE_NONE);
858 return updated_mod_order || updated_promises;
862 * Process a fence ModelAction
863 * @param curr The ModelAction to process
864 * @return True if synchronization was updated
866 bool ModelChecker::process_fence(ModelAction *curr)
869 * fence-relaxed: no-op
870 * fence-release: only log the occurence (not in this function), for
871 * use in later synchronization
872 * fence-acquire (this function): search for hypothetical release
875 bool updated = false;
876 if (curr->is_acquire()) {
877 action_list_t *list = action_trace;
878 action_list_t::reverse_iterator rit;
879 /* Find X : is_read(X) && X --sb-> curr */
880 for (rit = list->rbegin(); rit != list->rend(); rit++) {
881 ModelAction *act = *rit;
882 if (act->get_tid() != curr->get_tid())
884 /* Stop at the beginning of the thread */
885 if (act->is_thread_start())
887 /* Stop once we reach a prior fence-acquire */
888 if (act->is_fence() && act->is_acquire())
892 /* read-acquire will find its own release sequences */
893 if (act->is_acquire())
896 /* Establish hypothetical release sequences */
897 rel_heads_list_t release_heads;
898 get_release_seq_heads(curr, act, &release_heads);
899 for (unsigned int i = 0; i < release_heads.size(); i++)
900 if (!act->synchronize_with(release_heads[i]))
901 set_bad_synchronization();
902 if (release_heads.size() != 0)
910 * @brief Process the current action for thread-related activity
912 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
913 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
914 * synchronization, etc. This function is a no-op for non-THREAD actions
915 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
917 * @param curr The current action
918 * @return True if synchronization was updated or a thread completed
920 bool ModelChecker::process_thread_action(ModelAction *curr)
922 bool updated = false;
924 switch (curr->get_type()) {
925 case THREAD_CREATE: {
926 Thread *th = (Thread *)curr->get_location();
927 th->set_creation(curr);
931 Thread *blocking = (Thread *)curr->get_location();
932 ModelAction *act = get_last_action(blocking->get_id());
933 curr->synchronize_with(act);
934 updated = true; /* trigger rel-seq checks */
937 case THREAD_FINISH: {
938 Thread *th = get_thread(curr);
939 while (!th->wait_list_empty()) {
940 ModelAction *act = th->pop_wait_list();
941 scheduler->wake(get_thread(act));
944 updated = true; /* trigger rel-seq checks */
948 check_promises(curr->get_tid(), NULL, curr->get_cv());
959 * @brief Process the current action for release sequence fixup activity
961 * Performs model-checker release sequence fixups for the current action,
962 * forcing a single pending release sequence to break (with a given, potential
963 * "loose" write) or to complete (i.e., synchronize). If a pending release
964 * sequence forms a complete release sequence, then we must perform the fixup
965 * synchronization, mo_graph additions, etc.
967 * @param curr The current action; must be a release sequence fixup action
968 * @param work_queue The work queue to which to add work items as they are
971 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
973 const ModelAction *write = curr->get_node()->get_relseq_break();
974 struct release_seq *sequence = pending_rel_seqs->back();
975 pending_rel_seqs->pop_back();
977 ModelAction *acquire = sequence->acquire;
978 const ModelAction *rf = sequence->rf;
979 const ModelAction *release = sequence->release;
983 ASSERT(release->same_thread(rf));
987 * @todo Forcing a synchronization requires that we set
988 * modification order constraints. For instance, we can't allow
989 * a fixup sequence in which two separate read-acquire
990 * operations read from the same sequence, where the first one
991 * synchronizes and the other doesn't. Essentially, we can't
992 * allow any writes to insert themselves between 'release' and
996 /* Must synchronize */
997 if (!acquire->synchronize_with(release)) {
998 set_bad_synchronization();
1001 /* Re-check all pending release sequences */
1002 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1003 /* Re-check act for mo_graph edges */
1004 work_queue->push_back(MOEdgeWorkEntry(acquire));
1006 /* propagate synchronization to later actions */
1007 action_list_t::reverse_iterator rit = action_trace->rbegin();
1008 for (; (*rit) != acquire; rit++) {
1009 ModelAction *propagate = *rit;
1010 if (acquire->happens_before(propagate)) {
1011 propagate->synchronize_with(acquire);
1012 /* Re-check 'propagate' for mo_graph edges */
1013 work_queue->push_back(MOEdgeWorkEntry(propagate));
1017 /* Break release sequence with new edges:
1018 * release --mo--> write --mo--> rf */
1019 mo_graph->addEdge(release, write);
1020 mo_graph->addEdge(write, rf);
1023 /* See if we have realized a data race */
1028 * Initialize the current action by performing one or more of the following
1029 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1030 * in the NodeStack, manipulating backtracking sets, allocating and
1031 * initializing clock vectors, and computing the promises to fulfill.
1033 * @param curr The current action, as passed from the user context; may be
1034 * freed/invalidated after the execution of this function, with a different
1035 * action "returned" its place (pass-by-reference)
1036 * @return True if curr is a newly-explored action; false otherwise
1038 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1040 ModelAction *newcurr;
1042 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1043 newcurr = process_rmw(*curr);
1046 if (newcurr->is_rmw())
1047 compute_promises(newcurr);
1053 (*curr)->set_seq_number(get_next_seq_num());
1055 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1057 /* First restore type and order in case of RMW operation */
1058 if ((*curr)->is_rmwr())
1059 newcurr->copy_typeandorder(*curr);
1061 ASSERT((*curr)->get_location() == newcurr->get_location());
1062 newcurr->copy_from_new(*curr);
1064 /* Discard duplicate ModelAction; use action from NodeStack */
1067 /* Always compute new clock vector */
1068 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1071 return false; /* Action was explored previously */
1075 /* Always compute new clock vector */
1076 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1078 /* Assign most recent release fence */
1079 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1082 * Perform one-time actions when pushing new ModelAction onto
1085 if (newcurr->is_write())
1086 compute_promises(newcurr);
1087 else if (newcurr->is_relseq_fixup())
1088 compute_relseq_breakwrites(newcurr);
1089 else if (newcurr->is_wait())
1090 newcurr->get_node()->set_misc_max(2);
1091 else if (newcurr->is_notify_one()) {
1092 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1094 return true; /* This was a new ModelAction */
1099 * @brief Establish reads-from relation between two actions
1101 * Perform basic operations involved with establishing a concrete rf relation,
1102 * including setting the ModelAction data and checking for release sequences.
1104 * @param act The action that is reading (must be a read)
1105 * @param rf The action from which we are reading (must be a write)
1107 * @return True if this read established synchronization
1109 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1111 act->set_read_from(rf);
1112 if (rf != NULL && act->is_acquire()) {
1113 rel_heads_list_t release_heads;
1114 get_release_seq_heads(act, act, &release_heads);
1115 int num_heads = release_heads.size();
1116 for (unsigned int i = 0; i < release_heads.size(); i++)
1117 if (!act->synchronize_with(release_heads[i])) {
1118 set_bad_synchronization();
1121 return num_heads > 0;
1127 * @brief Check whether a model action is enabled.
1129 * Checks whether a lock or join operation would be successful (i.e., is the
1130 * lock already locked, or is the joined thread already complete). If not, put
1131 * the action in a waiter list.
1133 * @param curr is the ModelAction to check whether it is enabled.
1134 * @return a bool that indicates whether the action is enabled.
1136 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1137 if (curr->is_lock()) {
1138 std::mutex * lock = (std::mutex *)curr->get_location();
1139 struct std::mutex_state * state = lock->get_state();
1140 if (state->islocked) {
1141 //Stick the action in the appropriate waiting queue
1142 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
1145 } else if (curr->get_type() == THREAD_JOIN) {
1146 Thread *blocking = (Thread *)curr->get_location();
1147 if (!blocking->is_complete()) {
1148 blocking->push_wait_list(curr);
1157 * Stores the ModelAction for the current thread action. Call this
1158 * immediately before switching from user- to system-context to pass
1159 * data between them.
1160 * @param act The ModelAction created by the user-thread action
1162 void ModelChecker::set_current_action(ModelAction *act) {
1163 priv->current_action = act;
1167 * This is the heart of the model checker routine. It performs model-checking
1168 * actions corresponding to a given "current action." Among other processes, it
1169 * calculates reads-from relationships, updates synchronization clock vectors,
1170 * forms a memory_order constraints graph, and handles replay/backtrack
1171 * execution when running permutations of previously-observed executions.
1173 * @param curr The current action to process
1174 * @return The next Thread that must be executed. May be NULL if ModelChecker
1175 * makes no choice (e.g., according to replay execution, combining RMW actions,
1178 Thread * ModelChecker::check_current_action(ModelAction *curr)
1181 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1183 if (!check_action_enabled(curr)) {
1184 /* Make the execution look like we chose to run this action
1185 * much later, when a lock/join can succeed */
1186 get_current_thread()->set_pending(curr);
1187 scheduler->sleep(get_current_thread());
1188 return get_next_thread(NULL);
1191 bool newly_explored = initialize_curr_action(&curr);
1193 wake_up_sleeping_actions(curr);
1195 /* Add the action to lists before any other model-checking tasks */
1196 if (!second_part_of_rmw)
1197 add_action_to_lists(curr);
1199 /* Build may_read_from set for newly-created actions */
1200 if (newly_explored && curr->is_read())
1201 build_reads_from_past(curr);
1203 /* Initialize work_queue with the "current action" work */
1204 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1205 while (!work_queue.empty() && !has_asserted()) {
1206 WorkQueueEntry work = work_queue.front();
1207 work_queue.pop_front();
1209 switch (work.type) {
1210 case WORK_CHECK_CURR_ACTION: {
1211 ModelAction *act = work.action;
1212 bool update = false; /* update this location's release seq's */
1213 bool update_all = false; /* update all release seq's */
1215 if (process_thread_action(curr))
1218 if (act->is_read() && process_read(act, second_part_of_rmw))
1221 if (act->is_write() && process_write(act))
1224 if (act->is_fence() && process_fence(act))
1227 if (act->is_mutex_op() && process_mutex(act))
1230 if (act->is_relseq_fixup())
1231 process_relseq_fixup(curr, &work_queue);
1234 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1236 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1239 case WORK_CHECK_RELEASE_SEQ:
1240 resolve_release_sequences(work.location, &work_queue);
1242 case WORK_CHECK_MO_EDGES: {
1243 /** @todo Complete verification of work_queue */
1244 ModelAction *act = work.action;
1245 bool updated = false;
1247 if (act->is_read()) {
1248 const ModelAction *rf = act->get_reads_from();
1249 if (rf != NULL && r_modification_order(act, rf))
1252 if (act->is_write()) {
1253 if (w_modification_order(act))
1256 mo_graph->commitChanges();
1259 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1268 check_curr_backtracking(curr);
1269 set_backtracking(curr);
1270 return get_next_thread(curr);
1273 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
1274 Node *currnode = curr->get_node();
1275 Node *parnode = currnode->get_parent();
1277 if ((!parnode->backtrack_empty() ||
1278 !currnode->misc_empty() ||
1279 !currnode->read_from_empty() ||
1280 !currnode->future_value_empty() ||
1281 !currnode->promise_empty() ||
1282 !currnode->relseq_break_empty())
1283 && (!priv->next_backtrack ||
1284 *curr > *priv->next_backtrack)) {
1285 priv->next_backtrack = curr;
1289 bool ModelChecker::promises_expired() const
1291 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
1292 Promise *promise = (*promises)[promise_index];
1293 if (promise->get_expiration()<priv->used_sequence_numbers) {
1301 * This is the strongest feasibility check available.
1302 * @return whether the current trace (partial or complete) must be a prefix of
1305 bool ModelChecker::isfeasibleprefix() const
1307 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1311 * Returns whether the current completed trace is feasible, except for pending
1312 * release sequences.
1314 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1316 if (DBG_ENABLED() && promises->size() != 0)
1317 DEBUG("Infeasible: unrevolved promises\n");
1319 return !is_infeasible() && promises->size() == 0;
1323 * Check if the current partial trace is infeasible. Does not check any
1324 * end-of-execution flags, which might rule out the execution. Thus, this is
1325 * useful only for ruling an execution as infeasible.
1326 * @return whether the current partial trace is infeasible.
1328 bool ModelChecker::is_infeasible() const
1330 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
1331 DEBUG("Infeasible: RMW violation\n");
1333 return mo_graph->checkForRMWViolation() || is_infeasible_ignoreRMW();
1337 * Check If the current partial trace is infeasible, while ignoring
1338 * infeasibility related to 2 RMW's reading from the same store. It does not
1339 * check end-of-execution feasibility.
1340 * @see ModelChecker::is_infeasible
1341 * @return whether the current partial trace is infeasible, ignoring multiple
1342 * RMWs reading from the same store.
1344 bool ModelChecker::is_infeasible_ignoreRMW() const
1346 if (DBG_ENABLED()) {
1347 if (mo_graph->checkForCycles())
1348 DEBUG("Infeasible: modification order cycles\n");
1349 if (priv->failed_promise)
1350 DEBUG("Infeasible: failed promise\n");
1351 if (priv->too_many_reads)
1352 DEBUG("Infeasible: too many reads\n");
1353 if (priv->bad_synchronization)
1354 DEBUG("Infeasible: bad synchronization ordering\n");
1355 if (promises_expired())
1356 DEBUG("Infeasible: promises expired\n");
1358 return mo_graph->checkForCycles() || priv->failed_promise ||
1359 priv->too_many_reads || priv->bad_synchronization ||
1363 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1364 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1365 ModelAction *lastread = get_last_action(act->get_tid());
1366 lastread->process_rmw(act);
1367 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1368 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1369 mo_graph->commitChanges();
1375 * Checks whether a thread has read from the same write for too many times
1376 * without seeing the effects of a later write.
1379 * 1) there must a different write that we could read from that would satisfy the modification order,
1380 * 2) we must have read from the same value in excess of maxreads times, and
1381 * 3) that other write must have been in the reads_from set for maxreads times.
1383 * If so, we decide that the execution is no longer feasible.
1385 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1386 if (params.maxreads != 0) {
1388 if (curr->get_node()->get_read_from_size() <= 1)
1390 //Must make sure that execution is currently feasible... We could
1391 //accidentally clear by rolling back
1392 if (is_infeasible())
1394 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1395 int tid = id_to_int(curr->get_tid());
1398 if ((int)thrd_lists->size() <= tid)
1400 action_list_t *list = &(*thrd_lists)[tid];
1402 action_list_t::reverse_iterator rit = list->rbegin();
1403 /* Skip past curr */
1404 for (; (*rit) != curr; rit++)
1406 /* go past curr now */
1409 action_list_t::reverse_iterator ritcopy = rit;
1410 //See if we have enough reads from the same value
1412 for (; count < params.maxreads; rit++,count++) {
1413 if (rit==list->rend())
1415 ModelAction *act = *rit;
1416 if (!act->is_read())
1419 if (act->get_reads_from() != rf)
1421 if (act->get_node()->get_read_from_size() <= 1)
1424 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1426 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1428 //Need a different write
1432 /* Test to see whether this is a feasible write to read from*/
1433 mo_graph->startChanges();
1434 r_modification_order(curr, write);
1435 bool feasiblereadfrom = !is_infeasible();
1436 mo_graph->rollbackChanges();
1438 if (!feasiblereadfrom)
1442 bool feasiblewrite = true;
1443 //new we need to see if this write works for everyone
1445 for (int loop = count; loop>0; loop--,rit++) {
1446 ModelAction *act=*rit;
1447 bool foundvalue = false;
1448 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1449 if (act->get_node()->get_read_from_at(j)==write) {
1455 feasiblewrite = false;
1459 if (feasiblewrite) {
1460 priv->too_many_reads = true;
1468 * Updates the mo_graph with the constraints imposed from the current
1471 * Basic idea is the following: Go through each other thread and find
1472 * the lastest action that happened before our read. Two cases:
1474 * (1) The action is a write => that write must either occur before
1475 * the write we read from or be the write we read from.
1477 * (2) The action is a read => the write that that action read from
1478 * must occur before the write we read from or be the same write.
1480 * @param curr The current action. Must be a read.
1481 * @param rf The action that curr reads from. Must be a write.
1482 * @return True if modification order edges were added; false otherwise
1484 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1486 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1489 ASSERT(curr->is_read());
1491 /* Last SC fence in the current thread */
1492 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1494 /* Iterate over all threads */
1495 for (i = 0; i < thrd_lists->size(); i++) {
1496 /* Last SC fence in thread i */
1497 ModelAction *last_sc_fence_thread_local = NULL;
1498 if (int_to_id((int)i) != curr->get_tid())
1499 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1501 /* Last SC fence in thread i, before last SC fence in current thread */
1502 ModelAction *last_sc_fence_thread_before = NULL;
1503 if (last_sc_fence_local)
1504 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1506 /* Iterate over actions in thread, starting from most recent */
1507 action_list_t *list = &(*thrd_lists)[i];
1508 action_list_t::reverse_iterator rit;
1509 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1510 ModelAction *act = *rit;
1512 if (act->is_write() && act != rf && act != curr) {
1513 /* C++, Section 29.3 statement 5 */
1514 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1515 *act < *last_sc_fence_thread_local) {
1516 mo_graph->addEdge(act, rf);
1519 /* C++, Section 29.3 statement 4 */
1520 else if (act->is_seqcst() && last_sc_fence_local &&
1521 *act < *last_sc_fence_local) {
1522 mo_graph->addEdge(act, rf);
1525 /* C++, Section 29.3 statement 6 */
1526 else if (last_sc_fence_thread_before &&
1527 *act < *last_sc_fence_thread_before) {
1528 mo_graph->addEdge(act, rf);
1534 * Include at most one act per-thread that "happens
1535 * before" curr. Don't consider reflexively.
1537 if (act->happens_before(curr) && act != curr) {
1538 if (act->is_write()) {
1540 mo_graph->addEdge(act, rf);
1544 const ModelAction *prevreadfrom = act->get_reads_from();
1545 //if the previous read is unresolved, keep going...
1546 if (prevreadfrom == NULL)
1549 if (rf != prevreadfrom) {
1550 mo_graph->addEdge(prevreadfrom, rf);
1562 /** This method fixes up the modification order when we resolve a
1563 * promises. The basic problem is that actions that occur after the
1564 * read curr could not property add items to the modification order
1567 * So for each thread, we find the earliest item that happens after
1568 * the read curr. This is the item we have to fix up with additional
1569 * constraints. If that action is write, we add a MO edge between
1570 * the Action rf and that action. If the action is a read, we add a
1571 * MO edge between the Action rf, and whatever the read accessed.
1573 * @param curr is the read ModelAction that we are fixing up MO edges for.
1574 * @param rf is the write ModelAction that curr reads from.
1577 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1579 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1581 ASSERT(curr->is_read());
1583 /* Iterate over all threads */
1584 for (i = 0; i < thrd_lists->size(); i++) {
1585 /* Iterate over actions in thread, starting from most recent */
1586 action_list_t *list = &(*thrd_lists)[i];
1587 action_list_t::reverse_iterator rit;
1588 ModelAction *lastact = NULL;
1590 /* Find last action that happens after curr that is either not curr or a rmw */
1591 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1592 ModelAction *act = *rit;
1593 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1599 /* Include at most one act per-thread that "happens before" curr */
1600 if (lastact != NULL) {
1601 if (lastact==curr) {
1602 //Case 1: The resolved read is a RMW, and we need to make sure
1603 //that the write portion of the RMW mod order after rf
1605 mo_graph->addEdge(rf, lastact);
1606 } else if (lastact->is_read()) {
1607 //Case 2: The resolved read is a normal read and the next
1608 //operation is a read, and we need to make sure the value read
1609 //is mod ordered after rf
1611 const ModelAction *postreadfrom = lastact->get_reads_from();
1612 if (postreadfrom != NULL&&rf != postreadfrom)
1613 mo_graph->addEdge(rf, postreadfrom);
1615 //Case 3: The resolved read is a normal read and the next
1616 //operation is a write, and we need to make sure that the
1617 //write is mod ordered after rf
1619 mo_graph->addEdge(rf, lastact);
1627 * Updates the mo_graph with the constraints imposed from the current write.
1629 * Basic idea is the following: Go through each other thread and find
1630 * the lastest action that happened before our write. Two cases:
1632 * (1) The action is a write => that write must occur before
1635 * (2) The action is a read => the write that that action read from
1636 * must occur before the current write.
1638 * This method also handles two other issues:
1640 * (I) Sequential Consistency: Making sure that if the current write is
1641 * seq_cst, that it occurs after the previous seq_cst write.
1643 * (II) Sending the write back to non-synchronizing reads.
1645 * @param curr The current action. Must be a write.
1646 * @return True if modification order edges were added; false otherwise
1648 bool ModelChecker::w_modification_order(ModelAction *curr)
1650 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1653 ASSERT(curr->is_write());
1655 if (curr->is_seqcst()) {
1656 /* We have to at least see the last sequentially consistent write,
1657 so we are initialized. */
1658 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1659 if (last_seq_cst != NULL) {
1660 mo_graph->addEdge(last_seq_cst, curr);
1665 /* Last SC fence in the current thread */
1666 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1668 /* Iterate over all threads */
1669 for (i = 0; i < thrd_lists->size(); i++) {
1670 /* Last SC fence in thread i, before last SC fence in current thread */
1671 ModelAction *last_sc_fence_thread_before = NULL;
1672 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1673 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1675 /* Iterate over actions in thread, starting from most recent */
1676 action_list_t *list = &(*thrd_lists)[i];
1677 action_list_t::reverse_iterator rit;
1678 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1679 ModelAction *act = *rit;
1682 * 1) If RMW and it actually read from something, then we
1683 * already have all relevant edges, so just skip to next
1686 * 2) If RMW and it didn't read from anything, we should
1687 * whatever edge we can get to speed up convergence.
1689 * 3) If normal write, we need to look at earlier actions, so
1690 * continue processing list.
1692 if (curr->is_rmw()) {
1693 if (curr->get_reads_from()!=NULL)
1701 /* C++, Section 29.3 statement 7 */
1702 if (last_sc_fence_thread_before && act->is_write() &&
1703 *act < *last_sc_fence_thread_before) {
1704 mo_graph->addEdge(act, curr);
1709 * Include at most one act per-thread that "happens
1712 if (act->happens_before(curr)) {
1714 * Note: if act is RMW, just add edge:
1716 * The following edge should be handled elsewhere:
1717 * readfrom(act) --mo--> act
1719 if (act->is_write())
1720 mo_graph->addEdge(act, curr);
1721 else if (act->is_read()) {
1722 //if previous read accessed a null, just keep going
1723 if (act->get_reads_from() == NULL)
1725 mo_graph->addEdge(act->get_reads_from(), curr);
1729 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1730 !act->same_thread(curr)) {
1731 /* We have an action that:
1732 (1) did not happen before us
1733 (2) is a read and we are a write
1734 (3) cannot synchronize with us
1735 (4) is in a different thread
1737 that read could potentially read from our write. Note that
1738 these checks are overly conservative at this point, we'll
1739 do more checks before actually removing the
1743 if (thin_air_constraint_may_allow(curr, act)) {
1744 if (!is_infeasible() ||
1745 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && !is_infeasible_ignoreRMW())) {
1746 struct PendingFutureValue pfv = {curr,act};
1747 futurevalues->push_back(pfv);
1757 /** Arbitrary reads from the future are not allowed. Section 29.3
1758 * part 9 places some constraints. This method checks one result of constraint
1759 * constraint. Others require compiler support. */
1760 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1761 if (!writer->is_rmw())
1764 if (!reader->is_rmw())
1767 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1768 if (search == reader)
1770 if (search->get_tid() == reader->get_tid() &&
1771 search->happens_before(reader))
1779 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1780 * some constraints. This method checks one the following constraint (others
1781 * require compiler support):
1783 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1785 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1787 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1789 /* Iterate over all threads */
1790 for (i = 0; i < thrd_lists->size(); i++) {
1791 const ModelAction *write_after_read = NULL;
1793 /* Iterate over actions in thread, starting from most recent */
1794 action_list_t *list = &(*thrd_lists)[i];
1795 action_list_t::reverse_iterator rit;
1796 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1797 ModelAction *act = *rit;
1799 if (!reader->happens_before(act))
1801 else if (act->is_write())
1802 write_after_read = act;
1803 else if (act->is_read() && act->get_reads_from() != NULL && act != reader) {
1804 write_after_read = act->get_reads_from();
1808 if (write_after_read && write_after_read!=writer && mo_graph->checkReachable(write_after_read, writer))
1815 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1816 * The ModelAction under consideration is expected to be taking part in
1817 * release/acquire synchronization as an object of the "reads from" relation.
1818 * Note that this can only provide release sequence support for RMW chains
1819 * which do not read from the future, as those actions cannot be traced until
1820 * their "promise" is fulfilled. Similarly, we may not even establish the
1821 * presence of a release sequence with certainty, as some modification order
1822 * constraints may be decided further in the future. Thus, this function
1823 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1824 * and a boolean representing certainty.
1826 * @param rf The action that might be part of a release sequence. Must be a
1828 * @param release_heads A pass-by-reference style return parameter. After
1829 * execution of this function, release_heads will contain the heads of all the
1830 * relevant release sequences, if any exists with certainty
1831 * @param pending A pass-by-reference style return parameter which is only used
1832 * when returning false (i.e., uncertain). Returns most information regarding
1833 * an uncertain release sequence, including any write operations that might
1834 * break the sequence.
1835 * @return true, if the ModelChecker is certain that release_heads is complete;
1838 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1839 rel_heads_list_t *release_heads,
1840 struct release_seq *pending) const
1842 /* Only check for release sequences if there are no cycles */
1843 if (mo_graph->checkForCycles())
1847 ASSERT(rf->is_write());
1849 if (rf->is_release())
1850 release_heads->push_back(rf);
1851 else if (rf->get_last_fence_release())
1852 release_heads->push_back(rf->get_last_fence_release());
1854 break; /* End of RMW chain */
1856 /** @todo Need to be smarter here... In the linux lock
1857 * example, this will run to the beginning of the program for
1859 /** @todo The way to be smarter here is to keep going until 1
1860 * thread has a release preceded by an acquire and you've seen
1863 /* acq_rel RMW is a sufficient stopping condition */
1864 if (rf->is_acquire() && rf->is_release())
1865 return true; /* complete */
1867 rf = rf->get_reads_from();
1870 /* read from future: need to settle this later */
1872 return false; /* incomplete */
1875 if (rf->is_release())
1876 return true; /* complete */
1878 /* else relaxed write
1879 * - check for fence-release in the same thread (29.8, stmt. 3)
1880 * - check modification order for contiguous subsequence
1881 * -> rf must be same thread as release */
1883 const ModelAction *fence_release = rf->get_last_fence_release();
1884 /* Synchronize with a fence-release unconditionally; we don't need to
1885 * find any more "contiguous subsequence..." for it */
1887 release_heads->push_back(fence_release);
1889 int tid = id_to_int(rf->get_tid());
1890 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1891 action_list_t *list = &(*thrd_lists)[tid];
1892 action_list_t::const_reverse_iterator rit;
1894 /* Find rf in the thread list */
1895 rit = std::find(list->rbegin(), list->rend(), rf);
1896 ASSERT(rit != list->rend());
1898 /* Find the last {write,fence}-release */
1899 for (; rit != list->rend(); rit++) {
1900 if (fence_release && *(*rit) < *fence_release)
1902 if ((*rit)->is_release())
1905 if (rit == list->rend()) {
1906 /* No write-release in this thread */
1907 return true; /* complete */
1908 } else if (fence_release && *(*rit) < *fence_release) {
1909 /* The fence-release is more recent (and so, "stronger") than
1910 * the most recent write-release */
1911 return true; /* complete */
1912 } /* else, need to establish contiguous release sequence */
1913 ModelAction *release = *rit;
1915 ASSERT(rf->same_thread(release));
1917 pending->writes.clear();
1919 bool certain = true;
1920 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1921 if (id_to_int(rf->get_tid()) == (int)i)
1923 list = &(*thrd_lists)[i];
1925 /* Can we ensure no future writes from this thread may break
1926 * the release seq? */
1927 bool future_ordered = false;
1929 ModelAction *last = get_last_action(int_to_id(i));
1930 Thread *th = get_thread(int_to_id(i));
1931 if ((last && rf->happens_before(last)) ||
1934 future_ordered = true;
1936 ASSERT(!th->is_model_thread() || future_ordered);
1938 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1939 const ModelAction *act = *rit;
1940 /* Reach synchronization -> this thread is complete */
1941 if (act->happens_before(release))
1943 if (rf->happens_before(act)) {
1944 future_ordered = true;
1948 /* Only non-RMW writes can break release sequences */
1949 if (!act->is_write() || act->is_rmw())
1952 /* Check modification order */
1953 if (mo_graph->checkReachable(rf, act)) {
1954 /* rf --mo--> act */
1955 future_ordered = true;
1958 if (mo_graph->checkReachable(act, release))
1959 /* act --mo--> release */
1961 if (mo_graph->checkReachable(release, act) &&
1962 mo_graph->checkReachable(act, rf)) {
1963 /* release --mo-> act --mo--> rf */
1964 return true; /* complete */
1966 /* act may break release sequence */
1967 pending->writes.push_back(act);
1970 if (!future_ordered)
1971 certain = false; /* This thread is uncertain */
1975 release_heads->push_back(release);
1976 pending->writes.clear();
1978 pending->release = release;
1985 * An interface for getting the release sequence head(s) with which a
1986 * given ModelAction must synchronize. This function only returns a non-empty
1987 * result when it can locate a release sequence head with certainty. Otherwise,
1988 * it may mark the internal state of the ModelChecker so that it will handle
1989 * the release sequence at a later time, causing @a acquire to update its
1990 * synchronization at some later point in execution.
1992 * @param acquire The 'acquire' action that may synchronize with a release
1994 * @param read The read action that may read from a release sequence; this may
1995 * be the same as acquire, or else an earlier action in the same thread (i.e.,
1996 * when 'acquire' is a fence-acquire)
1997 * @param release_heads A pass-by-reference return parameter. Will be filled
1998 * with the head(s) of the release sequence(s), if they exists with certainty.
1999 * @see ModelChecker::release_seq_heads
2001 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2002 ModelAction *read, rel_heads_list_t *release_heads)
2004 const ModelAction *rf = read->get_reads_from();
2005 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2006 sequence->acquire = acquire;
2007 sequence->read = read;
2009 if (!release_seq_heads(rf, release_heads, sequence)) {
2010 /* add act to 'lazy checking' list */
2011 pending_rel_seqs->push_back(sequence);
2013 snapshot_free(sequence);
2018 * Attempt to resolve all stashed operations that might synchronize with a
2019 * release sequence for a given location. This implements the "lazy" portion of
2020 * determining whether or not a release sequence was contiguous, since not all
2021 * modification order information is present at the time an action occurs.
2023 * @param location The location/object that should be checked for release
2024 * sequence resolutions. A NULL value means to check all locations.
2025 * @param work_queue The work queue to which to add work items as they are
2027 * @return True if any updates occurred (new synchronization, new mo_graph
2030 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2032 bool updated = false;
2033 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
2034 while (it != pending_rel_seqs->end()) {
2035 struct release_seq *pending = *it;
2036 ModelAction *acquire = pending->acquire;
2037 const ModelAction *read = pending->read;
2039 /* Only resolve sequences on the given location, if provided */
2040 if (location && read->get_location() != location) {
2045 const ModelAction *rf = read->get_reads_from();
2046 rel_heads_list_t release_heads;
2048 complete = release_seq_heads(rf, &release_heads, pending);
2049 for (unsigned int i = 0; i < release_heads.size(); i++) {
2050 if (!acquire->has_synchronized_with(release_heads[i])) {
2051 if (acquire->synchronize_with(release_heads[i]))
2054 set_bad_synchronization();
2059 /* Re-check all pending release sequences */
2060 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2061 /* Re-check read-acquire for mo_graph edges */
2062 if (acquire->is_read())
2063 work_queue->push_back(MOEdgeWorkEntry(acquire));
2065 /* propagate synchronization to later actions */
2066 action_list_t::reverse_iterator rit = action_trace->rbegin();
2067 for (; (*rit) != acquire; rit++) {
2068 ModelAction *propagate = *rit;
2069 if (acquire->happens_before(propagate)) {
2070 propagate->synchronize_with(acquire);
2071 /* Re-check 'propagate' for mo_graph edges */
2072 work_queue->push_back(MOEdgeWorkEntry(propagate));
2077 it = pending_rel_seqs->erase(it);
2078 snapshot_free(pending);
2084 // If we resolved promises or data races, see if we have realized a data race.
2091 * Performs various bookkeeping operations for the current ModelAction. For
2092 * instance, adds action to the per-object, per-thread action vector and to the
2093 * action trace list of all thread actions.
2095 * @param act is the ModelAction to add.
2097 void ModelChecker::add_action_to_lists(ModelAction *act)
2099 int tid = id_to_int(act->get_tid());
2100 action_trace->push_back(act);
2102 get_safe_ptr_action(obj_map, act->get_location())->push_back(act);
2104 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2105 if (tid >= (int)vec->size())
2106 vec->resize(priv->next_thread_id);
2107 (*vec)[tid].push_back(act);
2109 if ((int)thrd_last_action->size() <= tid)
2110 thrd_last_action->resize(get_num_threads());
2111 (*thrd_last_action)[tid] = act;
2113 if (act->is_fence() && act->is_release()) {
2114 if ((int)thrd_last_fence_release->size() <= tid)
2115 thrd_last_fence_release->resize(get_num_threads());
2116 (*thrd_last_fence_release)[tid] = act;
2119 if (act->is_wait()) {
2120 void *mutex_loc=(void *) act->get_value();
2121 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2123 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2124 if (tid >= (int)vec->size())
2125 vec->resize(priv->next_thread_id);
2126 (*vec)[tid].push_back(act);
2131 * @brief Get the last action performed by a particular Thread
2132 * @param tid The thread ID of the Thread in question
2133 * @return The last action in the thread
2135 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2137 int threadid = id_to_int(tid);
2138 if (threadid < (int)thrd_last_action->size())
2139 return (*thrd_last_action)[id_to_int(tid)];
2145 * @brief Get the last fence release performed by a particular Thread
2146 * @param tid The thread ID of the Thread in question
2147 * @return The last fence release in the thread, if one exists; NULL otherwise
2149 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2151 int threadid = id_to_int(tid);
2152 if (threadid < (int)thrd_last_fence_release->size())
2153 return (*thrd_last_fence_release)[id_to_int(tid)];
2159 * Gets the last memory_order_seq_cst write (in the total global sequence)
2160 * performed on a particular object (i.e., memory location), not including the
2162 * @param curr The current ModelAction; also denotes the object location to
2164 * @return The last seq_cst write
2166 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2168 void *location = curr->get_location();
2169 action_list_t *list = get_safe_ptr_action(obj_map, location);
2170 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2171 action_list_t::reverse_iterator rit;
2172 for (rit = list->rbegin(); rit != list->rend(); rit++)
2173 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
2179 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2180 * performed in a particular thread, prior to a particular fence.
2181 * @param tid The ID of the thread to check
2182 * @param before_fence The fence from which to begin the search; if NULL, then
2183 * search for the most recent fence in the thread.
2184 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2186 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2188 /* All fences should have NULL location */
2189 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2190 action_list_t::reverse_iterator rit = list->rbegin();
2193 for (; rit != list->rend(); rit++)
2194 if (*rit == before_fence)
2197 ASSERT(*rit == before_fence);
2201 for (; rit != list->rend(); rit++)
2202 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2208 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2209 * location). This function identifies the mutex according to the current
2210 * action, which is presumed to perform on the same mutex.
2211 * @param curr The current ModelAction; also denotes the object location to
2213 * @return The last unlock operation
2215 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2217 void *location = curr->get_location();
2218 action_list_t *list = get_safe_ptr_action(obj_map, location);
2219 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2220 action_list_t::reverse_iterator rit;
2221 for (rit = list->rbegin(); rit != list->rend(); rit++)
2222 if ((*rit)->is_unlock() || (*rit)->is_wait())
2227 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2229 ModelAction *parent = get_last_action(tid);
2231 parent = get_thread(tid)->get_creation();
2236 * Returns the clock vector for a given thread.
2237 * @param tid The thread whose clock vector we want
2238 * @return Desired clock vector
2240 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2242 return get_parent_action(tid)->get_cv();
2246 * Resolve a set of Promises with a current write. The set is provided in the
2247 * Node corresponding to @a write.
2248 * @param write The ModelAction that is fulfilling Promises
2249 * @return True if promises were resolved; false otherwise
2251 bool ModelChecker::resolve_promises(ModelAction *write)
2253 bool resolved = false;
2254 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
2256 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
2257 Promise *promise = (*promises)[promise_index];
2258 if (write->get_node()->get_promise(i)) {
2259 ModelAction *read = promise->get_action();
2260 if (read->is_rmw()) {
2261 mo_graph->addRMWEdge(write, read);
2263 read_from(read, write);
2264 //First fix up the modification order for actions that happened
2266 r_modification_order(read, write);
2267 //Next fix up the modification order for actions that happened
2269 post_r_modification_order(read, write);
2270 //Make sure the promise's value matches the write's value
2271 ASSERT(promise->get_value() == write->get_value());
2274 promises->erase(promises->begin() + promise_index);
2275 threads_to_check.push_back(read->get_tid());
2282 //Check whether reading these writes has made threads unable to
2285 for(unsigned int i=0;i<threads_to_check.size();i++)
2286 mo_check_promises(threads_to_check[i], write);
2292 * Compute the set of promises that could potentially be satisfied by this
2293 * action. Note that the set computation actually appears in the Node, not in
2295 * @param curr The ModelAction that may satisfy promises
2297 void ModelChecker::compute_promises(ModelAction *curr)
2299 for (unsigned int i = 0; i < promises->size(); i++) {
2300 Promise *promise = (*promises)[i];
2301 const ModelAction *act = promise->get_action();
2302 if (!act->happens_before(curr) &&
2304 !act->could_synchronize_with(curr) &&
2305 !act->same_thread(curr) &&
2306 act->get_location() == curr->get_location() &&
2307 promise->get_value() == curr->get_value()) {
2308 curr->get_node()->set_promise(i, act->is_rmw());
2313 /** Checks promises in response to change in ClockVector Threads. */
2314 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2316 for (unsigned int i = 0; i < promises->size(); i++) {
2317 Promise *promise = (*promises)[i];
2318 const ModelAction *act = promise->get_action();
2319 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
2320 merge_cv->synchronized_since(act)) {
2321 if (promise->increment_threads(tid)) {
2322 //Promise has failed
2323 priv->failed_promise = true;
2330 void ModelChecker::check_promises_thread_disabled() {
2331 for (unsigned int i = 0; i < promises->size(); i++) {
2332 Promise *promise = (*promises)[i];
2333 if (promise->check_promise()) {
2334 priv->failed_promise = true;
2340 /** Checks promises in response to addition to modification order for threads.
2342 * pthread is the thread that performed the read that created the promise
2344 * pread is the read that created the promise
2346 * pwrite is either the first write to same location as pread by
2347 * pthread that is sequenced after pread or the value read by the
2348 * first read to the same lcoation as pread by pthread that is
2349 * sequenced after pread..
2351 * 1. If tid=pthread, then we check what other threads are reachable
2352 * through the mode order starting with pwrite. Those threads cannot
2353 * perform a write that will resolve the promise due to modification
2354 * order constraints.
2356 * 2. If the tid is not pthread, we check whether pwrite can reach the
2357 * action write through the modification order. If so, that thread
2358 * cannot perform a future write that will resolve the promise due to
2359 * modificatin order constraints.
2361 * @parem tid The thread that either read from the model action
2362 * write, or actually did the model action write.
2364 * @parem write The ModelAction representing the relevant write.
2367 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
2368 void * location = write->get_location();
2369 for (unsigned int i = 0; i < promises->size(); i++) {
2370 Promise *promise = (*promises)[i];
2371 const ModelAction *act = promise->get_action();
2373 //Is this promise on the same location?
2374 if ( act->get_location() != location )
2377 //same thread as the promise
2378 if ( act->get_tid()==tid ) {
2380 //do we have a pwrite for the promise, if not, set it
2381 if (promise->get_write() == NULL ) {
2382 promise->set_write(write);
2383 //The pwrite cannot happen before the promise
2384 if (write->happens_before(act) && (write != act)) {
2385 priv->failed_promise = true;
2389 if (mo_graph->checkPromise(write, promise)) {
2390 priv->failed_promise = true;
2395 //Don't do any lookups twice for the same thread
2396 if (promise->has_sync_thread(tid))
2399 if (promise->get_write()&&mo_graph->checkReachable(promise->get_write(), write)) {
2400 if (promise->increment_threads(tid)) {
2401 priv->failed_promise = true;
2409 * Compute the set of writes that may break the current pending release
2410 * sequence. This information is extracted from previou release sequence
2413 * @param curr The current ModelAction. Must be a release sequence fixup
2416 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2418 if (pending_rel_seqs->empty())
2421 struct release_seq *pending = pending_rel_seqs->back();
2422 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2423 const ModelAction *write = pending->writes[i];
2424 curr->get_node()->add_relseq_break(write);
2427 /* NULL means don't break the sequence; just synchronize */
2428 curr->get_node()->add_relseq_break(NULL);
2432 * Build up an initial set of all past writes that this 'read' action may read
2433 * from. This set is determined by the clock vector's "happens before"
2435 * @param curr is the current ModelAction that we are exploring; it must be a
2438 void ModelChecker::build_reads_from_past(ModelAction *curr)
2440 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2442 ASSERT(curr->is_read());
2444 ModelAction *last_sc_write = NULL;
2446 /* Track whether this object has been initialized */
2447 bool initialized = false;
2449 if (curr->is_seqcst()) {
2450 last_sc_write = get_last_seq_cst_write(curr);
2451 /* We have to at least see the last sequentially consistent write,
2452 so we are initialized. */
2453 if (last_sc_write != NULL)
2457 /* Iterate over all threads */
2458 for (i = 0; i < thrd_lists->size(); i++) {
2459 /* Iterate over actions in thread, starting from most recent */
2460 action_list_t *list = &(*thrd_lists)[i];
2461 action_list_t::reverse_iterator rit;
2462 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2463 ModelAction *act = *rit;
2465 /* Only consider 'write' actions */
2466 if (!act->is_write() || act == curr)
2469 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2470 bool allow_read = true;
2472 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2474 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2478 DEBUG("Adding action to may_read_from:\n");
2479 if (DBG_ENABLED()) {
2483 curr->get_node()->add_read_from(act);
2486 /* Include at most one act per-thread that "happens before" curr */
2487 if (act->happens_before(curr)) {
2495 assert_bug("May read from uninitialized atomic");
2497 if (DBG_ENABLED() || !initialized) {
2498 model_print("Reached read action:\n");
2500 model_print("Printing may_read_from\n");
2501 curr->get_node()->print_may_read_from();
2502 model_print("End printing may_read_from\n");
2506 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2508 Node *prevnode=write->get_node()->get_parent();
2510 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2511 if (write->is_release()&&thread_sleep)
2513 if (!write->is_rmw()) {
2516 if (write->get_reads_from()==NULL)
2518 write=write->get_reads_from();
2522 static void print_list(action_list_t *list, int exec_num = -1)
2524 action_list_t::iterator it;
2526 model_print("---------------------------------------------------------------------\n");
2528 model_print("Execution %d:\n", exec_num);
2530 unsigned int hash=0;
2532 for (it = list->begin(); it != list->end(); it++) {
2534 hash=hash^(hash<<3)^((*it)->hash());
2536 model_print("HASH %u\n", hash);
2537 model_print("---------------------------------------------------------------------\n");
2540 #if SUPPORT_MOD_ORDER_DUMP
2541 void ModelChecker::dumpGraph(char *filename) {
2543 sprintf(buffer, "%s.dot",filename);
2544 FILE *file=fopen(buffer, "w");
2545 fprintf(file, "digraph %s {\n",filename);
2546 mo_graph->dumpNodes(file);
2547 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2549 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2550 ModelAction *action=*it;
2551 if (action->is_read()) {
2552 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2553 if (action->get_reads_from()!=NULL)
2554 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2556 if (thread_array[action->get_tid()] != NULL) {
2557 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2560 thread_array[action->get_tid()]=action;
2562 fprintf(file,"}\n");
2563 model_free(thread_array);
2568 /** @brief Prints an execution trace summary. */
2569 void ModelChecker::print_summary() const
2571 #if SUPPORT_MOD_ORDER_DUMP
2573 char buffername[100];
2574 sprintf(buffername, "exec%04u", stats.num_total);
2575 mo_graph->dumpGraphToFile(buffername);
2576 sprintf(buffername, "graph%04u", stats.num_total);
2577 dumpGraph(buffername);
2580 if (!isfeasibleprefix())
2581 model_print("INFEASIBLE EXECUTION!\n");
2582 print_list(action_trace, stats.num_total);
2587 * Add a Thread to the system for the first time. Should only be called once
2589 * @param t The Thread to add
2591 void ModelChecker::add_thread(Thread *t)
2593 thread_map->put(id_to_int(t->get_id()), t);
2594 scheduler->add_thread(t);
2598 * Removes a thread from the scheduler.
2599 * @param the thread to remove.
2601 void ModelChecker::remove_thread(Thread *t)
2603 scheduler->remove_thread(t);
2607 * @brief Get a Thread reference by its ID
2608 * @param tid The Thread's ID
2609 * @return A Thread reference
2611 Thread * ModelChecker::get_thread(thread_id_t tid) const
2613 return thread_map->get(id_to_int(tid));
2617 * @brief Get a reference to the Thread in which a ModelAction was executed
2618 * @param act The ModelAction
2619 * @return A Thread reference
2621 Thread * ModelChecker::get_thread(ModelAction *act) const
2623 return get_thread(act->get_tid());
2627 * @brief Check if a Thread is currently enabled
2628 * @param t The Thread to check
2629 * @return True if the Thread is currently enabled
2631 bool ModelChecker::is_enabled(Thread *t) const
2633 return scheduler->is_enabled(t);
2637 * @brief Check if a Thread is currently enabled
2638 * @param tid The ID of the Thread to check
2639 * @return True if the Thread is currently enabled
2641 bool ModelChecker::is_enabled(thread_id_t tid) const
2643 return scheduler->is_enabled(tid);
2647 * Switch from a user-context to the "master thread" context (a.k.a. system
2648 * context). This switch is made with the intention of exploring a particular
2649 * model-checking action (described by a ModelAction object). Must be called
2650 * from a user-thread context.
2652 * @param act The current action that will be explored. May be NULL only if
2653 * trace is exiting via an assertion (see ModelChecker::set_assert and
2654 * ModelChecker::has_asserted).
2655 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2657 int ModelChecker::switch_to_master(ModelAction *act)
2660 Thread *old = thread_current();
2661 set_current_action(act);
2662 old->set_state(THREAD_READY);
2663 return Thread::swap(old, &system_context);
2667 * Takes the next step in the execution, if possible.
2668 * @return Returns true (success) if a step was taken and false otherwise.
2670 bool ModelChecker::take_step() {
2674 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2676 if (curr->get_state() == THREAD_READY) {
2677 ASSERT(priv->current_action);
2679 priv->nextThread = check_current_action(priv->current_action);
2680 priv->current_action = NULL;
2682 if (curr->is_blocked() || curr->is_complete())
2683 scheduler->remove_thread(curr);
2688 Thread *next = scheduler->next_thread(priv->nextThread);
2690 /* Infeasible -> don't take any more steps */
2691 if (is_infeasible())
2693 else if (isfeasibleprefix() && have_bug_reports()) {
2698 if (params.bound != 0) {
2699 if (priv->used_sequence_numbers > params.bound) {
2704 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2705 next ? id_to_int(next->get_id()) : -1);
2708 * Launch end-of-execution release sequence fixups only when there are:
2710 * (1) no more user threads to run (or when execution replay chooses
2711 * the 'model_thread')
2712 * (2) pending release sequences
2713 * (3) pending assertions (i.e., data races)
2714 * (4) no pending promises
2716 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2717 is_feasible_prefix_ignore_relseq() && !unrealizedraces.empty()) {
2718 model_print("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2719 pending_rel_seqs->size());
2720 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2721 std::memory_order_seq_cst, NULL, VALUE_NONE,
2723 set_current_action(fixup);
2727 /* next == NULL -> don't take any more steps */
2731 next->set_state(THREAD_RUNNING);
2733 if (next->get_pending() != NULL) {
2734 /* restart a pending action */
2735 set_current_action(next->get_pending());
2736 next->set_pending(NULL);
2737 next->set_state(THREAD_READY);
2741 /* Return false only if swap fails with an error */
2742 return (Thread::swap(&system_context, next) == 0);
2745 /** Wrapper to run the user's main function, with appropriate arguments */
2746 void user_main_wrapper(void *)
2748 user_main(model->params.argc, model->params.argv);
2751 /** @brief Run ModelChecker for the user program */
2752 void ModelChecker::run()
2757 /* Start user program */
2758 add_thread(new Thread(&user_thread, &user_main_wrapper, NULL));
2760 /* Wait for all threads to complete */
2761 while (take_step());
2762 } while (next_execution());