10 #include "snapshot-interface.h"
12 #include "clockvector.h"
13 #include "cyclegraph.h"
16 #include "threads-model.h"
19 #define INITIAL_THREAD_ID 0
24 bug_message(const char *str) {
25 const char *fmt = " [BUG] %s\n";
26 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
27 sprintf(msg, fmt, str);
29 ~bug_message() { if (msg) snapshot_free(msg); }
32 void print() { model_print("%s", msg); }
38 * Structure for holding small ModelChecker members that should be snapshotted
40 struct model_snapshot_members {
41 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),
48 failed_promise(false),
49 too_many_reads(false),
50 no_valid_reads(false),
51 bad_synchronization(false),
55 ~model_snapshot_members() {
56 for (unsigned int i = 0; i < bugs.size(); i++)
61 unsigned int next_thread_id;
62 modelclock_t used_sequence_numbers;
63 ModelAction *next_backtrack;
64 SnapVector<bug_message *> bugs;
65 struct execution_stats stats;
69 /** @brief Incorrectly-ordered synchronization was made */
70 bool bad_synchronization;
76 /** @brief Constructor */
77 ModelChecker::ModelChecker(struct model_params params) :
78 /* Initialize default scheduler */
80 scheduler(new Scheduler()),
82 earliest_diverge(NULL),
83 action_trace(new action_list_t()),
84 thread_map(new HashTable<int, Thread *, int>()),
85 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
86 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
87 obj_thrd_map(new HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4 >()),
88 promises(new SnapVector<Promise *>()),
89 futurevalues(new SnapVector<struct PendingFutureValue>()),
90 pending_rel_seqs(new SnapVector<struct release_seq *>()),
91 thrd_last_action(new SnapVector<ModelAction *>(1)),
92 thrd_last_fence_release(new SnapVector<ModelAction *>()),
93 node_stack(new NodeStack()),
94 priv(new struct model_snapshot_members()),
95 mo_graph(new CycleGraph())
97 /* Initialize a model-checker thread, for special ModelActions */
98 model_thread = new Thread(get_next_id());
99 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
102 /** @brief Destructor */
103 ModelChecker::~ModelChecker()
105 for (unsigned int i = 0; i < get_num_threads(); i++)
106 delete thread_map->get(i);
111 delete condvar_waiters_map;
114 for (unsigned int i = 0; i < promises->size(); i++)
115 delete (*promises)[i];
118 delete pending_rel_seqs;
120 delete thrd_last_action;
121 delete thrd_last_fence_release;
128 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
130 action_list_t *tmp = hash->get(ptr);
132 tmp = new action_list_t();
138 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
140 SnapVector<action_list_t> *tmp = hash->get(ptr);
142 tmp = new SnapVector<action_list_t>();
149 * Restores user program to initial state and resets all model-checker data
152 void ModelChecker::reset_to_initial_state()
154 DEBUG("+++ Resetting to initial state +++\n");
155 node_stack->reset_execution();
157 /* Print all model-checker output before rollback */
161 * FIXME: if we utilize partial rollback, we will need to free only
162 * those pending actions which were NOT pending before the rollback
165 for (unsigned int i = 0; i < get_num_threads(); i++)
166 delete get_thread(int_to_id(i))->get_pending();
168 snapshot_backtrack_before(0);
171 /** @return a thread ID for a new Thread */
172 thread_id_t ModelChecker::get_next_id()
174 return priv->next_thread_id++;
177 /** @return the number of user threads created during this execution */
178 unsigned int ModelChecker::get_num_threads() const
180 return priv->next_thread_id;
184 * Must be called from user-thread context (e.g., through the global
185 * thread_current() interface)
187 * @return The currently executing Thread.
189 Thread * ModelChecker::get_current_thread() const
191 return scheduler->get_current_thread();
194 /** @return a sequence number for a new ModelAction */
195 modelclock_t ModelChecker::get_next_seq_num()
197 return ++priv->used_sequence_numbers;
200 Node * ModelChecker::get_curr_node() const
202 return node_stack->get_head();
206 * @brief Select the next thread to execute based on the curren action
208 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
209 * actions should be followed by the execution of their child thread. In either
210 * case, the current action should determine the next thread schedule.
212 * @param curr The current action
213 * @return The next thread to run, if the current action will determine this
214 * selection; otherwise NULL
216 Thread * ModelChecker::action_select_next_thread(const ModelAction *curr) const
218 /* Do not split atomic RMW */
220 return get_thread(curr);
221 /* Follow CREATE with the created thread */
222 if (curr->get_type() == THREAD_CREATE)
223 return curr->get_thread_operand();
228 * @brief Choose the next thread to execute.
230 * This function chooses the next thread that should execute. It can enforce
231 * execution replay/backtracking or, if the model-checker has no preference
232 * regarding the next thread (i.e., when exploring a new execution ordering),
233 * we defer to the scheduler.
235 * @return The next chosen thread to run, if any exist. Or else if the current
236 * execution should terminate, return NULL.
238 Thread * ModelChecker::get_next_thread()
243 * Have we completed exploring the preselected path? Then let the
247 return scheduler->select_next_thread();
249 /* Else, we are trying to replay an execution */
250 ModelAction *next = node_stack->get_next()->get_action();
252 if (next == diverge) {
253 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
254 earliest_diverge = diverge;
256 Node *nextnode = next->get_node();
257 Node *prevnode = nextnode->get_parent();
258 scheduler->update_sleep_set(prevnode);
260 /* Reached divergence point */
261 if (nextnode->increment_behaviors()) {
262 /* Execute the same thread with a new behavior */
263 tid = next->get_tid();
264 node_stack->pop_restofstack(2);
267 /* Make a different thread execute for next step */
268 scheduler->add_sleep(get_thread(next->get_tid()));
269 tid = prevnode->get_next_backtrack();
270 /* Make sure the backtracked thread isn't sleeping. */
271 node_stack->pop_restofstack(1);
272 if (diverge == earliest_diverge) {
273 earliest_diverge = prevnode->get_action();
276 /* Start the round robin scheduler from this thread id */
277 scheduler->set_scheduler_thread(tid);
278 /* The correct sleep set is in the parent node. */
281 DEBUG("*** Divergence point ***\n");
285 tid = next->get_tid();
287 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
288 ASSERT(tid != THREAD_ID_T_NONE);
289 return thread_map->get(id_to_int(tid));
293 * We need to know what the next actions of all threads in the sleep
294 * set will be. This method computes them and stores the actions at
295 * the corresponding thread object's pending action.
298 void ModelChecker::execute_sleep_set()
300 for (unsigned int i = 0; i < get_num_threads(); i++) {
301 thread_id_t tid = int_to_id(i);
302 Thread *thr = get_thread(tid);
303 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
304 thr->get_pending()->set_sleep_flag();
310 * @brief Should the current action wake up a given thread?
312 * @param curr The current action
313 * @param thread The thread that we might wake up
314 * @return True, if we should wake up the sleeping thread; false otherwise
316 bool ModelChecker::should_wake_up(const ModelAction *curr, const Thread *thread) const
318 const ModelAction *asleep = thread->get_pending();
319 /* Don't allow partial RMW to wake anyone up */
322 /* Synchronizing actions may have been backtracked */
323 if (asleep->could_synchronize_with(curr))
325 /* All acquire/release fences and fence-acquire/store-release */
326 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
328 /* Fence-release + store can awake load-acquire on the same location */
329 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
330 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
331 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
337 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
339 for (unsigned int i = 0; i < get_num_threads(); i++) {
340 Thread *thr = get_thread(int_to_id(i));
341 if (scheduler->is_sleep_set(thr)) {
342 if (should_wake_up(curr, thr))
343 /* Remove this thread from sleep set */
344 scheduler->remove_sleep(thr);
349 /** @brief Alert the model-checker that an incorrectly-ordered
350 * synchronization was made */
351 void ModelChecker::set_bad_synchronization()
353 priv->bad_synchronization = true;
357 * Check whether the current trace has triggered an assertion which should halt
360 * @return True, if the execution should be aborted; false otherwise
362 bool ModelChecker::has_asserted() const
364 return priv->asserted;
368 * Trigger a trace assertion which should cause this execution to be halted.
369 * This can be due to a detected bug or due to an infeasibility that should
372 void ModelChecker::set_assert()
374 priv->asserted = true;
378 * Check if we are in a deadlock. Should only be called at the end of an
379 * execution, although it should not give false positives in the middle of an
380 * execution (there should be some ENABLED thread).
382 * @return True if program is in a deadlock; false otherwise
384 bool ModelChecker::is_deadlocked() const
386 bool blocking_threads = false;
387 for (unsigned int i = 0; i < get_num_threads(); i++) {
388 thread_id_t tid = int_to_id(i);
391 Thread *t = get_thread(tid);
392 if (!t->is_model_thread() && t->get_pending())
393 blocking_threads = true;
395 return blocking_threads;
399 * Check if this is a complete execution. That is, have all thread completed
400 * execution (rather than exiting because sleep sets have forced a redundant
403 * @return True if the execution is complete.
405 bool ModelChecker::is_complete_execution() const
407 for (unsigned int i = 0; i < get_num_threads(); i++)
408 if (is_enabled(int_to_id(i)))
414 * @brief Assert a bug in the executing program.
416 * Use this function to assert any sort of bug in the user program. If the
417 * current trace is feasible (actually, a prefix of some feasible execution),
418 * then this execution will be aborted, printing the appropriate message. If
419 * the current trace is not yet feasible, the error message will be stashed and
420 * printed if the execution ever becomes feasible.
422 * @param msg Descriptive message for the bug (do not include newline char)
423 * @return True if bug is immediately-feasible
425 bool ModelChecker::assert_bug(const char *msg)
427 priv->bugs.push_back(new bug_message(msg));
429 if (isfeasibleprefix()) {
437 * @brief Assert a bug in the executing program, asserted by a user thread
438 * @see ModelChecker::assert_bug
439 * @param msg Descriptive message for the bug (do not include newline char)
441 void ModelChecker::assert_user_bug(const char *msg)
443 /* If feasible bug, bail out now */
445 switch_to_master(NULL);
448 /** @return True, if any bugs have been reported for this execution */
449 bool ModelChecker::have_bug_reports() const
451 return priv->bugs.size() != 0;
454 /** @brief Print bug report listing for this execution (if any bugs exist) */
455 void ModelChecker::print_bugs() const
457 if (have_bug_reports()) {
458 model_print("Bug report: %zu bug%s detected\n",
460 priv->bugs.size() > 1 ? "s" : "");
461 for (unsigned int i = 0; i < priv->bugs.size(); i++)
462 priv->bugs[i]->print();
467 * @brief Record end-of-execution stats
469 * Must be run when exiting an execution. Records various stats.
470 * @see struct execution_stats
472 void ModelChecker::record_stats()
475 if (!isfeasibleprefix())
476 stats.num_infeasible++;
477 else if (have_bug_reports())
478 stats.num_buggy_executions++;
479 else if (is_complete_execution())
480 stats.num_complete++;
482 stats.num_redundant++;
485 * @todo We can violate this ASSERT() when fairness/sleep sets
486 * conflict to cause an execution to terminate, e.g. with:
487 * Scheduler: [0: disabled][1: disabled][2: sleep][3: current, enabled]
489 //ASSERT(scheduler->all_threads_sleeping());
493 /** @brief Print execution stats */
494 void ModelChecker::print_stats() const
496 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
497 model_print("Number of redundant executions: %d\n", stats.num_redundant);
498 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
499 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
500 model_print("Total executions: %d\n", stats.num_total);
501 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
505 * @brief End-of-exeuction print
506 * @param printbugs Should any existing bugs be printed?
508 void ModelChecker::print_execution(bool printbugs) const
510 print_program_output();
512 if (params.verbose) {
513 model_print("Earliest divergence point since last feasible execution:\n");
514 if (earliest_diverge)
515 earliest_diverge->print();
517 model_print("(Not set)\n");
523 /* Don't print invalid bugs */
532 * Queries the model-checker for more executions to explore and, if one
533 * exists, resets the model-checker state to execute a new execution.
535 * @return If there are more executions to explore, return true. Otherwise,
538 bool ModelChecker::next_execution()
541 /* Is this execution a feasible execution that's worth bug-checking? */
542 bool complete = isfeasibleprefix() && (is_complete_execution() ||
545 /* End-of-execution bug checks */
548 assert_bug("Deadlock detected");
556 if (params.verbose || (complete && have_bug_reports()))
557 print_execution(complete);
559 clear_program_output();
562 earliest_diverge = NULL;
564 if ((diverge = get_next_backtrack()) == NULL)
568 model_print("Next execution will diverge at:\n");
572 reset_to_initial_state();
577 * @brief Find the last fence-related backtracking conflict for a ModelAction
579 * This function performs the search for the most recent conflicting action
580 * against which we should perform backtracking, as affected by fence
581 * operations. This includes pairs of potentially-synchronizing actions which
582 * occur due to fence-acquire or fence-release, and hence should be explored in
583 * the opposite execution order.
585 * @param act The current action
586 * @return The most recent action which conflicts with act due to fences
588 ModelAction * ModelChecker::get_last_fence_conflict(ModelAction *act) const
590 /* Only perform release/acquire fence backtracking for stores */
591 if (!act->is_write())
594 /* Find a fence-release (or, act is a release) */
595 ModelAction *last_release;
596 if (act->is_release())
599 last_release = get_last_fence_release(act->get_tid());
603 /* Skip past the release */
604 action_list_t *list = action_trace;
605 action_list_t::reverse_iterator rit;
606 for (rit = list->rbegin(); rit != list->rend(); rit++)
607 if (*rit == last_release)
609 ASSERT(rit != list->rend());
614 * load --sb-> fence-acquire */
615 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
616 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
617 bool found_acquire_fences = false;
618 for ( ; rit != list->rend(); rit++) {
619 ModelAction *prev = *rit;
620 if (act->same_thread(prev))
623 int tid = id_to_int(prev->get_tid());
625 if (prev->is_read() && act->same_var(prev)) {
626 if (prev->is_acquire()) {
627 /* Found most recent load-acquire, don't need
628 * to search for more fences */
629 if (!found_acquire_fences)
632 prior_loads[tid] = prev;
635 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
636 found_acquire_fences = true;
637 acquire_fences[tid] = prev;
641 ModelAction *latest_backtrack = NULL;
642 for (unsigned int i = 0; i < acquire_fences.size(); i++)
643 if (acquire_fences[i] && prior_loads[i])
644 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
645 latest_backtrack = acquire_fences[i];
646 return latest_backtrack;
650 * @brief Find the last backtracking conflict for a ModelAction
652 * This function performs the search for the most recent conflicting action
653 * against which we should perform backtracking. This primary includes pairs of
654 * synchronizing actions which should be explored in the opposite execution
657 * @param act The current action
658 * @return The most recent action which conflicts with act
660 ModelAction * ModelChecker::get_last_conflict(ModelAction *act) const
662 switch (act->get_type()) {
663 /* case ATOMIC_FENCE: fences don't directly cause backtracking */
667 ModelAction *ret = NULL;
669 /* linear search: from most recent to oldest */
670 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
671 action_list_t::reverse_iterator rit;
672 for (rit = list->rbegin(); rit != list->rend(); rit++) {
673 ModelAction *prev = *rit;
674 if (prev->could_synchronize_with(act)) {
680 ModelAction *ret2 = get_last_fence_conflict(act);
690 case ATOMIC_TRYLOCK: {
691 /* linear search: from most recent to oldest */
692 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
693 action_list_t::reverse_iterator rit;
694 for (rit = list->rbegin(); rit != list->rend(); rit++) {
695 ModelAction *prev = *rit;
696 if (act->is_conflicting_lock(prev))
701 case ATOMIC_UNLOCK: {
702 /* linear search: from most recent to oldest */
703 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
704 action_list_t::reverse_iterator rit;
705 for (rit = list->rbegin(); rit != list->rend(); rit++) {
706 ModelAction *prev = *rit;
707 if (!act->same_thread(prev) && prev->is_failed_trylock())
713 /* linear search: from most recent to oldest */
714 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
715 action_list_t::reverse_iterator rit;
716 for (rit = list->rbegin(); rit != list->rend(); rit++) {
717 ModelAction *prev = *rit;
718 if (!act->same_thread(prev) && prev->is_failed_trylock())
720 if (!act->same_thread(prev) && prev->is_notify())
726 case ATOMIC_NOTIFY_ALL:
727 case ATOMIC_NOTIFY_ONE: {
728 /* linear search: from most recent to oldest */
729 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
730 action_list_t::reverse_iterator rit;
731 for (rit = list->rbegin(); rit != list->rend(); rit++) {
732 ModelAction *prev = *rit;
733 if (!act->same_thread(prev) && prev->is_wait())
744 /** This method finds backtracking points where we should try to
745 * reorder the parameter ModelAction against.
747 * @param the ModelAction to find backtracking points for.
749 void ModelChecker::set_backtracking(ModelAction *act)
751 Thread *t = get_thread(act);
752 ModelAction *prev = get_last_conflict(act);
756 Node *node = prev->get_node()->get_parent();
758 int low_tid, high_tid;
759 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
760 low_tid = id_to_int(act->get_tid());
761 high_tid = low_tid + 1;
764 high_tid = get_num_threads();
767 for (int i = low_tid; i < high_tid; i++) {
768 thread_id_t tid = int_to_id(i);
770 /* Make sure this thread can be enabled here. */
771 if (i >= node->get_num_threads())
774 /* Don't backtrack into a point where the thread is disabled or sleeping. */
775 if (node->enabled_status(tid) != THREAD_ENABLED)
778 /* Check if this has been explored already */
779 if (node->has_been_explored(tid))
782 /* See if fairness allows */
783 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
785 for (int t = 0; t < node->get_num_threads(); t++) {
786 thread_id_t tother = int_to_id(t);
787 if (node->is_enabled(tother) && node->has_priority(tother)) {
796 /* See if CHESS-like yield fairness allows */
797 if (model->params.yieldon) {
799 for (int t = 0; t < node->get_num_threads(); t++) {
800 thread_id_t tother = int_to_id(t);
801 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
810 /* Cache the latest backtracking point */
811 set_latest_backtrack(prev);
813 /* If this is a new backtracking point, mark the tree */
814 if (!node->set_backtrack(tid))
816 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
817 id_to_int(prev->get_tid()),
818 id_to_int(t->get_id()));
827 * @brief Cache the a backtracking point as the "most recent", if eligible
829 * Note that this does not prepare the NodeStack for this backtracking
830 * operation, it only caches the action on a per-execution basis
832 * @param act The operation at which we should explore a different next action
833 * (i.e., backtracking point)
834 * @return True, if this action is now the most recent backtracking point;
837 bool ModelChecker::set_latest_backtrack(ModelAction *act)
839 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
840 priv->next_backtrack = act;
847 * Returns last backtracking point. The model checker will explore a different
848 * path for this point in the next execution.
849 * @return The ModelAction at which the next execution should diverge.
851 ModelAction * ModelChecker::get_next_backtrack()
853 ModelAction *next = priv->next_backtrack;
854 priv->next_backtrack = NULL;
859 * Processes a read model action.
860 * @param curr is the read model action to process.
861 * @return True if processing this read updates the mo_graph.
863 bool ModelChecker::process_read(ModelAction *curr)
865 Node *node = curr->get_node();
867 bool updated = false;
868 switch (node->get_read_from_status()) {
869 case READ_FROM_PAST: {
870 const ModelAction *rf = node->get_read_from_past();
873 mo_graph->startChanges();
875 ASSERT(!is_infeasible());
876 if (!check_recency(curr, rf)) {
877 if (node->increment_read_from()) {
878 mo_graph->rollbackChanges();
881 priv->too_many_reads = true;
885 updated = r_modification_order(curr, rf);
887 mo_graph->commitChanges();
888 mo_check_promises(curr, true);
891 case READ_FROM_PROMISE: {
892 Promise *promise = curr->get_node()->get_read_from_promise();
893 if (promise->add_reader(curr))
894 priv->failed_promise = true;
895 curr->set_read_from_promise(promise);
896 mo_graph->startChanges();
897 if (!check_recency(curr, promise))
898 priv->too_many_reads = true;
899 updated = r_modification_order(curr, promise);
900 mo_graph->commitChanges();
903 case READ_FROM_FUTURE: {
904 /* Read from future value */
905 struct future_value fv = node->get_future_value();
906 Promise *promise = new Promise(curr, fv);
907 curr->set_read_from_promise(promise);
908 promises->push_back(promise);
909 mo_graph->startChanges();
910 updated = r_modification_order(curr, promise);
911 mo_graph->commitChanges();
917 get_thread(curr)->set_return_value(curr->get_return_value());
923 * Processes a lock, trylock, or unlock model action. @param curr is
924 * the read model action to process.
926 * The try lock operation checks whether the lock is taken. If not,
927 * it falls to the normal lock operation case. If so, it returns
930 * The lock operation has already been checked that it is enabled, so
931 * it just grabs the lock and synchronizes with the previous unlock.
933 * The unlock operation has to re-enable all of the threads that are
934 * waiting on the lock.
936 * @return True if synchronization was updated; false otherwise
938 bool ModelChecker::process_mutex(ModelAction *curr)
940 std::mutex *mutex = curr->get_mutex();
941 struct std::mutex_state *state = NULL;
944 state = mutex->get_state();
946 switch (curr->get_type()) {
947 case ATOMIC_TRYLOCK: {
948 bool success = !state->locked;
949 curr->set_try_lock(success);
951 get_thread(curr)->set_return_value(0);
954 get_thread(curr)->set_return_value(1);
956 //otherwise fall into the lock case
958 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
959 assert_bug("Lock access before initialization");
960 state->locked = get_thread(curr);
961 ModelAction *unlock = get_last_unlock(curr);
962 //synchronize with the previous unlock statement
963 if (unlock != NULL) {
964 curr->synchronize_with(unlock);
970 case ATOMIC_UNLOCK: {
971 /* wake up the other threads */
972 for (unsigned int i = 0; i < get_num_threads(); i++) {
973 Thread *t = get_thread(int_to_id(i));
974 Thread *curr_thrd = get_thread(curr);
975 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
979 /* unlock the lock - after checking who was waiting on it */
980 state->locked = NULL;
982 if (!curr->is_wait())
983 break; /* The rest is only for ATOMIC_WAIT */
985 /* Should we go to sleep? (simulate spurious failures) */
986 if (curr->get_node()->get_misc() == 0) {
987 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
989 scheduler->sleep(get_thread(curr));
993 case ATOMIC_NOTIFY_ALL: {
994 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
995 //activate all the waiting threads
996 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
997 scheduler->wake(get_thread(*rit));
1002 case ATOMIC_NOTIFY_ONE: {
1003 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1004 int wakeupthread = curr->get_node()->get_misc();
1005 action_list_t::iterator it = waiters->begin();
1006 advance(it, wakeupthread);
1007 scheduler->wake(get_thread(*it));
1019 * @brief Check if the current pending promises allow a future value to be sent
1021 * If one of the following is true:
1022 * (a) there are no pending promises
1023 * (b) the reader and writer do not cross any promises
1024 * Then, it is safe to pass a future value back now.
1026 * Otherwise, we must save the pending future value until (a) or (b) is true
1028 * @param writer The operation which sends the future value. Must be a write.
1029 * @param reader The operation which will observe the value. Must be a read.
1030 * @return True if the future value can be sent now; false if it must wait.
1032 bool ModelChecker::promises_may_allow(const ModelAction *writer,
1033 const ModelAction *reader) const
1035 if (promises->empty())
1037 for(int i=promises->size()-1;i>=0;i--) {
1038 ModelAction *pr=(*promises)[i]->get_reader(0);
1039 //reader is after promise...doesn't cross any promise
1042 //writer is after promise, reader before...bad...
1050 * @brief Add a future value to a reader
1052 * This function performs a few additional checks to ensure that the future
1053 * value can be feasibly observed by the reader
1055 * @param writer The operation whose value is sent. Must be a write.
1056 * @param reader The read operation which may read the future value. Must be a read.
1058 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
1060 /* Do more ambitious checks now that mo is more complete */
1061 if (!mo_may_allow(writer, reader))
1064 Node *node = reader->get_node();
1066 /* Find an ancestor thread which exists at the time of the reader */
1067 Thread *write_thread = get_thread(writer);
1068 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
1069 write_thread = write_thread->get_parent();
1071 struct future_value fv = {
1072 writer->get_write_value(),
1073 writer->get_seq_number() + params.maxfuturedelay,
1074 write_thread->get_id(),
1076 if (node->add_future_value(fv))
1077 set_latest_backtrack(reader);
1081 * Process a write ModelAction
1082 * @param curr The ModelAction to process
1083 * @return True if the mo_graph was updated or promises were resolved
1085 bool ModelChecker::process_write(ModelAction *curr)
1087 /* Readers to which we may send our future value */
1088 ModelVector<ModelAction *> send_fv;
1090 const ModelAction *earliest_promise_reader;
1091 bool updated_promises = false;
1093 bool updated_mod_order = w_modification_order(curr, &send_fv);
1094 Promise *promise = pop_promise_to_resolve(curr);
1097 earliest_promise_reader = promise->get_reader(0);
1098 updated_promises = resolve_promise(curr, promise);
1100 earliest_promise_reader = NULL;
1102 for (unsigned int i = 0; i < send_fv.size(); i++) {
1103 ModelAction *read = send_fv[i];
1105 /* Don't send future values to reads after the Promise we resolve */
1106 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
1107 /* Check if future value can be sent immediately */
1108 if (promises_may_allow(curr, read)) {
1109 add_future_value(curr, read);
1111 futurevalues->push_back(PendingFutureValue(curr, read));
1116 /* Check the pending future values */
1117 for (int i = (int)futurevalues->size() - 1; i >= 0; i--) {
1118 struct PendingFutureValue pfv = (*futurevalues)[i];
1119 if (promises_may_allow(pfv.writer, pfv.reader)) {
1120 add_future_value(pfv.writer, pfv.reader);
1121 futurevalues->erase(futurevalues->begin() + i);
1125 mo_graph->commitChanges();
1126 mo_check_promises(curr, false);
1128 get_thread(curr)->set_return_value(VALUE_NONE);
1129 return updated_mod_order || updated_promises;
1133 * Process a fence ModelAction
1134 * @param curr The ModelAction to process
1135 * @return True if synchronization was updated
1137 bool ModelChecker::process_fence(ModelAction *curr)
1140 * fence-relaxed: no-op
1141 * fence-release: only log the occurence (not in this function), for
1142 * use in later synchronization
1143 * fence-acquire (this function): search for hypothetical release
1145 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
1147 bool updated = false;
1148 if (curr->is_acquire()) {
1149 action_list_t *list = action_trace;
1150 action_list_t::reverse_iterator rit;
1151 /* Find X : is_read(X) && X --sb-> curr */
1152 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1153 ModelAction *act = *rit;
1156 if (act->get_tid() != curr->get_tid())
1158 /* Stop at the beginning of the thread */
1159 if (act->is_thread_start())
1161 /* Stop once we reach a prior fence-acquire */
1162 if (act->is_fence() && act->is_acquire())
1164 if (!act->is_read())
1166 /* read-acquire will find its own release sequences */
1167 if (act->is_acquire())
1170 /* Establish hypothetical release sequences */
1171 rel_heads_list_t release_heads;
1172 get_release_seq_heads(curr, act, &release_heads);
1173 for (unsigned int i = 0; i < release_heads.size(); i++)
1174 if (!curr->synchronize_with(release_heads[i]))
1175 set_bad_synchronization();
1176 if (release_heads.size() != 0)
1184 * @brief Process the current action for thread-related activity
1186 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
1187 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
1188 * synchronization, etc. This function is a no-op for non-THREAD actions
1189 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
1191 * @param curr The current action
1192 * @return True if synchronization was updated or a thread completed
1194 bool ModelChecker::process_thread_action(ModelAction *curr)
1196 bool updated = false;
1198 switch (curr->get_type()) {
1199 case THREAD_CREATE: {
1200 thrd_t *thrd = (thrd_t *)curr->get_location();
1201 struct thread_params *params = (struct thread_params *)curr->get_value();
1202 Thread *th = new Thread(thrd, params->func, params->arg, get_thread(curr));
1204 th->set_creation(curr);
1205 /* Promises can be satisfied by children */
1206 for (unsigned int i = 0; i < promises->size(); i++) {
1207 Promise *promise = (*promises)[i];
1208 if (promise->thread_is_available(curr->get_tid()))
1209 promise->add_thread(th->get_id());
1214 Thread *blocking = curr->get_thread_operand();
1215 ModelAction *act = get_last_action(blocking->get_id());
1216 curr->synchronize_with(act);
1217 updated = true; /* trigger rel-seq checks */
1220 case THREAD_FINISH: {
1221 Thread *th = get_thread(curr);
1222 /* Wake up any joining threads */
1223 for (unsigned int i = 0; i < get_num_threads(); i++) {
1224 Thread *waiting = get_thread(int_to_id(i));
1225 if (waiting->waiting_on() == th &&
1226 waiting->get_pending()->is_thread_join())
1227 scheduler->wake(waiting);
1230 /* Completed thread can't satisfy promises */
1231 for (unsigned int i = 0; i < promises->size(); i++) {
1232 Promise *promise = (*promises)[i];
1233 if (promise->thread_is_available(th->get_id()))
1234 if (promise->eliminate_thread(th->get_id()))
1235 priv->failed_promise = true;
1237 updated = true; /* trigger rel-seq checks */
1240 case THREAD_START: {
1241 check_promises(curr->get_tid(), NULL, curr->get_cv());
1252 * @brief Process the current action for release sequence fixup activity
1254 * Performs model-checker release sequence fixups for the current action,
1255 * forcing a single pending release sequence to break (with a given, potential
1256 * "loose" write) or to complete (i.e., synchronize). If a pending release
1257 * sequence forms a complete release sequence, then we must perform the fixup
1258 * synchronization, mo_graph additions, etc.
1260 * @param curr The current action; must be a release sequence fixup action
1261 * @param work_queue The work queue to which to add work items as they are
1264 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1266 const ModelAction *write = curr->get_node()->get_relseq_break();
1267 struct release_seq *sequence = pending_rel_seqs->back();
1268 pending_rel_seqs->pop_back();
1270 ModelAction *acquire = sequence->acquire;
1271 const ModelAction *rf = sequence->rf;
1272 const ModelAction *release = sequence->release;
1276 ASSERT(release->same_thread(rf));
1278 if (write == NULL) {
1280 * @todo Forcing a synchronization requires that we set
1281 * modification order constraints. For instance, we can't allow
1282 * a fixup sequence in which two separate read-acquire
1283 * operations read from the same sequence, where the first one
1284 * synchronizes and the other doesn't. Essentially, we can't
1285 * allow any writes to insert themselves between 'release' and
1289 /* Must synchronize */
1290 if (!acquire->synchronize_with(release)) {
1291 set_bad_synchronization();
1294 /* Re-check all pending release sequences */
1295 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1296 /* Re-check act for mo_graph edges */
1297 work_queue->push_back(MOEdgeWorkEntry(acquire));
1299 /* propagate synchronization to later actions */
1300 action_list_t::reverse_iterator rit = action_trace->rbegin();
1301 for (; (*rit) != acquire; rit++) {
1302 ModelAction *propagate = *rit;
1303 if (acquire->happens_before(propagate)) {
1304 propagate->synchronize_with(acquire);
1305 /* Re-check 'propagate' for mo_graph edges */
1306 work_queue->push_back(MOEdgeWorkEntry(propagate));
1310 /* Break release sequence with new edges:
1311 * release --mo--> write --mo--> rf */
1312 mo_graph->addEdge(release, write);
1313 mo_graph->addEdge(write, rf);
1316 /* See if we have realized a data race */
1321 * Initialize the current action by performing one or more of the following
1322 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1323 * in the NodeStack, manipulating backtracking sets, allocating and
1324 * initializing clock vectors, and computing the promises to fulfill.
1326 * @param curr The current action, as passed from the user context; may be
1327 * freed/invalidated after the execution of this function, with a different
1328 * action "returned" its place (pass-by-reference)
1329 * @return True if curr is a newly-explored action; false otherwise
1331 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1333 ModelAction *newcurr;
1335 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1336 newcurr = process_rmw(*curr);
1339 if (newcurr->is_rmw())
1340 compute_promises(newcurr);
1346 (*curr)->set_seq_number(get_next_seq_num());
1348 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1350 /* First restore type and order in case of RMW operation */
1351 if ((*curr)->is_rmwr())
1352 newcurr->copy_typeandorder(*curr);
1354 ASSERT((*curr)->get_location() == newcurr->get_location());
1355 newcurr->copy_from_new(*curr);
1357 /* Discard duplicate ModelAction; use action from NodeStack */
1360 /* Always compute new clock vector */
1361 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1364 return false; /* Action was explored previously */
1368 /* Always compute new clock vector */
1369 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1371 /* Assign most recent release fence */
1372 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1375 * Perform one-time actions when pushing new ModelAction onto
1378 if (newcurr->is_write())
1379 compute_promises(newcurr);
1380 else if (newcurr->is_relseq_fixup())
1381 compute_relseq_breakwrites(newcurr);
1382 else if (newcurr->is_wait())
1383 newcurr->get_node()->set_misc_max(2);
1384 else if (newcurr->is_notify_one()) {
1385 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1387 return true; /* This was a new ModelAction */
1392 * @brief Establish reads-from relation between two actions
1394 * Perform basic operations involved with establishing a concrete rf relation,
1395 * including setting the ModelAction data and checking for release sequences.
1397 * @param act The action that is reading (must be a read)
1398 * @param rf The action from which we are reading (must be a write)
1400 * @return True if this read established synchronization
1402 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1405 ASSERT(rf->is_write());
1407 act->set_read_from(rf);
1408 if (act->is_acquire()) {
1409 rel_heads_list_t release_heads;
1410 get_release_seq_heads(act, act, &release_heads);
1411 int num_heads = release_heads.size();
1412 for (unsigned int i = 0; i < release_heads.size(); i++)
1413 if (!act->synchronize_with(release_heads[i])) {
1414 set_bad_synchronization();
1417 return num_heads > 0;
1423 * Check promises and eliminate potentially-satisfying threads when a thread is
1424 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1425 * no longer satisfy a promise generated from that thread.
1427 * @param blocker The thread on which a thread is waiting
1428 * @param waiting The waiting thread
1430 void ModelChecker::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1432 for (unsigned int i = 0; i < promises->size(); i++) {
1433 Promise *promise = (*promises)[i];
1434 if (!promise->thread_is_available(waiting->get_id()))
1436 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1437 ModelAction *reader = promise->get_reader(j);
1438 if (reader->get_tid() != blocker->get_id())
1440 if (promise->eliminate_thread(waiting->get_id())) {
1441 /* Promise has failed */
1442 priv->failed_promise = true;
1444 /* Only eliminate the 'waiting' thread once */
1452 * @brief Check whether a model action is enabled.
1454 * Checks whether a lock or join operation would be successful (i.e., is the
1455 * lock already locked, or is the joined thread already complete). If not, put
1456 * the action in a waiter list.
1458 * @param curr is the ModelAction to check whether it is enabled.
1459 * @return a bool that indicates whether the action is enabled.
1461 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1462 if (curr->is_lock()) {
1463 std::mutex *lock = curr->get_mutex();
1464 struct std::mutex_state *state = lock->get_state();
1467 } else if (curr->is_thread_join()) {
1468 Thread *blocking = curr->get_thread_operand();
1469 if (!blocking->is_complete()) {
1470 thread_blocking_check_promises(blocking, get_thread(curr));
1479 * This is the heart of the model checker routine. It performs model-checking
1480 * actions corresponding to a given "current action." Among other processes, it
1481 * calculates reads-from relationships, updates synchronization clock vectors,
1482 * forms a memory_order constraints graph, and handles replay/backtrack
1483 * execution when running permutations of previously-observed executions.
1485 * @param curr The current action to process
1486 * @return The ModelAction that is actually executed; may be different than
1487 * curr; may be NULL, if the current action is not enabled to run
1489 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1492 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1494 if (!check_action_enabled(curr)) {
1495 /* Make the execution look like we chose to run this action
1496 * much later, when a lock/join can succeed */
1497 get_thread(curr)->set_pending(curr);
1498 scheduler->sleep(get_thread(curr));
1502 bool newly_explored = initialize_curr_action(&curr);
1508 wake_up_sleeping_actions(curr);
1510 /* Compute fairness information for CHESS yield algorithm */
1511 if (model->params.yieldon) {
1512 curr->get_node()->update_yield(scheduler);
1515 /* Add the action to lists before any other model-checking tasks */
1516 if (!second_part_of_rmw)
1517 add_action_to_lists(curr);
1519 /* Build may_read_from set for newly-created actions */
1520 if (newly_explored && curr->is_read())
1521 build_may_read_from(curr);
1523 /* Initialize work_queue with the "current action" work */
1524 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1525 while (!work_queue.empty() && !has_asserted()) {
1526 WorkQueueEntry work = work_queue.front();
1527 work_queue.pop_front();
1529 switch (work.type) {
1530 case WORK_CHECK_CURR_ACTION: {
1531 ModelAction *act = work.action;
1532 bool update = false; /* update this location's release seq's */
1533 bool update_all = false; /* update all release seq's */
1535 if (process_thread_action(curr))
1538 if (act->is_read() && !second_part_of_rmw && process_read(act))
1541 if (act->is_write() && process_write(act))
1544 if (act->is_fence() && process_fence(act))
1547 if (act->is_mutex_op() && process_mutex(act))
1550 if (act->is_relseq_fixup())
1551 process_relseq_fixup(curr, &work_queue);
1554 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1556 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1559 case WORK_CHECK_RELEASE_SEQ:
1560 resolve_release_sequences(work.location, &work_queue);
1562 case WORK_CHECK_MO_EDGES: {
1563 /** @todo Complete verification of work_queue */
1564 ModelAction *act = work.action;
1565 bool updated = false;
1567 if (act->is_read()) {
1568 const ModelAction *rf = act->get_reads_from();
1569 const Promise *promise = act->get_reads_from_promise();
1571 if (r_modification_order(act, rf))
1573 } else if (promise) {
1574 if (r_modification_order(act, promise))
1578 if (act->is_write()) {
1579 if (w_modification_order(act, NULL))
1582 mo_graph->commitChanges();
1585 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1594 check_curr_backtracking(curr);
1595 set_backtracking(curr);
1599 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1601 Node *currnode = curr->get_node();
1602 Node *parnode = currnode->get_parent();
1604 if ((parnode && !parnode->backtrack_empty()) ||
1605 !currnode->misc_empty() ||
1606 !currnode->read_from_empty() ||
1607 !currnode->promise_empty() ||
1608 !currnode->relseq_break_empty()) {
1609 set_latest_backtrack(curr);
1613 bool ModelChecker::promises_expired() const
1615 for (unsigned int i = 0; i < promises->size(); i++) {
1616 Promise *promise = (*promises)[i];
1617 if (promise->get_expiration() < priv->used_sequence_numbers)
1624 * This is the strongest feasibility check available.
1625 * @return whether the current trace (partial or complete) must be a prefix of
1628 bool ModelChecker::isfeasibleprefix() const
1630 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1634 * Print disagnostic information about an infeasible execution
1635 * @param prefix A string to prefix the output with; if NULL, then a default
1636 * message prefix will be provided
1638 void ModelChecker::print_infeasibility(const char *prefix) const
1642 if (mo_graph->checkForCycles())
1643 ptr += sprintf(ptr, "[mo cycle]");
1644 if (priv->failed_promise)
1645 ptr += sprintf(ptr, "[failed promise]");
1646 if (priv->too_many_reads)
1647 ptr += sprintf(ptr, "[too many reads]");
1648 if (priv->no_valid_reads)
1649 ptr += sprintf(ptr, "[no valid reads-from]");
1650 if (priv->bad_synchronization)
1651 ptr += sprintf(ptr, "[bad sw ordering]");
1652 if (promises_expired())
1653 ptr += sprintf(ptr, "[promise expired]");
1654 if (promises->size() != 0)
1655 ptr += sprintf(ptr, "[unresolved promise]");
1657 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1661 * Returns whether the current completed trace is feasible, except for pending
1662 * release sequences.
1664 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1666 return !is_infeasible() && promises->size() == 0;
1670 * Check if the current partial trace is infeasible. Does not check any
1671 * end-of-execution flags, which might rule out the execution. Thus, this is
1672 * useful only for ruling an execution as infeasible.
1673 * @return whether the current partial trace is infeasible.
1675 bool ModelChecker::is_infeasible() const
1677 return mo_graph->checkForCycles() ||
1678 priv->no_valid_reads ||
1679 priv->failed_promise ||
1680 priv->too_many_reads ||
1681 priv->bad_synchronization ||
1685 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1686 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1687 ModelAction *lastread = get_last_action(act->get_tid());
1688 lastread->process_rmw(act);
1689 if (act->is_rmw()) {
1690 if (lastread->get_reads_from())
1691 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1693 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1694 mo_graph->commitChanges();
1700 * A helper function for ModelChecker::check_recency, to check if the current
1701 * thread is able to read from a different write/promise for 'params.maxreads'
1702 * number of steps and if that write/promise should become visible (i.e., is
1703 * ordered later in the modification order). This helps model memory liveness.
1705 * @param curr The current action. Must be a read.
1706 * @param rf The write/promise from which we plan to read
1707 * @param other_rf The write/promise from which we may read
1708 * @return True if we were able to read from other_rf for params.maxreads steps
1710 template <typename T, typename U>
1711 bool ModelChecker::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1713 /* Need a different write/promise */
1714 if (other_rf->equals(rf))
1717 /* Only look for "newer" writes/promises */
1718 if (!mo_graph->checkReachable(rf, other_rf))
1721 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1722 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1723 action_list_t::reverse_iterator rit = list->rbegin();
1724 ASSERT((*rit) == curr);
1725 /* Skip past curr */
1728 /* Does this write/promise work for everyone? */
1729 for (int i = 0; i < params.maxreads; i++, rit++) {
1730 ModelAction *act = *rit;
1731 if (!act->may_read_from(other_rf))
1738 * Checks whether a thread has read from the same write or Promise for too many
1739 * times without seeing the effects of a later write/Promise.
1742 * 1) there must a different write/promise that we could read from,
1743 * 2) we must have read from the same write/promise in excess of maxreads times,
1744 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1745 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1747 * If so, we decide that the execution is no longer feasible.
1749 * @param curr The current action. Must be a read.
1750 * @param rf The ModelAction/Promise from which we might read.
1751 * @return True if the read should succeed; false otherwise
1753 template <typename T>
1754 bool ModelChecker::check_recency(ModelAction *curr, const T *rf) const
1756 if (!params.maxreads)
1759 //NOTE: Next check is just optimization, not really necessary....
1760 if (curr->get_node()->get_read_from_past_size() +
1761 curr->get_node()->get_read_from_promise_size() <= 1)
1764 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1765 int tid = id_to_int(curr->get_tid());
1766 ASSERT(tid < (int)thrd_lists->size());
1767 action_list_t *list = &(*thrd_lists)[tid];
1768 action_list_t::reverse_iterator rit = list->rbegin();
1769 ASSERT((*rit) == curr);
1770 /* Skip past curr */
1773 action_list_t::reverse_iterator ritcopy = rit;
1774 /* See if we have enough reads from the same value */
1775 for (int count = 0; count < params.maxreads; ritcopy++, count++) {
1776 if (ritcopy == list->rend())
1778 ModelAction *act = *ritcopy;
1779 if (!act->is_read())
1781 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1783 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1785 if (act->get_node()->get_read_from_past_size() +
1786 act->get_node()->get_read_from_promise_size() <= 1)
1789 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1790 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1791 if (should_read_instead(curr, rf, write))
1792 return false; /* liveness failure */
1794 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1795 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1796 if (should_read_instead(curr, rf, promise))
1797 return false; /* liveness failure */
1803 * Updates the mo_graph with the constraints imposed from the current
1806 * Basic idea is the following: Go through each other thread and find
1807 * the last action that happened before our read. Two cases:
1809 * (1) The action is a write => that write must either occur before
1810 * the write we read from or be the write we read from.
1812 * (2) The action is a read => the write that that action read from
1813 * must occur before the write we read from or be the same write.
1815 * @param curr The current action. Must be a read.
1816 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1817 * @return True if modification order edges were added; false otherwise
1819 template <typename rf_type>
1820 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1822 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1825 ASSERT(curr->is_read());
1827 /* Last SC fence in the current thread */
1828 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1829 ModelAction *last_sc_write = NULL;
1830 if (curr->is_seqcst())
1831 last_sc_write = get_last_seq_cst_write(curr);
1833 /* Iterate over all threads */
1834 for (i = 0; i < thrd_lists->size(); i++) {
1835 /* Last SC fence in thread i */
1836 ModelAction *last_sc_fence_thread_local = NULL;
1837 if (int_to_id((int)i) != curr->get_tid())
1838 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1840 /* Last SC fence in thread i, before last SC fence in current thread */
1841 ModelAction *last_sc_fence_thread_before = NULL;
1842 if (last_sc_fence_local)
1843 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1845 /* Iterate over actions in thread, starting from most recent */
1846 action_list_t *list = &(*thrd_lists)[i];
1847 action_list_t::reverse_iterator rit;
1848 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1849 ModelAction *act = *rit;
1854 /* Don't want to add reflexive edges on 'rf' */
1855 if (act->equals(rf)) {
1856 if (act->happens_before(curr))
1862 if (act->is_write()) {
1863 /* C++, Section 29.3 statement 5 */
1864 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1865 *act < *last_sc_fence_thread_local) {
1866 added = mo_graph->addEdge(act, rf) || added;
1869 /* C++, Section 29.3 statement 4 */
1870 else if (act->is_seqcst() && last_sc_fence_local &&
1871 *act < *last_sc_fence_local) {
1872 added = mo_graph->addEdge(act, rf) || added;
1875 /* C++, Section 29.3 statement 6 */
1876 else if (last_sc_fence_thread_before &&
1877 *act < *last_sc_fence_thread_before) {
1878 added = mo_graph->addEdge(act, rf) || added;
1883 /* C++, Section 29.3 statement 3 (second subpoint) */
1884 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1885 added = mo_graph->addEdge(act, rf) || added;
1890 * Include at most one act per-thread that "happens
1893 if (act->happens_before(curr)) {
1894 if (act->is_write()) {
1895 added = mo_graph->addEdge(act, rf) || added;
1897 const ModelAction *prevrf = act->get_reads_from();
1898 const Promise *prevrf_promise = act->get_reads_from_promise();
1900 if (!prevrf->equals(rf))
1901 added = mo_graph->addEdge(prevrf, rf) || added;
1902 } else if (!prevrf_promise->equals(rf)) {
1903 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1912 * All compatible, thread-exclusive promises must be ordered after any
1913 * concrete loads from the same thread
1915 for (unsigned int i = 0; i < promises->size(); i++)
1916 if ((*promises)[i]->is_compatible_exclusive(curr))
1917 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1923 * Updates the mo_graph with the constraints imposed from the current write.
1925 * Basic idea is the following: Go through each other thread and find
1926 * the lastest action that happened before our write. Two cases:
1928 * (1) The action is a write => that write must occur before
1931 * (2) The action is a read => the write that that action read from
1932 * must occur before the current write.
1934 * This method also handles two other issues:
1936 * (I) Sequential Consistency: Making sure that if the current write is
1937 * seq_cst, that it occurs after the previous seq_cst write.
1939 * (II) Sending the write back to non-synchronizing reads.
1941 * @param curr The current action. Must be a write.
1942 * @param send_fv A vector for stashing reads to which we may pass our future
1943 * value. If NULL, then don't record any future values.
1944 * @return True if modification order edges were added; false otherwise
1946 bool ModelChecker::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1948 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1951 ASSERT(curr->is_write());
1953 if (curr->is_seqcst()) {
1954 /* We have to at least see the last sequentially consistent write,
1955 so we are initialized. */
1956 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1957 if (last_seq_cst != NULL) {
1958 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1962 /* Last SC fence in the current thread */
1963 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1965 /* Iterate over all threads */
1966 for (i = 0; i < thrd_lists->size(); i++) {
1967 /* Last SC fence in thread i, before last SC fence in current thread */
1968 ModelAction *last_sc_fence_thread_before = NULL;
1969 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1970 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1972 /* Iterate over actions in thread, starting from most recent */
1973 action_list_t *list = &(*thrd_lists)[i];
1974 action_list_t::reverse_iterator rit;
1975 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1976 ModelAction *act = *rit;
1979 * 1) If RMW and it actually read from something, then we
1980 * already have all relevant edges, so just skip to next
1983 * 2) If RMW and it didn't read from anything, we should
1984 * whatever edge we can get to speed up convergence.
1986 * 3) If normal write, we need to look at earlier actions, so
1987 * continue processing list.
1989 if (curr->is_rmw()) {
1990 if (curr->get_reads_from() != NULL)
1998 /* C++, Section 29.3 statement 7 */
1999 if (last_sc_fence_thread_before && act->is_write() &&
2000 *act < *last_sc_fence_thread_before) {
2001 added = mo_graph->addEdge(act, curr) || added;
2006 * Include at most one act per-thread that "happens
2009 if (act->happens_before(curr)) {
2011 * Note: if act is RMW, just add edge:
2013 * The following edge should be handled elsewhere:
2014 * readfrom(act) --mo--> act
2016 if (act->is_write())
2017 added = mo_graph->addEdge(act, curr) || added;
2018 else if (act->is_read()) {
2019 //if previous read accessed a null, just keep going
2020 if (act->get_reads_from() == NULL)
2022 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
2025 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
2026 !act->same_thread(curr)) {
2027 /* We have an action that:
2028 (1) did not happen before us
2029 (2) is a read and we are a write
2030 (3) cannot synchronize with us
2031 (4) is in a different thread
2033 that read could potentially read from our write. Note that
2034 these checks are overly conservative at this point, we'll
2035 do more checks before actually removing the
2039 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
2040 if (!is_infeasible())
2041 send_fv->push_back(act);
2042 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
2043 add_future_value(curr, act);
2050 * All compatible, thread-exclusive promises must be ordered after any
2051 * concrete stores to the same thread, or else they can be merged with
2054 for (unsigned int i = 0; i < promises->size(); i++)
2055 if ((*promises)[i]->is_compatible_exclusive(curr))
2056 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
2061 /** Arbitrary reads from the future are not allowed. Section 29.3
2062 * part 9 places some constraints. This method checks one result of constraint
2063 * constraint. Others require compiler support. */
2064 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
2066 if (!writer->is_rmw())
2069 if (!reader->is_rmw())
2072 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
2073 if (search == reader)
2075 if (search->get_tid() == reader->get_tid() &&
2076 search->happens_before(reader))
2084 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
2085 * some constraints. This method checks one the following constraint (others
2086 * require compiler support):
2088 * If X --hb-> Y --mo-> Z, then X should not read from Z.
2090 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
2092 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
2094 /* Iterate over all threads */
2095 for (i = 0; i < thrd_lists->size(); i++) {
2096 const ModelAction *write_after_read = NULL;
2098 /* Iterate over actions in thread, starting from most recent */
2099 action_list_t *list = &(*thrd_lists)[i];
2100 action_list_t::reverse_iterator rit;
2101 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2102 ModelAction *act = *rit;
2104 /* Don't disallow due to act == reader */
2105 if (!reader->happens_before(act) || reader == act)
2107 else if (act->is_write())
2108 write_after_read = act;
2109 else if (act->is_read() && act->get_reads_from() != NULL)
2110 write_after_read = act->get_reads_from();
2113 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
2120 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
2121 * The ModelAction under consideration is expected to be taking part in
2122 * release/acquire synchronization as an object of the "reads from" relation.
2123 * Note that this can only provide release sequence support for RMW chains
2124 * which do not read from the future, as those actions cannot be traced until
2125 * their "promise" is fulfilled. Similarly, we may not even establish the
2126 * presence of a release sequence with certainty, as some modification order
2127 * constraints may be decided further in the future. Thus, this function
2128 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
2129 * and a boolean representing certainty.
2131 * @param rf The action that might be part of a release sequence. Must be a
2133 * @param release_heads A pass-by-reference style return parameter. After
2134 * execution of this function, release_heads will contain the heads of all the
2135 * relevant release sequences, if any exists with certainty
2136 * @param pending A pass-by-reference style return parameter which is only used
2137 * when returning false (i.e., uncertain). Returns most information regarding
2138 * an uncertain release sequence, including any write operations that might
2139 * break the sequence.
2140 * @return true, if the ModelChecker is certain that release_heads is complete;
2143 bool ModelChecker::release_seq_heads(const ModelAction *rf,
2144 rel_heads_list_t *release_heads,
2145 struct release_seq *pending) const
2147 /* Only check for release sequences if there are no cycles */
2148 if (mo_graph->checkForCycles())
2151 for ( ; rf != NULL; rf = rf->get_reads_from()) {
2152 ASSERT(rf->is_write());
2154 if (rf->is_release())
2155 release_heads->push_back(rf);
2156 else if (rf->get_last_fence_release())
2157 release_heads->push_back(rf->get_last_fence_release());
2159 break; /* End of RMW chain */
2161 /** @todo Need to be smarter here... In the linux lock
2162 * example, this will run to the beginning of the program for
2164 /** @todo The way to be smarter here is to keep going until 1
2165 * thread has a release preceded by an acquire and you've seen
2168 /* acq_rel RMW is a sufficient stopping condition */
2169 if (rf->is_acquire() && rf->is_release())
2170 return true; /* complete */
2173 /* read from future: need to settle this later */
2175 return false; /* incomplete */
2178 if (rf->is_release())
2179 return true; /* complete */
2181 /* else relaxed write
2182 * - check for fence-release in the same thread (29.8, stmt. 3)
2183 * - check modification order for contiguous subsequence
2184 * -> rf must be same thread as release */
2186 const ModelAction *fence_release = rf->get_last_fence_release();
2187 /* Synchronize with a fence-release unconditionally; we don't need to
2188 * find any more "contiguous subsequence..." for it */
2190 release_heads->push_back(fence_release);
2192 int tid = id_to_int(rf->get_tid());
2193 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
2194 action_list_t *list = &(*thrd_lists)[tid];
2195 action_list_t::const_reverse_iterator rit;
2197 /* Find rf in the thread list */
2198 rit = std::find(list->rbegin(), list->rend(), rf);
2199 ASSERT(rit != list->rend());
2201 /* Find the last {write,fence}-release */
2202 for (; rit != list->rend(); rit++) {
2203 if (fence_release && *(*rit) < *fence_release)
2205 if ((*rit)->is_release())
2208 if (rit == list->rend()) {
2209 /* No write-release in this thread */
2210 return true; /* complete */
2211 } else if (fence_release && *(*rit) < *fence_release) {
2212 /* The fence-release is more recent (and so, "stronger") than
2213 * the most recent write-release */
2214 return true; /* complete */
2215 } /* else, need to establish contiguous release sequence */
2216 ModelAction *release = *rit;
2218 ASSERT(rf->same_thread(release));
2220 pending->writes.clear();
2222 bool certain = true;
2223 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2224 if (id_to_int(rf->get_tid()) == (int)i)
2226 list = &(*thrd_lists)[i];
2228 /* Can we ensure no future writes from this thread may break
2229 * the release seq? */
2230 bool future_ordered = false;
2232 ModelAction *last = get_last_action(int_to_id(i));
2233 Thread *th = get_thread(int_to_id(i));
2234 if ((last && rf->happens_before(last)) ||
2237 future_ordered = true;
2239 ASSERT(!th->is_model_thread() || future_ordered);
2241 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2242 const ModelAction *act = *rit;
2243 /* Reach synchronization -> this thread is complete */
2244 if (act->happens_before(release))
2246 if (rf->happens_before(act)) {
2247 future_ordered = true;
2251 /* Only non-RMW writes can break release sequences */
2252 if (!act->is_write() || act->is_rmw())
2255 /* Check modification order */
2256 if (mo_graph->checkReachable(rf, act)) {
2257 /* rf --mo--> act */
2258 future_ordered = true;
2261 if (mo_graph->checkReachable(act, release))
2262 /* act --mo--> release */
2264 if (mo_graph->checkReachable(release, act) &&
2265 mo_graph->checkReachable(act, rf)) {
2266 /* release --mo-> act --mo--> rf */
2267 return true; /* complete */
2269 /* act may break release sequence */
2270 pending->writes.push_back(act);
2273 if (!future_ordered)
2274 certain = false; /* This thread is uncertain */
2278 release_heads->push_back(release);
2279 pending->writes.clear();
2281 pending->release = release;
2288 * An interface for getting the release sequence head(s) with which a
2289 * given ModelAction must synchronize. This function only returns a non-empty
2290 * result when it can locate a release sequence head with certainty. Otherwise,
2291 * it may mark the internal state of the ModelChecker so that it will handle
2292 * the release sequence at a later time, causing @a acquire to update its
2293 * synchronization at some later point in execution.
2295 * @param acquire The 'acquire' action that may synchronize with a release
2297 * @param read The read action that may read from a release sequence; this may
2298 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2299 * when 'acquire' is a fence-acquire)
2300 * @param release_heads A pass-by-reference return parameter. Will be filled
2301 * with the head(s) of the release sequence(s), if they exists with certainty.
2302 * @see ModelChecker::release_seq_heads
2304 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2305 ModelAction *read, rel_heads_list_t *release_heads)
2307 const ModelAction *rf = read->get_reads_from();
2308 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2309 sequence->acquire = acquire;
2310 sequence->read = read;
2312 if (!release_seq_heads(rf, release_heads, sequence)) {
2313 /* add act to 'lazy checking' list */
2314 pending_rel_seqs->push_back(sequence);
2316 snapshot_free(sequence);
2321 * Attempt to resolve all stashed operations that might synchronize with a
2322 * release sequence for a given location. This implements the "lazy" portion of
2323 * determining whether or not a release sequence was contiguous, since not all
2324 * modification order information is present at the time an action occurs.
2326 * @param location The location/object that should be checked for release
2327 * sequence resolutions. A NULL value means to check all locations.
2328 * @param work_queue The work queue to which to add work items as they are
2330 * @return True if any updates occurred (new synchronization, new mo_graph
2333 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2335 bool updated = false;
2336 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs->begin();
2337 while (it != pending_rel_seqs->end()) {
2338 struct release_seq *pending = *it;
2339 ModelAction *acquire = pending->acquire;
2340 const ModelAction *read = pending->read;
2342 /* Only resolve sequences on the given location, if provided */
2343 if (location && read->get_location() != location) {
2348 const ModelAction *rf = read->get_reads_from();
2349 rel_heads_list_t release_heads;
2351 complete = release_seq_heads(rf, &release_heads, pending);
2352 for (unsigned int i = 0; i < release_heads.size(); i++) {
2353 if (!acquire->has_synchronized_with(release_heads[i])) {
2354 if (acquire->synchronize_with(release_heads[i]))
2357 set_bad_synchronization();
2362 /* Re-check all pending release sequences */
2363 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2364 /* Re-check read-acquire for mo_graph edges */
2365 if (acquire->is_read())
2366 work_queue->push_back(MOEdgeWorkEntry(acquire));
2368 /* propagate synchronization to later actions */
2369 action_list_t::reverse_iterator rit = action_trace->rbegin();
2370 for (; (*rit) != acquire; rit++) {
2371 ModelAction *propagate = *rit;
2372 if (acquire->happens_before(propagate)) {
2373 propagate->synchronize_with(acquire);
2374 /* Re-check 'propagate' for mo_graph edges */
2375 work_queue->push_back(MOEdgeWorkEntry(propagate));
2380 it = pending_rel_seqs->erase(it);
2381 snapshot_free(pending);
2387 // If we resolved promises or data races, see if we have realized a data race.
2394 * Performs various bookkeeping operations for the current ModelAction. For
2395 * instance, adds action to the per-object, per-thread action vector and to the
2396 * action trace list of all thread actions.
2398 * @param act is the ModelAction to add.
2400 void ModelChecker::add_action_to_lists(ModelAction *act)
2402 int tid = id_to_int(act->get_tid());
2403 ModelAction *uninit = NULL;
2405 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2406 if (list->empty() && act->is_atomic_var()) {
2407 uninit = get_uninitialized_action(act);
2408 uninit_id = id_to_int(uninit->get_tid());
2409 list->push_front(uninit);
2411 list->push_back(act);
2413 action_trace->push_back(act);
2415 action_trace->push_front(uninit);
2417 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2418 if (tid >= (int)vec->size())
2419 vec->resize(priv->next_thread_id);
2420 (*vec)[tid].push_back(act);
2422 (*vec)[uninit_id].push_front(uninit);
2424 if ((int)thrd_last_action->size() <= tid)
2425 thrd_last_action->resize(get_num_threads());
2426 (*thrd_last_action)[tid] = act;
2428 (*thrd_last_action)[uninit_id] = uninit;
2430 if (act->is_fence() && act->is_release()) {
2431 if ((int)thrd_last_fence_release->size() <= tid)
2432 thrd_last_fence_release->resize(get_num_threads());
2433 (*thrd_last_fence_release)[tid] = act;
2436 if (act->is_wait()) {
2437 void *mutex_loc = (void *) act->get_value();
2438 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2440 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2441 if (tid >= (int)vec->size())
2442 vec->resize(priv->next_thread_id);
2443 (*vec)[tid].push_back(act);
2448 * @brief Get the last action performed by a particular Thread
2449 * @param tid The thread ID of the Thread in question
2450 * @return The last action in the thread
2452 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2454 int threadid = id_to_int(tid);
2455 if (threadid < (int)thrd_last_action->size())
2456 return (*thrd_last_action)[id_to_int(tid)];
2462 * @brief Get the last fence release performed by a particular Thread
2463 * @param tid The thread ID of the Thread in question
2464 * @return The last fence release in the thread, if one exists; NULL otherwise
2466 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2468 int threadid = id_to_int(tid);
2469 if (threadid < (int)thrd_last_fence_release->size())
2470 return (*thrd_last_fence_release)[id_to_int(tid)];
2476 * Gets the last memory_order_seq_cst write (in the total global sequence)
2477 * performed on a particular object (i.e., memory location), not including the
2479 * @param curr The current ModelAction; also denotes the object location to
2481 * @return The last seq_cst write
2483 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2485 void *location = curr->get_location();
2486 action_list_t *list = get_safe_ptr_action(obj_map, location);
2487 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2488 action_list_t::reverse_iterator rit;
2489 for (rit = list->rbegin(); (*rit) != curr; rit++)
2491 rit++; /* Skip past curr */
2492 for ( ; rit != list->rend(); rit++)
2493 if ((*rit)->is_write() && (*rit)->is_seqcst())
2499 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2500 * performed in a particular thread, prior to a particular fence.
2501 * @param tid The ID of the thread to check
2502 * @param before_fence The fence from which to begin the search; if NULL, then
2503 * search for the most recent fence in the thread.
2504 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2506 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2508 /* All fences should have NULL location */
2509 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2510 action_list_t::reverse_iterator rit = list->rbegin();
2513 for (; rit != list->rend(); rit++)
2514 if (*rit == before_fence)
2517 ASSERT(*rit == before_fence);
2521 for (; rit != list->rend(); rit++)
2522 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2528 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2529 * location). This function identifies the mutex according to the current
2530 * action, which is presumed to perform on the same mutex.
2531 * @param curr The current ModelAction; also denotes the object location to
2533 * @return The last unlock operation
2535 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2537 void *location = curr->get_location();
2538 action_list_t *list = get_safe_ptr_action(obj_map, location);
2539 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2540 action_list_t::reverse_iterator rit;
2541 for (rit = list->rbegin(); rit != list->rend(); rit++)
2542 if ((*rit)->is_unlock() || (*rit)->is_wait())
2547 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2549 ModelAction *parent = get_last_action(tid);
2551 parent = get_thread(tid)->get_creation();
2556 * Returns the clock vector for a given thread.
2557 * @param tid The thread whose clock vector we want
2558 * @return Desired clock vector
2560 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2562 return get_parent_action(tid)->get_cv();
2566 * @brief Find the promise (if any) to resolve for the current action and
2567 * remove it from the pending promise vector
2568 * @param curr The current ModelAction. Should be a write.
2569 * @return The Promise to resolve, if any; otherwise NULL
2571 Promise * ModelChecker::pop_promise_to_resolve(const ModelAction *curr)
2573 for (unsigned int i = 0; i < promises->size(); i++)
2574 if (curr->get_node()->get_promise(i)) {
2575 Promise *ret = (*promises)[i];
2576 promises->erase(promises->begin() + i);
2583 * Resolve a Promise with a current write.
2584 * @param write The ModelAction that is fulfilling Promises
2585 * @param promise The Promise to resolve
2586 * @return True if the Promise was successfully resolved; false otherwise
2588 bool ModelChecker::resolve_promise(ModelAction *write, Promise *promise)
2590 ModelVector<ModelAction *> actions_to_check;
2592 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2593 ModelAction *read = promise->get_reader(i);
2594 read_from(read, write);
2595 actions_to_check.push_back(read);
2597 /* Make sure the promise's value matches the write's value */
2598 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2599 if (!mo_graph->resolvePromise(promise, write))
2600 priv->failed_promise = true;
2603 * @todo It is possible to end up in an inconsistent state, where a
2604 * "resolved" promise may still be referenced if
2605 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2607 * Note that the inconsistency only matters when dumping mo_graph to
2613 //Check whether reading these writes has made threads unable to
2615 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2616 ModelAction *read = actions_to_check[i];
2617 mo_check_promises(read, true);
2624 * Compute the set of promises that could potentially be satisfied by this
2625 * action. Note that the set computation actually appears in the Node, not in
2627 * @param curr The ModelAction that may satisfy promises
2629 void ModelChecker::compute_promises(ModelAction *curr)
2631 for (unsigned int i = 0; i < promises->size(); i++) {
2632 Promise *promise = (*promises)[i];
2633 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2636 bool satisfy = true;
2637 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2638 const ModelAction *act = promise->get_reader(j);
2639 if (act->happens_before(curr) ||
2640 act->could_synchronize_with(curr)) {
2646 curr->get_node()->set_promise(i);
2650 /** Checks promises in response to change in ClockVector Threads. */
2651 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2653 for (unsigned int i = 0; i < promises->size(); i++) {
2654 Promise *promise = (*promises)[i];
2655 if (!promise->thread_is_available(tid))
2657 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2658 const ModelAction *act = promise->get_reader(j);
2659 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2660 merge_cv->synchronized_since(act)) {
2661 if (promise->eliminate_thread(tid)) {
2662 /* Promise has failed */
2663 priv->failed_promise = true;
2671 void ModelChecker::check_promises_thread_disabled()
2673 for (unsigned int i = 0; i < promises->size(); i++) {
2674 Promise *promise = (*promises)[i];
2675 if (promise->has_failed()) {
2676 priv->failed_promise = true;
2683 * @brief Checks promises in response to addition to modification order for
2686 * We test whether threads are still available for satisfying promises after an
2687 * addition to our modification order constraints. Those that are unavailable
2688 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2689 * that promise has failed.
2691 * @param act The ModelAction which updated the modification order
2692 * @param is_read_check Should be true if act is a read and we must check for
2693 * updates to the store from which it read (there is a distinction here for
2694 * RMW's, which are both a load and a store)
2696 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2698 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2700 for (unsigned int i = 0; i < promises->size(); i++) {
2701 Promise *promise = (*promises)[i];
2703 // Is this promise on the same location?
2704 if (!promise->same_location(write))
2707 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2708 const ModelAction *pread = promise->get_reader(j);
2709 if (!pread->happens_before(act))
2711 if (mo_graph->checkPromise(write, promise)) {
2712 priv->failed_promise = true;
2718 // Don't do any lookups twice for the same thread
2719 if (!promise->thread_is_available(act->get_tid()))
2722 if (mo_graph->checkReachable(promise, write)) {
2723 if (mo_graph->checkPromise(write, promise)) {
2724 priv->failed_promise = true;
2732 * Compute the set of writes that may break the current pending release
2733 * sequence. This information is extracted from previou release sequence
2736 * @param curr The current ModelAction. Must be a release sequence fixup
2739 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2741 if (pending_rel_seqs->empty())
2744 struct release_seq *pending = pending_rel_seqs->back();
2745 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2746 const ModelAction *write = pending->writes[i];
2747 curr->get_node()->add_relseq_break(write);
2750 /* NULL means don't break the sequence; just synchronize */
2751 curr->get_node()->add_relseq_break(NULL);
2755 * Build up an initial set of all past writes that this 'read' action may read
2756 * from, as well as any previously-observed future values that must still be valid.
2758 * @param curr is the current ModelAction that we are exploring; it must be a
2761 void ModelChecker::build_may_read_from(ModelAction *curr)
2763 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2765 ASSERT(curr->is_read());
2767 ModelAction *last_sc_write = NULL;
2769 if (curr->is_seqcst())
2770 last_sc_write = get_last_seq_cst_write(curr);
2772 /* Iterate over all threads */
2773 for (i = 0; i < thrd_lists->size(); i++) {
2774 /* Iterate over actions in thread, starting from most recent */
2775 action_list_t *list = &(*thrd_lists)[i];
2776 action_list_t::reverse_iterator rit;
2777 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2778 ModelAction *act = *rit;
2780 /* Only consider 'write' actions */
2781 if (!act->is_write() || act == curr)
2784 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2785 bool allow_read = true;
2787 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2789 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2793 /* Only add feasible reads */
2794 mo_graph->startChanges();
2795 r_modification_order(curr, act);
2796 if (!is_infeasible())
2797 curr->get_node()->add_read_from_past(act);
2798 mo_graph->rollbackChanges();
2801 /* Include at most one act per-thread that "happens before" curr */
2802 if (act->happens_before(curr))
2807 /* Inherit existing, promised future values */
2808 for (i = 0; i < promises->size(); i++) {
2809 const Promise *promise = (*promises)[i];
2810 const ModelAction *promise_read = promise->get_reader(0);
2811 if (promise_read->same_var(curr)) {
2812 /* Only add feasible future-values */
2813 mo_graph->startChanges();
2814 r_modification_order(curr, promise);
2815 if (!is_infeasible())
2816 curr->get_node()->add_read_from_promise(promise_read);
2817 mo_graph->rollbackChanges();
2821 /* We may find no valid may-read-from only if the execution is doomed */
2822 if (!curr->get_node()->read_from_size()) {
2823 priv->no_valid_reads = true;
2827 if (DBG_ENABLED()) {
2828 model_print("Reached read action:\n");
2830 model_print("Printing read_from_past\n");
2831 curr->get_node()->print_read_from_past();
2832 model_print("End printing read_from_past\n");
2836 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2838 for ( ; write != NULL; write = write->get_reads_from()) {
2839 /* UNINIT actions don't have a Node, and they never sleep */
2840 if (write->is_uninitialized())
2842 Node *prevnode = write->get_node()->get_parent();
2844 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2845 if (write->is_release() && thread_sleep)
2847 if (!write->is_rmw())
2854 * @brief Get an action representing an uninitialized atomic
2856 * This function may create a new one or try to retrieve one from the NodeStack
2858 * @param curr The current action, which prompts the creation of an UNINIT action
2859 * @return A pointer to the UNINIT ModelAction
2861 ModelAction * ModelChecker::get_uninitialized_action(const ModelAction *curr) const
2863 Node *node = curr->get_node();
2864 ModelAction *act = node->get_uninit_action();
2866 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), model->params.uninitvalue, model_thread);
2867 node->set_uninit_action(act);
2869 act->create_cv(NULL);
2873 static void print_list(action_list_t *list)
2875 action_list_t::iterator it;
2877 model_print("---------------------------------------------------------------------\n");
2879 unsigned int hash = 0;
2881 for (it = list->begin(); it != list->end(); it++) {
2882 const ModelAction *act = *it;
2883 if (act->get_seq_number() > 0)
2885 hash = hash^(hash<<3)^((*it)->hash());
2887 model_print("HASH %u\n", hash);
2888 model_print("---------------------------------------------------------------------\n");
2891 #if SUPPORT_MOD_ORDER_DUMP
2892 void ModelChecker::dumpGraph(char *filename) const
2895 sprintf(buffer, "%s.dot", filename);
2896 FILE *file = fopen(buffer, "w");
2897 fprintf(file, "digraph %s {\n", filename);
2898 mo_graph->dumpNodes(file);
2899 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2901 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2902 ModelAction *act = *it;
2903 if (act->is_read()) {
2904 mo_graph->dot_print_node(file, act);
2905 if (act->get_reads_from())
2906 mo_graph->dot_print_edge(file,
2907 act->get_reads_from(),
2909 "label=\"rf\", color=red, weight=2");
2911 mo_graph->dot_print_edge(file,
2912 act->get_reads_from_promise(),
2914 "label=\"rf\", color=red");
2916 if (thread_array[act->get_tid()]) {
2917 mo_graph->dot_print_edge(file,
2918 thread_array[id_to_int(act->get_tid())],
2920 "label=\"sb\", color=blue, weight=400");
2923 thread_array[act->get_tid()] = act;
2925 fprintf(file, "}\n");
2926 model_free(thread_array);
2931 /** @brief Prints an execution trace summary. */
2932 void ModelChecker::print_summary() const
2934 #if SUPPORT_MOD_ORDER_DUMP
2935 char buffername[100];
2936 sprintf(buffername, "exec%04u", stats.num_total);
2937 mo_graph->dumpGraphToFile(buffername);
2938 sprintf(buffername, "graph%04u", stats.num_total);
2939 dumpGraph(buffername);
2942 model_print("Execution %d:", stats.num_total);
2943 if (isfeasibleprefix()) {
2944 if (scheduler->all_threads_sleeping())
2945 model_print(" SLEEP-SET REDUNDANT");
2948 print_infeasibility(" INFEASIBLE");
2949 print_list(action_trace);
2951 if (!promises->empty()) {
2952 model_print("Pending promises:\n");
2953 for (unsigned int i = 0; i < promises->size(); i++) {
2954 model_print(" [P%u] ", i);
2955 (*promises)[i]->print();
2962 * Add a Thread to the system for the first time. Should only be called once
2964 * @param t The Thread to add
2966 void ModelChecker::add_thread(Thread *t)
2968 thread_map->put(id_to_int(t->get_id()), t);
2969 scheduler->add_thread(t);
2973 * @brief Get a Thread reference by its ID
2974 * @param tid The Thread's ID
2975 * @return A Thread reference
2977 Thread * ModelChecker::get_thread(thread_id_t tid) const
2979 return thread_map->get(id_to_int(tid));
2983 * @brief Get a reference to the Thread in which a ModelAction was executed
2984 * @param act The ModelAction
2985 * @return A Thread reference
2987 Thread * ModelChecker::get_thread(const ModelAction *act) const
2989 return get_thread(act->get_tid());
2993 * @brief Get a Promise's "promise number"
2995 * A "promise number" is an index number that is unique to a promise, valid
2996 * only for a specific snapshot of an execution trace. Promises may come and go
2997 * as they are generated an resolved, so an index only retains meaning for the
3000 * @param promise The Promise to check
3001 * @return The promise index, if the promise still is valid; otherwise -1
3003 int ModelChecker::get_promise_number(const Promise *promise) const
3005 for (unsigned int i = 0; i < promises->size(); i++)
3006 if ((*promises)[i] == promise)
3013 * @brief Check if a Thread is currently enabled
3014 * @param t The Thread to check
3015 * @return True if the Thread is currently enabled
3017 bool ModelChecker::is_enabled(Thread *t) const
3019 return scheduler->is_enabled(t);
3023 * @brief Check if a Thread is currently enabled
3024 * @param tid The ID of the Thread to check
3025 * @return True if the Thread is currently enabled
3027 bool ModelChecker::is_enabled(thread_id_t tid) const
3029 return scheduler->is_enabled(tid);
3033 * Switch from a model-checker context to a user-thread context. This is the
3034 * complement of ModelChecker::switch_to_master and must be called from the
3035 * model-checker context
3037 * @param thread The user-thread to switch to
3039 void ModelChecker::switch_from_master(Thread *thread)
3041 scheduler->set_current_thread(thread);
3042 Thread::swap(&system_context, thread);
3046 * Switch from a user-context to the "master thread" context (a.k.a. system
3047 * context). This switch is made with the intention of exploring a particular
3048 * model-checking action (described by a ModelAction object). Must be called
3049 * from a user-thread context.
3051 * @param act The current action that will be explored. May be NULL only if
3052 * trace is exiting via an assertion (see ModelChecker::set_assert and
3053 * ModelChecker::has_asserted).
3054 * @return Return the value returned by the current action
3056 uint64_t ModelChecker::switch_to_master(ModelAction *act)
3059 Thread *old = thread_current();
3060 scheduler->set_current_thread(NULL);
3061 ASSERT(!old->get_pending());
3062 old->set_pending(act);
3063 if (Thread::swap(old, &system_context) < 0) {
3064 perror("swap threads");
3067 return old->get_return_value();
3071 * Takes the next step in the execution, if possible.
3072 * @param curr The current step to take
3073 * @return Returns the next Thread to run, if any; NULL if this execution
3076 Thread * ModelChecker::take_step(ModelAction *curr)
3078 Thread *curr_thrd = get_thread(curr);
3079 ASSERT(curr_thrd->get_state() == THREAD_READY);
3081 curr = check_current_action(curr);
3083 /* Infeasible -> don't take any more steps */
3084 if (is_infeasible())
3086 else if (isfeasibleprefix() && have_bug_reports()) {
3091 if (params.bound != 0 && priv->used_sequence_numbers > params.bound)
3094 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
3095 scheduler->remove_thread(curr_thrd);
3097 Thread *next_thrd = NULL;
3099 next_thrd = action_select_next_thread(curr);
3101 next_thrd = get_next_thread();
3103 DEBUG("(%d, %d)\n", curr_thrd ? id_to_int(curr_thrd->get_id()) : -1,
3104 next_thrd ? id_to_int(next_thrd->get_id()) : -1);
3109 /** Wrapper to run the user's main function, with appropriate arguments */
3110 void user_main_wrapper(void *)
3112 user_main(model->params.argc, model->params.argv);
3115 /** @brief Run ModelChecker for the user program */
3116 void ModelChecker::run()
3120 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL, NULL);
3125 * Stash next pending action(s) for thread(s). There
3126 * should only need to stash one thread's action--the
3127 * thread which just took a step--plus the first step
3128 * for any newly-created thread
3130 for (unsigned int i = 0; i < get_num_threads(); i++) {
3131 thread_id_t tid = int_to_id(i);
3132 Thread *thr = get_thread(tid);
3133 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
3134 switch_from_master(thr);
3135 if (thr->is_waiting_on(thr))
3136 assert_bug("Deadlock detected");
3140 /* Catch assertions from prior take_step or from
3141 * between-ModelAction bugs (e.g., data races) */
3145 /* Consume the next action for a Thread */
3146 ModelAction *curr = t->get_pending();
3147 t->set_pending(NULL);
3148 t = take_step(curr);
3149 } while (t && !t->is_model_thread());
3152 * Launch end-of-execution release sequence fixups only when
3153 * the execution is otherwise feasible AND there are:
3155 * (1) pending release sequences
3156 * (2) pending assertions that could be invalidated by a change
3157 * in clock vectors (i.e., data races)
3158 * (3) no pending promises
3160 while (!pending_rel_seqs->empty() &&
3161 is_feasible_prefix_ignore_relseq() &&
3162 !unrealizedraces.empty()) {
3163 model_print("*** WARNING: release sequence fixup action "
3164 "(%zu pending release seuqence(s)) ***\n",
3165 pending_rel_seqs->size());
3166 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
3167 std::memory_order_seq_cst, NULL, VALUE_NONE,
3171 } while (next_execution());
3173 model_print("******* Model-checking complete: *******\n");