10 #include "nodestack.h"
13 #include "clockvector.h"
14 #include "cyclegraph.h"
17 #include "threads-model.h"
18 #include "bugmessage.h"
20 #define INITIAL_THREAD_ID 0
23 * Structure for holding small ModelChecker members that should be snapshotted
25 struct model_snapshot_members {
26 model_snapshot_members() :
27 /* First thread created will have id INITIAL_THREAD_ID */
28 next_thread_id(INITIAL_THREAD_ID),
29 used_sequence_numbers(0),
32 failed_promise(false),
33 hard_failed_promise(false),
34 too_many_reads(false),
35 no_valid_reads(false),
36 bad_synchronization(false),
40 ~model_snapshot_members() {
41 for (unsigned int i = 0; i < bugs.size(); i++)
46 unsigned int next_thread_id;
47 modelclock_t used_sequence_numbers;
48 ModelAction *next_backtrack;
49 SnapVector<bug_message *> bugs;
51 bool hard_failed_promise;
54 /** @brief Incorrectly-ordered synchronization was made */
55 bool bad_synchronization;
61 /** @brief Constructor */
62 ModelExecution::ModelExecution(ModelChecker *m,
63 const struct model_params *params,
65 NodeStack *node_stack) :
70 thread_map(2), /* We'll always need at least 2 threads */
72 condvar_waiters_map(),
78 thrd_last_fence_release(),
79 node_stack(node_stack),
80 priv(new struct model_snapshot_members()),
81 mo_graph(new CycleGraph())
83 /* Initialize a model-checker thread, for special ModelActions */
84 model_thread = new Thread(get_next_id());
85 add_thread(model_thread);
86 scheduler->register_engine(this);
87 node_stack->register_engine(this);
90 /** @brief Destructor */
91 ModelExecution::~ModelExecution()
93 for (unsigned int i = 0; i < get_num_threads(); i++)
94 delete get_thread(int_to_id(i));
96 for (unsigned int i = 0; i < promises.size(); i++)
103 int ModelExecution::get_execution_number() const
105 return model->get_execution_number();
108 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
110 action_list_t *tmp = hash->get(ptr);
112 tmp = new action_list_t();
118 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
120 SnapVector<action_list_t> *tmp = hash->get(ptr);
122 tmp = new SnapVector<action_list_t>();
128 action_list_t * ModelExecution::get_actions_on_obj(void * obj, thread_id_t tid) const
130 SnapVector<action_list_t> *wrv = obj_thrd_map.get(obj);
133 unsigned int thread=id_to_int(tid);
134 if (thread < wrv->size())
135 return &(*wrv)[thread];
140 /** @return a thread ID for a new Thread */
141 thread_id_t ModelExecution::get_next_id()
143 return priv->next_thread_id++;
146 /** @return the number of user threads created during this execution */
147 unsigned int ModelExecution::get_num_threads() const
149 return priv->next_thread_id;
152 /** @return a sequence number for a new ModelAction */
153 modelclock_t ModelExecution::get_next_seq_num()
155 return ++priv->used_sequence_numbers;
159 * @brief Should the current action wake up a given thread?
161 * @param curr The current action
162 * @param thread The thread that we might wake up
163 * @return True, if we should wake up the sleeping thread; false otherwise
165 bool ModelExecution::should_wake_up(const ModelAction *curr, const Thread *thread) const
167 const ModelAction *asleep = thread->get_pending();
168 /* Don't allow partial RMW to wake anyone up */
171 /* Synchronizing actions may have been backtracked */
172 if (asleep->could_synchronize_with(curr))
174 /* All acquire/release fences and fence-acquire/store-release */
175 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
177 /* Fence-release + store can awake load-acquire on the same location */
178 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
179 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
180 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
186 void ModelExecution::wake_up_sleeping_actions(ModelAction *curr)
188 for (unsigned int i = 0; i < get_num_threads(); i++) {
189 Thread *thr = get_thread(int_to_id(i));
190 if (scheduler->is_sleep_set(thr)) {
191 if (should_wake_up(curr, thr))
192 /* Remove this thread from sleep set */
193 scheduler->remove_sleep(thr);
198 /** @brief Alert the model-checker that an incorrectly-ordered
199 * synchronization was made */
200 void ModelExecution::set_bad_synchronization()
202 priv->bad_synchronization = true;
205 bool ModelExecution::assert_bug(const char *msg)
207 priv->bugs.push_back(new bug_message(msg));
209 if (isfeasibleprefix()) {
216 /** @return True, if any bugs have been reported for this execution */
217 bool ModelExecution::have_bug_reports() const
219 return priv->bugs.size() != 0;
222 SnapVector<bug_message *> * ModelExecution::get_bugs() const
228 * Check whether the current trace has triggered an assertion which should halt
231 * @return True, if the execution should be aborted; false otherwise
233 bool ModelExecution::has_asserted() const
235 return priv->asserted;
239 * Trigger a trace assertion which should cause this execution to be halted.
240 * This can be due to a detected bug or due to an infeasibility that should
243 void ModelExecution::set_assert()
245 priv->asserted = true;
249 * Check if we are in a deadlock. Should only be called at the end of an
250 * execution, although it should not give false positives in the middle of an
251 * execution (there should be some ENABLED thread).
253 * @return True if program is in a deadlock; false otherwise
255 bool ModelExecution::is_deadlocked() const
257 bool blocking_threads = false;
258 for (unsigned int i = 0; i < get_num_threads(); i++) {
259 thread_id_t tid = int_to_id(i);
262 Thread *t = get_thread(tid);
263 if (!t->is_model_thread() && t->get_pending())
264 blocking_threads = true;
266 return blocking_threads;
270 * @brief Check if we are yield-blocked
272 * A program can be "yield-blocked" if all threads are ready to execute a
275 * @return True if the program is yield-blocked; false otherwise
277 bool ModelExecution::is_yieldblocked() const
279 if (!params->yieldblock)
282 for (unsigned int i = 0; i < get_num_threads(); i++) {
283 thread_id_t tid = int_to_id(i);
284 Thread *t = get_thread(tid);
285 if (t->get_pending() && t->get_pending()->is_yield())
292 * Check if this is a complete execution. That is, have all thread completed
293 * execution (rather than exiting because sleep sets have forced a redundant
296 * @return True if the execution is complete.
298 bool ModelExecution::is_complete_execution() const
300 if (is_yieldblocked())
302 for (unsigned int i = 0; i < get_num_threads(); i++)
303 if (is_enabled(int_to_id(i)))
309 * @brief Find the last fence-related backtracking conflict for a ModelAction
311 * This function performs the search for the most recent conflicting action
312 * against which we should perform backtracking, as affected by fence
313 * operations. This includes pairs of potentially-synchronizing actions which
314 * occur due to fence-acquire or fence-release, and hence should be explored in
315 * the opposite execution order.
317 * @param act The current action
318 * @return The most recent action which conflicts with act due to fences
320 ModelAction * ModelExecution::get_last_fence_conflict(ModelAction *act) const
322 /* Only perform release/acquire fence backtracking for stores */
323 if (!act->is_write())
326 /* Find a fence-release (or, act is a release) */
327 ModelAction *last_release;
328 if (act->is_release())
331 last_release = get_last_fence_release(act->get_tid());
335 /* Skip past the release */
336 const action_list_t *list = &action_trace;
337 action_list_t::const_reverse_iterator rit;
338 for (rit = list->rbegin(); rit != list->rend(); rit++)
339 if (*rit == last_release)
341 ASSERT(rit != list->rend());
346 * load --sb-> fence-acquire */
347 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
348 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
349 bool found_acquire_fences = false;
350 for ( ; rit != list->rend(); rit++) {
351 ModelAction *prev = *rit;
352 if (act->same_thread(prev))
355 int tid = id_to_int(prev->get_tid());
357 if (prev->is_read() && act->same_var(prev)) {
358 if (prev->is_acquire()) {
359 /* Found most recent load-acquire, don't need
360 * to search for more fences */
361 if (!found_acquire_fences)
364 prior_loads[tid] = prev;
367 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
368 found_acquire_fences = true;
369 acquire_fences[tid] = prev;
373 ModelAction *latest_backtrack = NULL;
374 for (unsigned int i = 0; i < acquire_fences.size(); i++)
375 if (acquire_fences[i] && prior_loads[i])
376 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
377 latest_backtrack = acquire_fences[i];
378 return latest_backtrack;
382 * @brief Find the last backtracking conflict for a ModelAction
384 * This function performs the search for the most recent conflicting action
385 * against which we should perform backtracking. This primary includes pairs of
386 * synchronizing actions which should be explored in the opposite execution
389 * @param act The current action
390 * @return The most recent action which conflicts with act
392 ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
394 switch (act->get_type()) {
396 /* Only seq-cst fences can (directly) cause backtracking */
397 if (!act->is_seqcst())
402 ModelAction *ret = NULL;
404 /* linear search: from most recent to oldest */
405 action_list_t *list = obj_map.get(act->get_location());
406 action_list_t::reverse_iterator rit;
407 for (rit = list->rbegin(); rit != list->rend(); rit++) {
408 ModelAction *prev = *rit;
411 if (prev->could_synchronize_with(act)) {
417 ModelAction *ret2 = get_last_fence_conflict(act);
427 case ATOMIC_TRYLOCK: {
428 /* linear search: from most recent to oldest */
429 action_list_t *list = obj_map.get(act->get_location());
430 action_list_t::reverse_iterator rit;
431 for (rit = list->rbegin(); rit != list->rend(); rit++) {
432 ModelAction *prev = *rit;
433 if (act->is_conflicting_lock(prev))
438 case ATOMIC_UNLOCK: {
439 /* linear search: from most recent to oldest */
440 action_list_t *list = obj_map.get(act->get_location());
441 action_list_t::reverse_iterator rit;
442 for (rit = list->rbegin(); rit != list->rend(); rit++) {
443 ModelAction *prev = *rit;
444 if (!act->same_thread(prev) && prev->is_failed_trylock())
450 /* linear search: from most recent to oldest */
451 action_list_t *list = obj_map.get(act->get_location());
452 action_list_t::reverse_iterator rit;
453 for (rit = list->rbegin(); rit != list->rend(); rit++) {
454 ModelAction *prev = *rit;
455 if (!act->same_thread(prev) && prev->is_failed_trylock())
457 if (!act->same_thread(prev) && prev->is_notify())
463 case ATOMIC_NOTIFY_ALL:
464 case ATOMIC_NOTIFY_ONE: {
465 /* linear search: from most recent to oldest */
466 action_list_t *list = obj_map.get(act->get_location());
467 action_list_t::reverse_iterator rit;
468 for (rit = list->rbegin(); rit != list->rend(); rit++) {
469 ModelAction *prev = *rit;
470 if (!act->same_thread(prev) && prev->is_wait())
481 /** This method finds backtracking points where we should try to
482 * reorder the parameter ModelAction against.
484 * @param the ModelAction to find backtracking points for.
486 void ModelExecution::set_backtracking(ModelAction *act)
488 Thread *t = get_thread(act);
489 ModelAction *prev = get_last_conflict(act);
493 Node *node = prev->get_node()->get_parent();
495 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
496 int low_tid, high_tid;
497 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
498 low_tid = id_to_int(act->get_tid());
499 high_tid = low_tid + 1;
502 high_tid = get_num_threads();
505 for (int i = low_tid; i < high_tid; i++) {
506 thread_id_t tid = int_to_id(i);
508 /* Make sure this thread can be enabled here. */
509 if (i >= node->get_num_threads())
512 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
513 /* Don't backtrack into a point where the thread is disabled or sleeping. */
514 if (node->enabled_status(tid) != THREAD_ENABLED)
517 /* Check if this has been explored already */
518 if (node->has_been_explored(tid))
521 /* See if fairness allows */
522 if (params->fairwindow != 0 && !node->has_priority(tid)) {
524 for (int t = 0; t < node->get_num_threads(); t++) {
525 thread_id_t tother = int_to_id(t);
526 if (node->is_enabled(tother) && node->has_priority(tother)) {
535 /* See if CHESS-like yield fairness allows */
536 if (params->yieldon) {
538 for (int t = 0; t < node->get_num_threads(); t++) {
539 thread_id_t tother = int_to_id(t);
540 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
549 /* Cache the latest backtracking point */
550 set_latest_backtrack(prev);
552 /* If this is a new backtracking point, mark the tree */
553 if (!node->set_backtrack(tid))
555 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
556 id_to_int(prev->get_tid()),
557 id_to_int(t->get_id()));
566 * @brief Cache the a backtracking point as the "most recent", if eligible
568 * Note that this does not prepare the NodeStack for this backtracking
569 * operation, it only caches the action on a per-execution basis
571 * @param act The operation at which we should explore a different next action
572 * (i.e., backtracking point)
573 * @return True, if this action is now the most recent backtracking point;
576 bool ModelExecution::set_latest_backtrack(ModelAction *act)
578 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
579 priv->next_backtrack = act;
586 * Returns last backtracking point. The model checker will explore a different
587 * path for this point in the next execution.
588 * @return The ModelAction at which the next execution should diverge.
590 ModelAction * ModelExecution::get_next_backtrack()
592 ModelAction *next = priv->next_backtrack;
593 priv->next_backtrack = NULL;
598 * Processes a read model action.
599 * @param curr is the read model action to process.
600 * @return True if processing this read updates the mo_graph.
602 bool ModelExecution::process_read(ModelAction *curr)
604 Node *node = curr->get_node();
606 bool updated = false;
607 switch (node->get_read_from_status()) {
608 case READ_FROM_PAST: {
609 const ModelAction *rf = node->get_read_from_past();
612 mo_graph->startChanges();
614 ASSERT(!is_infeasible());
615 if (!check_recency(curr, rf)) {
616 if (node->increment_read_from()) {
617 mo_graph->rollbackChanges();
620 priv->too_many_reads = true;
624 updated = r_modification_order(curr, rf);
626 mo_graph->commitChanges();
627 mo_check_promises(curr, true);
630 case READ_FROM_PROMISE: {
631 Promise *promise = curr->get_node()->get_read_from_promise();
632 if (promise->add_reader(curr))
633 priv->failed_promise = true;
634 curr->set_read_from_promise(promise);
635 mo_graph->startChanges();
636 if (!check_recency(curr, promise))
637 priv->too_many_reads = true;
638 updated = r_modification_order(curr, promise);
639 mo_graph->commitChanges();
642 case READ_FROM_FUTURE: {
643 /* Read from future value */
644 struct future_value fv = node->get_future_value();
645 Promise *promise = new Promise(this, curr, fv);
646 curr->set_read_from_promise(promise);
647 promises.push_back(promise);
648 mo_graph->startChanges();
649 updated = r_modification_order(curr, promise);
650 mo_graph->commitChanges();
656 get_thread(curr)->set_return_value(curr->get_return_value());
662 * Processes a lock, trylock, or unlock model action. @param curr is
663 * the read model action to process.
665 * The try lock operation checks whether the lock is taken. If not,
666 * it falls to the normal lock operation case. If so, it returns
669 * The lock operation has already been checked that it is enabled, so
670 * it just grabs the lock and synchronizes with the previous unlock.
672 * The unlock operation has to re-enable all of the threads that are
673 * waiting on the lock.
675 * @return True if synchronization was updated; false otherwise
677 bool ModelExecution::process_mutex(ModelAction *curr)
679 std::mutex *mutex = curr->get_mutex();
680 struct std::mutex_state *state = NULL;
683 state = mutex->get_state();
685 switch (curr->get_type()) {
686 case ATOMIC_TRYLOCK: {
687 bool success = !state->locked;
688 curr->set_try_lock(success);
690 get_thread(curr)->set_return_value(0);
693 get_thread(curr)->set_return_value(1);
695 //otherwise fall into the lock case
697 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
698 assert_bug("Lock access before initialization");
699 state->locked = get_thread(curr);
700 ModelAction *unlock = get_last_unlock(curr);
701 //synchronize with the previous unlock statement
702 if (unlock != NULL) {
703 synchronize(unlock, curr);
709 case ATOMIC_UNLOCK: {
710 /* wake up the other threads */
711 for (unsigned int i = 0; i < get_num_threads(); i++) {
712 Thread *t = get_thread(int_to_id(i));
713 Thread *curr_thrd = get_thread(curr);
714 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
718 /* unlock the lock - after checking who was waiting on it */
719 state->locked = NULL;
721 if (!curr->is_wait())
722 break; /* The rest is only for ATOMIC_WAIT */
724 /* Should we go to sleep? (simulate spurious failures) */
725 if (curr->get_node()->get_misc() == 0) {
726 get_safe_ptr_action(&condvar_waiters_map, curr->get_location())->push_back(curr);
728 scheduler->sleep(get_thread(curr));
732 case ATOMIC_NOTIFY_ALL: {
733 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
734 //activate all the waiting threads
735 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
736 scheduler->wake(get_thread(*rit));
741 case ATOMIC_NOTIFY_ONE: {
742 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
743 int wakeupthread = curr->get_node()->get_misc();
744 action_list_t::iterator it = waiters->begin();
745 advance(it, wakeupthread);
746 scheduler->wake(get_thread(*it));
758 * @brief Check if the current pending promises allow a future value to be sent
760 * It is unsafe to pass a future value back if there exists a pending promise Pr
763 * reader --exec-> Pr --exec-> writer
765 * If such Pr exists, we must save the pending future value until Pr is
768 * @param writer The operation which sends the future value. Must be a write.
769 * @param reader The operation which will observe the value. Must be a read.
770 * @return True if the future value can be sent now; false if it must wait.
772 bool ModelExecution::promises_may_allow(const ModelAction *writer,
773 const ModelAction *reader) const
775 for (int i = promises.size() - 1; i >= 0; i--) {
776 ModelAction *pr = promises[i]->get_reader(0);
777 //reader is after promise...doesn't cross any promise
780 //writer is after promise, reader before...bad...
788 * @brief Add a future value to a reader
790 * This function performs a few additional checks to ensure that the future
791 * value can be feasibly observed by the reader
793 * @param writer The operation whose value is sent. Must be a write.
794 * @param reader The read operation which may read the future value. Must be a read.
796 void ModelExecution::add_future_value(const ModelAction *writer, ModelAction *reader)
798 /* Do more ambitious checks now that mo is more complete */
799 if (!mo_may_allow(writer, reader))
802 Node *node = reader->get_node();
804 /* Find an ancestor thread which exists at the time of the reader */
805 Thread *write_thread = get_thread(writer);
806 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
807 write_thread = write_thread->get_parent();
809 struct future_value fv = {
810 writer->get_write_value(),
811 writer->get_seq_number() + params->maxfuturedelay,
812 write_thread->get_id(),
814 if (node->add_future_value(fv))
815 set_latest_backtrack(reader);
819 * Process a write ModelAction
820 * @param curr The ModelAction to process
821 * @param work The work queue, for adding fixup work
822 * @return True if the mo_graph was updated or promises were resolved
824 bool ModelExecution::process_write(ModelAction *curr, work_queue_t *work)
826 /* Readers to which we may send our future value */
827 ModelVector<ModelAction *> send_fv;
829 const ModelAction *earliest_promise_reader;
830 bool updated_promises = false;
832 bool updated_mod_order = w_modification_order(curr, &send_fv);
833 Promise *promise = pop_promise_to_resolve(curr);
836 earliest_promise_reader = promise->get_reader(0);
837 updated_promises = resolve_promise(curr, promise, work);
839 earliest_promise_reader = NULL;
841 for (unsigned int i = 0; i < send_fv.size(); i++) {
842 ModelAction *read = send_fv[i];
844 /* Don't send future values to reads after the Promise we resolve */
845 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
846 /* Check if future value can be sent immediately */
847 if (promises_may_allow(curr, read)) {
848 add_future_value(curr, read);
850 futurevalues.push_back(PendingFutureValue(curr, read));
855 /* Check the pending future values */
856 for (int i = (int)futurevalues.size() - 1; i >= 0; i--) {
857 struct PendingFutureValue pfv = futurevalues[i];
858 if (promises_may_allow(pfv.writer, pfv.reader)) {
859 add_future_value(pfv.writer, pfv.reader);
860 futurevalues.erase(futurevalues.begin() + i);
864 mo_graph->commitChanges();
865 mo_check_promises(curr, false);
867 get_thread(curr)->set_return_value(VALUE_NONE);
868 return updated_mod_order || updated_promises;
872 * Process a fence ModelAction
873 * @param curr The ModelAction to process
874 * @return True if synchronization was updated
876 bool ModelExecution::process_fence(ModelAction *curr)
879 * fence-relaxed: no-op
880 * fence-release: only log the occurence (not in this function), for
881 * use in later synchronization
882 * fence-acquire (this function): search for hypothetical release
884 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
886 bool updated = false;
887 if (curr->is_acquire()) {
888 action_list_t *list = &action_trace;
889 action_list_t::reverse_iterator rit;
890 /* Find X : is_read(X) && X --sb-> curr */
891 for (rit = list->rbegin(); rit != list->rend(); rit++) {
892 ModelAction *act = *rit;
895 if (act->get_tid() != curr->get_tid())
897 /* Stop at the beginning of the thread */
898 if (act->is_thread_start())
900 /* Stop once we reach a prior fence-acquire */
901 if (act->is_fence() && act->is_acquire())
905 /* read-acquire will find its own release sequences */
906 if (act->is_acquire())
909 /* Establish hypothetical release sequences */
910 rel_heads_list_t release_heads;
911 get_release_seq_heads(curr, act, &release_heads);
912 for (unsigned int i = 0; i < release_heads.size(); i++)
913 synchronize(release_heads[i], curr);
914 if (release_heads.size() != 0)
922 * @brief Process the current action for thread-related activity
924 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
925 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
926 * synchronization, etc. This function is a no-op for non-THREAD actions
927 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
929 * @param curr The current action
930 * @return True if synchronization was updated or a thread completed
932 bool ModelExecution::process_thread_action(ModelAction *curr)
934 bool updated = false;
936 switch (curr->get_type()) {
937 case THREAD_CREATE: {
938 thrd_t *thrd = (thrd_t *)curr->get_location();
939 struct thread_params *params = (struct thread_params *)curr->get_value();
940 Thread *th = new Thread(get_next_id(), thrd, params->func, params->arg, get_thread(curr));
942 th->set_creation(curr);
943 /* Promises can be satisfied by children */
944 for (unsigned int i = 0; i < promises.size(); i++) {
945 Promise *promise = promises[i];
946 if (promise->thread_is_available(curr->get_tid()))
947 promise->add_thread(th->get_id());
952 Thread *blocking = curr->get_thread_operand();
953 ModelAction *act = get_last_action(blocking->get_id());
954 synchronize(act, curr);
955 updated = true; /* trigger rel-seq checks */
958 case THREAD_FINISH: {
959 Thread *th = get_thread(curr);
960 /* Wake up any joining threads */
961 for (unsigned int i = 0; i < get_num_threads(); i++) {
962 Thread *waiting = get_thread(int_to_id(i));
963 if (waiting->waiting_on() == th &&
964 waiting->get_pending()->is_thread_join())
965 scheduler->wake(waiting);
968 /* Completed thread can't satisfy promises */
969 for (unsigned int i = 0; i < promises.size(); i++) {
970 Promise *promise = promises[i];
971 if (promise->thread_is_available(th->get_id()))
972 if (promise->eliminate_thread(th->get_id()))
973 priv->failed_promise = true;
975 updated = true; /* trigger rel-seq checks */
979 check_promises(curr->get_tid(), NULL, curr->get_cv());
990 * @brief Process the current action for release sequence fixup activity
992 * Performs model-checker release sequence fixups for the current action,
993 * forcing a single pending release sequence to break (with a given, potential
994 * "loose" write) or to complete (i.e., synchronize). If a pending release
995 * sequence forms a complete release sequence, then we must perform the fixup
996 * synchronization, mo_graph additions, etc.
998 * @param curr The current action; must be a release sequence fixup action
999 * @param work_queue The work queue to which to add work items as they are
1002 void ModelExecution::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1004 const ModelAction *write = curr->get_node()->get_relseq_break();
1005 struct release_seq *sequence = pending_rel_seqs.back();
1006 pending_rel_seqs.pop_back();
1008 ModelAction *acquire = sequence->acquire;
1009 const ModelAction *rf = sequence->rf;
1010 const ModelAction *release = sequence->release;
1014 ASSERT(release->same_thread(rf));
1016 if (write == NULL) {
1018 * @todo Forcing a synchronization requires that we set
1019 * modification order constraints. For instance, we can't allow
1020 * a fixup sequence in which two separate read-acquire
1021 * operations read from the same sequence, where the first one
1022 * synchronizes and the other doesn't. Essentially, we can't
1023 * allow any writes to insert themselves between 'release' and
1027 /* Must synchronize */
1028 if (!synchronize(release, acquire))
1031 /* Propagate the changed clock vector */
1032 propagate_clockvector(acquire, work_queue);
1034 /* Break release sequence with new edges:
1035 * release --mo--> write --mo--> rf */
1036 mo_graph->addEdge(release, write);
1037 mo_graph->addEdge(write, rf);
1040 /* See if we have realized a data race */
1045 * Initialize the current action by performing one or more of the following
1046 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1047 * in the NodeStack, manipulating backtracking sets, allocating and
1048 * initializing clock vectors, and computing the promises to fulfill.
1050 * @param curr The current action, as passed from the user context; may be
1051 * freed/invalidated after the execution of this function, with a different
1052 * action "returned" its place (pass-by-reference)
1053 * @return True if curr is a newly-explored action; false otherwise
1055 bool ModelExecution::initialize_curr_action(ModelAction **curr)
1057 ModelAction *newcurr;
1059 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1060 newcurr = process_rmw(*curr);
1063 if (newcurr->is_rmw())
1064 compute_promises(newcurr);
1070 (*curr)->set_seq_number(get_next_seq_num());
1072 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1074 /* First restore type and order in case of RMW operation */
1075 if ((*curr)->is_rmwr())
1076 newcurr->copy_typeandorder(*curr);
1078 ASSERT((*curr)->get_location() == newcurr->get_location());
1079 newcurr->copy_from_new(*curr);
1081 /* Discard duplicate ModelAction; use action from NodeStack */
1084 /* Always compute new clock vector */
1085 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1088 return false; /* Action was explored previously */
1092 /* Always compute new clock vector */
1093 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1095 /* Assign most recent release fence */
1096 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1099 * Perform one-time actions when pushing new ModelAction onto
1102 if (newcurr->is_write())
1103 compute_promises(newcurr);
1104 else if (newcurr->is_relseq_fixup())
1105 compute_relseq_breakwrites(newcurr);
1106 else if (newcurr->is_wait())
1107 newcurr->get_node()->set_misc_max(2);
1108 else if (newcurr->is_notify_one()) {
1109 newcurr->get_node()->set_misc_max(get_safe_ptr_action(&condvar_waiters_map, newcurr->get_location())->size());
1111 return true; /* This was a new ModelAction */
1116 * @brief Establish reads-from relation between two actions
1118 * Perform basic operations involved with establishing a concrete rf relation,
1119 * including setting the ModelAction data and checking for release sequences.
1121 * @param act The action that is reading (must be a read)
1122 * @param rf The action from which we are reading (must be a write)
1124 * @return True if this read established synchronization
1126 bool ModelExecution::read_from(ModelAction *act, const ModelAction *rf)
1129 ASSERT(rf->is_write());
1131 act->set_read_from(rf);
1132 if (act->is_acquire()) {
1133 rel_heads_list_t release_heads;
1134 get_release_seq_heads(act, act, &release_heads);
1135 int num_heads = release_heads.size();
1136 for (unsigned int i = 0; i < release_heads.size(); i++)
1137 if (!synchronize(release_heads[i], act))
1139 return num_heads > 0;
1145 * @brief Synchronizes two actions
1147 * When A synchronizes with B (or A --sw-> B), B inherits A's clock vector.
1148 * This function performs the synchronization as well as providing other hooks
1149 * for other checks along with synchronization.
1151 * @param first The left-hand side of the synchronizes-with relation
1152 * @param second The right-hand side of the synchronizes-with relation
1153 * @return True if the synchronization was successful (i.e., was consistent
1154 * with the execution order); false otherwise
1156 bool ModelExecution::synchronize(const ModelAction *first, ModelAction *second)
1158 if (*second < *first) {
1159 set_bad_synchronization();
1162 check_promises(first->get_tid(), second->get_cv(), first->get_cv());
1163 return second->synchronize_with(first);
1167 * Check promises and eliminate potentially-satisfying threads when a thread is
1168 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1169 * no longer satisfy a promise generated from that thread.
1171 * @param blocker The thread on which a thread is waiting
1172 * @param waiting The waiting thread
1174 void ModelExecution::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1176 for (unsigned int i = 0; i < promises.size(); i++) {
1177 Promise *promise = promises[i];
1178 if (!promise->thread_is_available(waiting->get_id()))
1180 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1181 ModelAction *reader = promise->get_reader(j);
1182 if (reader->get_tid() != blocker->get_id())
1184 if (promise->eliminate_thread(waiting->get_id())) {
1185 /* Promise has failed */
1186 priv->failed_promise = true;
1188 /* Only eliminate the 'waiting' thread once */
1196 * @brief Check whether a model action is enabled.
1198 * Checks whether an operation would be successful (i.e., is a lock already
1199 * locked, or is the joined thread already complete).
1201 * For yield-blocking, yields are never enabled.
1203 * @param curr is the ModelAction to check whether it is enabled.
1204 * @return a bool that indicates whether the action is enabled.
1206 bool ModelExecution::check_action_enabled(ModelAction *curr) {
1207 if (curr->is_lock()) {
1208 std::mutex *lock = curr->get_mutex();
1209 struct std::mutex_state *state = lock->get_state();
1212 } else if (curr->is_thread_join()) {
1213 Thread *blocking = curr->get_thread_operand();
1214 if (!blocking->is_complete()) {
1215 thread_blocking_check_promises(blocking, get_thread(curr));
1218 } else if (params->yieldblock && curr->is_yield()) {
1226 * This is the heart of the model checker routine. It performs model-checking
1227 * actions corresponding to a given "current action." Among other processes, it
1228 * calculates reads-from relationships, updates synchronization clock vectors,
1229 * forms a memory_order constraints graph, and handles replay/backtrack
1230 * execution when running permutations of previously-observed executions.
1232 * @param curr The current action to process
1233 * @return The ModelAction that is actually executed; may be different than
1236 ModelAction * ModelExecution::check_current_action(ModelAction *curr)
1239 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1240 bool newly_explored = initialize_curr_action(&curr);
1244 wake_up_sleeping_actions(curr);
1246 /* Compute fairness information for CHESS yield algorithm */
1247 if (params->yieldon) {
1248 curr->get_node()->update_yield(scheduler);
1251 /* Add the action to lists before any other model-checking tasks */
1252 if (!second_part_of_rmw)
1253 add_action_to_lists(curr);
1255 /* Build may_read_from set for newly-created actions */
1256 if (newly_explored && curr->is_read())
1257 build_may_read_from(curr);
1259 /* Initialize work_queue with the "current action" work */
1260 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1261 while (!work_queue.empty() && !has_asserted()) {
1262 WorkQueueEntry work = work_queue.front();
1263 work_queue.pop_front();
1265 switch (work.type) {
1266 case WORK_CHECK_CURR_ACTION: {
1267 ModelAction *act = work.action;
1268 bool update = false; /* update this location's release seq's */
1269 bool update_all = false; /* update all release seq's */
1271 if (process_thread_action(curr))
1274 if (act->is_read() && !second_part_of_rmw && process_read(act))
1277 if (act->is_write() && process_write(act, &work_queue))
1280 if (act->is_fence() && process_fence(act))
1283 if (act->is_mutex_op() && process_mutex(act))
1286 if (act->is_relseq_fixup())
1287 process_relseq_fixup(curr, &work_queue);
1290 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1292 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1295 case WORK_CHECK_RELEASE_SEQ:
1296 resolve_release_sequences(work.location, &work_queue);
1298 case WORK_CHECK_MO_EDGES: {
1299 /** @todo Complete verification of work_queue */
1300 ModelAction *act = work.action;
1301 bool updated = false;
1303 if (act->is_read()) {
1304 const ModelAction *rf = act->get_reads_from();
1305 const Promise *promise = act->get_reads_from_promise();
1307 if (r_modification_order(act, rf))
1309 } else if (promise) {
1310 if (r_modification_order(act, promise))
1314 if (act->is_write()) {
1315 if (w_modification_order(act, NULL))
1318 mo_graph->commitChanges();
1321 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1330 check_curr_backtracking(curr);
1331 set_backtracking(curr);
1335 void ModelExecution::check_curr_backtracking(ModelAction *curr)
1337 Node *currnode = curr->get_node();
1338 Node *parnode = currnode->get_parent();
1340 if ((parnode && !parnode->backtrack_empty()) ||
1341 !currnode->misc_empty() ||
1342 !currnode->read_from_empty() ||
1343 !currnode->promise_empty() ||
1344 !currnode->relseq_break_empty()) {
1345 set_latest_backtrack(curr);
1349 bool ModelExecution::promises_expired() const
1351 for (unsigned int i = 0; i < promises.size(); i++) {
1352 Promise *promise = promises[i];
1353 if (promise->get_expiration() < priv->used_sequence_numbers)
1360 * This is the strongest feasibility check available.
1361 * @return whether the current trace (partial or complete) must be a prefix of
1364 bool ModelExecution::isfeasibleprefix() const
1366 return pending_rel_seqs.size() == 0 && is_feasible_prefix_ignore_relseq();
1370 * Print disagnostic information about an infeasible execution
1371 * @param prefix A string to prefix the output with; if NULL, then a default
1372 * message prefix will be provided
1374 void ModelExecution::print_infeasibility(const char *prefix) const
1378 if (mo_graph->checkForCycles())
1379 ptr += sprintf(ptr, "[mo cycle]");
1380 if (priv->failed_promise || priv->hard_failed_promise)
1381 ptr += sprintf(ptr, "[failed promise]");
1382 if (priv->too_many_reads)
1383 ptr += sprintf(ptr, "[too many reads]");
1384 if (priv->no_valid_reads)
1385 ptr += sprintf(ptr, "[no valid reads-from]");
1386 if (priv->bad_synchronization)
1387 ptr += sprintf(ptr, "[bad sw ordering]");
1388 if (promises_expired())
1389 ptr += sprintf(ptr, "[promise expired]");
1390 if (promises.size() != 0)
1391 ptr += sprintf(ptr, "[unresolved promise]");
1393 model_print("%s: %s", prefix ? prefix : "Infeasible", buf);
1397 * Returns whether the current completed trace is feasible, except for pending
1398 * release sequences.
1400 bool ModelExecution::is_feasible_prefix_ignore_relseq() const
1402 return !is_infeasible() && promises.size() == 0 && ! priv->failed_promise;
1407 * Check if the current partial trace is infeasible. Does not check any
1408 * end-of-execution flags, which might rule out the execution. Thus, this is
1409 * useful only for ruling an execution as infeasible.
1410 * @return whether the current partial trace is infeasible.
1412 bool ModelExecution::is_infeasible() const
1414 return mo_graph->checkForCycles() ||
1415 priv->no_valid_reads ||
1416 priv->too_many_reads ||
1417 priv->bad_synchronization ||
1418 priv->hard_failed_promise ||
1422 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1423 ModelAction * ModelExecution::process_rmw(ModelAction *act) {
1424 ModelAction *lastread = get_last_action(act->get_tid());
1425 lastread->process_rmw(act);
1426 if (act->is_rmw()) {
1427 if (lastread->get_reads_from())
1428 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1430 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1431 mo_graph->commitChanges();
1437 * A helper function for ModelExecution::check_recency, to check if the current
1438 * thread is able to read from a different write/promise for 'params.maxreads'
1439 * number of steps and if that write/promise should become visible (i.e., is
1440 * ordered later in the modification order). This helps model memory liveness.
1442 * @param curr The current action. Must be a read.
1443 * @param rf The write/promise from which we plan to read
1444 * @param other_rf The write/promise from which we may read
1445 * @return True if we were able to read from other_rf for params.maxreads steps
1447 template <typename T, typename U>
1448 bool ModelExecution::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1450 /* Need a different write/promise */
1451 if (other_rf->equals(rf))
1454 /* Only look for "newer" writes/promises */
1455 if (!mo_graph->checkReachable(rf, other_rf))
1458 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1459 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1460 action_list_t::reverse_iterator rit = list->rbegin();
1461 ASSERT((*rit) == curr);
1462 /* Skip past curr */
1465 /* Does this write/promise work for everyone? */
1466 for (int i = 0; i < params->maxreads; i++, rit++) {
1467 ModelAction *act = *rit;
1468 if (!act->may_read_from(other_rf))
1475 * Checks whether a thread has read from the same write or Promise for too many
1476 * times without seeing the effects of a later write/Promise.
1479 * 1) there must a different write/promise that we could read from,
1480 * 2) we must have read from the same write/promise in excess of maxreads times,
1481 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1482 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1484 * If so, we decide that the execution is no longer feasible.
1486 * @param curr The current action. Must be a read.
1487 * @param rf The ModelAction/Promise from which we might read.
1488 * @return True if the read should succeed; false otherwise
1490 template <typename T>
1491 bool ModelExecution::check_recency(ModelAction *curr, const T *rf) const
1493 if (!params->maxreads)
1496 //NOTE: Next check is just optimization, not really necessary....
1497 if (curr->get_node()->get_read_from_past_size() +
1498 curr->get_node()->get_read_from_promise_size() <= 1)
1501 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1502 int tid = id_to_int(curr->get_tid());
1503 ASSERT(tid < (int)thrd_lists->size());
1504 action_list_t *list = &(*thrd_lists)[tid];
1505 action_list_t::reverse_iterator rit = list->rbegin();
1506 ASSERT((*rit) == curr);
1507 /* Skip past curr */
1510 action_list_t::reverse_iterator ritcopy = rit;
1511 /* See if we have enough reads from the same value */
1512 for (int count = 0; count < params->maxreads; ritcopy++, count++) {
1513 if (ritcopy == list->rend())
1515 ModelAction *act = *ritcopy;
1516 if (!act->is_read())
1518 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1520 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1522 if (act->get_node()->get_read_from_past_size() +
1523 act->get_node()->get_read_from_promise_size() <= 1)
1526 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1527 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1528 if (should_read_instead(curr, rf, write))
1529 return false; /* liveness failure */
1531 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1532 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1533 if (should_read_instead(curr, rf, promise))
1534 return false; /* liveness failure */
1540 * @brief Updates the mo_graph with the constraints imposed from the current
1543 * Basic idea is the following: Go through each other thread and find
1544 * the last action that happened before our read. Two cases:
1546 * -# The action is a write: that write must either occur before
1547 * the write we read from or be the write we read from.
1548 * -# The action is a read: the write that that action read from
1549 * must occur before the write we read from or be the same write.
1551 * @param curr The current action. Must be a read.
1552 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1553 * @return True if modification order edges were added; false otherwise
1555 template <typename rf_type>
1556 bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
1558 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1561 ASSERT(curr->is_read());
1563 /* Last SC fence in the current thread */
1564 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1565 ModelAction *last_sc_write = NULL;
1566 if (curr->is_seqcst())
1567 last_sc_write = get_last_seq_cst_write(curr);
1569 /* Iterate over all threads */
1570 for (i = 0; i < thrd_lists->size(); i++) {
1571 /* Last SC fence in thread i */
1572 ModelAction *last_sc_fence_thread_local = NULL;
1573 if (int_to_id((int)i) != curr->get_tid())
1574 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1576 /* Last SC fence in thread i, before last SC fence in current thread */
1577 ModelAction *last_sc_fence_thread_before = NULL;
1578 if (last_sc_fence_local)
1579 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1581 /* Iterate over actions in thread, starting from most recent */
1582 action_list_t *list = &(*thrd_lists)[i];
1583 action_list_t::reverse_iterator rit;
1584 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1585 ModelAction *act = *rit;
1590 /* Don't want to add reflexive edges on 'rf' */
1591 if (act->equals(rf)) {
1592 if (act->happens_before(curr))
1598 if (act->is_write()) {
1599 /* C++, Section 29.3 statement 5 */
1600 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1601 *act < *last_sc_fence_thread_local) {
1602 added = mo_graph->addEdge(act, rf) || added;
1605 /* C++, Section 29.3 statement 4 */
1606 else if (act->is_seqcst() && last_sc_fence_local &&
1607 *act < *last_sc_fence_local) {
1608 added = mo_graph->addEdge(act, rf) || added;
1611 /* C++, Section 29.3 statement 6 */
1612 else if (last_sc_fence_thread_before &&
1613 *act < *last_sc_fence_thread_before) {
1614 added = mo_graph->addEdge(act, rf) || added;
1619 /* C++, Section 29.3 statement 3 (second subpoint) */
1620 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1621 added = mo_graph->addEdge(act, rf) || added;
1626 * Include at most one act per-thread that "happens
1629 if (act->happens_before(curr)) {
1630 if (act->is_write()) {
1631 added = mo_graph->addEdge(act, rf) || added;
1633 const ModelAction *prevrf = act->get_reads_from();
1634 const Promise *prevrf_promise = act->get_reads_from_promise();
1636 if (!prevrf->equals(rf))
1637 added = mo_graph->addEdge(prevrf, rf) || added;
1638 } else if (!prevrf_promise->equals(rf)) {
1639 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1648 * All compatible, thread-exclusive promises must be ordered after any
1649 * concrete loads from the same thread
1651 for (unsigned int i = 0; i < promises.size(); i++)
1652 if (promises[i]->is_compatible_exclusive(curr))
1653 added = mo_graph->addEdge(rf, promises[i]) || added;
1659 * Updates the mo_graph with the constraints imposed from the current write.
1661 * Basic idea is the following: Go through each other thread and find
1662 * the lastest action that happened before our write. Two cases:
1664 * (1) The action is a write => that write must occur before
1667 * (2) The action is a read => the write that that action read from
1668 * must occur before the current write.
1670 * This method also handles two other issues:
1672 * (I) Sequential Consistency: Making sure that if the current write is
1673 * seq_cst, that it occurs after the previous seq_cst write.
1675 * (II) Sending the write back to non-synchronizing reads.
1677 * @param curr The current action. Must be a write.
1678 * @param send_fv A vector for stashing reads to which we may pass our future
1679 * value. If NULL, then don't record any future values.
1680 * @return True if modification order edges were added; false otherwise
1682 bool ModelExecution::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1684 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1687 ASSERT(curr->is_write());
1689 if (curr->is_seqcst()) {
1690 /* We have to at least see the last sequentially consistent write,
1691 so we are initialized. */
1692 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1693 if (last_seq_cst != NULL) {
1694 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1698 /* Last SC fence in the current thread */
1699 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1701 /* Iterate over all threads */
1702 for (i = 0; i < thrd_lists->size(); i++) {
1703 /* Last SC fence in thread i, before last SC fence in current thread */
1704 ModelAction *last_sc_fence_thread_before = NULL;
1705 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1706 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1708 /* Iterate over actions in thread, starting from most recent */
1709 action_list_t *list = &(*thrd_lists)[i];
1710 action_list_t::reverse_iterator rit;
1711 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1712 ModelAction *act = *rit;
1715 * 1) If RMW and it actually read from something, then we
1716 * already have all relevant edges, so just skip to next
1719 * 2) If RMW and it didn't read from anything, we should
1720 * whatever edge we can get to speed up convergence.
1722 * 3) If normal write, we need to look at earlier actions, so
1723 * continue processing list.
1725 if (curr->is_rmw()) {
1726 if (curr->get_reads_from() != NULL)
1734 /* C++, Section 29.3 statement 7 */
1735 if (last_sc_fence_thread_before && act->is_write() &&
1736 *act < *last_sc_fence_thread_before) {
1737 added = mo_graph->addEdge(act, curr) || added;
1742 * Include at most one act per-thread that "happens
1745 if (act->happens_before(curr)) {
1747 * Note: if act is RMW, just add edge:
1749 * The following edge should be handled elsewhere:
1750 * readfrom(act) --mo--> act
1752 if (act->is_write())
1753 added = mo_graph->addEdge(act, curr) || added;
1754 else if (act->is_read()) {
1755 //if previous read accessed a null, just keep going
1756 if (act->get_reads_from() == NULL)
1758 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
1761 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1762 !act->same_thread(curr)) {
1763 /* We have an action that:
1764 (1) did not happen before us
1765 (2) is a read and we are a write
1766 (3) cannot synchronize with us
1767 (4) is in a different thread
1769 that read could potentially read from our write. Note that
1770 these checks are overly conservative at this point, we'll
1771 do more checks before actually removing the
1776 if (send_fv && thin_air_constraint_may_allow(curr, act) && check_coherence_promise(curr, act)) {
1777 if (!is_infeasible())
1778 send_fv->push_back(act);
1779 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
1780 add_future_value(curr, act);
1787 * All compatible, thread-exclusive promises must be ordered after any
1788 * concrete stores to the same thread, or else they can be merged with
1791 for (unsigned int i = 0; i < promises.size(); i++)
1792 if (promises[i]->is_compatible_exclusive(curr))
1793 added = mo_graph->addEdge(curr, promises[i]) || added;
1798 //This procedure uses cohere to prune future values that are
1799 //guaranteed to generate a coherence violation.
1801 //need to see if there is (1) a promise for thread write, (2)
1802 //the promise is sb before write, (3) the promise can only be
1803 //resolved by the thread read, and (4) the promise has same
1804 //location as read/write
1806 bool ModelExecution::check_coherence_promise(const ModelAction * write, const ModelAction *read) {
1807 thread_id_t write_tid=write->get_tid();
1808 for(unsigned int i = promises.size(); i>0; i--) {
1809 Promise *pr=promises[i-1];
1810 if (!pr->same_location(write))
1812 //the reading thread is the only thread that can resolve the promise
1813 if (pr->get_num_was_available_threads()==1 && pr->thread_was_available(read->get_tid())) {
1814 for(unsigned int j=0;j<pr->get_num_readers();j++) {
1815 ModelAction *prreader=pr->get_reader(j);
1816 //the writing thread reads from the promise before the write
1817 if (prreader->get_tid()==write_tid &&
1818 (*prreader)<(*write)) {
1819 if ((*read)>(*prreader)) {
1820 //check that we don't have a read between the read and promise
1821 //from the same thread as read
1823 for(const ModelAction *tmp=read;tmp!=prreader;) {
1824 tmp=tmp->get_node()->get_parent()->get_action();
1825 if (tmp->is_read() && tmp->same_thread(read)) {
1842 /** Arbitrary reads from the future are not allowed. Section 29.3
1843 * part 9 places some constraints. This method checks one result of constraint
1844 * constraint. Others require compiler support. */
1845 bool ModelExecution::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
1847 if (!writer->is_rmw())
1850 if (!reader->is_rmw())
1853 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1854 if (search == reader)
1856 if (search->get_tid() == reader->get_tid() &&
1857 search->happens_before(reader))
1865 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1866 * some constraints. This method checks one the following constraint (others
1867 * require compiler support):
1869 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1870 * If X --hb-> Y, A --rf-> Y, and A --mo-> Z, then X should not read from Z.
1872 bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1874 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(reader->get_location());
1876 /* Iterate over all threads */
1877 for (i = 0; i < thrd_lists->size(); i++) {
1878 const ModelAction *write_after_read = NULL;
1880 /* Iterate over actions in thread, starting from most recent */
1881 action_list_t *list = &(*thrd_lists)[i];
1882 action_list_t::reverse_iterator rit;
1883 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1884 ModelAction *act = *rit;
1886 /* Don't disallow due to act == reader */
1887 if (!reader->happens_before(act) || reader == act)
1889 else if (act->is_write())
1890 write_after_read = act;
1891 else if (act->is_read() && act->get_reads_from() != NULL)
1892 write_after_read = act->get_reads_from();
1895 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
1902 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1903 * The ModelAction under consideration is expected to be taking part in
1904 * release/acquire synchronization as an object of the "reads from" relation.
1905 * Note that this can only provide release sequence support for RMW chains
1906 * which do not read from the future, as those actions cannot be traced until
1907 * their "promise" is fulfilled. Similarly, we may not even establish the
1908 * presence of a release sequence with certainty, as some modification order
1909 * constraints may be decided further in the future. Thus, this function
1910 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1911 * and a boolean representing certainty.
1913 * @param rf The action that might be part of a release sequence. Must be a
1915 * @param release_heads A pass-by-reference style return parameter. After
1916 * execution of this function, release_heads will contain the heads of all the
1917 * relevant release sequences, if any exists with certainty
1918 * @param pending A pass-by-reference style return parameter which is only used
1919 * when returning false (i.e., uncertain). Returns most information regarding
1920 * an uncertain release sequence, including any write operations that might
1921 * break the sequence.
1922 * @return true, if the ModelExecution is certain that release_heads is complete;
1925 bool ModelExecution::release_seq_heads(const ModelAction *rf,
1926 rel_heads_list_t *release_heads,
1927 struct release_seq *pending) const
1929 /* Only check for release sequences if there are no cycles */
1930 if (mo_graph->checkForCycles())
1933 for ( ; rf != NULL; rf = rf->get_reads_from()) {
1934 ASSERT(rf->is_write());
1936 if (rf->is_release())
1937 release_heads->push_back(rf);
1938 else if (rf->get_last_fence_release())
1939 release_heads->push_back(rf->get_last_fence_release());
1941 break; /* End of RMW chain */
1943 /** @todo Need to be smarter here... In the linux lock
1944 * example, this will run to the beginning of the program for
1946 /** @todo The way to be smarter here is to keep going until 1
1947 * thread has a release preceded by an acquire and you've seen
1950 /* acq_rel RMW is a sufficient stopping condition */
1951 if (rf->is_acquire() && rf->is_release())
1952 return true; /* complete */
1955 /* read from future: need to settle this later */
1957 return false; /* incomplete */
1960 if (rf->is_release())
1961 return true; /* complete */
1963 /* else relaxed write
1964 * - check for fence-release in the same thread (29.8, stmt. 3)
1965 * - check modification order for contiguous subsequence
1966 * -> rf must be same thread as release */
1968 const ModelAction *fence_release = rf->get_last_fence_release();
1969 /* Synchronize with a fence-release unconditionally; we don't need to
1970 * find any more "contiguous subsequence..." for it */
1972 release_heads->push_back(fence_release);
1974 int tid = id_to_int(rf->get_tid());
1975 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(rf->get_location());
1976 action_list_t *list = &(*thrd_lists)[tid];
1977 action_list_t::const_reverse_iterator rit;
1979 /* Find rf in the thread list */
1980 rit = std::find(list->rbegin(), list->rend(), rf);
1981 ASSERT(rit != list->rend());
1983 /* Find the last {write,fence}-release */
1984 for (; rit != list->rend(); rit++) {
1985 if (fence_release && *(*rit) < *fence_release)
1987 if ((*rit)->is_release())
1990 if (rit == list->rend()) {
1991 /* No write-release in this thread */
1992 return true; /* complete */
1993 } else if (fence_release && *(*rit) < *fence_release) {
1994 /* The fence-release is more recent (and so, "stronger") than
1995 * the most recent write-release */
1996 return true; /* complete */
1997 } /* else, need to establish contiguous release sequence */
1998 ModelAction *release = *rit;
2000 ASSERT(rf->same_thread(release));
2002 pending->writes.clear();
2004 bool certain = true;
2005 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2006 if (id_to_int(rf->get_tid()) == (int)i)
2008 list = &(*thrd_lists)[i];
2010 /* Can we ensure no future writes from this thread may break
2011 * the release seq? */
2012 bool future_ordered = false;
2014 ModelAction *last = get_last_action(int_to_id(i));
2015 Thread *th = get_thread(int_to_id(i));
2016 if ((last && rf->happens_before(last)) ||
2019 future_ordered = true;
2021 ASSERT(!th->is_model_thread() || future_ordered);
2023 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2024 const ModelAction *act = *rit;
2025 /* Reach synchronization -> this thread is complete */
2026 if (act->happens_before(release))
2028 if (rf->happens_before(act)) {
2029 future_ordered = true;
2033 /* Only non-RMW writes can break release sequences */
2034 if (!act->is_write() || act->is_rmw())
2037 /* Check modification order */
2038 if (mo_graph->checkReachable(rf, act)) {
2039 /* rf --mo--> act */
2040 future_ordered = true;
2043 if (mo_graph->checkReachable(act, release))
2044 /* act --mo--> release */
2046 if (mo_graph->checkReachable(release, act) &&
2047 mo_graph->checkReachable(act, rf)) {
2048 /* release --mo-> act --mo--> rf */
2049 return true; /* complete */
2051 /* act may break release sequence */
2052 pending->writes.push_back(act);
2055 if (!future_ordered)
2056 certain = false; /* This thread is uncertain */
2060 release_heads->push_back(release);
2061 pending->writes.clear();
2063 pending->release = release;
2070 * An interface for getting the release sequence head(s) with which a
2071 * given ModelAction must synchronize. This function only returns a non-empty
2072 * result when it can locate a release sequence head with certainty. Otherwise,
2073 * it may mark the internal state of the ModelExecution so that it will handle
2074 * the release sequence at a later time, causing @a acquire to update its
2075 * synchronization at some later point in execution.
2077 * @param acquire The 'acquire' action that may synchronize with a release
2079 * @param read The read action that may read from a release sequence; this may
2080 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2081 * when 'acquire' is a fence-acquire)
2082 * @param release_heads A pass-by-reference return parameter. Will be filled
2083 * with the head(s) of the release sequence(s), if they exists with certainty.
2084 * @see ModelExecution::release_seq_heads
2086 void ModelExecution::get_release_seq_heads(ModelAction *acquire,
2087 ModelAction *read, rel_heads_list_t *release_heads)
2089 const ModelAction *rf = read->get_reads_from();
2090 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2091 sequence->acquire = acquire;
2092 sequence->read = read;
2094 if (!release_seq_heads(rf, release_heads, sequence)) {
2095 /* add act to 'lazy checking' list */
2096 pending_rel_seqs.push_back(sequence);
2098 snapshot_free(sequence);
2103 * @brief Propagate a modified clock vector to actions later in the execution
2106 * After an acquire operation lazily completes a release-sequence
2107 * synchronization, we must update all clock vectors for operations later than
2108 * the acquire in the execution order.
2110 * @param acquire The ModelAction whose clock vector must be propagated
2111 * @param work The work queue to which we can add work items, if this
2112 * propagation triggers more updates (e.g., to the modification order)
2114 void ModelExecution::propagate_clockvector(ModelAction *acquire, work_queue_t *work)
2116 /* Re-check all pending release sequences */
2117 work->push_back(CheckRelSeqWorkEntry(NULL));
2118 /* Re-check read-acquire for mo_graph edges */
2119 work->push_back(MOEdgeWorkEntry(acquire));
2121 /* propagate synchronization to later actions */
2122 action_list_t::reverse_iterator rit = action_trace.rbegin();
2123 for (; (*rit) != acquire; rit++) {
2124 ModelAction *propagate = *rit;
2125 if (acquire->happens_before(propagate)) {
2126 synchronize(acquire, propagate);
2127 /* Re-check 'propagate' for mo_graph edges */
2128 work->push_back(MOEdgeWorkEntry(propagate));
2134 * Attempt to resolve all stashed operations that might synchronize with a
2135 * release sequence for a given location. This implements the "lazy" portion of
2136 * determining whether or not a release sequence was contiguous, since not all
2137 * modification order information is present at the time an action occurs.
2139 * @param location The location/object that should be checked for release
2140 * sequence resolutions. A NULL value means to check all locations.
2141 * @param work_queue The work queue to which to add work items as they are
2143 * @return True if any updates occurred (new synchronization, new mo_graph
2146 bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *work_queue)
2148 bool updated = false;
2149 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs.begin();
2150 while (it != pending_rel_seqs.end()) {
2151 struct release_seq *pending = *it;
2152 ModelAction *acquire = pending->acquire;
2153 const ModelAction *read = pending->read;
2155 /* Only resolve sequences on the given location, if provided */
2156 if (location && read->get_location() != location) {
2161 const ModelAction *rf = read->get_reads_from();
2162 rel_heads_list_t release_heads;
2164 complete = release_seq_heads(rf, &release_heads, pending);
2165 for (unsigned int i = 0; i < release_heads.size(); i++)
2166 if (!acquire->has_synchronized_with(release_heads[i]))
2167 if (synchronize(release_heads[i], acquire))
2171 /* Propagate the changed clock vector */
2172 propagate_clockvector(acquire, work_queue);
2175 it = pending_rel_seqs.erase(it);
2176 snapshot_free(pending);
2182 // If we resolved promises or data races, see if we have realized a data race.
2189 * Performs various bookkeeping operations for the current ModelAction. For
2190 * instance, adds action to the per-object, per-thread action vector and to the
2191 * action trace list of all thread actions.
2193 * @param act is the ModelAction to add.
2195 void ModelExecution::add_action_to_lists(ModelAction *act)
2197 int tid = id_to_int(act->get_tid());
2198 ModelAction *uninit = NULL;
2200 action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
2201 if (list->empty() && act->is_atomic_var()) {
2202 uninit = get_uninitialized_action(act);
2203 uninit_id = id_to_int(uninit->get_tid());
2204 list->push_front(uninit);
2206 list->push_back(act);
2208 action_trace.push_back(act);
2210 action_trace.push_front(uninit);
2212 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
2213 if (tid >= (int)vec->size())
2214 vec->resize(priv->next_thread_id);
2215 (*vec)[tid].push_back(act);
2217 (*vec)[uninit_id].push_front(uninit);
2219 if ((int)thrd_last_action.size() <= tid)
2220 thrd_last_action.resize(get_num_threads());
2221 thrd_last_action[tid] = act;
2223 thrd_last_action[uninit_id] = uninit;
2225 if (act->is_fence() && act->is_release()) {
2226 if ((int)thrd_last_fence_release.size() <= tid)
2227 thrd_last_fence_release.resize(get_num_threads());
2228 thrd_last_fence_release[tid] = act;
2231 if (act->is_wait()) {
2232 void *mutex_loc = (void *) act->get_value();
2233 get_safe_ptr_action(&obj_map, mutex_loc)->push_back(act);
2235 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
2236 if (tid >= (int)vec->size())
2237 vec->resize(priv->next_thread_id);
2238 (*vec)[tid].push_back(act);
2243 * @brief Get the last action performed by a particular Thread
2244 * @param tid The thread ID of the Thread in question
2245 * @return The last action in the thread
2247 ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
2249 int threadid = id_to_int(tid);
2250 if (threadid < (int)thrd_last_action.size())
2251 return thrd_last_action[id_to_int(tid)];
2257 * @brief Get the last fence release performed by a particular Thread
2258 * @param tid The thread ID of the Thread in question
2259 * @return The last fence release in the thread, if one exists; NULL otherwise
2261 ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
2263 int threadid = id_to_int(tid);
2264 if (threadid < (int)thrd_last_fence_release.size())
2265 return thrd_last_fence_release[id_to_int(tid)];
2271 * Gets the last memory_order_seq_cst write (in the total global sequence)
2272 * performed on a particular object (i.e., memory location), not including the
2274 * @param curr The current ModelAction; also denotes the object location to
2276 * @return The last seq_cst write
2278 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
2280 void *location = curr->get_location();
2281 action_list_t *list = obj_map.get(location);
2282 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2283 action_list_t::reverse_iterator rit;
2284 for (rit = list->rbegin(); (*rit) != curr; rit++)
2286 rit++; /* Skip past curr */
2287 for ( ; rit != list->rend(); rit++)
2288 if ((*rit)->is_write() && (*rit)->is_seqcst())
2294 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2295 * performed in a particular thread, prior to a particular fence.
2296 * @param tid The ID of the thread to check
2297 * @param before_fence The fence from which to begin the search; if NULL, then
2298 * search for the most recent fence in the thread.
2299 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2301 ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2303 /* All fences should have location FENCE_LOCATION */
2304 action_list_t *list = obj_map.get(FENCE_LOCATION);
2309 action_list_t::reverse_iterator rit = list->rbegin();
2312 for (; rit != list->rend(); rit++)
2313 if (*rit == before_fence)
2316 ASSERT(*rit == before_fence);
2320 for (; rit != list->rend(); rit++)
2321 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2327 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2328 * location). This function identifies the mutex according to the current
2329 * action, which is presumed to perform on the same mutex.
2330 * @param curr The current ModelAction; also denotes the object location to
2332 * @return The last unlock operation
2334 ModelAction * ModelExecution::get_last_unlock(ModelAction *curr) const
2336 void *location = curr->get_location();
2337 action_list_t *list = obj_map.get(location);
2338 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2339 action_list_t::reverse_iterator rit;
2340 for (rit = list->rbegin(); rit != list->rend(); rit++)
2341 if ((*rit)->is_unlock() || (*rit)->is_wait())
2346 ModelAction * ModelExecution::get_parent_action(thread_id_t tid) const
2348 ModelAction *parent = get_last_action(tid);
2350 parent = get_thread(tid)->get_creation();
2355 * Returns the clock vector for a given thread.
2356 * @param tid The thread whose clock vector we want
2357 * @return Desired clock vector
2359 ClockVector * ModelExecution::get_cv(thread_id_t tid) const
2361 return get_parent_action(tid)->get_cv();
2365 * @brief Find the promise (if any) to resolve for the current action and
2366 * remove it from the pending promise vector
2367 * @param curr The current ModelAction. Should be a write.
2368 * @return The Promise to resolve, if any; otherwise NULL
2370 Promise * ModelExecution::pop_promise_to_resolve(const ModelAction *curr)
2372 for (unsigned int i = 0; i < promises.size(); i++)
2373 if (curr->get_node()->get_promise(i)) {
2374 Promise *ret = promises[i];
2375 promises.erase(promises.begin() + i);
2382 * Resolve a Promise with a current write.
2383 * @param write The ModelAction that is fulfilling Promises
2384 * @param promise The Promise to resolve
2385 * @param work The work queue, for adding new fixup work
2386 * @return True if the Promise was successfully resolved; false otherwise
2388 bool ModelExecution::resolve_promise(ModelAction *write, Promise *promise,
2391 ModelVector<ModelAction *> actions_to_check;
2393 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2394 ModelAction *read = promise->get_reader(i);
2395 if (read_from(read, write)) {
2396 /* Propagate the changed clock vector */
2397 propagate_clockvector(read, work);
2399 actions_to_check.push_back(read);
2401 /* Make sure the promise's value matches the write's value */
2402 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2403 if (!mo_graph->resolvePromise(promise, write))
2404 priv->hard_failed_promise = true;
2407 * @todo It is possible to end up in an inconsistent state, where a
2408 * "resolved" promise may still be referenced if
2409 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2411 * Note that the inconsistency only matters when dumping mo_graph to
2417 //Check whether reading these writes has made threads unable to
2419 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2420 ModelAction *read = actions_to_check[i];
2421 mo_check_promises(read, true);
2428 * Compute the set of promises that could potentially be satisfied by this
2429 * action. Note that the set computation actually appears in the Node, not in
2431 * @param curr The ModelAction that may satisfy promises
2433 void ModelExecution::compute_promises(ModelAction *curr)
2435 for (unsigned int i = 0; i < promises.size(); i++) {
2436 Promise *promise = promises[i];
2437 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2440 bool satisfy = true;
2441 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2442 const ModelAction *act = promise->get_reader(j);
2443 if (act->happens_before(curr) ||
2444 act->could_synchronize_with(curr)) {
2450 curr->get_node()->set_promise(i);
2454 /** Checks promises in response to change in ClockVector Threads. */
2455 void ModelExecution::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2457 for (unsigned int i = 0; i < promises.size(); i++) {
2458 Promise *promise = promises[i];
2459 if (!promise->thread_is_available(tid))
2461 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2462 const ModelAction *act = promise->get_reader(j);
2463 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2464 merge_cv->synchronized_since(act)) {
2465 if (promise->eliminate_thread(tid)) {
2466 /* Promise has failed */
2467 priv->failed_promise = true;
2475 void ModelExecution::check_promises_thread_disabled()
2477 for (unsigned int i = 0; i < promises.size(); i++) {
2478 Promise *promise = promises[i];
2479 if (promise->has_failed()) {
2480 priv->failed_promise = true;
2487 * @brief Checks promises in response to addition to modification order for
2490 * We test whether threads are still available for satisfying promises after an
2491 * addition to our modification order constraints. Those that are unavailable
2492 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2493 * that promise has failed.
2495 * @param act The ModelAction which updated the modification order
2496 * @param is_read_check Should be true if act is a read and we must check for
2497 * updates to the store from which it read (there is a distinction here for
2498 * RMW's, which are both a load and a store)
2500 void ModelExecution::mo_check_promises(const ModelAction *act, bool is_read_check)
2502 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2504 for (unsigned int i = 0; i < promises.size(); i++) {
2505 Promise *promise = promises[i];
2507 // Is this promise on the same location?
2508 if (!promise->same_location(write))
2511 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2512 const ModelAction *pread = promise->get_reader(j);
2513 if (!pread->happens_before(act))
2515 if (mo_graph->checkPromise(write, promise)) {
2516 priv->hard_failed_promise = true;
2522 // Don't do any lookups twice for the same thread
2523 if (!promise->thread_is_available(act->get_tid()))
2526 if (mo_graph->checkReachable(promise, write)) {
2527 if (mo_graph->checkPromise(write, promise)) {
2528 priv->hard_failed_promise = true;
2536 * Compute the set of writes that may break the current pending release
2537 * sequence. This information is extracted from previou release sequence
2540 * @param curr The current ModelAction. Must be a release sequence fixup
2543 void ModelExecution::compute_relseq_breakwrites(ModelAction *curr)
2545 if (pending_rel_seqs.empty())
2548 struct release_seq *pending = pending_rel_seqs.back();
2549 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2550 const ModelAction *write = pending->writes[i];
2551 curr->get_node()->add_relseq_break(write);
2554 /* NULL means don't break the sequence; just synchronize */
2555 curr->get_node()->add_relseq_break(NULL);
2559 * Build up an initial set of all past writes that this 'read' action may read
2560 * from, as well as any previously-observed future values that must still be valid.
2562 * @param curr is the current ModelAction that we are exploring; it must be a
2565 void ModelExecution::build_may_read_from(ModelAction *curr)
2567 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
2569 ASSERT(curr->is_read());
2571 ModelAction *last_sc_write = NULL;
2573 if (curr->is_seqcst())
2574 last_sc_write = get_last_seq_cst_write(curr);
2576 /* Iterate over all threads */
2577 for (i = 0; i < thrd_lists->size(); i++) {
2578 /* Iterate over actions in thread, starting from most recent */
2579 action_list_t *list = &(*thrd_lists)[i];
2580 action_list_t::reverse_iterator rit;
2581 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2582 ModelAction *act = *rit;
2584 /* Only consider 'write' actions */
2585 if (!act->is_write() || act == curr)
2588 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2589 bool allow_read = true;
2591 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2593 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2597 /* Only add feasible reads */
2598 mo_graph->startChanges();
2599 r_modification_order(curr, act);
2600 if (!is_infeasible())
2601 curr->get_node()->add_read_from_past(act);
2602 mo_graph->rollbackChanges();
2605 /* Include at most one act per-thread that "happens before" curr */
2606 if (act->happens_before(curr))
2611 /* Inherit existing, promised future values */
2612 for (i = 0; i < promises.size(); i++) {
2613 const Promise *promise = promises[i];
2614 const ModelAction *promise_read = promise->get_reader(0);
2615 if (promise_read->same_var(curr)) {
2616 /* Only add feasible future-values */
2617 mo_graph->startChanges();
2618 r_modification_order(curr, promise);
2619 if (!is_infeasible())
2620 curr->get_node()->add_read_from_promise(promise_read);
2621 mo_graph->rollbackChanges();
2625 /* We may find no valid may-read-from only if the execution is doomed */
2626 if (!curr->get_node()->read_from_size()) {
2627 priv->no_valid_reads = true;
2631 if (DBG_ENABLED()) {
2632 model_print("Reached read action:\n");
2634 model_print("Printing read_from_past\n");
2635 curr->get_node()->print_read_from_past();
2636 model_print("End printing read_from_past\n");
2640 bool ModelExecution::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2642 for ( ; write != NULL; write = write->get_reads_from()) {
2643 /* UNINIT actions don't have a Node, and they never sleep */
2644 if (write->is_uninitialized())
2646 Node *prevnode = write->get_node()->get_parent();
2648 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2649 if (write->is_release() && thread_sleep)
2651 if (!write->is_rmw())
2658 * @brief Get an action representing an uninitialized atomic
2660 * This function may create a new one or try to retrieve one from the NodeStack
2662 * @param curr The current action, which prompts the creation of an UNINIT action
2663 * @return A pointer to the UNINIT ModelAction
2665 ModelAction * ModelExecution::get_uninitialized_action(const ModelAction *curr) const
2667 Node *node = curr->get_node();
2668 ModelAction *act = node->get_uninit_action();
2670 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), params->uninitvalue, model_thread);
2671 node->set_uninit_action(act);
2673 act->create_cv(NULL);
2677 static void print_list(const action_list_t *list)
2679 action_list_t::const_iterator it;
2681 model_print("------------------------------------------------------------------------------------\n");
2682 model_print("# t Action type MO Location Value Rf CV\n");
2683 model_print("------------------------------------------------------------------------------------\n");
2685 unsigned int hash = 0;
2687 for (it = list->begin(); it != list->end(); it++) {
2688 const ModelAction *act = *it;
2689 if (act->get_seq_number() > 0)
2691 hash = hash^(hash<<3)^((*it)->hash());
2693 model_print("HASH %u\n", hash);
2694 model_print("------------------------------------------------------------------------------------\n");
2697 #if SUPPORT_MOD_ORDER_DUMP
2698 void ModelExecution::dumpGraph(char *filename) const
2701 sprintf(buffer, "%s.dot", filename);
2702 FILE *file = fopen(buffer, "w");
2703 fprintf(file, "digraph %s {\n", filename);
2704 mo_graph->dumpNodes(file);
2705 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2707 for (action_list_t::const_iterator it = action_trace.begin(); it != action_trace.end(); it++) {
2708 ModelAction *act = *it;
2709 if (act->is_read()) {
2710 mo_graph->dot_print_node(file, act);
2711 if (act->get_reads_from())
2712 mo_graph->dot_print_edge(file,
2713 act->get_reads_from(),
2715 "label=\"rf\", color=red, weight=2");
2717 mo_graph->dot_print_edge(file,
2718 act->get_reads_from_promise(),
2720 "label=\"rf\", color=red");
2722 if (thread_array[act->get_tid()]) {
2723 mo_graph->dot_print_edge(file,
2724 thread_array[id_to_int(act->get_tid())],
2726 "label=\"sb\", color=blue, weight=400");
2729 thread_array[act->get_tid()] = act;
2731 fprintf(file, "}\n");
2732 model_free(thread_array);
2737 /** @brief Prints an execution trace summary. */
2738 void ModelExecution::print_summary() const
2740 #if SUPPORT_MOD_ORDER_DUMP
2741 char buffername[100];
2742 sprintf(buffername, "exec%04u", get_execution_number());
2743 mo_graph->dumpGraphToFile(buffername);
2744 sprintf(buffername, "graph%04u", get_execution_number());
2745 dumpGraph(buffername);
2748 model_print("Execution trace %d:", get_execution_number());
2749 if (isfeasibleprefix()) {
2750 if (is_yieldblocked())
2751 model_print(" YIELD BLOCKED");
2752 if (scheduler->all_threads_sleeping())
2753 model_print(" SLEEP-SET REDUNDANT");
2754 if (have_bug_reports())
2755 model_print(" DETECTED BUG(S)");
2757 print_infeasibility(" INFEASIBLE");
2760 print_list(&action_trace);
2763 if (!promises.empty()) {
2764 model_print("Pending promises:\n");
2765 for (unsigned int i = 0; i < promises.size(); i++) {
2766 model_print(" [P%u] ", i);
2767 promises[i]->print();
2774 * Add a Thread to the system for the first time. Should only be called once
2776 * @param t The Thread to add
2778 void ModelExecution::add_thread(Thread *t)
2780 unsigned int i = id_to_int(t->get_id());
2781 if (i >= thread_map.size())
2782 thread_map.resize(i + 1);
2784 if (!t->is_model_thread())
2785 scheduler->add_thread(t);
2789 * @brief Get a Thread reference by its ID
2790 * @param tid The Thread's ID
2791 * @return A Thread reference
2793 Thread * ModelExecution::get_thread(thread_id_t tid) const
2795 unsigned int i = id_to_int(tid);
2796 if (i < thread_map.size())
2797 return thread_map[i];
2802 * @brief Get a reference to the Thread in which a ModelAction was executed
2803 * @param act The ModelAction
2804 * @return A Thread reference
2806 Thread * ModelExecution::get_thread(const ModelAction *act) const
2808 return get_thread(act->get_tid());
2812 * @brief Get a Promise's "promise number"
2814 * A "promise number" is an index number that is unique to a promise, valid
2815 * only for a specific snapshot of an execution trace. Promises may come and go
2816 * as they are generated an resolved, so an index only retains meaning for the
2819 * @param promise The Promise to check
2820 * @return The promise index, if the promise still is valid; otherwise -1
2822 int ModelExecution::get_promise_number(const Promise *promise) const
2824 for (unsigned int i = 0; i < promises.size(); i++)
2825 if (promises[i] == promise)
2832 * @brief Check if a Thread is currently enabled
2833 * @param t The Thread to check
2834 * @return True if the Thread is currently enabled
2836 bool ModelExecution::is_enabled(Thread *t) const
2838 return scheduler->is_enabled(t);
2842 * @brief Check if a Thread is currently enabled
2843 * @param tid The ID of the Thread to check
2844 * @return True if the Thread is currently enabled
2846 bool ModelExecution::is_enabled(thread_id_t tid) const
2848 return scheduler->is_enabled(tid);
2852 * @brief Select the next thread to execute based on the curren action
2854 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
2855 * actions should be followed by the execution of their child thread. In either
2856 * case, the current action should determine the next thread schedule.
2858 * @param curr The current action
2859 * @return The next thread to run, if the current action will determine this
2860 * selection; otherwise NULL
2862 Thread * ModelExecution::action_select_next_thread(const ModelAction *curr) const
2864 /* Do not split atomic RMW */
2865 if (curr->is_rmwr())
2866 return get_thread(curr);
2867 /* Follow CREATE with the created thread */
2868 if (curr->get_type() == THREAD_CREATE)
2869 return curr->get_thread_operand();
2873 /** @return True if the execution has taken too many steps */
2874 bool ModelExecution::too_many_steps() const
2876 return params->bound != 0 && priv->used_sequence_numbers > params->bound;
2880 * Takes the next step in the execution, if possible.
2881 * @param curr The current step to take
2882 * @return Returns the next Thread to run, if any; NULL if this execution
2885 Thread * ModelExecution::take_step(ModelAction *curr)
2887 Thread *curr_thrd = get_thread(curr);
2888 ASSERT(curr_thrd->get_state() == THREAD_READY);
2890 ASSERT(check_action_enabled(curr)); /* May have side effects? */
2891 curr = check_current_action(curr);
2894 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
2895 scheduler->remove_thread(curr_thrd);
2897 return action_select_next_thread(curr);
2901 * Launch end-of-execution release sequence fixups only when
2902 * the execution is otherwise feasible AND there are:
2904 * (1) pending release sequences
2905 * (2) pending assertions that could be invalidated by a change
2906 * in clock vectors (i.e., data races)
2907 * (3) no pending promises
2909 void ModelExecution::fixup_release_sequences()
2911 while (!pending_rel_seqs.empty() &&
2912 is_feasible_prefix_ignore_relseq() &&
2913 haveUnrealizedRaces()) {
2914 model_print("*** WARNING: release sequence fixup action "
2915 "(%zu pending release seuqence(s)) ***\n",
2916 pending_rel_seqs.size());
2917 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2918 std::memory_order_seq_cst, NULL, VALUE_NONE,