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 too_many_reads(false),
34 no_valid_reads(false),
35 bad_synchronization(false),
39 ~model_snapshot_members() {
40 for (unsigned int i = 0; i < bugs.size(); i++)
45 unsigned int next_thread_id;
46 modelclock_t used_sequence_numbers;
47 ModelAction *next_backtrack;
48 SnapVector<bug_message *> bugs;
52 /** @brief Incorrectly-ordered synchronization was made */
53 bool bad_synchronization;
59 /** @brief Constructor */
60 ModelExecution::ModelExecution(ModelChecker *m,
61 struct model_params *params,
63 NodeStack *node_stack) :
68 thread_map(2), /* We'll always need at least 2 threads */
70 condvar_waiters_map(),
76 thrd_last_fence_release(),
77 node_stack(node_stack),
78 priv(new struct model_snapshot_members()),
79 mo_graph(new CycleGraph())
81 /* Initialize a model-checker thread, for special ModelActions */
82 model_thread = new Thread(get_next_id());
83 add_thread(model_thread);
84 scheduler->register_engine(this);
85 node_stack->register_engine(this);
88 /** @brief Destructor */
89 ModelExecution::~ModelExecution()
91 for (unsigned int i = 0; i < get_num_threads(); i++)
92 delete get_thread(int_to_id(i));
94 for (unsigned int i = 0; i < promises.size(); i++)
101 int ModelExecution::get_execution_number() const
103 return model->get_execution_number();
106 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
108 action_list_t *tmp = hash->get(ptr);
110 tmp = new action_list_t();
116 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
118 SnapVector<action_list_t> *tmp = hash->get(ptr);
120 tmp = new SnapVector<action_list_t>();
126 action_list_t * ModelExecution::get_actions_on_obj(void * obj, thread_id_t tid) const
128 SnapVector<action_list_t> *wrv = obj_thrd_map.get(obj);
131 unsigned int thread=id_to_int(tid);
132 if (thread < wrv->size())
133 return &(*wrv)[thread];
138 /** @return a thread ID for a new Thread */
139 thread_id_t ModelExecution::get_next_id()
141 return priv->next_thread_id++;
144 /** @return the number of user threads created during this execution */
145 unsigned int ModelExecution::get_num_threads() const
147 return priv->next_thread_id;
150 /** @return a sequence number for a new ModelAction */
151 modelclock_t ModelExecution::get_next_seq_num()
153 return ++priv->used_sequence_numbers;
157 * @brief Should the current action wake up a given thread?
159 * @param curr The current action
160 * @param thread The thread that we might wake up
161 * @return True, if we should wake up the sleeping thread; false otherwise
163 bool ModelExecution::should_wake_up(const ModelAction *curr, const Thread *thread) const
165 const ModelAction *asleep = thread->get_pending();
166 /* Don't allow partial RMW to wake anyone up */
169 /* Synchronizing actions may have been backtracked */
170 if (asleep->could_synchronize_with(curr))
172 /* All acquire/release fences and fence-acquire/store-release */
173 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
175 /* Fence-release + store can awake load-acquire on the same location */
176 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
177 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
178 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
184 void ModelExecution::wake_up_sleeping_actions(ModelAction *curr)
186 for (unsigned int i = 0; i < get_num_threads(); i++) {
187 Thread *thr = get_thread(int_to_id(i));
188 if (scheduler->is_sleep_set(thr)) {
189 if (should_wake_up(curr, thr))
190 /* Remove this thread from sleep set */
191 scheduler->remove_sleep(thr);
196 /** @brief Alert the model-checker that an incorrectly-ordered
197 * synchronization was made */
198 void ModelExecution::set_bad_synchronization()
200 priv->bad_synchronization = true;
203 bool ModelExecution::assert_bug(const char *msg)
205 priv->bugs.push_back(new bug_message(msg));
207 if (isfeasibleprefix()) {
214 /** @return True, if any bugs have been reported for this execution */
215 bool ModelExecution::have_bug_reports() const
217 return priv->bugs.size() != 0;
220 SnapVector<bug_message *> * ModelExecution::get_bugs() const
226 * Check whether the current trace has triggered an assertion which should halt
229 * @return True, if the execution should be aborted; false otherwise
231 bool ModelExecution::has_asserted() const
233 return priv->asserted;
237 * Trigger a trace assertion which should cause this execution to be halted.
238 * This can be due to a detected bug or due to an infeasibility that should
241 void ModelExecution::set_assert()
243 priv->asserted = true;
247 * Check if we are in a deadlock. Should only be called at the end of an
248 * execution, although it should not give false positives in the middle of an
249 * execution (there should be some ENABLED thread).
251 * @return True if program is in a deadlock; false otherwise
253 bool ModelExecution::is_deadlocked() const
255 bool blocking_threads = false;
256 for (unsigned int i = 0; i < get_num_threads(); i++) {
257 thread_id_t tid = int_to_id(i);
260 Thread *t = get_thread(tid);
261 if (!t->is_model_thread() && t->get_pending())
262 blocking_threads = true;
264 return blocking_threads;
268 * Check if this is a complete execution. That is, have all thread completed
269 * execution (rather than exiting because sleep sets have forced a redundant
272 * @return True if the execution is complete.
274 bool ModelExecution::is_complete_execution() const
276 for (unsigned int i = 0; i < get_num_threads(); i++)
277 if (is_enabled(int_to_id(i)))
283 * @brief Find the last fence-related backtracking conflict for a ModelAction
285 * This function performs the search for the most recent conflicting action
286 * against which we should perform backtracking, as affected by fence
287 * operations. This includes pairs of potentially-synchronizing actions which
288 * occur due to fence-acquire or fence-release, and hence should be explored in
289 * the opposite execution order.
291 * @param act The current action
292 * @return The most recent action which conflicts with act due to fences
294 ModelAction * ModelExecution::get_last_fence_conflict(ModelAction *act) const
296 /* Only perform release/acquire fence backtracking for stores */
297 if (!act->is_write())
300 /* Find a fence-release (or, act is a release) */
301 ModelAction *last_release;
302 if (act->is_release())
305 last_release = get_last_fence_release(act->get_tid());
309 /* Skip past the release */
310 const action_list_t *list = &action_trace;
311 action_list_t::const_reverse_iterator rit;
312 for (rit = list->rbegin(); rit != list->rend(); rit++)
313 if (*rit == last_release)
315 ASSERT(rit != list->rend());
320 * load --sb-> fence-acquire */
321 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
322 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
323 bool found_acquire_fences = false;
324 for ( ; rit != list->rend(); rit++) {
325 ModelAction *prev = *rit;
326 if (act->same_thread(prev))
329 int tid = id_to_int(prev->get_tid());
331 if (prev->is_read() && act->same_var(prev)) {
332 if (prev->is_acquire()) {
333 /* Found most recent load-acquire, don't need
334 * to search for more fences */
335 if (!found_acquire_fences)
338 prior_loads[tid] = prev;
341 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
342 found_acquire_fences = true;
343 acquire_fences[tid] = prev;
347 ModelAction *latest_backtrack = NULL;
348 for (unsigned int i = 0; i < acquire_fences.size(); i++)
349 if (acquire_fences[i] && prior_loads[i])
350 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
351 latest_backtrack = acquire_fences[i];
352 return latest_backtrack;
356 * @brief Find the last backtracking conflict for a ModelAction
358 * This function performs the search for the most recent conflicting action
359 * against which we should perform backtracking. This primary includes pairs of
360 * synchronizing actions which should be explored in the opposite execution
363 * @param act The current action
364 * @return The most recent action which conflicts with act
366 ModelAction * ModelExecution::get_last_conflict(ModelAction *act) const
368 switch (act->get_type()) {
370 /* Only seq-cst fences can (directly) cause backtracking */
371 if (!act->is_seqcst())
376 ModelAction *ret = NULL;
378 /* linear search: from most recent to oldest */
379 action_list_t *list = obj_map.get(act->get_location());
380 action_list_t::reverse_iterator rit;
381 for (rit = list->rbegin(); rit != list->rend(); rit++) {
382 ModelAction *prev = *rit;
383 if (prev->could_synchronize_with(act)) {
389 ModelAction *ret2 = get_last_fence_conflict(act);
399 case ATOMIC_TRYLOCK: {
400 /* linear search: from most recent to oldest */
401 action_list_t *list = obj_map.get(act->get_location());
402 action_list_t::reverse_iterator rit;
403 for (rit = list->rbegin(); rit != list->rend(); rit++) {
404 ModelAction *prev = *rit;
405 if (act->is_conflicting_lock(prev))
410 case ATOMIC_UNLOCK: {
411 /* linear search: from most recent to oldest */
412 action_list_t *list = obj_map.get(act->get_location());
413 action_list_t::reverse_iterator rit;
414 for (rit = list->rbegin(); rit != list->rend(); rit++) {
415 ModelAction *prev = *rit;
416 if (!act->same_thread(prev) && prev->is_failed_trylock())
422 /* linear search: from most recent to oldest */
423 action_list_t *list = obj_map.get(act->get_location());
424 action_list_t::reverse_iterator rit;
425 for (rit = list->rbegin(); rit != list->rend(); rit++) {
426 ModelAction *prev = *rit;
427 if (!act->same_thread(prev) && prev->is_failed_trylock())
429 if (!act->same_thread(prev) && prev->is_notify())
435 case ATOMIC_NOTIFY_ALL:
436 case ATOMIC_NOTIFY_ONE: {
437 /* linear search: from most recent to oldest */
438 action_list_t *list = obj_map.get(act->get_location());
439 action_list_t::reverse_iterator rit;
440 for (rit = list->rbegin(); rit != list->rend(); rit++) {
441 ModelAction *prev = *rit;
442 if (!act->same_thread(prev) && prev->is_wait())
453 /** This method finds backtracking points where we should try to
454 * reorder the parameter ModelAction against.
456 * @param the ModelAction to find backtracking points for.
458 void ModelExecution::set_backtracking(ModelAction *act)
460 Thread *t = get_thread(act);
461 ModelAction *prev = get_last_conflict(act);
465 Node *node = prev->get_node()->get_parent();
467 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
468 int low_tid, high_tid;
469 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
470 low_tid = id_to_int(act->get_tid());
471 high_tid = low_tid + 1;
474 high_tid = get_num_threads();
477 for (int i = low_tid; i < high_tid; i++) {
478 thread_id_t tid = int_to_id(i);
480 /* Make sure this thread can be enabled here. */
481 if (i >= node->get_num_threads())
484 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
485 /* Don't backtrack into a point where the thread is disabled or sleeping. */
486 if (node->enabled_status(tid) != THREAD_ENABLED)
489 /* Check if this has been explored already */
490 if (node->has_been_explored(tid))
493 /* See if fairness allows */
494 if (params->fairwindow != 0 && !node->has_priority(tid)) {
496 for (int t = 0; t < node->get_num_threads(); t++) {
497 thread_id_t tother = int_to_id(t);
498 if (node->is_enabled(tother) && node->has_priority(tother)) {
507 /* See if CHESS-like yield fairness allows */
508 if (params->yieldon) {
510 for (int t = 0; t < node->get_num_threads(); t++) {
511 thread_id_t tother = int_to_id(t);
512 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
521 /* Cache the latest backtracking point */
522 set_latest_backtrack(prev);
524 /* If this is a new backtracking point, mark the tree */
525 if (!node->set_backtrack(tid))
527 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
528 id_to_int(prev->get_tid()),
529 id_to_int(t->get_id()));
538 * @brief Cache the a backtracking point as the "most recent", if eligible
540 * Note that this does not prepare the NodeStack for this backtracking
541 * operation, it only caches the action on a per-execution basis
543 * @param act The operation at which we should explore a different next action
544 * (i.e., backtracking point)
545 * @return True, if this action is now the most recent backtracking point;
548 bool ModelExecution::set_latest_backtrack(ModelAction *act)
550 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
551 priv->next_backtrack = act;
558 * Returns last backtracking point. The model checker will explore a different
559 * path for this point in the next execution.
560 * @return The ModelAction at which the next execution should diverge.
562 ModelAction * ModelExecution::get_next_backtrack()
564 ModelAction *next = priv->next_backtrack;
565 priv->next_backtrack = NULL;
570 * Processes a read model action.
571 * @param curr is the read model action to process.
572 * @return True if processing this read updates the mo_graph.
574 bool ModelExecution::process_read(ModelAction *curr)
576 Node *node = curr->get_node();
578 bool updated = false;
579 switch (node->get_read_from_status()) {
580 case READ_FROM_PAST: {
581 const ModelAction *rf = node->get_read_from_past();
584 mo_graph->startChanges();
586 ASSERT(!is_infeasible());
587 if (!check_recency(curr, rf)) {
588 if (node->increment_read_from()) {
589 mo_graph->rollbackChanges();
592 priv->too_many_reads = true;
596 updated = r_modification_order(curr, rf);
598 mo_graph->commitChanges();
599 mo_check_promises(curr, true);
602 case READ_FROM_PROMISE: {
603 Promise *promise = curr->get_node()->get_read_from_promise();
604 if (promise->add_reader(curr))
605 priv->failed_promise = true;
606 curr->set_read_from_promise(promise);
607 mo_graph->startChanges();
608 if (!check_recency(curr, promise))
609 priv->too_many_reads = true;
610 updated = r_modification_order(curr, promise);
611 mo_graph->commitChanges();
614 case READ_FROM_FUTURE: {
615 /* Read from future value */
616 struct future_value fv = node->get_future_value();
617 Promise *promise = new Promise(this, curr, fv);
618 curr->set_read_from_promise(promise);
619 promises.push_back(promise);
620 mo_graph->startChanges();
621 updated = r_modification_order(curr, promise);
622 mo_graph->commitChanges();
628 get_thread(curr)->set_return_value(curr->get_return_value());
634 * Processes a lock, trylock, or unlock model action. @param curr is
635 * the read model action to process.
637 * The try lock operation checks whether the lock is taken. If not,
638 * it falls to the normal lock operation case. If so, it returns
641 * The lock operation has already been checked that it is enabled, so
642 * it just grabs the lock and synchronizes with the previous unlock.
644 * The unlock operation has to re-enable all of the threads that are
645 * waiting on the lock.
647 * @return True if synchronization was updated; false otherwise
649 bool ModelExecution::process_mutex(ModelAction *curr)
651 std::mutex *mutex = curr->get_mutex();
652 struct std::mutex_state *state = NULL;
655 state = mutex->get_state();
657 switch (curr->get_type()) {
658 case ATOMIC_TRYLOCK: {
659 bool success = !state->locked;
660 curr->set_try_lock(success);
662 get_thread(curr)->set_return_value(0);
665 get_thread(curr)->set_return_value(1);
667 //otherwise fall into the lock case
669 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
670 assert_bug("Lock access before initialization");
671 state->locked = get_thread(curr);
672 ModelAction *unlock = get_last_unlock(curr);
673 //synchronize with the previous unlock statement
674 if (unlock != NULL) {
675 synchronize(unlock, curr);
681 case ATOMIC_UNLOCK: {
682 /* wake up the other threads */
683 for (unsigned int i = 0; i < get_num_threads(); i++) {
684 Thread *t = get_thread(int_to_id(i));
685 Thread *curr_thrd = get_thread(curr);
686 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
690 /* unlock the lock - after checking who was waiting on it */
691 state->locked = NULL;
693 if (!curr->is_wait())
694 break; /* The rest is only for ATOMIC_WAIT */
696 /* Should we go to sleep? (simulate spurious failures) */
697 if (curr->get_node()->get_misc() == 0) {
698 get_safe_ptr_action(&condvar_waiters_map, curr->get_location())->push_back(curr);
700 scheduler->sleep(get_thread(curr));
704 case ATOMIC_NOTIFY_ALL: {
705 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
706 //activate all the waiting threads
707 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
708 scheduler->wake(get_thread(*rit));
713 case ATOMIC_NOTIFY_ONE: {
714 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
715 int wakeupthread = curr->get_node()->get_misc();
716 action_list_t::iterator it = waiters->begin();
717 advance(it, wakeupthread);
718 scheduler->wake(get_thread(*it));
730 * @brief Check if the current pending promises allow a future value to be sent
732 * If one of the following is true:
733 * (a) there are no pending promises
734 * (b) the reader and writer do not cross any promises
735 * Then, it is safe to pass a future value back now.
737 * Otherwise, we must save the pending future value until (a) or (b) is true
739 * @param writer The operation which sends the future value. Must be a write.
740 * @param reader The operation which will observe the value. Must be a read.
741 * @return True if the future value can be sent now; false if it must wait.
743 bool ModelExecution::promises_may_allow(const ModelAction *writer,
744 const ModelAction *reader) const
746 if (promises.empty())
748 for (int i = promises.size() - 1; i >= 0; i--) {
749 ModelAction *pr = promises[i]->get_reader(0);
750 //reader is after promise...doesn't cross any promise
753 //writer is after promise, reader before...bad...
761 * @brief Add a future value to a reader
763 * This function performs a few additional checks to ensure that the future
764 * value can be feasibly observed by the reader
766 * @param writer The operation whose value is sent. Must be a write.
767 * @param reader The read operation which may read the future value. Must be a read.
769 void ModelExecution::add_future_value(const ModelAction *writer, ModelAction *reader)
771 /* Do more ambitious checks now that mo is more complete */
772 if (!mo_may_allow(writer, reader))
775 Node *node = reader->get_node();
777 /* Find an ancestor thread which exists at the time of the reader */
778 Thread *write_thread = get_thread(writer);
779 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
780 write_thread = write_thread->get_parent();
782 struct future_value fv = {
783 writer->get_write_value(),
784 writer->get_seq_number() + params->maxfuturedelay,
785 write_thread->get_id(),
787 if (node->add_future_value(fv))
788 set_latest_backtrack(reader);
792 * Process a write ModelAction
793 * @param curr The ModelAction to process
794 * @return True if the mo_graph was updated or promises were resolved
796 bool ModelExecution::process_write(ModelAction *curr)
798 /* Readers to which we may send our future value */
799 ModelVector<ModelAction *> send_fv;
801 const ModelAction *earliest_promise_reader;
802 bool updated_promises = false;
804 bool updated_mod_order = w_modification_order(curr, &send_fv);
805 Promise *promise = pop_promise_to_resolve(curr);
808 earliest_promise_reader = promise->get_reader(0);
809 updated_promises = resolve_promise(curr, promise);
811 earliest_promise_reader = NULL;
813 for (unsigned int i = 0; i < send_fv.size(); i++) {
814 ModelAction *read = send_fv[i];
816 /* Don't send future values to reads after the Promise we resolve */
817 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
818 /* Check if future value can be sent immediately */
819 if (promises_may_allow(curr, read)) {
820 add_future_value(curr, read);
822 futurevalues.push_back(PendingFutureValue(curr, read));
827 /* Check the pending future values */
828 for (int i = (int)futurevalues.size() - 1; i >= 0; i--) {
829 struct PendingFutureValue pfv = futurevalues[i];
830 if (promises_may_allow(pfv.writer, pfv.reader)) {
831 add_future_value(pfv.writer, pfv.reader);
832 futurevalues.erase(futurevalues.begin() + i);
836 mo_graph->commitChanges();
837 mo_check_promises(curr, false);
839 get_thread(curr)->set_return_value(VALUE_NONE);
840 return updated_mod_order || updated_promises;
844 * Process a fence ModelAction
845 * @param curr The ModelAction to process
846 * @return True if synchronization was updated
848 bool ModelExecution::process_fence(ModelAction *curr)
851 * fence-relaxed: no-op
852 * fence-release: only log the occurence (not in this function), for
853 * use in later synchronization
854 * fence-acquire (this function): search for hypothetical release
856 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
858 bool updated = false;
859 if (curr->is_acquire()) {
860 action_list_t *list = &action_trace;
861 action_list_t::reverse_iterator rit;
862 /* Find X : is_read(X) && X --sb-> curr */
863 for (rit = list->rbegin(); rit != list->rend(); rit++) {
864 ModelAction *act = *rit;
867 if (act->get_tid() != curr->get_tid())
869 /* Stop at the beginning of the thread */
870 if (act->is_thread_start())
872 /* Stop once we reach a prior fence-acquire */
873 if (act->is_fence() && act->is_acquire())
877 /* read-acquire will find its own release sequences */
878 if (act->is_acquire())
881 /* Establish hypothetical release sequences */
882 rel_heads_list_t release_heads;
883 get_release_seq_heads(curr, act, &release_heads);
884 for (unsigned int i = 0; i < release_heads.size(); i++)
885 synchronize(release_heads[i], curr);
886 if (release_heads.size() != 0)
894 * @brief Process the current action for thread-related activity
896 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
897 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
898 * synchronization, etc. This function is a no-op for non-THREAD actions
899 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
901 * @param curr The current action
902 * @return True if synchronization was updated or a thread completed
904 bool ModelExecution::process_thread_action(ModelAction *curr)
906 bool updated = false;
908 switch (curr->get_type()) {
909 case THREAD_CREATE: {
910 thrd_t *thrd = (thrd_t *)curr->get_location();
911 struct thread_params *params = (struct thread_params *)curr->get_value();
912 Thread *th = new Thread(get_next_id(), thrd, params->func, params->arg, get_thread(curr));
914 th->set_creation(curr);
915 /* Promises can be satisfied by children */
916 for (unsigned int i = 0; i < promises.size(); i++) {
917 Promise *promise = promises[i];
918 if (promise->thread_is_available(curr->get_tid()))
919 promise->add_thread(th->get_id());
924 Thread *blocking = curr->get_thread_operand();
925 ModelAction *act = get_last_action(blocking->get_id());
926 synchronize(act, curr);
927 updated = true; /* trigger rel-seq checks */
930 case THREAD_FINISH: {
931 Thread *th = get_thread(curr);
932 /* Wake up any joining threads */
933 for (unsigned int i = 0; i < get_num_threads(); i++) {
934 Thread *waiting = get_thread(int_to_id(i));
935 if (waiting->waiting_on() == th &&
936 waiting->get_pending()->is_thread_join())
937 scheduler->wake(waiting);
940 /* Completed thread can't satisfy promises */
941 for (unsigned int i = 0; i < promises.size(); i++) {
942 Promise *promise = promises[i];
943 if (promise->thread_is_available(th->get_id()))
944 if (promise->eliminate_thread(th->get_id()))
945 priv->failed_promise = true;
947 updated = true; /* trigger rel-seq checks */
951 check_promises(curr->get_tid(), NULL, curr->get_cv());
962 * @brief Process the current action for release sequence fixup activity
964 * Performs model-checker release sequence fixups for the current action,
965 * forcing a single pending release sequence to break (with a given, potential
966 * "loose" write) or to complete (i.e., synchronize). If a pending release
967 * sequence forms a complete release sequence, then we must perform the fixup
968 * synchronization, mo_graph additions, etc.
970 * @param curr The current action; must be a release sequence fixup action
971 * @param work_queue The work queue to which to add work items as they are
974 void ModelExecution::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
976 const ModelAction *write = curr->get_node()->get_relseq_break();
977 struct release_seq *sequence = pending_rel_seqs.back();
978 pending_rel_seqs.pop_back();
980 ModelAction *acquire = sequence->acquire;
981 const ModelAction *rf = sequence->rf;
982 const ModelAction *release = sequence->release;
986 ASSERT(release->same_thread(rf));
990 * @todo Forcing a synchronization requires that we set
991 * modification order constraints. For instance, we can't allow
992 * a fixup sequence in which two separate read-acquire
993 * operations read from the same sequence, where the first one
994 * synchronizes and the other doesn't. Essentially, we can't
995 * allow any writes to insert themselves between 'release' and
999 /* Must synchronize */
1000 if (!synchronize(release, acquire))
1002 /* Re-check all pending release sequences */
1003 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1004 /* Re-check act for mo_graph edges */
1005 work_queue->push_back(MOEdgeWorkEntry(acquire));
1007 /* propagate synchronization to later actions */
1008 action_list_t::reverse_iterator rit = action_trace.rbegin();
1009 for (; (*rit) != acquire; rit++) {
1010 ModelAction *propagate = *rit;
1011 if (acquire->happens_before(propagate)) {
1012 synchronize(acquire, propagate);
1013 /* Re-check 'propagate' for mo_graph edges */
1014 work_queue->push_back(MOEdgeWorkEntry(propagate));
1018 /* Break release sequence with new edges:
1019 * release --mo--> write --mo--> rf */
1020 mo_graph->addEdge(release, write);
1021 mo_graph->addEdge(write, rf);
1024 /* See if we have realized a data race */
1029 * Initialize the current action by performing one or more of the following
1030 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1031 * in the NodeStack, manipulating backtracking sets, allocating and
1032 * initializing clock vectors, and computing the promises to fulfill.
1034 * @param curr The current action, as passed from the user context; may be
1035 * freed/invalidated after the execution of this function, with a different
1036 * action "returned" its place (pass-by-reference)
1037 * @return True if curr is a newly-explored action; false otherwise
1039 bool ModelExecution::initialize_curr_action(ModelAction **curr)
1041 ModelAction *newcurr;
1043 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1044 newcurr = process_rmw(*curr);
1047 if (newcurr->is_rmw())
1048 compute_promises(newcurr);
1054 (*curr)->set_seq_number(get_next_seq_num());
1056 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1058 /* First restore type and order in case of RMW operation */
1059 if ((*curr)->is_rmwr())
1060 newcurr->copy_typeandorder(*curr);
1062 ASSERT((*curr)->get_location() == newcurr->get_location());
1063 newcurr->copy_from_new(*curr);
1065 /* Discard duplicate ModelAction; use action from NodeStack */
1068 /* Always compute new clock vector */
1069 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1072 return false; /* Action was explored previously */
1076 /* Always compute new clock vector */
1077 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1079 /* Assign most recent release fence */
1080 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1083 * Perform one-time actions when pushing new ModelAction onto
1086 if (newcurr->is_write())
1087 compute_promises(newcurr);
1088 else if (newcurr->is_relseq_fixup())
1089 compute_relseq_breakwrites(newcurr);
1090 else if (newcurr->is_wait())
1091 newcurr->get_node()->set_misc_max(2);
1092 else if (newcurr->is_notify_one()) {
1093 newcurr->get_node()->set_misc_max(get_safe_ptr_action(&condvar_waiters_map, newcurr->get_location())->size());
1095 return true; /* This was a new ModelAction */
1100 * @brief Establish reads-from relation between two actions
1102 * Perform basic operations involved with establishing a concrete rf relation,
1103 * including setting the ModelAction data and checking for release sequences.
1105 * @param act The action that is reading (must be a read)
1106 * @param rf The action from which we are reading (must be a write)
1108 * @return True if this read established synchronization
1110 bool ModelExecution::read_from(ModelAction *act, const ModelAction *rf)
1113 ASSERT(rf->is_write());
1115 act->set_read_from(rf);
1116 if (act->is_acquire()) {
1117 rel_heads_list_t release_heads;
1118 get_release_seq_heads(act, act, &release_heads);
1119 int num_heads = release_heads.size();
1120 for (unsigned int i = 0; i < release_heads.size(); i++)
1121 if (!synchronize(release_heads[i], act))
1123 return num_heads > 0;
1129 * @brief Synchronizes two actions
1131 * When A synchronizes with B (or A --sw-> B), B inherits A's clock vector.
1132 * This function performs the synchronization as well as providing other hooks
1133 * for other checks along with synchronization.
1135 * @param first The left-hand side of the synchronizes-with relation
1136 * @param second The right-hand side of the synchronizes-with relation
1137 * @return True if the synchronization was successful (i.e., was consistent
1138 * with the execution order); false otherwise
1140 bool ModelExecution::synchronize(const ModelAction *first, ModelAction *second)
1142 if (*second < *first) {
1143 set_bad_synchronization();
1146 check_promises(first->get_tid(), second->get_cv(), first->get_cv());
1147 return second->synchronize_with(first);
1151 * Check promises and eliminate potentially-satisfying threads when a thread is
1152 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1153 * no longer satisfy a promise generated from that thread.
1155 * @param blocker The thread on which a thread is waiting
1156 * @param waiting The waiting thread
1158 void ModelExecution::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1160 for (unsigned int i = 0; i < promises.size(); i++) {
1161 Promise *promise = promises[i];
1162 if (!promise->thread_is_available(waiting->get_id()))
1164 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1165 ModelAction *reader = promise->get_reader(j);
1166 if (reader->get_tid() != blocker->get_id())
1168 if (promise->eliminate_thread(waiting->get_id())) {
1169 /* Promise has failed */
1170 priv->failed_promise = true;
1172 /* Only eliminate the 'waiting' thread once */
1180 * @brief Check whether a model action is enabled.
1182 * Checks whether a lock or join operation would be successful (i.e., is the
1183 * lock already locked, or is the joined thread already complete). If not, put
1184 * the action in a waiter list.
1186 * @param curr is the ModelAction to check whether it is enabled.
1187 * @return a bool that indicates whether the action is enabled.
1189 bool ModelExecution::check_action_enabled(ModelAction *curr) {
1190 if (curr->is_lock()) {
1191 std::mutex *lock = curr->get_mutex();
1192 struct std::mutex_state *state = lock->get_state();
1195 } else if (curr->is_thread_join()) {
1196 Thread *blocking = curr->get_thread_operand();
1197 if (!blocking->is_complete()) {
1198 thread_blocking_check_promises(blocking, get_thread(curr));
1207 * This is the heart of the model checker routine. It performs model-checking
1208 * actions corresponding to a given "current action." Among other processes, it
1209 * calculates reads-from relationships, updates synchronization clock vectors,
1210 * forms a memory_order constraints graph, and handles replay/backtrack
1211 * execution when running permutations of previously-observed executions.
1213 * @param curr The current action to process
1214 * @return The ModelAction that is actually executed; may be different than
1217 ModelAction * ModelExecution::check_current_action(ModelAction *curr)
1220 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1221 bool newly_explored = initialize_curr_action(&curr);
1225 wake_up_sleeping_actions(curr);
1227 /* Compute fairness information for CHESS yield algorithm */
1228 if (params->yieldon) {
1229 curr->get_node()->update_yield(scheduler);
1232 /* Add the action to lists before any other model-checking tasks */
1233 if (!second_part_of_rmw)
1234 add_action_to_lists(curr);
1236 /* Build may_read_from set for newly-created actions */
1237 if (newly_explored && curr->is_read())
1238 build_may_read_from(curr);
1240 /* Initialize work_queue with the "current action" work */
1241 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1242 while (!work_queue.empty() && !has_asserted()) {
1243 WorkQueueEntry work = work_queue.front();
1244 work_queue.pop_front();
1246 switch (work.type) {
1247 case WORK_CHECK_CURR_ACTION: {
1248 ModelAction *act = work.action;
1249 bool update = false; /* update this location's release seq's */
1250 bool update_all = false; /* update all release seq's */
1252 if (process_thread_action(curr))
1255 if (act->is_read() && !second_part_of_rmw && process_read(act))
1258 if (act->is_write() && process_write(act))
1261 if (act->is_fence() && process_fence(act))
1264 if (act->is_mutex_op() && process_mutex(act))
1267 if (act->is_relseq_fixup())
1268 process_relseq_fixup(curr, &work_queue);
1271 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1273 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1276 case WORK_CHECK_RELEASE_SEQ:
1277 resolve_release_sequences(work.location, &work_queue);
1279 case WORK_CHECK_MO_EDGES: {
1280 /** @todo Complete verification of work_queue */
1281 ModelAction *act = work.action;
1282 bool updated = false;
1284 if (act->is_read()) {
1285 const ModelAction *rf = act->get_reads_from();
1286 const Promise *promise = act->get_reads_from_promise();
1288 if (r_modification_order(act, rf))
1290 } else if (promise) {
1291 if (r_modification_order(act, promise))
1295 if (act->is_write()) {
1296 if (w_modification_order(act, NULL))
1299 mo_graph->commitChanges();
1302 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1311 check_curr_backtracking(curr);
1312 set_backtracking(curr);
1316 void ModelExecution::check_curr_backtracking(ModelAction *curr)
1318 Node *currnode = curr->get_node();
1319 Node *parnode = currnode->get_parent();
1321 if ((parnode && !parnode->backtrack_empty()) ||
1322 !currnode->misc_empty() ||
1323 !currnode->read_from_empty() ||
1324 !currnode->promise_empty() ||
1325 !currnode->relseq_break_empty()) {
1326 set_latest_backtrack(curr);
1330 bool ModelExecution::promises_expired() const
1332 for (unsigned int i = 0; i < promises.size(); i++) {
1333 Promise *promise = promises[i];
1334 if (promise->get_expiration() < priv->used_sequence_numbers)
1341 * This is the strongest feasibility check available.
1342 * @return whether the current trace (partial or complete) must be a prefix of
1345 bool ModelExecution::isfeasibleprefix() const
1347 return pending_rel_seqs.size() == 0 && is_feasible_prefix_ignore_relseq();
1351 * Print disagnostic information about an infeasible execution
1352 * @param prefix A string to prefix the output with; if NULL, then a default
1353 * message prefix will be provided
1355 void ModelExecution::print_infeasibility(const char *prefix) const
1359 if (mo_graph->checkForCycles())
1360 ptr += sprintf(ptr, "[mo cycle]");
1361 if (priv->failed_promise)
1362 ptr += sprintf(ptr, "[failed promise]");
1363 if (priv->too_many_reads)
1364 ptr += sprintf(ptr, "[too many reads]");
1365 if (priv->no_valid_reads)
1366 ptr += sprintf(ptr, "[no valid reads-from]");
1367 if (priv->bad_synchronization)
1368 ptr += sprintf(ptr, "[bad sw ordering]");
1369 if (promises_expired())
1370 ptr += sprintf(ptr, "[promise expired]");
1371 if (promises.size() != 0)
1372 ptr += sprintf(ptr, "[unresolved promise]");
1374 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1378 * Returns whether the current completed trace is feasible, except for pending
1379 * release sequences.
1381 bool ModelExecution::is_feasible_prefix_ignore_relseq() const
1383 return !is_infeasible() && promises.size() == 0;
1387 * Check if the current partial trace is infeasible. Does not check any
1388 * end-of-execution flags, which might rule out the execution. Thus, this is
1389 * useful only for ruling an execution as infeasible.
1390 * @return whether the current partial trace is infeasible.
1392 bool ModelExecution::is_infeasible() const
1394 return mo_graph->checkForCycles() ||
1395 priv->no_valid_reads ||
1396 priv->failed_promise ||
1397 priv->too_many_reads ||
1398 priv->bad_synchronization ||
1402 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1403 ModelAction * ModelExecution::process_rmw(ModelAction *act) {
1404 ModelAction *lastread = get_last_action(act->get_tid());
1405 lastread->process_rmw(act);
1406 if (act->is_rmw()) {
1407 if (lastread->get_reads_from())
1408 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1410 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1411 mo_graph->commitChanges();
1417 * A helper function for ModelExecution::check_recency, to check if the current
1418 * thread is able to read from a different write/promise for 'params.maxreads'
1419 * number of steps and if that write/promise should become visible (i.e., is
1420 * ordered later in the modification order). This helps model memory liveness.
1422 * @param curr The current action. Must be a read.
1423 * @param rf The write/promise from which we plan to read
1424 * @param other_rf The write/promise from which we may read
1425 * @return True if we were able to read from other_rf for params.maxreads steps
1427 template <typename T, typename U>
1428 bool ModelExecution::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1430 /* Need a different write/promise */
1431 if (other_rf->equals(rf))
1434 /* Only look for "newer" writes/promises */
1435 if (!mo_graph->checkReachable(rf, other_rf))
1438 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1439 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1440 action_list_t::reverse_iterator rit = list->rbegin();
1441 ASSERT((*rit) == curr);
1442 /* Skip past curr */
1445 /* Does this write/promise work for everyone? */
1446 for (int i = 0; i < params->maxreads; i++, rit++) {
1447 ModelAction *act = *rit;
1448 if (!act->may_read_from(other_rf))
1455 * Checks whether a thread has read from the same write or Promise for too many
1456 * times without seeing the effects of a later write/Promise.
1459 * 1) there must a different write/promise that we could read from,
1460 * 2) we must have read from the same write/promise in excess of maxreads times,
1461 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1462 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1464 * If so, we decide that the execution is no longer feasible.
1466 * @param curr The current action. Must be a read.
1467 * @param rf The ModelAction/Promise from which we might read.
1468 * @return True if the read should succeed; false otherwise
1470 template <typename T>
1471 bool ModelExecution::check_recency(ModelAction *curr, const T *rf) const
1473 if (!params->maxreads)
1476 //NOTE: Next check is just optimization, not really necessary....
1477 if (curr->get_node()->get_read_from_past_size() +
1478 curr->get_node()->get_read_from_promise_size() <= 1)
1481 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1482 int tid = id_to_int(curr->get_tid());
1483 ASSERT(tid < (int)thrd_lists->size());
1484 action_list_t *list = &(*thrd_lists)[tid];
1485 action_list_t::reverse_iterator rit = list->rbegin();
1486 ASSERT((*rit) == curr);
1487 /* Skip past curr */
1490 action_list_t::reverse_iterator ritcopy = rit;
1491 /* See if we have enough reads from the same value */
1492 for (int count = 0; count < params->maxreads; ritcopy++, count++) {
1493 if (ritcopy == list->rend())
1495 ModelAction *act = *ritcopy;
1496 if (!act->is_read())
1498 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1500 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1502 if (act->get_node()->get_read_from_past_size() +
1503 act->get_node()->get_read_from_promise_size() <= 1)
1506 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1507 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1508 if (should_read_instead(curr, rf, write))
1509 return false; /* liveness failure */
1511 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1512 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1513 if (should_read_instead(curr, rf, promise))
1514 return false; /* liveness failure */
1520 * @brief Updates the mo_graph with the constraints imposed from the current
1523 * Basic idea is the following: Go through each other thread and find
1524 * the last action that happened before our read. Two cases:
1526 * -# The action is a write: that write must either occur before
1527 * the write we read from or be the write we read from.
1528 * -# The action is a read: the write that that action read from
1529 * must occur before the write we read from or be the same write.
1531 * @param curr The current action. Must be a read.
1532 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1533 * @return True if modification order edges were added; false otherwise
1535 template <typename rf_type>
1536 bool ModelExecution::r_modification_order(ModelAction *curr, const rf_type *rf)
1538 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1541 ASSERT(curr->is_read());
1543 /* Last SC fence in the current thread */
1544 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1545 ModelAction *last_sc_write = NULL;
1546 if (curr->is_seqcst())
1547 last_sc_write = get_last_seq_cst_write(curr);
1549 /* Iterate over all threads */
1550 for (i = 0; i < thrd_lists->size(); i++) {
1551 /* Last SC fence in thread i */
1552 ModelAction *last_sc_fence_thread_local = NULL;
1553 if (int_to_id((int)i) != curr->get_tid())
1554 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1556 /* Last SC fence in thread i, before last SC fence in current thread */
1557 ModelAction *last_sc_fence_thread_before = NULL;
1558 if (last_sc_fence_local)
1559 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1561 /* Iterate over actions in thread, starting from most recent */
1562 action_list_t *list = &(*thrd_lists)[i];
1563 action_list_t::reverse_iterator rit;
1564 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1565 ModelAction *act = *rit;
1570 /* Don't want to add reflexive edges on 'rf' */
1571 if (act->equals(rf)) {
1572 if (act->happens_before(curr))
1578 if (act->is_write()) {
1579 /* C++, Section 29.3 statement 5 */
1580 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1581 *act < *last_sc_fence_thread_local) {
1582 added = mo_graph->addEdge(act, rf) || added;
1585 /* C++, Section 29.3 statement 4 */
1586 else if (act->is_seqcst() && last_sc_fence_local &&
1587 *act < *last_sc_fence_local) {
1588 added = mo_graph->addEdge(act, rf) || added;
1591 /* C++, Section 29.3 statement 6 */
1592 else if (last_sc_fence_thread_before &&
1593 *act < *last_sc_fence_thread_before) {
1594 added = mo_graph->addEdge(act, rf) || added;
1599 /* C++, Section 29.3 statement 3 (second subpoint) */
1600 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1601 added = mo_graph->addEdge(act, rf) || added;
1606 * Include at most one act per-thread that "happens
1609 if (act->happens_before(curr)) {
1610 if (act->is_write()) {
1611 added = mo_graph->addEdge(act, rf) || added;
1613 const ModelAction *prevrf = act->get_reads_from();
1614 const Promise *prevrf_promise = act->get_reads_from_promise();
1616 if (!prevrf->equals(rf))
1617 added = mo_graph->addEdge(prevrf, rf) || added;
1618 } else if (!prevrf_promise->equals(rf)) {
1619 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1628 * All compatible, thread-exclusive promises must be ordered after any
1629 * concrete loads from the same thread
1631 for (unsigned int i = 0; i < promises.size(); i++)
1632 if (promises[i]->is_compatible_exclusive(curr))
1633 added = mo_graph->addEdge(rf, promises[i]) || added;
1639 * Updates the mo_graph with the constraints imposed from the current write.
1641 * Basic idea is the following: Go through each other thread and find
1642 * the lastest action that happened before our write. Two cases:
1644 * (1) The action is a write => that write must occur before
1647 * (2) The action is a read => the write that that action read from
1648 * must occur before the current write.
1650 * This method also handles two other issues:
1652 * (I) Sequential Consistency: Making sure that if the current write is
1653 * seq_cst, that it occurs after the previous seq_cst write.
1655 * (II) Sending the write back to non-synchronizing reads.
1657 * @param curr The current action. Must be a write.
1658 * @param send_fv A vector for stashing reads to which we may pass our future
1659 * value. If NULL, then don't record any future values.
1660 * @return True if modification order edges were added; false otherwise
1662 bool ModelExecution::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1664 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1667 ASSERT(curr->is_write());
1669 if (curr->is_seqcst()) {
1670 /* We have to at least see the last sequentially consistent write,
1671 so we are initialized. */
1672 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1673 if (last_seq_cst != NULL) {
1674 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1678 /* Last SC fence in the current thread */
1679 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1681 /* Iterate over all threads */
1682 for (i = 0; i < thrd_lists->size(); i++) {
1683 /* Last SC fence in thread i, before last SC fence in current thread */
1684 ModelAction *last_sc_fence_thread_before = NULL;
1685 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1686 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1688 /* Iterate over actions in thread, starting from most recent */
1689 action_list_t *list = &(*thrd_lists)[i];
1690 action_list_t::reverse_iterator rit;
1691 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1692 ModelAction *act = *rit;
1695 * 1) If RMW and it actually read from something, then we
1696 * already have all relevant edges, so just skip to next
1699 * 2) If RMW and it didn't read from anything, we should
1700 * whatever edge we can get to speed up convergence.
1702 * 3) If normal write, we need to look at earlier actions, so
1703 * continue processing list.
1705 if (curr->is_rmw()) {
1706 if (curr->get_reads_from() != NULL)
1714 /* C++, Section 29.3 statement 7 */
1715 if (last_sc_fence_thread_before && act->is_write() &&
1716 *act < *last_sc_fence_thread_before) {
1717 added = mo_graph->addEdge(act, curr) || added;
1722 * Include at most one act per-thread that "happens
1725 if (act->happens_before(curr)) {
1727 * Note: if act is RMW, just add edge:
1729 * The following edge should be handled elsewhere:
1730 * readfrom(act) --mo--> act
1732 if (act->is_write())
1733 added = mo_graph->addEdge(act, curr) || added;
1734 else if (act->is_read()) {
1735 //if previous read accessed a null, just keep going
1736 if (act->get_reads_from() == NULL)
1738 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
1741 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1742 !act->same_thread(curr)) {
1743 /* We have an action that:
1744 (1) did not happen before us
1745 (2) is a read and we are a write
1746 (3) cannot synchronize with us
1747 (4) is in a different thread
1749 that read could potentially read from our write. Note that
1750 these checks are overly conservative at this point, we'll
1751 do more checks before actually removing the
1755 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
1756 if (!is_infeasible())
1757 send_fv->push_back(act);
1758 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
1759 add_future_value(curr, act);
1766 * All compatible, thread-exclusive promises must be ordered after any
1767 * concrete stores to the same thread, or else they can be merged with
1770 for (unsigned int i = 0; i < promises.size(); i++)
1771 if (promises[i]->is_compatible_exclusive(curr))
1772 added = mo_graph->addEdge(curr, promises[i]) || added;
1777 /** Arbitrary reads from the future are not allowed. Section 29.3
1778 * part 9 places some constraints. This method checks one result of constraint
1779 * constraint. Others require compiler support. */
1780 bool ModelExecution::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
1782 if (!writer->is_rmw())
1785 if (!reader->is_rmw())
1788 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1789 if (search == reader)
1791 if (search->get_tid() == reader->get_tid() &&
1792 search->happens_before(reader))
1800 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1801 * some constraints. This method checks one the following constraint (others
1802 * require compiler support):
1804 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1806 bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1808 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(reader->get_location());
1810 /* Iterate over all threads */
1811 for (i = 0; i < thrd_lists->size(); i++) {
1812 const ModelAction *write_after_read = NULL;
1814 /* Iterate over actions in thread, starting from most recent */
1815 action_list_t *list = &(*thrd_lists)[i];
1816 action_list_t::reverse_iterator rit;
1817 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1818 ModelAction *act = *rit;
1820 /* Don't disallow due to act == reader */
1821 if (!reader->happens_before(act) || reader == act)
1823 else if (act->is_write())
1824 write_after_read = act;
1825 else if (act->is_read() && act->get_reads_from() != NULL)
1826 write_after_read = act->get_reads_from();
1829 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
1836 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1837 * The ModelAction under consideration is expected to be taking part in
1838 * release/acquire synchronization as an object of the "reads from" relation.
1839 * Note that this can only provide release sequence support for RMW chains
1840 * which do not read from the future, as those actions cannot be traced until
1841 * their "promise" is fulfilled. Similarly, we may not even establish the
1842 * presence of a release sequence with certainty, as some modification order
1843 * constraints may be decided further in the future. Thus, this function
1844 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1845 * and a boolean representing certainty.
1847 * @param rf The action that might be part of a release sequence. Must be a
1849 * @param release_heads A pass-by-reference style return parameter. After
1850 * execution of this function, release_heads will contain the heads of all the
1851 * relevant release sequences, if any exists with certainty
1852 * @param pending A pass-by-reference style return parameter which is only used
1853 * when returning false (i.e., uncertain). Returns most information regarding
1854 * an uncertain release sequence, including any write operations that might
1855 * break the sequence.
1856 * @return true, if the ModelExecution is certain that release_heads is complete;
1859 bool ModelExecution::release_seq_heads(const ModelAction *rf,
1860 rel_heads_list_t *release_heads,
1861 struct release_seq *pending) const
1863 /* Only check for release sequences if there are no cycles */
1864 if (mo_graph->checkForCycles())
1867 for ( ; rf != NULL; rf = rf->get_reads_from()) {
1868 ASSERT(rf->is_write());
1870 if (rf->is_release())
1871 release_heads->push_back(rf);
1872 else if (rf->get_last_fence_release())
1873 release_heads->push_back(rf->get_last_fence_release());
1875 break; /* End of RMW chain */
1877 /** @todo Need to be smarter here... In the linux lock
1878 * example, this will run to the beginning of the program for
1880 /** @todo The way to be smarter here is to keep going until 1
1881 * thread has a release preceded by an acquire and you've seen
1884 /* acq_rel RMW is a sufficient stopping condition */
1885 if (rf->is_acquire() && rf->is_release())
1886 return true; /* complete */
1889 /* read from future: need to settle this later */
1891 return false; /* incomplete */
1894 if (rf->is_release())
1895 return true; /* complete */
1897 /* else relaxed write
1898 * - check for fence-release in the same thread (29.8, stmt. 3)
1899 * - check modification order for contiguous subsequence
1900 * -> rf must be same thread as release */
1902 const ModelAction *fence_release = rf->get_last_fence_release();
1903 /* Synchronize with a fence-release unconditionally; we don't need to
1904 * find any more "contiguous subsequence..." for it */
1906 release_heads->push_back(fence_release);
1908 int tid = id_to_int(rf->get_tid());
1909 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(rf->get_location());
1910 action_list_t *list = &(*thrd_lists)[tid];
1911 action_list_t::const_reverse_iterator rit;
1913 /* Find rf in the thread list */
1914 rit = std::find(list->rbegin(), list->rend(), rf);
1915 ASSERT(rit != list->rend());
1917 /* Find the last {write,fence}-release */
1918 for (; rit != list->rend(); rit++) {
1919 if (fence_release && *(*rit) < *fence_release)
1921 if ((*rit)->is_release())
1924 if (rit == list->rend()) {
1925 /* No write-release in this thread */
1926 return true; /* complete */
1927 } else if (fence_release && *(*rit) < *fence_release) {
1928 /* The fence-release is more recent (and so, "stronger") than
1929 * the most recent write-release */
1930 return true; /* complete */
1931 } /* else, need to establish contiguous release sequence */
1932 ModelAction *release = *rit;
1934 ASSERT(rf->same_thread(release));
1936 pending->writes.clear();
1938 bool certain = true;
1939 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1940 if (id_to_int(rf->get_tid()) == (int)i)
1942 list = &(*thrd_lists)[i];
1944 /* Can we ensure no future writes from this thread may break
1945 * the release seq? */
1946 bool future_ordered = false;
1948 ModelAction *last = get_last_action(int_to_id(i));
1949 Thread *th = get_thread(int_to_id(i));
1950 if ((last && rf->happens_before(last)) ||
1953 future_ordered = true;
1955 ASSERT(!th->is_model_thread() || future_ordered);
1957 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1958 const ModelAction *act = *rit;
1959 /* Reach synchronization -> this thread is complete */
1960 if (act->happens_before(release))
1962 if (rf->happens_before(act)) {
1963 future_ordered = true;
1967 /* Only non-RMW writes can break release sequences */
1968 if (!act->is_write() || act->is_rmw())
1971 /* Check modification order */
1972 if (mo_graph->checkReachable(rf, act)) {
1973 /* rf --mo--> act */
1974 future_ordered = true;
1977 if (mo_graph->checkReachable(act, release))
1978 /* act --mo--> release */
1980 if (mo_graph->checkReachable(release, act) &&
1981 mo_graph->checkReachable(act, rf)) {
1982 /* release --mo-> act --mo--> rf */
1983 return true; /* complete */
1985 /* act may break release sequence */
1986 pending->writes.push_back(act);
1989 if (!future_ordered)
1990 certain = false; /* This thread is uncertain */
1994 release_heads->push_back(release);
1995 pending->writes.clear();
1997 pending->release = release;
2004 * An interface for getting the release sequence head(s) with which a
2005 * given ModelAction must synchronize. This function only returns a non-empty
2006 * result when it can locate a release sequence head with certainty. Otherwise,
2007 * it may mark the internal state of the ModelExecution so that it will handle
2008 * the release sequence at a later time, causing @a acquire to update its
2009 * synchronization at some later point in execution.
2011 * @param acquire The 'acquire' action that may synchronize with a release
2013 * @param read The read action that may read from a release sequence; this may
2014 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2015 * when 'acquire' is a fence-acquire)
2016 * @param release_heads A pass-by-reference return parameter. Will be filled
2017 * with the head(s) of the release sequence(s), if they exists with certainty.
2018 * @see ModelExecution::release_seq_heads
2020 void ModelExecution::get_release_seq_heads(ModelAction *acquire,
2021 ModelAction *read, rel_heads_list_t *release_heads)
2023 const ModelAction *rf = read->get_reads_from();
2024 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2025 sequence->acquire = acquire;
2026 sequence->read = read;
2028 if (!release_seq_heads(rf, release_heads, sequence)) {
2029 /* add act to 'lazy checking' list */
2030 pending_rel_seqs.push_back(sequence);
2032 snapshot_free(sequence);
2037 * Attempt to resolve all stashed operations that might synchronize with a
2038 * release sequence for a given location. This implements the "lazy" portion of
2039 * determining whether or not a release sequence was contiguous, since not all
2040 * modification order information is present at the time an action occurs.
2042 * @param location The location/object that should be checked for release
2043 * sequence resolutions. A NULL value means to check all locations.
2044 * @param work_queue The work queue to which to add work items as they are
2046 * @return True if any updates occurred (new synchronization, new mo_graph
2049 bool ModelExecution::resolve_release_sequences(void *location, work_queue_t *work_queue)
2051 bool updated = false;
2052 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs.begin();
2053 while (it != pending_rel_seqs.end()) {
2054 struct release_seq *pending = *it;
2055 ModelAction *acquire = pending->acquire;
2056 const ModelAction *read = pending->read;
2058 /* Only resolve sequences on the given location, if provided */
2059 if (location && read->get_location() != location) {
2064 const ModelAction *rf = read->get_reads_from();
2065 rel_heads_list_t release_heads;
2067 complete = release_seq_heads(rf, &release_heads, pending);
2068 for (unsigned int i = 0; i < release_heads.size(); i++)
2069 if (!acquire->has_synchronized_with(release_heads[i]))
2070 if (synchronize(release_heads[i], acquire))
2074 /* Re-check all pending release sequences */
2075 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2076 /* Re-check read-acquire for mo_graph edges */
2077 if (acquire->is_read())
2078 work_queue->push_back(MOEdgeWorkEntry(acquire));
2080 /* propagate synchronization to later actions */
2081 action_list_t::reverse_iterator rit = action_trace.rbegin();
2082 for (; (*rit) != acquire; rit++) {
2083 ModelAction *propagate = *rit;
2084 if (acquire->happens_before(propagate)) {
2085 synchronize(acquire, propagate);
2086 /* Re-check 'propagate' for mo_graph edges */
2087 work_queue->push_back(MOEdgeWorkEntry(propagate));
2092 it = pending_rel_seqs.erase(it);
2093 snapshot_free(pending);
2099 // If we resolved promises or data races, see if we have realized a data race.
2106 * Performs various bookkeeping operations for the current ModelAction. For
2107 * instance, adds action to the per-object, per-thread action vector and to the
2108 * action trace list of all thread actions.
2110 * @param act is the ModelAction to add.
2112 void ModelExecution::add_action_to_lists(ModelAction *act)
2114 int tid = id_to_int(act->get_tid());
2115 ModelAction *uninit = NULL;
2117 action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
2118 if (list->empty() && act->is_atomic_var()) {
2119 uninit = get_uninitialized_action(act);
2120 uninit_id = id_to_int(uninit->get_tid());
2121 list->push_front(uninit);
2123 list->push_back(act);
2125 action_trace.push_back(act);
2127 action_trace.push_front(uninit);
2129 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
2130 if (tid >= (int)vec->size())
2131 vec->resize(priv->next_thread_id);
2132 (*vec)[tid].push_back(act);
2134 (*vec)[uninit_id].push_front(uninit);
2136 if ((int)thrd_last_action.size() <= tid)
2137 thrd_last_action.resize(get_num_threads());
2138 thrd_last_action[tid] = act;
2140 thrd_last_action[uninit_id] = uninit;
2142 if (act->is_fence() && act->is_release()) {
2143 if ((int)thrd_last_fence_release.size() <= tid)
2144 thrd_last_fence_release.resize(get_num_threads());
2145 thrd_last_fence_release[tid] = act;
2148 if (act->is_wait()) {
2149 void *mutex_loc = (void *) act->get_value();
2150 get_safe_ptr_action(&obj_map, mutex_loc)->push_back(act);
2152 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
2153 if (tid >= (int)vec->size())
2154 vec->resize(priv->next_thread_id);
2155 (*vec)[tid].push_back(act);
2160 * @brief Get the last action performed by a particular Thread
2161 * @param tid The thread ID of the Thread in question
2162 * @return The last action in the thread
2164 ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
2166 int threadid = id_to_int(tid);
2167 if (threadid < (int)thrd_last_action.size())
2168 return thrd_last_action[id_to_int(tid)];
2174 * @brief Get the last fence release performed by a particular Thread
2175 * @param tid The thread ID of the Thread in question
2176 * @return The last fence release in the thread, if one exists; NULL otherwise
2178 ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
2180 int threadid = id_to_int(tid);
2181 if (threadid < (int)thrd_last_fence_release.size())
2182 return thrd_last_fence_release[id_to_int(tid)];
2188 * Gets the last memory_order_seq_cst write (in the total global sequence)
2189 * performed on a particular object (i.e., memory location), not including the
2191 * @param curr The current ModelAction; also denotes the object location to
2193 * @return The last seq_cst write
2195 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
2197 void *location = curr->get_location();
2198 action_list_t *list = obj_map.get(location);
2199 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2200 action_list_t::reverse_iterator rit;
2201 for (rit = list->rbegin(); (*rit) != curr; rit++)
2203 rit++; /* Skip past curr */
2204 for ( ; rit != list->rend(); rit++)
2205 if ((*rit)->is_write() && (*rit)->is_seqcst())
2211 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2212 * performed in a particular thread, prior to a particular fence.
2213 * @param tid The ID of the thread to check
2214 * @param before_fence The fence from which to begin the search; if NULL, then
2215 * search for the most recent fence in the thread.
2216 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2218 ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2220 /* All fences should have location FENCE_LOCATION */
2221 action_list_t *list = obj_map.get(FENCE_LOCATION);
2226 action_list_t::reverse_iterator rit = list->rbegin();
2229 for (; rit != list->rend(); rit++)
2230 if (*rit == before_fence)
2233 ASSERT(*rit == before_fence);
2237 for (; rit != list->rend(); rit++)
2238 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2244 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2245 * location). This function identifies the mutex according to the current
2246 * action, which is presumed to perform on the same mutex.
2247 * @param curr The current ModelAction; also denotes the object location to
2249 * @return The last unlock operation
2251 ModelAction * ModelExecution::get_last_unlock(ModelAction *curr) const
2253 void *location = curr->get_location();
2254 action_list_t *list = obj_map.get(location);
2255 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2256 action_list_t::reverse_iterator rit;
2257 for (rit = list->rbegin(); rit != list->rend(); rit++)
2258 if ((*rit)->is_unlock() || (*rit)->is_wait())
2263 ModelAction * ModelExecution::get_parent_action(thread_id_t tid) const
2265 ModelAction *parent = get_last_action(tid);
2267 parent = get_thread(tid)->get_creation();
2272 * Returns the clock vector for a given thread.
2273 * @param tid The thread whose clock vector we want
2274 * @return Desired clock vector
2276 ClockVector * ModelExecution::get_cv(thread_id_t tid) const
2278 return get_parent_action(tid)->get_cv();
2282 * @brief Find the promise (if any) to resolve for the current action and
2283 * remove it from the pending promise vector
2284 * @param curr The current ModelAction. Should be a write.
2285 * @return The Promise to resolve, if any; otherwise NULL
2287 Promise * ModelExecution::pop_promise_to_resolve(const ModelAction *curr)
2289 for (unsigned int i = 0; i < promises.size(); i++)
2290 if (curr->get_node()->get_promise(i)) {
2291 Promise *ret = promises[i];
2292 promises.erase(promises.begin() + i);
2299 * Resolve a Promise with a current write.
2300 * @param write The ModelAction that is fulfilling Promises
2301 * @param promise The Promise to resolve
2302 * @return True if the Promise was successfully resolved; false otherwise
2304 bool ModelExecution::resolve_promise(ModelAction *write, Promise *promise)
2306 ModelVector<ModelAction *> actions_to_check;
2308 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2309 ModelAction *read = promise->get_reader(i);
2310 read_from(read, write);
2311 actions_to_check.push_back(read);
2313 /* Make sure the promise's value matches the write's value */
2314 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2315 if (!mo_graph->resolvePromise(promise, write))
2316 priv->failed_promise = true;
2319 * @todo It is possible to end up in an inconsistent state, where a
2320 * "resolved" promise may still be referenced if
2321 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2323 * Note that the inconsistency only matters when dumping mo_graph to
2329 //Check whether reading these writes has made threads unable to
2331 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2332 ModelAction *read = actions_to_check[i];
2333 mo_check_promises(read, true);
2340 * Compute the set of promises that could potentially be satisfied by this
2341 * action. Note that the set computation actually appears in the Node, not in
2343 * @param curr The ModelAction that may satisfy promises
2345 void ModelExecution::compute_promises(ModelAction *curr)
2347 for (unsigned int i = 0; i < promises.size(); i++) {
2348 Promise *promise = promises[i];
2349 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2352 bool satisfy = true;
2353 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2354 const ModelAction *act = promise->get_reader(j);
2355 if (act->happens_before(curr) ||
2356 act->could_synchronize_with(curr)) {
2362 curr->get_node()->set_promise(i);
2366 /** Checks promises in response to change in ClockVector Threads. */
2367 void ModelExecution::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2369 for (unsigned int i = 0; i < promises.size(); i++) {
2370 Promise *promise = promises[i];
2371 if (!promise->thread_is_available(tid))
2373 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2374 const ModelAction *act = promise->get_reader(j);
2375 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2376 merge_cv->synchronized_since(act)) {
2377 if (promise->eliminate_thread(tid)) {
2378 /* Promise has failed */
2379 priv->failed_promise = true;
2387 void ModelExecution::check_promises_thread_disabled()
2389 for (unsigned int i = 0; i < promises.size(); i++) {
2390 Promise *promise = promises[i];
2391 if (promise->has_failed()) {
2392 priv->failed_promise = true;
2399 * @brief Checks promises in response to addition to modification order for
2402 * We test whether threads are still available for satisfying promises after an
2403 * addition to our modification order constraints. Those that are unavailable
2404 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2405 * that promise has failed.
2407 * @param act The ModelAction which updated the modification order
2408 * @param is_read_check Should be true if act is a read and we must check for
2409 * updates to the store from which it read (there is a distinction here for
2410 * RMW's, which are both a load and a store)
2412 void ModelExecution::mo_check_promises(const ModelAction *act, bool is_read_check)
2414 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2416 for (unsigned int i = 0; i < promises.size(); i++) {
2417 Promise *promise = promises[i];
2419 // Is this promise on the same location?
2420 if (!promise->same_location(write))
2423 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2424 const ModelAction *pread = promise->get_reader(j);
2425 if (!pread->happens_before(act))
2427 if (mo_graph->checkPromise(write, promise)) {
2428 priv->failed_promise = true;
2434 // Don't do any lookups twice for the same thread
2435 if (!promise->thread_is_available(act->get_tid()))
2438 if (mo_graph->checkReachable(promise, write)) {
2439 if (mo_graph->checkPromise(write, promise)) {
2440 priv->failed_promise = true;
2448 * Compute the set of writes that may break the current pending release
2449 * sequence. This information is extracted from previou release sequence
2452 * @param curr The current ModelAction. Must be a release sequence fixup
2455 void ModelExecution::compute_relseq_breakwrites(ModelAction *curr)
2457 if (pending_rel_seqs.empty())
2460 struct release_seq *pending = pending_rel_seqs.back();
2461 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2462 const ModelAction *write = pending->writes[i];
2463 curr->get_node()->add_relseq_break(write);
2466 /* NULL means don't break the sequence; just synchronize */
2467 curr->get_node()->add_relseq_break(NULL);
2471 * Build up an initial set of all past writes that this 'read' action may read
2472 * from, as well as any previously-observed future values that must still be valid.
2474 * @param curr is the current ModelAction that we are exploring; it must be a
2477 void ModelExecution::build_may_read_from(ModelAction *curr)
2479 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
2481 ASSERT(curr->is_read());
2483 ModelAction *last_sc_write = NULL;
2485 if (curr->is_seqcst())
2486 last_sc_write = get_last_seq_cst_write(curr);
2488 /* Iterate over all threads */
2489 for (i = 0; i < thrd_lists->size(); i++) {
2490 /* Iterate over actions in thread, starting from most recent */
2491 action_list_t *list = &(*thrd_lists)[i];
2492 action_list_t::reverse_iterator rit;
2493 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2494 ModelAction *act = *rit;
2496 /* Only consider 'write' actions */
2497 if (!act->is_write() || act == curr)
2500 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2501 bool allow_read = true;
2503 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2505 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2509 /* Only add feasible reads */
2510 mo_graph->startChanges();
2511 r_modification_order(curr, act);
2512 if (!is_infeasible())
2513 curr->get_node()->add_read_from_past(act);
2514 mo_graph->rollbackChanges();
2517 /* Include at most one act per-thread that "happens before" curr */
2518 if (act->happens_before(curr))
2523 /* Inherit existing, promised future values */
2524 for (i = 0; i < promises.size(); i++) {
2525 const Promise *promise = promises[i];
2526 const ModelAction *promise_read = promise->get_reader(0);
2527 if (promise_read->same_var(curr)) {
2528 /* Only add feasible future-values */
2529 mo_graph->startChanges();
2530 r_modification_order(curr, promise);
2531 if (!is_infeasible())
2532 curr->get_node()->add_read_from_promise(promise_read);
2533 mo_graph->rollbackChanges();
2537 /* We may find no valid may-read-from only if the execution is doomed */
2538 if (!curr->get_node()->read_from_size()) {
2539 priv->no_valid_reads = true;
2543 if (DBG_ENABLED()) {
2544 model_print("Reached read action:\n");
2546 model_print("Printing read_from_past\n");
2547 curr->get_node()->print_read_from_past();
2548 model_print("End printing read_from_past\n");
2552 bool ModelExecution::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2554 for ( ; write != NULL; write = write->get_reads_from()) {
2555 /* UNINIT actions don't have a Node, and they never sleep */
2556 if (write->is_uninitialized())
2558 Node *prevnode = write->get_node()->get_parent();
2560 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2561 if (write->is_release() && thread_sleep)
2563 if (!write->is_rmw())
2570 * @brief Get an action representing an uninitialized atomic
2572 * This function may create a new one or try to retrieve one from the NodeStack
2574 * @param curr The current action, which prompts the creation of an UNINIT action
2575 * @return A pointer to the UNINIT ModelAction
2577 ModelAction * ModelExecution::get_uninitialized_action(const ModelAction *curr) const
2579 Node *node = curr->get_node();
2580 ModelAction *act = node->get_uninit_action();
2582 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), params->uninitvalue, model_thread);
2583 node->set_uninit_action(act);
2585 act->create_cv(NULL);
2589 static void print_list(const action_list_t *list)
2591 action_list_t::const_iterator it;
2593 model_print("---------------------------------------------------------------------\n");
2595 unsigned int hash = 0;
2597 for (it = list->begin(); it != list->end(); it++) {
2598 const ModelAction *act = *it;
2599 if (act->get_seq_number() > 0)
2601 hash = hash^(hash<<3)^((*it)->hash());
2603 model_print("HASH %u\n", hash);
2604 model_print("---------------------------------------------------------------------\n");
2607 #if SUPPORT_MOD_ORDER_DUMP
2608 void ModelExecution::dumpGraph(char *filename) const
2611 sprintf(buffer, "%s.dot", filename);
2612 FILE *file = fopen(buffer, "w");
2613 fprintf(file, "digraph %s {\n", filename);
2614 mo_graph->dumpNodes(file);
2615 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2617 for (action_list_t::iterator it = action_trace.begin(); it != action_trace.end(); it++) {
2618 ModelAction *act = *it;
2619 if (act->is_read()) {
2620 mo_graph->dot_print_node(file, act);
2621 if (act->get_reads_from())
2622 mo_graph->dot_print_edge(file,
2623 act->get_reads_from(),
2625 "label=\"rf\", color=red, weight=2");
2627 mo_graph->dot_print_edge(file,
2628 act->get_reads_from_promise(),
2630 "label=\"rf\", color=red");
2632 if (thread_array[act->get_tid()]) {
2633 mo_graph->dot_print_edge(file,
2634 thread_array[id_to_int(act->get_tid())],
2636 "label=\"sb\", color=blue, weight=400");
2639 thread_array[act->get_tid()] = act;
2641 fprintf(file, "}\n");
2642 model_free(thread_array);
2647 /** @brief Prints an execution trace summary. */
2648 void ModelExecution::print_summary() const
2650 #if SUPPORT_MOD_ORDER_DUMP
2651 char buffername[100];
2652 sprintf(buffername, "exec%04u", get_execution_number());
2653 mo_graph->dumpGraphToFile(buffername);
2654 sprintf(buffername, "graph%04u", get_execution_number());
2655 dumpGraph(buffername);
2658 model_print("Execution %d:", get_execution_number());
2659 if (isfeasibleprefix()) {
2660 if (scheduler->all_threads_sleeping())
2661 model_print(" SLEEP-SET REDUNDANT");
2664 print_infeasibility(" INFEASIBLE");
2665 print_list(&action_trace);
2667 if (!promises.empty()) {
2668 model_print("Pending promises:\n");
2669 for (unsigned int i = 0; i < promises.size(); i++) {
2670 model_print(" [P%u] ", i);
2671 promises[i]->print();
2678 * Add a Thread to the system for the first time. Should only be called once
2680 * @param t The Thread to add
2682 void ModelExecution::add_thread(Thread *t)
2684 unsigned int i = id_to_int(t->get_id());
2685 if (i >= thread_map.size())
2686 thread_map.resize(i + 1);
2688 if (!t->is_model_thread())
2689 scheduler->add_thread(t);
2693 * @brief Get a Thread reference by its ID
2694 * @param tid The Thread's ID
2695 * @return A Thread reference
2697 Thread * ModelExecution::get_thread(thread_id_t tid) const
2699 unsigned int i = id_to_int(tid);
2700 if (i < thread_map.size())
2701 return thread_map[i];
2706 * @brief Get a reference to the Thread in which a ModelAction was executed
2707 * @param act The ModelAction
2708 * @return A Thread reference
2710 Thread * ModelExecution::get_thread(const ModelAction *act) const
2712 return get_thread(act->get_tid());
2716 * @brief Get a Promise's "promise number"
2718 * A "promise number" is an index number that is unique to a promise, valid
2719 * only for a specific snapshot of an execution trace. Promises may come and go
2720 * as they are generated an resolved, so an index only retains meaning for the
2723 * @param promise The Promise to check
2724 * @return The promise index, if the promise still is valid; otherwise -1
2726 int ModelExecution::get_promise_number(const Promise *promise) const
2728 for (unsigned int i = 0; i < promises.size(); i++)
2729 if (promises[i] == promise)
2736 * @brief Check if a Thread is currently enabled
2737 * @param t The Thread to check
2738 * @return True if the Thread is currently enabled
2740 bool ModelExecution::is_enabled(Thread *t) const
2742 return scheduler->is_enabled(t);
2746 * @brief Check if a Thread is currently enabled
2747 * @param tid The ID of the Thread to check
2748 * @return True if the Thread is currently enabled
2750 bool ModelExecution::is_enabled(thread_id_t tid) const
2752 return scheduler->is_enabled(tid);
2756 * @brief Select the next thread to execute based on the curren action
2758 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
2759 * actions should be followed by the execution of their child thread. In either
2760 * case, the current action should determine the next thread schedule.
2762 * @param curr The current action
2763 * @return The next thread to run, if the current action will determine this
2764 * selection; otherwise NULL
2766 Thread * ModelExecution::action_select_next_thread(const ModelAction *curr) const
2768 /* Do not split atomic RMW */
2769 if (curr->is_rmwr())
2770 return get_thread(curr);
2771 /* Follow CREATE with the created thread */
2772 if (curr->get_type() == THREAD_CREATE)
2773 return curr->get_thread_operand();
2777 /** @return True if the execution has taken too many steps */
2778 bool ModelExecution::too_many_steps() const
2780 return params->bound != 0 && priv->used_sequence_numbers > params->bound;
2784 * Takes the next step in the execution, if possible.
2785 * @param curr The current step to take
2786 * @return Returns the next Thread to run, if any; NULL if this execution
2789 Thread * ModelExecution::take_step(ModelAction *curr)
2791 Thread *curr_thrd = get_thread(curr);
2792 ASSERT(curr_thrd->get_state() == THREAD_READY);
2794 ASSERT(check_action_enabled(curr)); /* May have side effects? */
2795 curr = check_current_action(curr);
2798 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
2799 scheduler->remove_thread(curr_thrd);
2801 return action_select_next_thread(curr);
2805 * Launch end-of-execution release sequence fixups only when
2806 * the execution is otherwise feasible AND there are:
2808 * (1) pending release sequences
2809 * (2) pending assertions that could be invalidated by a change
2810 * in clock vectors (i.e., data races)
2811 * (3) no pending promises
2813 void ModelExecution::fixup_release_sequences()
2815 while (!pending_rel_seqs.empty() &&
2816 is_feasible_prefix_ignore_relseq() &&
2817 !unrealizedraces.empty()) {
2818 model_print("*** WARNING: release sequence fixup action "
2819 "(%zu pending release seuqence(s)) ***\n",
2820 pending_rel_seqs.size());
2821 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2822 std::memory_order_seq_cst, NULL, VALUE_NONE,