10 #include "snapshot-interface.h"
12 #include "clockvector.h"
13 #include "cyclegraph.h"
16 #include "threads-model.h"
19 #define INITIAL_THREAD_ID 0
24 bug_message(const char *str) {
25 const char *fmt = " [BUG] %s\n";
26 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
27 sprintf(msg, fmt, str);
29 ~bug_message() { if (msg) snapshot_free(msg); }
32 void print() { model_print("%s", msg); }
38 * Structure for holding small ModelChecker members that should be snapshotted
40 struct model_snapshot_members {
41 model_snapshot_members() :
42 /* First thread created will have id INITIAL_THREAD_ID */
43 next_thread_id(INITIAL_THREAD_ID),
44 used_sequence_numbers(0),
48 failed_promise(false),
49 too_many_reads(false),
50 no_valid_reads(false),
51 bad_synchronization(false),
55 ~model_snapshot_members() {
56 for (unsigned int i = 0; i < bugs.size(); i++)
61 unsigned int next_thread_id;
62 modelclock_t used_sequence_numbers;
63 ModelAction *next_backtrack;
64 SnapVector<bug_message *> bugs;
65 struct execution_stats stats;
69 /** @brief Incorrectly-ordered synchronization was made */
70 bool bad_synchronization;
76 /** @brief Constructor */
77 ModelChecker::ModelChecker(struct model_params params) :
78 /* Initialize default scheduler */
80 scheduler(new Scheduler()),
82 earliest_diverge(NULL),
83 action_trace(new action_list_t()),
84 thread_map(new HashTable<int, Thread *, int>()),
85 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
86 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
87 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
88 obj_thrd_map(new HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4 >()),
89 promises(new SnapVector<Promise *>()),
90 futurevalues(new SnapVector<struct PendingFutureValue>()),
91 pending_rel_seqs(new SnapVector<struct release_seq *>()),
92 thrd_last_action(new SnapVector<ModelAction *>(1)),
93 thrd_last_fence_release(new SnapVector<ModelAction *>()),
94 node_stack(new NodeStack()),
95 priv(new struct model_snapshot_members()),
96 mo_graph(new CycleGraph())
98 /* Initialize a model-checker thread, for special ModelActions */
99 model_thread = new Thread(get_next_id());
100 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
103 /** @brief Destructor */
104 ModelChecker::~ModelChecker()
106 for (unsigned int i = 0; i < get_num_threads(); i++)
107 delete thread_map->get(i);
112 delete lock_waiters_map;
113 delete condvar_waiters_map;
116 for (unsigned int i = 0; i < promises->size(); i++)
117 delete (*promises)[i];
120 delete pending_rel_seqs;
122 delete thrd_last_action;
123 delete thrd_last_fence_release;
130 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
132 action_list_t *tmp = hash->get(ptr);
134 tmp = new action_list_t();
140 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
142 SnapVector<action_list_t> *tmp = hash->get(ptr);
144 tmp = new SnapVector<action_list_t>();
151 * Restores user program to initial state and resets all model-checker data
154 void ModelChecker::reset_to_initial_state()
156 DEBUG("+++ Resetting to initial state +++\n");
157 node_stack->reset_execution();
159 /* Print all model-checker output before rollback */
163 * FIXME: if we utilize partial rollback, we will need to free only
164 * those pending actions which were NOT pending before the rollback
167 for (unsigned int i = 0; i < get_num_threads(); i++)
168 delete get_thread(int_to_id(i))->get_pending();
170 snapshot_backtrack_before(0);
173 /** @return a thread ID for a new Thread */
174 thread_id_t ModelChecker::get_next_id()
176 return priv->next_thread_id++;
179 /** @return the number of user threads created during this execution */
180 unsigned int ModelChecker::get_num_threads() const
182 return priv->next_thread_id;
186 * Must be called from user-thread context (e.g., through the global
187 * thread_current() interface)
189 * @return The currently executing Thread.
191 Thread * ModelChecker::get_current_thread() const
193 return scheduler->get_current_thread();
196 /** @return a sequence number for a new ModelAction */
197 modelclock_t ModelChecker::get_next_seq_num()
199 return ++priv->used_sequence_numbers;
202 Node * ModelChecker::get_curr_node() const
204 return node_stack->get_head();
208 * @brief Select the next thread to execute based on the curren action
210 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
211 * actions should be followed by the execution of their child thread. In either
212 * case, the current action should determine the next thread schedule.
214 * @param curr The current action
215 * @return The next thread to run, if the current action will determine this
216 * selection; otherwise NULL
218 Thread * ModelChecker::action_select_next_thread(const ModelAction *curr) const
220 /* Do not split atomic RMW */
222 return get_thread(curr);
223 /* Follow CREATE with the created thread */
224 if (curr->get_type() == THREAD_CREATE)
225 return curr->get_thread_operand();
230 * @brief Choose the next thread to execute.
232 * This function chooses the next thread that should execute. It can enforce
233 * execution replay/backtracking or, if the model-checker has no preference
234 * regarding the next thread (i.e., when exploring a new execution ordering),
235 * we defer to the scheduler.
237 * @return The next chosen thread to run, if any exist. Or else if the current
238 * execution should terminate, return NULL.
240 Thread * ModelChecker::get_next_thread()
245 * Have we completed exploring the preselected path? Then let the
249 return scheduler->select_next_thread();
251 /* Else, we are trying to replay an execution */
252 ModelAction *next = node_stack->get_next()->get_action();
254 if (next == diverge) {
255 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
256 earliest_diverge = diverge;
258 Node *nextnode = next->get_node();
259 Node *prevnode = nextnode->get_parent();
260 scheduler->update_sleep_set(prevnode);
262 /* Reached divergence point */
263 if (nextnode->increment_misc()) {
264 /* The next node will try to satisfy a different misc_index values. */
265 tid = next->get_tid();
266 node_stack->pop_restofstack(2);
267 } else if (nextnode->increment_promise()) {
268 /* The next node will try to satisfy a different set of promises. */
269 tid = next->get_tid();
270 node_stack->pop_restofstack(2);
271 } else if (nextnode->increment_read_from()) {
272 /* The next node will read from a different value. */
273 tid = next->get_tid();
274 node_stack->pop_restofstack(2);
275 } else if (nextnode->increment_relseq_break()) {
276 /* The next node will try to resolve a release sequence differently */
277 tid = next->get_tid();
278 node_stack->pop_restofstack(2);
281 /* Make a different thread execute for next step */
282 scheduler->add_sleep(get_thread(next->get_tid()));
283 tid = prevnode->get_next_backtrack();
284 /* Make sure the backtracked thread isn't sleeping. */
285 node_stack->pop_restofstack(1);
286 if (diverge == earliest_diverge) {
287 earliest_diverge = prevnode->get_action();
290 /* Start the round robin scheduler from this thread id */
291 scheduler->set_scheduler_thread(tid);
292 /* The correct sleep set is in the parent node. */
295 DEBUG("*** Divergence point ***\n");
299 tid = next->get_tid();
301 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
302 ASSERT(tid != THREAD_ID_T_NONE);
303 return thread_map->get(id_to_int(tid));
307 * We need to know what the next actions of all threads in the sleep
308 * set will be. This method computes them and stores the actions at
309 * the corresponding thread object's pending action.
312 void ModelChecker::execute_sleep_set()
314 for (unsigned int i = 0; i < get_num_threads(); i++) {
315 thread_id_t tid = int_to_id(i);
316 Thread *thr = get_thread(tid);
317 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
318 thr->get_pending()->set_sleep_flag();
324 * @brief Should the current action wake up a given thread?
326 * @param curr The current action
327 * @param thread The thread that we might wake up
328 * @return True, if we should wake up the sleeping thread; false otherwise
330 bool ModelChecker::should_wake_up(const ModelAction *curr, const Thread *thread) const
332 const ModelAction *asleep = thread->get_pending();
333 /* Don't allow partial RMW to wake anyone up */
336 /* Synchronizing actions may have been backtracked */
337 if (asleep->could_synchronize_with(curr))
339 /* All acquire/release fences and fence-acquire/store-release */
340 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
342 /* Fence-release + store can awake load-acquire on the same location */
343 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
344 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
345 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
351 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
353 for (unsigned int i = 0; i < get_num_threads(); i++) {
354 Thread *thr = get_thread(int_to_id(i));
355 if (scheduler->is_sleep_set(thr)) {
356 if (should_wake_up(curr, thr))
357 /* Remove this thread from sleep set */
358 scheduler->remove_sleep(thr);
363 /** @brief Alert the model-checker that an incorrectly-ordered
364 * synchronization was made */
365 void ModelChecker::set_bad_synchronization()
367 priv->bad_synchronization = true;
371 * Check whether the current trace has triggered an assertion which should halt
374 * @return True, if the execution should be aborted; false otherwise
376 bool ModelChecker::has_asserted() const
378 return priv->asserted;
382 * Trigger a trace assertion which should cause this execution to be halted.
383 * This can be due to a detected bug or due to an infeasibility that should
386 void ModelChecker::set_assert()
388 priv->asserted = true;
392 * Check if we are in a deadlock. Should only be called at the end of an
393 * execution, although it should not give false positives in the middle of an
394 * execution (there should be some ENABLED thread).
396 * @return True if program is in a deadlock; false otherwise
398 bool ModelChecker::is_deadlocked() const
400 bool blocking_threads = false;
401 for (unsigned int i = 0; i < get_num_threads(); i++) {
402 thread_id_t tid = int_to_id(i);
405 Thread *t = get_thread(tid);
406 if (!t->is_model_thread() && t->get_pending())
407 blocking_threads = true;
409 return blocking_threads;
413 * Check if this is a complete execution. That is, have all thread completed
414 * execution (rather than exiting because sleep sets have forced a redundant
417 * @return True if the execution is complete.
419 bool ModelChecker::is_complete_execution() const
421 for (unsigned int i = 0; i < get_num_threads(); i++)
422 if (is_enabled(int_to_id(i)))
428 * @brief Assert a bug in the executing program.
430 * Use this function to assert any sort of bug in the user program. If the
431 * current trace is feasible (actually, a prefix of some feasible execution),
432 * then this execution will be aborted, printing the appropriate message. If
433 * the current trace is not yet feasible, the error message will be stashed and
434 * printed if the execution ever becomes feasible.
436 * @param msg Descriptive message for the bug (do not include newline char)
437 * @return True if bug is immediately-feasible
439 bool ModelChecker::assert_bug(const char *msg)
441 priv->bugs.push_back(new bug_message(msg));
443 if (isfeasibleprefix()) {
451 * @brief Assert a bug in the executing program, asserted by a user thread
452 * @see ModelChecker::assert_bug
453 * @param msg Descriptive message for the bug (do not include newline char)
455 void ModelChecker::assert_user_bug(const char *msg)
457 /* If feasible bug, bail out now */
459 switch_to_master(NULL);
462 /** @return True, if any bugs have been reported for this execution */
463 bool ModelChecker::have_bug_reports() const
465 return priv->bugs.size() != 0;
468 /** @brief Print bug report listing for this execution (if any bugs exist) */
469 void ModelChecker::print_bugs() const
471 if (have_bug_reports()) {
472 model_print("Bug report: %zu bug%s detected\n",
474 priv->bugs.size() > 1 ? "s" : "");
475 for (unsigned int i = 0; i < priv->bugs.size(); i++)
476 priv->bugs[i]->print();
481 * @brief Record end-of-execution stats
483 * Must be run when exiting an execution. Records various stats.
484 * @see struct execution_stats
486 void ModelChecker::record_stats()
489 if (!isfeasibleprefix())
490 stats.num_infeasible++;
491 else if (have_bug_reports())
492 stats.num_buggy_executions++;
493 else if (is_complete_execution())
494 stats.num_complete++;
496 stats.num_redundant++;
499 * @todo We can violate this ASSERT() when fairness/sleep sets
500 * conflict to cause an execution to terminate, e.g. with:
501 * Scheduler: [0: disabled][1: disabled][2: sleep][3: current, enabled]
503 //ASSERT(scheduler->all_threads_sleeping());
507 /** @brief Print execution stats */
508 void ModelChecker::print_stats() const
510 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
511 model_print("Number of redundant executions: %d\n", stats.num_redundant);
512 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
513 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
514 model_print("Total executions: %d\n", stats.num_total);
515 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
519 * @brief End-of-exeuction print
520 * @param printbugs Should any existing bugs be printed?
522 void ModelChecker::print_execution(bool printbugs) const
524 print_program_output();
526 if (params.verbose) {
527 model_print("Earliest divergence point since last feasible execution:\n");
528 if (earliest_diverge)
529 earliest_diverge->print();
531 model_print("(Not set)\n");
537 /* Don't print invalid bugs */
546 * Queries the model-checker for more executions to explore and, if one
547 * exists, resets the model-checker state to execute a new execution.
549 * @return If there are more executions to explore, return true. Otherwise,
552 bool ModelChecker::next_execution()
555 /* Is this execution a feasible execution that's worth bug-checking? */
556 bool complete = isfeasibleprefix() && (is_complete_execution() ||
559 /* End-of-execution bug checks */
562 assert_bug("Deadlock detected");
570 if (params.verbose || (complete && have_bug_reports()))
571 print_execution(complete);
573 clear_program_output();
576 earliest_diverge = NULL;
578 if ((diverge = get_next_backtrack()) == NULL)
582 model_print("Next execution will diverge at:\n");
586 reset_to_initial_state();
591 * @brief Find the last fence-related backtracking conflict for a ModelAction
593 * This function performs the search for the most recent conflicting action
594 * against which we should perform backtracking, as affected by fence
595 * operations. This includes pairs of potentially-synchronizing actions which
596 * occur due to fence-acquire or fence-release, and hence should be explored in
597 * the opposite execution order.
599 * @param act The current action
600 * @return The most recent action which conflicts with act due to fences
602 ModelAction * ModelChecker::get_last_fence_conflict(ModelAction *act) const
604 /* Only perform release/acquire fence backtracking for stores */
605 if (!act->is_write())
608 /* Find a fence-release (or, act is a release) */
609 ModelAction *last_release;
610 if (act->is_release())
613 last_release = get_last_fence_release(act->get_tid());
617 /* Skip past the release */
618 action_list_t *list = action_trace;
619 action_list_t::reverse_iterator rit;
620 for (rit = list->rbegin(); rit != list->rend(); rit++)
621 if (*rit == last_release)
623 ASSERT(rit != list->rend());
628 * load --sb-> fence-acquire */
629 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
630 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
631 bool found_acquire_fences = false;
632 for ( ; rit != list->rend(); rit++) {
633 ModelAction *prev = *rit;
634 if (act->same_thread(prev))
637 int tid = id_to_int(prev->get_tid());
639 if (prev->is_read() && act->same_var(prev)) {
640 if (prev->is_acquire()) {
641 /* Found most recent load-acquire, don't need
642 * to search for more fences */
643 if (!found_acquire_fences)
646 prior_loads[tid] = prev;
649 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
650 found_acquire_fences = true;
651 acquire_fences[tid] = prev;
655 ModelAction *latest_backtrack = NULL;
656 for (unsigned int i = 0; i < acquire_fences.size(); i++)
657 if (acquire_fences[i] && prior_loads[i])
658 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
659 latest_backtrack = acquire_fences[i];
660 return latest_backtrack;
664 * @brief Find the last backtracking conflict for a ModelAction
666 * This function performs the search for the most recent conflicting action
667 * against which we should perform backtracking. This primary includes pairs of
668 * synchronizing actions which should be explored in the opposite execution
671 * @param act The current action
672 * @return The most recent action which conflicts with act
674 ModelAction * ModelChecker::get_last_conflict(ModelAction *act) const
676 switch (act->get_type()) {
677 /* case ATOMIC_FENCE: fences don't directly cause backtracking */
681 ModelAction *ret = NULL;
683 /* linear search: from most recent to oldest */
684 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
685 action_list_t::reverse_iterator rit;
686 for (rit = list->rbegin(); rit != list->rend(); rit++) {
687 ModelAction *prev = *rit;
688 if (prev->could_synchronize_with(act)) {
694 ModelAction *ret2 = get_last_fence_conflict(act);
704 case ATOMIC_TRYLOCK: {
705 /* linear search: from most recent to oldest */
706 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
707 action_list_t::reverse_iterator rit;
708 for (rit = list->rbegin(); rit != list->rend(); rit++) {
709 ModelAction *prev = *rit;
710 if (act->is_conflicting_lock(prev))
715 case ATOMIC_UNLOCK: {
716 /* linear search: from most recent to oldest */
717 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
718 action_list_t::reverse_iterator rit;
719 for (rit = list->rbegin(); rit != list->rend(); rit++) {
720 ModelAction *prev = *rit;
721 if (!act->same_thread(prev) && prev->is_failed_trylock())
727 /* linear search: from most recent to oldest */
728 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
729 action_list_t::reverse_iterator rit;
730 for (rit = list->rbegin(); rit != list->rend(); rit++) {
731 ModelAction *prev = *rit;
732 if (!act->same_thread(prev) && prev->is_failed_trylock())
734 if (!act->same_thread(prev) && prev->is_notify())
740 case ATOMIC_NOTIFY_ALL:
741 case ATOMIC_NOTIFY_ONE: {
742 /* linear search: from most recent to oldest */
743 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
744 action_list_t::reverse_iterator rit;
745 for (rit = list->rbegin(); rit != list->rend(); rit++) {
746 ModelAction *prev = *rit;
747 if (!act->same_thread(prev) && prev->is_wait())
758 /** This method finds backtracking points where we should try to
759 * reorder the parameter ModelAction against.
761 * @param the ModelAction to find backtracking points for.
763 void ModelChecker::set_backtracking(ModelAction *act)
765 Thread *t = get_thread(act);
766 ModelAction *prev = get_last_conflict(act);
770 Node *node = prev->get_node()->get_parent();
772 int low_tid, high_tid;
773 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
774 low_tid = id_to_int(act->get_tid());
775 high_tid = low_tid + 1;
778 high_tid = get_num_threads();
781 for (int i = low_tid; i < high_tid; i++) {
782 thread_id_t tid = int_to_id(i);
784 /* Make sure this thread can be enabled here. */
785 if (i >= node->get_num_threads())
788 /* Don't backtrack into a point where the thread is disabled or sleeping. */
789 if (node->enabled_status(tid) != THREAD_ENABLED)
792 /* Check if this has been explored already */
793 if (node->has_been_explored(tid))
796 /* See if fairness allows */
797 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
799 for (int t = 0; t < node->get_num_threads(); t++) {
800 thread_id_t tother = int_to_id(t);
801 if (node->is_enabled(tother) && node->has_priority(tother)) {
810 /* See if CHESS-like yield fairness allows */
811 if (model->params.yieldon) {
813 for (int t = 0; t < node->get_num_threads(); t++) {
814 thread_id_t tother = int_to_id(t);
815 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
824 /* Cache the latest backtracking point */
825 set_latest_backtrack(prev);
827 /* If this is a new backtracking point, mark the tree */
828 if (!node->set_backtrack(tid))
830 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
831 id_to_int(prev->get_tid()),
832 id_to_int(t->get_id()));
841 * @brief Cache the a backtracking point as the "most recent", if eligible
843 * Note that this does not prepare the NodeStack for this backtracking
844 * operation, it only caches the action on a per-execution basis
846 * @param act The operation at which we should explore a different next action
847 * (i.e., backtracking point)
848 * @return True, if this action is now the most recent backtracking point;
851 bool ModelChecker::set_latest_backtrack(ModelAction *act)
853 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
854 priv->next_backtrack = act;
861 * Returns last backtracking point. The model checker will explore a different
862 * path for this point in the next execution.
863 * @return The ModelAction at which the next execution should diverge.
865 ModelAction * ModelChecker::get_next_backtrack()
867 ModelAction *next = priv->next_backtrack;
868 priv->next_backtrack = NULL;
873 * Processes a read model action.
874 * @param curr is the read model action to process.
875 * @return True if processing this read updates the mo_graph.
877 bool ModelChecker::process_read(ModelAction *curr)
879 Node *node = curr->get_node();
881 bool updated = false;
882 switch (node->get_read_from_status()) {
883 case READ_FROM_PAST: {
884 const ModelAction *rf = node->get_read_from_past();
887 mo_graph->startChanges();
889 ASSERT(!is_infeasible());
890 if (!check_recency(curr, rf)) {
891 if (node->increment_read_from()) {
892 mo_graph->rollbackChanges();
895 priv->too_many_reads = true;
899 updated = r_modification_order(curr, rf);
901 mo_graph->commitChanges();
902 mo_check_promises(curr, true);
905 case READ_FROM_PROMISE: {
906 Promise *promise = curr->get_node()->get_read_from_promise();
907 if (promise->add_reader(curr))
908 priv->failed_promise = true;
909 curr->set_read_from_promise(promise);
910 mo_graph->startChanges();
911 if (!check_recency(curr, promise))
912 priv->too_many_reads = true;
913 updated = r_modification_order(curr, promise);
914 mo_graph->commitChanges();
917 case READ_FROM_FUTURE: {
918 /* Read from future value */
919 struct future_value fv = node->get_future_value();
920 Promise *promise = new Promise(curr, fv);
921 curr->set_read_from_promise(promise);
922 promises->push_back(promise);
923 mo_graph->startChanges();
924 updated = r_modification_order(curr, promise);
925 mo_graph->commitChanges();
931 get_thread(curr)->set_return_value(curr->get_return_value());
937 * Processes a lock, trylock, or unlock model action. @param curr is
938 * the read model action to process.
940 * The try lock operation checks whether the lock is taken. If not,
941 * it falls to the normal lock operation case. If so, it returns
944 * The lock operation has already been checked that it is enabled, so
945 * it just grabs the lock and synchronizes with the previous unlock.
947 * The unlock operation has to re-enable all of the threads that are
948 * waiting on the lock.
950 * @return True if synchronization was updated; false otherwise
952 bool ModelChecker::process_mutex(ModelAction *curr)
954 std::mutex *mutex = curr->get_mutex();
955 struct std::mutex_state *state = NULL;
958 state = mutex->get_state();
960 switch (curr->get_type()) {
961 case ATOMIC_TRYLOCK: {
962 bool success = !state->locked;
963 curr->set_try_lock(success);
965 get_thread(curr)->set_return_value(0);
968 get_thread(curr)->set_return_value(1);
970 //otherwise fall into the lock case
972 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
973 assert_bug("Lock access before initialization");
974 state->locked = get_thread(curr);
975 ModelAction *unlock = get_last_unlock(curr);
976 //synchronize with the previous unlock statement
977 if (unlock != NULL) {
978 curr->synchronize_with(unlock);
983 case ATOMIC_UNLOCK: {
985 state->locked = NULL;
986 //wake up the other threads
987 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
988 //activate all the waiting threads
989 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
990 scheduler->wake(get_thread(*rit));
997 state->locked = NULL;
998 //wake up the other threads
999 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
1000 //activate all the waiting threads
1001 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
1002 scheduler->wake(get_thread(*rit));
1005 //check whether we should go to sleep or not...simulate spurious failures
1006 if (curr->get_node()->get_misc() == 0) {
1007 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
1009 scheduler->sleep(get_thread(curr));
1013 case ATOMIC_NOTIFY_ALL: {
1014 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1015 //activate all the waiting threads
1016 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
1017 scheduler->wake(get_thread(*rit));
1022 case ATOMIC_NOTIFY_ONE: {
1023 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1024 int wakeupthread = curr->get_node()->get_misc();
1025 action_list_t::iterator it = waiters->begin();
1026 advance(it, wakeupthread);
1027 scheduler->wake(get_thread(*it));
1039 * @brief Check if the current pending promises allow a future value to be sent
1041 * If one of the following is true:
1042 * (a) there are no pending promises
1043 * (b) the reader and writer do not cross any promises
1044 * Then, it is safe to pass a future value back now.
1046 * Otherwise, we must save the pending future value until (a) or (b) is true
1048 * @param writer The operation which sends the future value. Must be a write.
1049 * @param reader The operation which will observe the value. Must be a read.
1050 * @return True if the future value can be sent now; false if it must wait.
1052 bool ModelChecker::promises_may_allow(const ModelAction *writer,
1053 const ModelAction *reader) const
1055 if (promises->empty())
1057 for(int i=promises->size()-1;i>=0;i--) {
1058 ModelAction *pr=(*promises)[i]->get_reader(0);
1059 //reader is after promise...doesn't cross any promise
1062 //writer is after promise, reader before...bad...
1070 * @brief Add a future value to a reader
1072 * This function performs a few additional checks to ensure that the future
1073 * value can be feasibly observed by the reader
1075 * @param writer The operation whose value is sent. Must be a write.
1076 * @param reader The read operation which may read the future value. Must be a read.
1078 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
1080 /* Do more ambitious checks now that mo is more complete */
1081 if (!mo_may_allow(writer, reader))
1084 Node *node = reader->get_node();
1086 /* Find an ancestor thread which exists at the time of the reader */
1087 Thread *write_thread = get_thread(writer);
1088 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
1089 write_thread = write_thread->get_parent();
1091 struct future_value fv = {
1092 writer->get_write_value(),
1093 writer->get_seq_number() + params.maxfuturedelay,
1094 write_thread->get_id(),
1096 if (node->add_future_value(fv))
1097 set_latest_backtrack(reader);
1101 * Process a write ModelAction
1102 * @param curr The ModelAction to process
1103 * @return True if the mo_graph was updated or promises were resolved
1105 bool ModelChecker::process_write(ModelAction *curr)
1107 /* Readers to which we may send our future value */
1108 ModelVector<ModelAction *> send_fv;
1110 const ModelAction *earliest_promise_reader;
1111 bool updated_promises = false;
1113 bool updated_mod_order = w_modification_order(curr, &send_fv);
1114 Promise *promise = pop_promise_to_resolve(curr);
1117 earliest_promise_reader = promise->get_reader(0);
1118 updated_promises = resolve_promise(curr, promise);
1120 earliest_promise_reader = NULL;
1122 for (unsigned int i = 0; i < send_fv.size(); i++) {
1123 ModelAction *read = send_fv[i];
1125 /* Don't send future values to reads after the Promise we resolve */
1126 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
1127 /* Check if future value can be sent immediately */
1128 if (promises_may_allow(curr, read)) {
1129 add_future_value(curr, read);
1131 futurevalues->push_back(PendingFutureValue(curr, read));
1136 /* Check the pending future values */
1137 for (int i = (int)futurevalues->size() - 1; i >= 0; i--) {
1138 struct PendingFutureValue pfv = (*futurevalues)[i];
1139 if (promises_may_allow(pfv.writer, pfv.reader)) {
1140 add_future_value(pfv.writer, pfv.reader);
1141 futurevalues->erase(futurevalues->begin() + i);
1145 mo_graph->commitChanges();
1146 mo_check_promises(curr, false);
1148 get_thread(curr)->set_return_value(VALUE_NONE);
1149 return updated_mod_order || updated_promises;
1153 * Process a fence ModelAction
1154 * @param curr The ModelAction to process
1155 * @return True if synchronization was updated
1157 bool ModelChecker::process_fence(ModelAction *curr)
1160 * fence-relaxed: no-op
1161 * fence-release: only log the occurence (not in this function), for
1162 * use in later synchronization
1163 * fence-acquire (this function): search for hypothetical release
1165 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
1167 bool updated = false;
1168 if (curr->is_acquire()) {
1169 action_list_t *list = action_trace;
1170 action_list_t::reverse_iterator rit;
1171 /* Find X : is_read(X) && X --sb-> curr */
1172 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1173 ModelAction *act = *rit;
1176 if (act->get_tid() != curr->get_tid())
1178 /* Stop at the beginning of the thread */
1179 if (act->is_thread_start())
1181 /* Stop once we reach a prior fence-acquire */
1182 if (act->is_fence() && act->is_acquire())
1184 if (!act->is_read())
1186 /* read-acquire will find its own release sequences */
1187 if (act->is_acquire())
1190 /* Establish hypothetical release sequences */
1191 rel_heads_list_t release_heads;
1192 get_release_seq_heads(curr, act, &release_heads);
1193 for (unsigned int i = 0; i < release_heads.size(); i++)
1194 if (!curr->synchronize_with(release_heads[i]))
1195 set_bad_synchronization();
1196 if (release_heads.size() != 0)
1204 * @brief Process the current action for thread-related activity
1206 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
1207 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
1208 * synchronization, etc. This function is a no-op for non-THREAD actions
1209 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
1211 * @param curr The current action
1212 * @return True if synchronization was updated or a thread completed
1214 bool ModelChecker::process_thread_action(ModelAction *curr)
1216 bool updated = false;
1218 switch (curr->get_type()) {
1219 case THREAD_CREATE: {
1220 thrd_t *thrd = (thrd_t *)curr->get_location();
1221 struct thread_params *params = (struct thread_params *)curr->get_value();
1222 Thread *th = new Thread(thrd, params->func, params->arg, get_thread(curr));
1224 th->set_creation(curr);
1225 /* Promises can be satisfied by children */
1226 for (unsigned int i = 0; i < promises->size(); i++) {
1227 Promise *promise = (*promises)[i];
1228 if (promise->thread_is_available(curr->get_tid()))
1229 promise->add_thread(th->get_id());
1234 Thread *blocking = curr->get_thread_operand();
1235 ModelAction *act = get_last_action(blocking->get_id());
1236 curr->synchronize_with(act);
1237 updated = true; /* trigger rel-seq checks */
1240 case THREAD_FINISH: {
1241 Thread *th = get_thread(curr);
1242 while (!th->wait_list_empty()) {
1243 ModelAction *act = th->pop_wait_list();
1244 scheduler->wake(get_thread(act));
1247 /* Completed thread can't satisfy promises */
1248 for (unsigned int i = 0; i < promises->size(); i++) {
1249 Promise *promise = (*promises)[i];
1250 if (promise->thread_is_available(th->get_id()))
1251 if (promise->eliminate_thread(th->get_id()))
1252 priv->failed_promise = true;
1254 updated = true; /* trigger rel-seq checks */
1257 case THREAD_START: {
1258 check_promises(curr->get_tid(), NULL, curr->get_cv());
1269 * @brief Process the current action for release sequence fixup activity
1271 * Performs model-checker release sequence fixups for the current action,
1272 * forcing a single pending release sequence to break (with a given, potential
1273 * "loose" write) or to complete (i.e., synchronize). If a pending release
1274 * sequence forms a complete release sequence, then we must perform the fixup
1275 * synchronization, mo_graph additions, etc.
1277 * @param curr The current action; must be a release sequence fixup action
1278 * @param work_queue The work queue to which to add work items as they are
1281 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1283 const ModelAction *write = curr->get_node()->get_relseq_break();
1284 struct release_seq *sequence = pending_rel_seqs->back();
1285 pending_rel_seqs->pop_back();
1287 ModelAction *acquire = sequence->acquire;
1288 const ModelAction *rf = sequence->rf;
1289 const ModelAction *release = sequence->release;
1293 ASSERT(release->same_thread(rf));
1295 if (write == NULL) {
1297 * @todo Forcing a synchronization requires that we set
1298 * modification order constraints. For instance, we can't allow
1299 * a fixup sequence in which two separate read-acquire
1300 * operations read from the same sequence, where the first one
1301 * synchronizes and the other doesn't. Essentially, we can't
1302 * allow any writes to insert themselves between 'release' and
1306 /* Must synchronize */
1307 if (!acquire->synchronize_with(release)) {
1308 set_bad_synchronization();
1311 /* Re-check all pending release sequences */
1312 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1313 /* Re-check act for mo_graph edges */
1314 work_queue->push_back(MOEdgeWorkEntry(acquire));
1316 /* propagate synchronization to later actions */
1317 action_list_t::reverse_iterator rit = action_trace->rbegin();
1318 for (; (*rit) != acquire; rit++) {
1319 ModelAction *propagate = *rit;
1320 if (acquire->happens_before(propagate)) {
1321 propagate->synchronize_with(acquire);
1322 /* Re-check 'propagate' for mo_graph edges */
1323 work_queue->push_back(MOEdgeWorkEntry(propagate));
1327 /* Break release sequence with new edges:
1328 * release --mo--> write --mo--> rf */
1329 mo_graph->addEdge(release, write);
1330 mo_graph->addEdge(write, rf);
1333 /* See if we have realized a data race */
1338 * Initialize the current action by performing one or more of the following
1339 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1340 * in the NodeStack, manipulating backtracking sets, allocating and
1341 * initializing clock vectors, and computing the promises to fulfill.
1343 * @param curr The current action, as passed from the user context; may be
1344 * freed/invalidated after the execution of this function, with a different
1345 * action "returned" its place (pass-by-reference)
1346 * @return True if curr is a newly-explored action; false otherwise
1348 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1350 ModelAction *newcurr;
1352 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1353 newcurr = process_rmw(*curr);
1356 if (newcurr->is_rmw())
1357 compute_promises(newcurr);
1363 (*curr)->set_seq_number(get_next_seq_num());
1365 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1367 /* First restore type and order in case of RMW operation */
1368 if ((*curr)->is_rmwr())
1369 newcurr->copy_typeandorder(*curr);
1371 ASSERT((*curr)->get_location() == newcurr->get_location());
1372 newcurr->copy_from_new(*curr);
1374 /* Discard duplicate ModelAction; use action from NodeStack */
1377 /* Always compute new clock vector */
1378 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1381 return false; /* Action was explored previously */
1385 /* Always compute new clock vector */
1386 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1388 /* Assign most recent release fence */
1389 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1392 * Perform one-time actions when pushing new ModelAction onto
1395 if (newcurr->is_write())
1396 compute_promises(newcurr);
1397 else if (newcurr->is_relseq_fixup())
1398 compute_relseq_breakwrites(newcurr);
1399 else if (newcurr->is_wait())
1400 newcurr->get_node()->set_misc_max(2);
1401 else if (newcurr->is_notify_one()) {
1402 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1404 return true; /* This was a new ModelAction */
1409 * @brief Establish reads-from relation between two actions
1411 * Perform basic operations involved with establishing a concrete rf relation,
1412 * including setting the ModelAction data and checking for release sequences.
1414 * @param act The action that is reading (must be a read)
1415 * @param rf The action from which we are reading (must be a write)
1417 * @return True if this read established synchronization
1419 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1422 ASSERT(rf->is_write());
1424 act->set_read_from(rf);
1425 if (act->is_acquire()) {
1426 rel_heads_list_t release_heads;
1427 get_release_seq_heads(act, act, &release_heads);
1428 int num_heads = release_heads.size();
1429 for (unsigned int i = 0; i < release_heads.size(); i++)
1430 if (!act->synchronize_with(release_heads[i])) {
1431 set_bad_synchronization();
1434 return num_heads > 0;
1440 * Check promises and eliminate potentially-satisfying threads when a thread is
1441 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1442 * no longer satisfy a promise generated from that thread.
1444 * @param blocker The thread on which a thread is waiting
1445 * @param waiting The waiting thread
1447 void ModelChecker::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1449 for (unsigned int i = 0; i < promises->size(); i++) {
1450 Promise *promise = (*promises)[i];
1451 if (!promise->thread_is_available(waiting->get_id()))
1453 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1454 ModelAction *reader = promise->get_reader(j);
1455 if (reader->get_tid() != blocker->get_id())
1457 if (promise->eliminate_thread(waiting->get_id())) {
1458 /* Promise has failed */
1459 priv->failed_promise = true;
1461 /* Only eliminate the 'waiting' thread once */
1469 * @brief Check whether a model action is enabled.
1471 * Checks whether a lock or join operation would be successful (i.e., is the
1472 * lock already locked, or is the joined thread already complete). If not, put
1473 * the action in a waiter list.
1475 * @param curr is the ModelAction to check whether it is enabled.
1476 * @return a bool that indicates whether the action is enabled.
1478 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1479 if (curr->is_lock()) {
1480 std::mutex *lock = (std::mutex *)curr->get_location();
1481 struct std::mutex_state *state = lock->get_state();
1482 if (state->locked) {
1483 //Stick the action in the appropriate waiting queue
1484 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
1487 } else if (curr->get_type() == THREAD_JOIN) {
1488 Thread *blocking = (Thread *)curr->get_location();
1489 if (!blocking->is_complete()) {
1490 blocking->push_wait_list(curr);
1491 thread_blocking_check_promises(blocking, get_thread(curr));
1500 * This is the heart of the model checker routine. It performs model-checking
1501 * actions corresponding to a given "current action." Among other processes, it
1502 * calculates reads-from relationships, updates synchronization clock vectors,
1503 * forms a memory_order constraints graph, and handles replay/backtrack
1504 * execution when running permutations of previously-observed executions.
1506 * @param curr The current action to process
1507 * @return The ModelAction that is actually executed; may be different than
1508 * curr; may be NULL, if the current action is not enabled to run
1510 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1513 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1515 if (!check_action_enabled(curr)) {
1516 /* Make the execution look like we chose to run this action
1517 * much later, when a lock/join can succeed */
1518 get_thread(curr)->set_pending(curr);
1519 scheduler->sleep(get_thread(curr));
1523 bool newly_explored = initialize_curr_action(&curr);
1529 wake_up_sleeping_actions(curr);
1531 /* Compute fairness information for CHESS yield algorithm */
1532 if (model->params.yieldon) {
1533 curr->get_node()->update_yield(scheduler);
1536 /* Add the action to lists before any other model-checking tasks */
1537 if (!second_part_of_rmw)
1538 add_action_to_lists(curr);
1540 /* Build may_read_from set for newly-created actions */
1541 if (newly_explored && curr->is_read())
1542 build_may_read_from(curr);
1544 /* Initialize work_queue with the "current action" work */
1545 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1546 while (!work_queue.empty() && !has_asserted()) {
1547 WorkQueueEntry work = work_queue.front();
1548 work_queue.pop_front();
1550 switch (work.type) {
1551 case WORK_CHECK_CURR_ACTION: {
1552 ModelAction *act = work.action;
1553 bool update = false; /* update this location's release seq's */
1554 bool update_all = false; /* update all release seq's */
1556 if (process_thread_action(curr))
1559 if (act->is_read() && !second_part_of_rmw && process_read(act))
1562 if (act->is_write() && process_write(act))
1565 if (act->is_fence() && process_fence(act))
1568 if (act->is_mutex_op() && process_mutex(act))
1571 if (act->is_relseq_fixup())
1572 process_relseq_fixup(curr, &work_queue);
1575 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1577 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1580 case WORK_CHECK_RELEASE_SEQ:
1581 resolve_release_sequences(work.location, &work_queue);
1583 case WORK_CHECK_MO_EDGES: {
1584 /** @todo Complete verification of work_queue */
1585 ModelAction *act = work.action;
1586 bool updated = false;
1588 if (act->is_read()) {
1589 const ModelAction *rf = act->get_reads_from();
1590 const Promise *promise = act->get_reads_from_promise();
1592 if (r_modification_order(act, rf))
1594 } else if (promise) {
1595 if (r_modification_order(act, promise))
1599 if (act->is_write()) {
1600 if (w_modification_order(act, NULL))
1603 mo_graph->commitChanges();
1606 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1615 check_curr_backtracking(curr);
1616 set_backtracking(curr);
1620 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1622 Node *currnode = curr->get_node();
1623 Node *parnode = currnode->get_parent();
1625 if ((parnode && !parnode->backtrack_empty()) ||
1626 !currnode->misc_empty() ||
1627 !currnode->read_from_empty() ||
1628 !currnode->promise_empty() ||
1629 !currnode->relseq_break_empty()) {
1630 set_latest_backtrack(curr);
1634 bool ModelChecker::promises_expired() const
1636 for (unsigned int i = 0; i < promises->size(); i++) {
1637 Promise *promise = (*promises)[i];
1638 if (promise->get_expiration() < priv->used_sequence_numbers)
1645 * This is the strongest feasibility check available.
1646 * @return whether the current trace (partial or complete) must be a prefix of
1649 bool ModelChecker::isfeasibleprefix() const
1651 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1655 * Print disagnostic information about an infeasible execution
1656 * @param prefix A string to prefix the output with; if NULL, then a default
1657 * message prefix will be provided
1659 void ModelChecker::print_infeasibility(const char *prefix) const
1663 if (mo_graph->checkForCycles())
1664 ptr += sprintf(ptr, "[mo cycle]");
1665 if (priv->failed_promise)
1666 ptr += sprintf(ptr, "[failed promise]");
1667 if (priv->too_many_reads)
1668 ptr += sprintf(ptr, "[too many reads]");
1669 if (priv->no_valid_reads)
1670 ptr += sprintf(ptr, "[no valid reads-from]");
1671 if (priv->bad_synchronization)
1672 ptr += sprintf(ptr, "[bad sw ordering]");
1673 if (promises_expired())
1674 ptr += sprintf(ptr, "[promise expired]");
1675 if (promises->size() != 0)
1676 ptr += sprintf(ptr, "[unresolved promise]");
1678 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1682 * Returns whether the current completed trace is feasible, except for pending
1683 * release sequences.
1685 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1687 return !is_infeasible() && promises->size() == 0;
1691 * Check if the current partial trace is infeasible. Does not check any
1692 * end-of-execution flags, which might rule out the execution. Thus, this is
1693 * useful only for ruling an execution as infeasible.
1694 * @return whether the current partial trace is infeasible.
1696 bool ModelChecker::is_infeasible() const
1698 return mo_graph->checkForCycles() ||
1699 priv->no_valid_reads ||
1700 priv->failed_promise ||
1701 priv->too_many_reads ||
1702 priv->bad_synchronization ||
1706 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1707 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1708 ModelAction *lastread = get_last_action(act->get_tid());
1709 lastread->process_rmw(act);
1710 if (act->is_rmw()) {
1711 if (lastread->get_reads_from())
1712 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1714 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1715 mo_graph->commitChanges();
1721 * A helper function for ModelChecker::check_recency, to check if the current
1722 * thread is able to read from a different write/promise for 'params.maxreads'
1723 * number of steps and if that write/promise should become visible (i.e., is
1724 * ordered later in the modification order). This helps model memory liveness.
1726 * @param curr The current action. Must be a read.
1727 * @param rf The write/promise from which we plan to read
1728 * @param other_rf The write/promise from which we may read
1729 * @return True if we were able to read from other_rf for params.maxreads steps
1731 template <typename T, typename U>
1732 bool ModelChecker::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1734 /* Need a different write/promise */
1735 if (other_rf->equals(rf))
1738 /* Only look for "newer" writes/promises */
1739 if (!mo_graph->checkReachable(rf, other_rf))
1742 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1743 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1744 action_list_t::reverse_iterator rit = list->rbegin();
1745 ASSERT((*rit) == curr);
1746 /* Skip past curr */
1749 /* Does this write/promise work for everyone? */
1750 for (int i = 0; i < params.maxreads; i++, rit++) {
1751 ModelAction *act = *rit;
1752 if (!act->may_read_from(other_rf))
1759 * Checks whether a thread has read from the same write or Promise for too many
1760 * times without seeing the effects of a later write/Promise.
1763 * 1) there must a different write/promise that we could read from,
1764 * 2) we must have read from the same write/promise in excess of maxreads times,
1765 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1766 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1768 * If so, we decide that the execution is no longer feasible.
1770 * @param curr The current action. Must be a read.
1771 * @param rf The ModelAction/Promise from which we might read.
1772 * @return True if the read should succeed; false otherwise
1774 template <typename T>
1775 bool ModelChecker::check_recency(ModelAction *curr, const T *rf) const
1777 if (!params.maxreads)
1780 //NOTE: Next check is just optimization, not really necessary....
1781 if (curr->get_node()->get_read_from_past_size() +
1782 curr->get_node()->get_read_from_promise_size() <= 1)
1785 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1786 int tid = id_to_int(curr->get_tid());
1787 ASSERT(tid < (int)thrd_lists->size());
1788 action_list_t *list = &(*thrd_lists)[tid];
1789 action_list_t::reverse_iterator rit = list->rbegin();
1790 ASSERT((*rit) == curr);
1791 /* Skip past curr */
1794 action_list_t::reverse_iterator ritcopy = rit;
1795 /* See if we have enough reads from the same value */
1796 for (int count = 0; count < params.maxreads; ritcopy++, count++) {
1797 if (ritcopy == list->rend())
1799 ModelAction *act = *ritcopy;
1800 if (!act->is_read())
1802 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1804 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1806 if (act->get_node()->get_read_from_past_size() +
1807 act->get_node()->get_read_from_promise_size() <= 1)
1810 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1811 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1812 if (should_read_instead(curr, rf, write))
1813 return false; /* liveness failure */
1815 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1816 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1817 if (should_read_instead(curr, rf, promise))
1818 return false; /* liveness failure */
1824 * Updates the mo_graph with the constraints imposed from the current
1827 * Basic idea is the following: Go through each other thread and find
1828 * the last action that happened before our read. Two cases:
1830 * (1) The action is a write => that write must either occur before
1831 * the write we read from or be the write we read from.
1833 * (2) The action is a read => the write that that action read from
1834 * must occur before the write we read from or be the same write.
1836 * @param curr The current action. Must be a read.
1837 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1838 * @return True if modification order edges were added; false otherwise
1840 template <typename rf_type>
1841 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1843 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1846 ASSERT(curr->is_read());
1848 /* Last SC fence in the current thread */
1849 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1850 ModelAction *last_sc_write = NULL;
1851 if (curr->is_seqcst())
1852 last_sc_write = get_last_seq_cst_write(curr);
1854 /* Iterate over all threads */
1855 for (i = 0; i < thrd_lists->size(); i++) {
1856 /* Last SC fence in thread i */
1857 ModelAction *last_sc_fence_thread_local = NULL;
1858 if (int_to_id((int)i) != curr->get_tid())
1859 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1861 /* Last SC fence in thread i, before last SC fence in current thread */
1862 ModelAction *last_sc_fence_thread_before = NULL;
1863 if (last_sc_fence_local)
1864 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1866 /* Iterate over actions in thread, starting from most recent */
1867 action_list_t *list = &(*thrd_lists)[i];
1868 action_list_t::reverse_iterator rit;
1869 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1870 ModelAction *act = *rit;
1875 /* Don't want to add reflexive edges on 'rf' */
1876 if (act->equals(rf)) {
1877 if (act->happens_before(curr))
1883 if (act->is_write()) {
1884 /* C++, Section 29.3 statement 5 */
1885 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1886 *act < *last_sc_fence_thread_local) {
1887 added = mo_graph->addEdge(act, rf) || added;
1890 /* C++, Section 29.3 statement 4 */
1891 else if (act->is_seqcst() && last_sc_fence_local &&
1892 *act < *last_sc_fence_local) {
1893 added = mo_graph->addEdge(act, rf) || added;
1896 /* C++, Section 29.3 statement 6 */
1897 else if (last_sc_fence_thread_before &&
1898 *act < *last_sc_fence_thread_before) {
1899 added = mo_graph->addEdge(act, rf) || added;
1904 /* C++, Section 29.3 statement 3 (second subpoint) */
1905 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1906 added = mo_graph->addEdge(act, rf) || added;
1911 * Include at most one act per-thread that "happens
1914 if (act->happens_before(curr)) {
1915 if (act->is_write()) {
1916 added = mo_graph->addEdge(act, rf) || added;
1918 const ModelAction *prevrf = act->get_reads_from();
1919 const Promise *prevrf_promise = act->get_reads_from_promise();
1921 if (!prevrf->equals(rf))
1922 added = mo_graph->addEdge(prevrf, rf) || added;
1923 } else if (!prevrf_promise->equals(rf)) {
1924 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1933 * All compatible, thread-exclusive promises must be ordered after any
1934 * concrete loads from the same thread
1936 for (unsigned int i = 0; i < promises->size(); i++)
1937 if ((*promises)[i]->is_compatible_exclusive(curr))
1938 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1944 * Updates the mo_graph with the constraints imposed from the current write.
1946 * Basic idea is the following: Go through each other thread and find
1947 * the lastest action that happened before our write. Two cases:
1949 * (1) The action is a write => that write must occur before
1952 * (2) The action is a read => the write that that action read from
1953 * must occur before the current write.
1955 * This method also handles two other issues:
1957 * (I) Sequential Consistency: Making sure that if the current write is
1958 * seq_cst, that it occurs after the previous seq_cst write.
1960 * (II) Sending the write back to non-synchronizing reads.
1962 * @param curr The current action. Must be a write.
1963 * @param send_fv A vector for stashing reads to which we may pass our future
1964 * value. If NULL, then don't record any future values.
1965 * @return True if modification order edges were added; false otherwise
1967 bool ModelChecker::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1969 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1972 ASSERT(curr->is_write());
1974 if (curr->is_seqcst()) {
1975 /* We have to at least see the last sequentially consistent write,
1976 so we are initialized. */
1977 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1978 if (last_seq_cst != NULL) {
1979 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1983 /* Last SC fence in the current thread */
1984 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1986 /* Iterate over all threads */
1987 for (i = 0; i < thrd_lists->size(); i++) {
1988 /* Last SC fence in thread i, before last SC fence in current thread */
1989 ModelAction *last_sc_fence_thread_before = NULL;
1990 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1991 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1993 /* Iterate over actions in thread, starting from most recent */
1994 action_list_t *list = &(*thrd_lists)[i];
1995 action_list_t::reverse_iterator rit;
1996 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1997 ModelAction *act = *rit;
2000 * 1) If RMW and it actually read from something, then we
2001 * already have all relevant edges, so just skip to next
2004 * 2) If RMW and it didn't read from anything, we should
2005 * whatever edge we can get to speed up convergence.
2007 * 3) If normal write, we need to look at earlier actions, so
2008 * continue processing list.
2010 if (curr->is_rmw()) {
2011 if (curr->get_reads_from() != NULL)
2019 /* C++, Section 29.3 statement 7 */
2020 if (last_sc_fence_thread_before && act->is_write() &&
2021 *act < *last_sc_fence_thread_before) {
2022 added = mo_graph->addEdge(act, curr) || added;
2027 * Include at most one act per-thread that "happens
2030 if (act->happens_before(curr)) {
2032 * Note: if act is RMW, just add edge:
2034 * The following edge should be handled elsewhere:
2035 * readfrom(act) --mo--> act
2037 if (act->is_write())
2038 added = mo_graph->addEdge(act, curr) || added;
2039 else if (act->is_read()) {
2040 //if previous read accessed a null, just keep going
2041 if (act->get_reads_from() == NULL)
2043 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
2046 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
2047 !act->same_thread(curr)) {
2048 /* We have an action that:
2049 (1) did not happen before us
2050 (2) is a read and we are a write
2051 (3) cannot synchronize with us
2052 (4) is in a different thread
2054 that read could potentially read from our write. Note that
2055 these checks are overly conservative at this point, we'll
2056 do more checks before actually removing the
2060 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
2061 if (!is_infeasible())
2062 send_fv->push_back(act);
2063 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
2064 add_future_value(curr, act);
2071 * All compatible, thread-exclusive promises must be ordered after any
2072 * concrete stores to the same thread, or else they can be merged with
2075 for (unsigned int i = 0; i < promises->size(); i++)
2076 if ((*promises)[i]->is_compatible_exclusive(curr))
2077 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
2082 /** Arbitrary reads from the future are not allowed. Section 29.3
2083 * part 9 places some constraints. This method checks one result of constraint
2084 * constraint. Others require compiler support. */
2085 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
2087 if (!writer->is_rmw())
2090 if (!reader->is_rmw())
2093 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
2094 if (search == reader)
2096 if (search->get_tid() == reader->get_tid() &&
2097 search->happens_before(reader))
2105 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
2106 * some constraints. This method checks one the following constraint (others
2107 * require compiler support):
2109 * If X --hb-> Y --mo-> Z, then X should not read from Z.
2111 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
2113 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
2115 /* Iterate over all threads */
2116 for (i = 0; i < thrd_lists->size(); i++) {
2117 const ModelAction *write_after_read = NULL;
2119 /* Iterate over actions in thread, starting from most recent */
2120 action_list_t *list = &(*thrd_lists)[i];
2121 action_list_t::reverse_iterator rit;
2122 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2123 ModelAction *act = *rit;
2125 /* Don't disallow due to act == reader */
2126 if (!reader->happens_before(act) || reader == act)
2128 else if (act->is_write())
2129 write_after_read = act;
2130 else if (act->is_read() && act->get_reads_from() != NULL)
2131 write_after_read = act->get_reads_from();
2134 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
2141 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
2142 * The ModelAction under consideration is expected to be taking part in
2143 * release/acquire synchronization as an object of the "reads from" relation.
2144 * Note that this can only provide release sequence support for RMW chains
2145 * which do not read from the future, as those actions cannot be traced until
2146 * their "promise" is fulfilled. Similarly, we may not even establish the
2147 * presence of a release sequence with certainty, as some modification order
2148 * constraints may be decided further in the future. Thus, this function
2149 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
2150 * and a boolean representing certainty.
2152 * @param rf The action that might be part of a release sequence. Must be a
2154 * @param release_heads A pass-by-reference style return parameter. After
2155 * execution of this function, release_heads will contain the heads of all the
2156 * relevant release sequences, if any exists with certainty
2157 * @param pending A pass-by-reference style return parameter which is only used
2158 * when returning false (i.e., uncertain). Returns most information regarding
2159 * an uncertain release sequence, including any write operations that might
2160 * break the sequence.
2161 * @return true, if the ModelChecker is certain that release_heads is complete;
2164 bool ModelChecker::release_seq_heads(const ModelAction *rf,
2165 rel_heads_list_t *release_heads,
2166 struct release_seq *pending) const
2168 /* Only check for release sequences if there are no cycles */
2169 if (mo_graph->checkForCycles())
2172 for ( ; rf != NULL; rf = rf->get_reads_from()) {
2173 ASSERT(rf->is_write());
2175 if (rf->is_release())
2176 release_heads->push_back(rf);
2177 else if (rf->get_last_fence_release())
2178 release_heads->push_back(rf->get_last_fence_release());
2180 break; /* End of RMW chain */
2182 /** @todo Need to be smarter here... In the linux lock
2183 * example, this will run to the beginning of the program for
2185 /** @todo The way to be smarter here is to keep going until 1
2186 * thread has a release preceded by an acquire and you've seen
2189 /* acq_rel RMW is a sufficient stopping condition */
2190 if (rf->is_acquire() && rf->is_release())
2191 return true; /* complete */
2194 /* read from future: need to settle this later */
2196 return false; /* incomplete */
2199 if (rf->is_release())
2200 return true; /* complete */
2202 /* else relaxed write
2203 * - check for fence-release in the same thread (29.8, stmt. 3)
2204 * - check modification order for contiguous subsequence
2205 * -> rf must be same thread as release */
2207 const ModelAction *fence_release = rf->get_last_fence_release();
2208 /* Synchronize with a fence-release unconditionally; we don't need to
2209 * find any more "contiguous subsequence..." for it */
2211 release_heads->push_back(fence_release);
2213 int tid = id_to_int(rf->get_tid());
2214 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
2215 action_list_t *list = &(*thrd_lists)[tid];
2216 action_list_t::const_reverse_iterator rit;
2218 /* Find rf in the thread list */
2219 rit = std::find(list->rbegin(), list->rend(), rf);
2220 ASSERT(rit != list->rend());
2222 /* Find the last {write,fence}-release */
2223 for (; rit != list->rend(); rit++) {
2224 if (fence_release && *(*rit) < *fence_release)
2226 if ((*rit)->is_release())
2229 if (rit == list->rend()) {
2230 /* No write-release in this thread */
2231 return true; /* complete */
2232 } else if (fence_release && *(*rit) < *fence_release) {
2233 /* The fence-release is more recent (and so, "stronger") than
2234 * the most recent write-release */
2235 return true; /* complete */
2236 } /* else, need to establish contiguous release sequence */
2237 ModelAction *release = *rit;
2239 ASSERT(rf->same_thread(release));
2241 pending->writes.clear();
2243 bool certain = true;
2244 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2245 if (id_to_int(rf->get_tid()) == (int)i)
2247 list = &(*thrd_lists)[i];
2249 /* Can we ensure no future writes from this thread may break
2250 * the release seq? */
2251 bool future_ordered = false;
2253 ModelAction *last = get_last_action(int_to_id(i));
2254 Thread *th = get_thread(int_to_id(i));
2255 if ((last && rf->happens_before(last)) ||
2258 future_ordered = true;
2260 ASSERT(!th->is_model_thread() || future_ordered);
2262 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2263 const ModelAction *act = *rit;
2264 /* Reach synchronization -> this thread is complete */
2265 if (act->happens_before(release))
2267 if (rf->happens_before(act)) {
2268 future_ordered = true;
2272 /* Only non-RMW writes can break release sequences */
2273 if (!act->is_write() || act->is_rmw())
2276 /* Check modification order */
2277 if (mo_graph->checkReachable(rf, act)) {
2278 /* rf --mo--> act */
2279 future_ordered = true;
2282 if (mo_graph->checkReachable(act, release))
2283 /* act --mo--> release */
2285 if (mo_graph->checkReachable(release, act) &&
2286 mo_graph->checkReachable(act, rf)) {
2287 /* release --mo-> act --mo--> rf */
2288 return true; /* complete */
2290 /* act may break release sequence */
2291 pending->writes.push_back(act);
2294 if (!future_ordered)
2295 certain = false; /* This thread is uncertain */
2299 release_heads->push_back(release);
2300 pending->writes.clear();
2302 pending->release = release;
2309 * An interface for getting the release sequence head(s) with which a
2310 * given ModelAction must synchronize. This function only returns a non-empty
2311 * result when it can locate a release sequence head with certainty. Otherwise,
2312 * it may mark the internal state of the ModelChecker so that it will handle
2313 * the release sequence at a later time, causing @a acquire to update its
2314 * synchronization at some later point in execution.
2316 * @param acquire The 'acquire' action that may synchronize with a release
2318 * @param read The read action that may read from a release sequence; this may
2319 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2320 * when 'acquire' is a fence-acquire)
2321 * @param release_heads A pass-by-reference return parameter. Will be filled
2322 * with the head(s) of the release sequence(s), if they exists with certainty.
2323 * @see ModelChecker::release_seq_heads
2325 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2326 ModelAction *read, rel_heads_list_t *release_heads)
2328 const ModelAction *rf = read->get_reads_from();
2329 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2330 sequence->acquire = acquire;
2331 sequence->read = read;
2333 if (!release_seq_heads(rf, release_heads, sequence)) {
2334 /* add act to 'lazy checking' list */
2335 pending_rel_seqs->push_back(sequence);
2337 snapshot_free(sequence);
2342 * Attempt to resolve all stashed operations that might synchronize with a
2343 * release sequence for a given location. This implements the "lazy" portion of
2344 * determining whether or not a release sequence was contiguous, since not all
2345 * modification order information is present at the time an action occurs.
2347 * @param location The location/object that should be checked for release
2348 * sequence resolutions. A NULL value means to check all locations.
2349 * @param work_queue The work queue to which to add work items as they are
2351 * @return True if any updates occurred (new synchronization, new mo_graph
2354 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2356 bool updated = false;
2357 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs->begin();
2358 while (it != pending_rel_seqs->end()) {
2359 struct release_seq *pending = *it;
2360 ModelAction *acquire = pending->acquire;
2361 const ModelAction *read = pending->read;
2363 /* Only resolve sequences on the given location, if provided */
2364 if (location && read->get_location() != location) {
2369 const ModelAction *rf = read->get_reads_from();
2370 rel_heads_list_t release_heads;
2372 complete = release_seq_heads(rf, &release_heads, pending);
2373 for (unsigned int i = 0; i < release_heads.size(); i++) {
2374 if (!acquire->has_synchronized_with(release_heads[i])) {
2375 if (acquire->synchronize_with(release_heads[i]))
2378 set_bad_synchronization();
2383 /* Re-check all pending release sequences */
2384 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2385 /* Re-check read-acquire for mo_graph edges */
2386 if (acquire->is_read())
2387 work_queue->push_back(MOEdgeWorkEntry(acquire));
2389 /* propagate synchronization to later actions */
2390 action_list_t::reverse_iterator rit = action_trace->rbegin();
2391 for (; (*rit) != acquire; rit++) {
2392 ModelAction *propagate = *rit;
2393 if (acquire->happens_before(propagate)) {
2394 propagate->synchronize_with(acquire);
2395 /* Re-check 'propagate' for mo_graph edges */
2396 work_queue->push_back(MOEdgeWorkEntry(propagate));
2401 it = pending_rel_seqs->erase(it);
2402 snapshot_free(pending);
2408 // If we resolved promises or data races, see if we have realized a data race.
2415 * Performs various bookkeeping operations for the current ModelAction. For
2416 * instance, adds action to the per-object, per-thread action vector and to the
2417 * action trace list of all thread actions.
2419 * @param act is the ModelAction to add.
2421 void ModelChecker::add_action_to_lists(ModelAction *act)
2423 int tid = id_to_int(act->get_tid());
2424 ModelAction *uninit = NULL;
2426 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2427 if (list->empty() && act->is_atomic_var()) {
2428 uninit = get_uninitialized_action(act);
2429 uninit_id = id_to_int(uninit->get_tid());
2430 list->push_front(uninit);
2432 list->push_back(act);
2434 action_trace->push_back(act);
2436 action_trace->push_front(uninit);
2438 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2439 if (tid >= (int)vec->size())
2440 vec->resize(priv->next_thread_id);
2441 (*vec)[tid].push_back(act);
2443 (*vec)[uninit_id].push_front(uninit);
2445 if ((int)thrd_last_action->size() <= tid)
2446 thrd_last_action->resize(get_num_threads());
2447 (*thrd_last_action)[tid] = act;
2449 (*thrd_last_action)[uninit_id] = uninit;
2451 if (act->is_fence() && act->is_release()) {
2452 if ((int)thrd_last_fence_release->size() <= tid)
2453 thrd_last_fence_release->resize(get_num_threads());
2454 (*thrd_last_fence_release)[tid] = act;
2457 if (act->is_wait()) {
2458 void *mutex_loc = (void *) act->get_value();
2459 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2461 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2462 if (tid >= (int)vec->size())
2463 vec->resize(priv->next_thread_id);
2464 (*vec)[tid].push_back(act);
2469 * @brief Get the last action performed by a particular Thread
2470 * @param tid The thread ID of the Thread in question
2471 * @return The last action in the thread
2473 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2475 int threadid = id_to_int(tid);
2476 if (threadid < (int)thrd_last_action->size())
2477 return (*thrd_last_action)[id_to_int(tid)];
2483 * @brief Get the last fence release performed by a particular Thread
2484 * @param tid The thread ID of the Thread in question
2485 * @return The last fence release in the thread, if one exists; NULL otherwise
2487 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2489 int threadid = id_to_int(tid);
2490 if (threadid < (int)thrd_last_fence_release->size())
2491 return (*thrd_last_fence_release)[id_to_int(tid)];
2497 * Gets the last memory_order_seq_cst write (in the total global sequence)
2498 * performed on a particular object (i.e., memory location), not including the
2500 * @param curr The current ModelAction; also denotes the object location to
2502 * @return The last seq_cst write
2504 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2506 void *location = curr->get_location();
2507 action_list_t *list = get_safe_ptr_action(obj_map, location);
2508 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2509 action_list_t::reverse_iterator rit;
2510 for (rit = list->rbegin(); (*rit) != curr; rit++)
2512 rit++; /* Skip past curr */
2513 for ( ; rit != list->rend(); rit++)
2514 if ((*rit)->is_write() && (*rit)->is_seqcst())
2520 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2521 * performed in a particular thread, prior to a particular fence.
2522 * @param tid The ID of the thread to check
2523 * @param before_fence The fence from which to begin the search; if NULL, then
2524 * search for the most recent fence in the thread.
2525 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2527 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2529 /* All fences should have NULL location */
2530 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2531 action_list_t::reverse_iterator rit = list->rbegin();
2534 for (; rit != list->rend(); rit++)
2535 if (*rit == before_fence)
2538 ASSERT(*rit == before_fence);
2542 for (; rit != list->rend(); rit++)
2543 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2549 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2550 * location). This function identifies the mutex according to the current
2551 * action, which is presumed to perform on the same mutex.
2552 * @param curr The current ModelAction; also denotes the object location to
2554 * @return The last unlock operation
2556 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2558 void *location = curr->get_location();
2559 action_list_t *list = get_safe_ptr_action(obj_map, location);
2560 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2561 action_list_t::reverse_iterator rit;
2562 for (rit = list->rbegin(); rit != list->rend(); rit++)
2563 if ((*rit)->is_unlock() || (*rit)->is_wait())
2568 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2570 ModelAction *parent = get_last_action(tid);
2572 parent = get_thread(tid)->get_creation();
2577 * Returns the clock vector for a given thread.
2578 * @param tid The thread whose clock vector we want
2579 * @return Desired clock vector
2581 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2583 return get_parent_action(tid)->get_cv();
2587 * @brief Find the promise (if any) to resolve for the current action and
2588 * remove it from the pending promise vector
2589 * @param curr The current ModelAction. Should be a write.
2590 * @return The Promise to resolve, if any; otherwise NULL
2592 Promise * ModelChecker::pop_promise_to_resolve(const ModelAction *curr)
2594 for (unsigned int i = 0; i < promises->size(); i++)
2595 if (curr->get_node()->get_promise(i)) {
2596 Promise *ret = (*promises)[i];
2597 promises->erase(promises->begin() + i);
2604 * Resolve a Promise with a current write.
2605 * @param write The ModelAction that is fulfilling Promises
2606 * @param promise The Promise to resolve
2607 * @return True if the Promise was successfully resolved; false otherwise
2609 bool ModelChecker::resolve_promise(ModelAction *write, Promise *promise)
2611 ModelVector<ModelAction *> actions_to_check;
2613 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2614 ModelAction *read = promise->get_reader(i);
2615 read_from(read, write);
2616 actions_to_check.push_back(read);
2618 /* Make sure the promise's value matches the write's value */
2619 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2620 if (!mo_graph->resolvePromise(promise, write))
2621 priv->failed_promise = true;
2624 * @todo It is possible to end up in an inconsistent state, where a
2625 * "resolved" promise may still be referenced if
2626 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2628 * Note that the inconsistency only matters when dumping mo_graph to
2634 //Check whether reading these writes has made threads unable to
2636 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2637 ModelAction *read = actions_to_check[i];
2638 mo_check_promises(read, true);
2645 * Compute the set of promises that could potentially be satisfied by this
2646 * action. Note that the set computation actually appears in the Node, not in
2648 * @param curr The ModelAction that may satisfy promises
2650 void ModelChecker::compute_promises(ModelAction *curr)
2652 for (unsigned int i = 0; i < promises->size(); i++) {
2653 Promise *promise = (*promises)[i];
2654 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2657 bool satisfy = true;
2658 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2659 const ModelAction *act = promise->get_reader(j);
2660 if (act->happens_before(curr) ||
2661 act->could_synchronize_with(curr)) {
2667 curr->get_node()->set_promise(i);
2671 /** Checks promises in response to change in ClockVector Threads. */
2672 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2674 for (unsigned int i = 0; i < promises->size(); i++) {
2675 Promise *promise = (*promises)[i];
2676 if (!promise->thread_is_available(tid))
2678 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2679 const ModelAction *act = promise->get_reader(j);
2680 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2681 merge_cv->synchronized_since(act)) {
2682 if (promise->eliminate_thread(tid)) {
2683 /* Promise has failed */
2684 priv->failed_promise = true;
2692 void ModelChecker::check_promises_thread_disabled()
2694 for (unsigned int i = 0; i < promises->size(); i++) {
2695 Promise *promise = (*promises)[i];
2696 if (promise->has_failed()) {
2697 priv->failed_promise = true;
2704 * @brief Checks promises in response to addition to modification order for
2707 * We test whether threads are still available for satisfying promises after an
2708 * addition to our modification order constraints. Those that are unavailable
2709 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2710 * that promise has failed.
2712 * @param act The ModelAction which updated the modification order
2713 * @param is_read_check Should be true if act is a read and we must check for
2714 * updates to the store from which it read (there is a distinction here for
2715 * RMW's, which are both a load and a store)
2717 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2719 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2721 for (unsigned int i = 0; i < promises->size(); i++) {
2722 Promise *promise = (*promises)[i];
2724 // Is this promise on the same location?
2725 if (!promise->same_location(write))
2728 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2729 const ModelAction *pread = promise->get_reader(j);
2730 if (!pread->happens_before(act))
2732 if (mo_graph->checkPromise(write, promise)) {
2733 priv->failed_promise = true;
2739 // Don't do any lookups twice for the same thread
2740 if (!promise->thread_is_available(act->get_tid()))
2743 if (mo_graph->checkReachable(promise, write)) {
2744 if (mo_graph->checkPromise(write, promise)) {
2745 priv->failed_promise = true;
2753 * Compute the set of writes that may break the current pending release
2754 * sequence. This information is extracted from previou release sequence
2757 * @param curr The current ModelAction. Must be a release sequence fixup
2760 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2762 if (pending_rel_seqs->empty())
2765 struct release_seq *pending = pending_rel_seqs->back();
2766 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2767 const ModelAction *write = pending->writes[i];
2768 curr->get_node()->add_relseq_break(write);
2771 /* NULL means don't break the sequence; just synchronize */
2772 curr->get_node()->add_relseq_break(NULL);
2776 * Build up an initial set of all past writes that this 'read' action may read
2777 * from, as well as any previously-observed future values that must still be valid.
2779 * @param curr is the current ModelAction that we are exploring; it must be a
2782 void ModelChecker::build_may_read_from(ModelAction *curr)
2784 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2786 ASSERT(curr->is_read());
2788 ModelAction *last_sc_write = NULL;
2790 if (curr->is_seqcst())
2791 last_sc_write = get_last_seq_cst_write(curr);
2793 /* Iterate over all threads */
2794 for (i = 0; i < thrd_lists->size(); i++) {
2795 /* Iterate over actions in thread, starting from most recent */
2796 action_list_t *list = &(*thrd_lists)[i];
2797 action_list_t::reverse_iterator rit;
2798 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2799 ModelAction *act = *rit;
2801 /* Only consider 'write' actions */
2802 if (!act->is_write() || act == curr)
2805 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2806 bool allow_read = true;
2808 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2810 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2814 /* Only add feasible reads */
2815 mo_graph->startChanges();
2816 r_modification_order(curr, act);
2817 if (!is_infeasible())
2818 curr->get_node()->add_read_from_past(act);
2819 mo_graph->rollbackChanges();
2822 /* Include at most one act per-thread that "happens before" curr */
2823 if (act->happens_before(curr))
2828 /* Inherit existing, promised future values */
2829 for (i = 0; i < promises->size(); i++) {
2830 const Promise *promise = (*promises)[i];
2831 const ModelAction *promise_read = promise->get_reader(0);
2832 if (promise_read->same_var(curr)) {
2833 /* Only add feasible future-values */
2834 mo_graph->startChanges();
2835 r_modification_order(curr, promise);
2836 if (!is_infeasible())
2837 curr->get_node()->add_read_from_promise(promise_read);
2838 mo_graph->rollbackChanges();
2842 /* We may find no valid may-read-from only if the execution is doomed */
2843 if (!curr->get_node()->read_from_size()) {
2844 priv->no_valid_reads = true;
2848 if (DBG_ENABLED()) {
2849 model_print("Reached read action:\n");
2851 model_print("Printing read_from_past\n");
2852 curr->get_node()->print_read_from_past();
2853 model_print("End printing read_from_past\n");
2857 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2859 for ( ; write != NULL; write = write->get_reads_from()) {
2860 /* UNINIT actions don't have a Node, and they never sleep */
2861 if (write->is_uninitialized())
2863 Node *prevnode = write->get_node()->get_parent();
2865 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2866 if (write->is_release() && thread_sleep)
2868 if (!write->is_rmw())
2875 * @brief Get an action representing an uninitialized atomic
2877 * This function may create a new one or try to retrieve one from the NodeStack
2879 * @param curr The current action, which prompts the creation of an UNINIT action
2880 * @return A pointer to the UNINIT ModelAction
2882 ModelAction * ModelChecker::get_uninitialized_action(const ModelAction *curr) const
2884 Node *node = curr->get_node();
2885 ModelAction *act = node->get_uninit_action();
2887 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), model->params.uninitvalue, model_thread);
2888 node->set_uninit_action(act);
2890 act->create_cv(NULL);
2894 static void print_list(action_list_t *list)
2896 action_list_t::iterator it;
2898 model_print("---------------------------------------------------------------------\n");
2900 unsigned int hash = 0;
2902 for (it = list->begin(); it != list->end(); it++) {
2903 const ModelAction *act = *it;
2904 if (act->get_seq_number() > 0)
2906 hash = hash^(hash<<3)^((*it)->hash());
2908 model_print("HASH %u\n", hash);
2909 model_print("---------------------------------------------------------------------\n");
2912 #if SUPPORT_MOD_ORDER_DUMP
2913 void ModelChecker::dumpGraph(char *filename) const
2916 sprintf(buffer, "%s.dot", filename);
2917 FILE *file = fopen(buffer, "w");
2918 fprintf(file, "digraph %s {\n", filename);
2919 mo_graph->dumpNodes(file);
2920 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2922 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2923 ModelAction *act = *it;
2924 if (act->is_read()) {
2925 mo_graph->dot_print_node(file, act);
2926 if (act->get_reads_from())
2927 mo_graph->dot_print_edge(file,
2928 act->get_reads_from(),
2930 "label=\"rf\", color=red, weight=2");
2932 mo_graph->dot_print_edge(file,
2933 act->get_reads_from_promise(),
2935 "label=\"rf\", color=red");
2937 if (thread_array[act->get_tid()]) {
2938 mo_graph->dot_print_edge(file,
2939 thread_array[id_to_int(act->get_tid())],
2941 "label=\"sb\", color=blue, weight=400");
2944 thread_array[act->get_tid()] = act;
2946 fprintf(file, "}\n");
2947 model_free(thread_array);
2952 /** @brief Prints an execution trace summary. */
2953 void ModelChecker::print_summary() const
2955 #if SUPPORT_MOD_ORDER_DUMP
2956 char buffername[100];
2957 sprintf(buffername, "exec%04u", stats.num_total);
2958 mo_graph->dumpGraphToFile(buffername);
2959 sprintf(buffername, "graph%04u", stats.num_total);
2960 dumpGraph(buffername);
2963 model_print("Execution %d:", stats.num_total);
2964 if (isfeasibleprefix()) {
2965 if (scheduler->all_threads_sleeping())
2966 model_print(" SLEEP-SET REDUNDANT");
2969 print_infeasibility(" INFEASIBLE");
2970 print_list(action_trace);
2972 if (!promises->empty()) {
2973 model_print("Pending promises:\n");
2974 for (unsigned int i = 0; i < promises->size(); i++) {
2975 model_print(" [P%u] ", i);
2976 (*promises)[i]->print();
2983 * Add a Thread to the system for the first time. Should only be called once
2985 * @param t The Thread to add
2987 void ModelChecker::add_thread(Thread *t)
2989 thread_map->put(id_to_int(t->get_id()), t);
2990 scheduler->add_thread(t);
2994 * @brief Get a Thread reference by its ID
2995 * @param tid The Thread's ID
2996 * @return A Thread reference
2998 Thread * ModelChecker::get_thread(thread_id_t tid) const
3000 return thread_map->get(id_to_int(tid));
3004 * @brief Get a reference to the Thread in which a ModelAction was executed
3005 * @param act The ModelAction
3006 * @return A Thread reference
3008 Thread * ModelChecker::get_thread(const ModelAction *act) const
3010 return get_thread(act->get_tid());
3014 * @brief Get a Promise's "promise number"
3016 * A "promise number" is an index number that is unique to a promise, valid
3017 * only for a specific snapshot of an execution trace. Promises may come and go
3018 * as they are generated an resolved, so an index only retains meaning for the
3021 * @param promise The Promise to check
3022 * @return The promise index, if the promise still is valid; otherwise -1
3024 int ModelChecker::get_promise_number(const Promise *promise) const
3026 for (unsigned int i = 0; i < promises->size(); i++)
3027 if ((*promises)[i] == promise)
3034 * @brief Check if a Thread is currently enabled
3035 * @param t The Thread to check
3036 * @return True if the Thread is currently enabled
3038 bool ModelChecker::is_enabled(Thread *t) const
3040 return scheduler->is_enabled(t);
3044 * @brief Check if a Thread is currently enabled
3045 * @param tid The ID of the Thread to check
3046 * @return True if the Thread is currently enabled
3048 bool ModelChecker::is_enabled(thread_id_t tid) const
3050 return scheduler->is_enabled(tid);
3054 * Switch from a model-checker context to a user-thread context. This is the
3055 * complement of ModelChecker::switch_to_master and must be called from the
3056 * model-checker context
3058 * @param thread The user-thread to switch to
3060 void ModelChecker::switch_from_master(Thread *thread)
3062 scheduler->set_current_thread(thread);
3063 Thread::swap(&system_context, thread);
3067 * Switch from a user-context to the "master thread" context (a.k.a. system
3068 * context). This switch is made with the intention of exploring a particular
3069 * model-checking action (described by a ModelAction object). Must be called
3070 * from a user-thread context.
3072 * @param act The current action that will be explored. May be NULL only if
3073 * trace is exiting via an assertion (see ModelChecker::set_assert and
3074 * ModelChecker::has_asserted).
3075 * @return Return the value returned by the current action
3077 uint64_t ModelChecker::switch_to_master(ModelAction *act)
3080 Thread *old = thread_current();
3081 scheduler->set_current_thread(NULL);
3082 ASSERT(!old->get_pending());
3083 old->set_pending(act);
3084 if (Thread::swap(old, &system_context) < 0) {
3085 perror("swap threads");
3088 return old->get_return_value();
3092 * Takes the next step in the execution, if possible.
3093 * @param curr The current step to take
3094 * @return Returns the next Thread to run, if any; NULL if this execution
3097 Thread * ModelChecker::take_step(ModelAction *curr)
3099 Thread *curr_thrd = get_thread(curr);
3100 ASSERT(curr_thrd->get_state() == THREAD_READY);
3102 curr = check_current_action(curr);
3104 /* Infeasible -> don't take any more steps */
3105 if (is_infeasible())
3107 else if (isfeasibleprefix() && have_bug_reports()) {
3112 if (params.bound != 0 && priv->used_sequence_numbers > params.bound)
3115 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
3116 scheduler->remove_thread(curr_thrd);
3118 Thread *next_thrd = NULL;
3120 next_thrd = action_select_next_thread(curr);
3122 next_thrd = get_next_thread();
3124 DEBUG("(%d, %d)\n", curr_thrd ? id_to_int(curr_thrd->get_id()) : -1,
3125 next_thrd ? id_to_int(next_thrd->get_id()) : -1);
3130 /** Wrapper to run the user's main function, with appropriate arguments */
3131 void user_main_wrapper(void *)
3133 user_main(model->params.argc, model->params.argv);
3136 /** @brief Run ModelChecker for the user program */
3137 void ModelChecker::run()
3141 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL, NULL);
3146 * Stash next pending action(s) for thread(s). There
3147 * should only need to stash one thread's action--the
3148 * thread which just took a step--plus the first step
3149 * for any newly-created thread
3151 for (unsigned int i = 0; i < get_num_threads(); i++) {
3152 thread_id_t tid = int_to_id(i);
3153 Thread *thr = get_thread(tid);
3154 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
3155 switch_from_master(thr);
3156 if (thr->is_waiting_on(thr))
3157 assert_bug("Deadlock detected");
3161 /* Catch assertions from prior take_step or from
3162 * between-ModelAction bugs (e.g., data races) */
3166 /* Consume the next action for a Thread */
3167 ModelAction *curr = t->get_pending();
3168 t->set_pending(NULL);
3169 t = take_step(curr);
3170 } while (t && !t->is_model_thread());
3173 * Launch end-of-execution release sequence fixups only when
3174 * the execution is otherwise feasible AND there are:
3176 * (1) pending release sequences
3177 * (2) pending assertions that could be invalidated by a change
3178 * in clock vectors (i.e., data races)
3179 * (3) no pending promises
3181 while (!pending_rel_seqs->empty() &&
3182 is_feasible_prefix_ignore_relseq() &&
3183 !unrealizedraces.empty()) {
3184 model_print("*** WARNING: release sequence fixup action "
3185 "(%zu pending release seuqence(s)) ***\n",
3186 pending_rel_seqs->size());
3187 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
3188 std::memory_order_seq_cst, NULL, VALUE_NONE,
3192 } while (next_execution());
3194 model_print("******* Model-checking complete: *******\n");