11 #include "snapshot-interface.h"
13 #include "clockvector.h"
14 #include "cyclegraph.h"
17 #include "threads-model.h"
19 #include "traceanalysis.h"
21 #define INITIAL_THREAD_ID 0
26 bug_message(const char *str) {
27 const char *fmt = " [BUG] %s\n";
28 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
29 sprintf(msg, fmt, str);
31 ~bug_message() { if (msg) snapshot_free(msg); }
34 void print() { model_print("%s", msg); }
40 * Structure for holding small ModelChecker members that should be snapshotted
42 struct model_snapshot_members {
43 model_snapshot_members() :
44 /* First thread created will have id INITIAL_THREAD_ID */
45 next_thread_id(INITIAL_THREAD_ID),
46 used_sequence_numbers(0),
50 failed_promise(false),
51 too_many_reads(false),
52 no_valid_reads(false),
53 bad_synchronization(false),
57 ~model_snapshot_members() {
58 for (unsigned int i = 0; i < bugs.size(); i++)
63 unsigned int next_thread_id;
64 modelclock_t used_sequence_numbers;
65 ModelAction *next_backtrack;
66 SnapVector<bug_message *> bugs;
67 struct execution_stats stats;
71 /** @brief Incorrectly-ordered synchronization was made */
72 bool bad_synchronization;
78 /** @brief Constructor */
79 ModelChecker::ModelChecker(struct model_params params) :
80 /* Initialize default scheduler */
82 scheduler(new Scheduler()),
84 earliest_diverge(NULL),
85 action_trace(new action_list_t()),
86 thread_map(new HashTable<int, Thread *, int>()),
87 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
88 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
89 obj_thrd_map(new HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4 >()),
90 promises(new SnapVector<Promise *>()),
91 futurevalues(new SnapVector<struct PendingFutureValue>()),
92 pending_rel_seqs(new SnapVector<struct release_seq *>()),
93 thrd_last_action(new SnapVector<ModelAction *>(1)),
94 thrd_last_fence_release(new SnapVector<ModelAction *>()),
95 node_stack(new NodeStack()),
96 trace_analyses(new ModelVector<Trace_Analysis *>()),
97 priv(new struct model_snapshot_members()),
98 mo_graph(new CycleGraph())
100 /* Initialize a model-checker thread, for special ModelActions */
101 model_thread = new Thread(get_next_id());
102 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
105 /** @brief Destructor */
106 ModelChecker::~ModelChecker()
108 for (unsigned int i = 0; i < get_num_threads(); i++)
109 delete thread_map->get(i);
114 delete condvar_waiters_map;
117 for (unsigned int i = 0; i < promises->size(); i++)
118 delete (*promises)[i];
121 delete pending_rel_seqs;
123 delete thrd_last_action;
124 delete thrd_last_fence_release;
126 for (unsigned int i = 0; i <trace_analyses->size();i++)
127 delete (*trace_analyses)[i];
128 delete trace_analyses;
134 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
136 action_list_t *tmp = hash->get(ptr);
138 tmp = new action_list_t();
144 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
146 SnapVector<action_list_t> *tmp = hash->get(ptr);
148 tmp = new SnapVector<action_list_t>();
155 * Restores user program to initial state and resets all model-checker data
158 void ModelChecker::reset_to_initial_state()
160 DEBUG("+++ Resetting to initial state +++\n");
161 node_stack->reset_execution();
164 * FIXME: if we utilize partial rollback, we will need to free only
165 * those pending actions which were NOT pending before the rollback
168 for (unsigned int i = 0; i < get_num_threads(); i++)
169 delete get_thread(int_to_id(i))->get_pending();
171 snapshot_backtrack_before(0);
174 /** @return a thread ID for a new Thread */
175 thread_id_t ModelChecker::get_next_id()
177 return priv->next_thread_id++;
180 /** @return the number of user threads created during this execution */
181 unsigned int ModelChecker::get_num_threads() const
183 return priv->next_thread_id;
187 * Must be called from user-thread context (e.g., through the global
188 * thread_current() interface)
190 * @return The currently executing Thread.
192 Thread * ModelChecker::get_current_thread() const
194 return scheduler->get_current_thread();
197 /** @return a sequence number for a new ModelAction */
198 modelclock_t ModelChecker::get_next_seq_num()
200 return ++priv->used_sequence_numbers;
203 Node * ModelChecker::get_curr_node() const
205 return node_stack->get_head();
209 * @brief Select the next thread to execute based on the curren action
211 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
212 * actions should be followed by the execution of their child thread. In either
213 * case, the current action should determine the next thread schedule.
215 * @param curr The current action
216 * @return The next thread to run, if the current action will determine this
217 * selection; otherwise NULL
219 Thread * ModelChecker::action_select_next_thread(const ModelAction *curr) const
221 /* Do not split atomic RMW */
223 return get_thread(curr);
224 /* Follow CREATE with the created thread */
225 if (curr->get_type() == THREAD_CREATE)
226 return curr->get_thread_operand();
231 * @brief Choose the next thread to execute.
233 * This function chooses the next thread that should execute. It can enforce
234 * execution replay/backtracking or, if the model-checker has no preference
235 * regarding the next thread (i.e., when exploring a new execution ordering),
236 * we defer to the scheduler.
238 * @return The next chosen thread to run, if any exist. Or else if the current
239 * execution should terminate, return NULL.
241 Thread * ModelChecker::get_next_thread()
246 * Have we completed exploring the preselected path? Then let the
250 return scheduler->select_next_thread();
252 /* Else, we are trying to replay an execution */
253 ModelAction *next = node_stack->get_next()->get_action();
255 if (next == diverge) {
256 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
257 earliest_diverge = diverge;
259 Node *nextnode = next->get_node();
260 Node *prevnode = nextnode->get_parent();
261 scheduler->update_sleep_set(prevnode);
263 /* Reached divergence point */
264 if (nextnode->increment_behaviors()) {
265 /* Execute the same thread with a new behavior */
266 tid = next->get_tid();
267 node_stack->pop_restofstack(2);
270 /* Make a different thread execute for next step */
271 scheduler->add_sleep(get_thread(next->get_tid()));
272 tid = prevnode->get_next_backtrack();
273 /* Make sure the backtracked thread isn't sleeping. */
274 node_stack->pop_restofstack(1);
275 if (diverge == earliest_diverge) {
276 earliest_diverge = prevnode->get_action();
279 /* Start the round robin scheduler from this thread id */
280 scheduler->set_scheduler_thread(tid);
281 /* The correct sleep set is in the parent node. */
284 DEBUG("*** Divergence point ***\n");
288 tid = next->get_tid();
290 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
291 ASSERT(tid != THREAD_ID_T_NONE);
292 return get_thread(id_to_int(tid));
296 * We need to know what the next actions of all threads in the sleep
297 * set will be. This method computes them and stores the actions at
298 * the corresponding thread object's pending action.
301 void ModelChecker::execute_sleep_set()
303 for (unsigned int i = 0; i < get_num_threads(); i++) {
304 thread_id_t tid = int_to_id(i);
305 Thread *thr = get_thread(tid);
306 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
307 thr->get_pending()->set_sleep_flag();
313 * @brief Should the current action wake up a given thread?
315 * @param curr The current action
316 * @param thread The thread that we might wake up
317 * @return True, if we should wake up the sleeping thread; false otherwise
319 bool ModelChecker::should_wake_up(const ModelAction *curr, const Thread *thread) const
321 const ModelAction *asleep = thread->get_pending();
322 /* Don't allow partial RMW to wake anyone up */
325 /* Synchronizing actions may have been backtracked */
326 if (asleep->could_synchronize_with(curr))
328 /* All acquire/release fences and fence-acquire/store-release */
329 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
331 /* Fence-release + store can awake load-acquire on the same location */
332 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
333 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
334 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
340 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
342 for (unsigned int i = 0; i < get_num_threads(); i++) {
343 Thread *thr = get_thread(int_to_id(i));
344 if (scheduler->is_sleep_set(thr)) {
345 if (should_wake_up(curr, thr))
346 /* Remove this thread from sleep set */
347 scheduler->remove_sleep(thr);
352 /** @brief Alert the model-checker that an incorrectly-ordered
353 * synchronization was made */
354 void ModelChecker::set_bad_synchronization()
356 priv->bad_synchronization = true;
360 * Check whether the current trace has triggered an assertion which should halt
363 * @return True, if the execution should be aborted; false otherwise
365 bool ModelChecker::has_asserted() const
367 return priv->asserted;
371 * Trigger a trace assertion which should cause this execution to be halted.
372 * This can be due to a detected bug or due to an infeasibility that should
375 void ModelChecker::set_assert()
377 priv->asserted = true;
381 * Check if we are in a deadlock. Should only be called at the end of an
382 * execution, although it should not give false positives in the middle of an
383 * execution (there should be some ENABLED thread).
385 * @return True if program is in a deadlock; false otherwise
387 bool ModelChecker::is_deadlocked() const
389 bool blocking_threads = false;
390 for (unsigned int i = 0; i < get_num_threads(); i++) {
391 thread_id_t tid = int_to_id(i);
394 Thread *t = get_thread(tid);
395 if (!t->is_model_thread() && t->get_pending())
396 blocking_threads = true;
398 return blocking_threads;
402 * Check if this is a complete execution. That is, have all thread completed
403 * execution (rather than exiting because sleep sets have forced a redundant
406 * @return True if the execution is complete.
408 bool ModelChecker::is_complete_execution() const
410 for (unsigned int i = 0; i < get_num_threads(); i++)
411 if (is_enabled(int_to_id(i)))
417 * @brief Assert a bug in the executing program.
419 * Use this function to assert any sort of bug in the user program. If the
420 * current trace is feasible (actually, a prefix of some feasible execution),
421 * then this execution will be aborted, printing the appropriate message. If
422 * the current trace is not yet feasible, the error message will be stashed and
423 * printed if the execution ever becomes feasible.
425 * @param msg Descriptive message for the bug (do not include newline char)
426 * @return True if bug is immediately-feasible
428 bool ModelChecker::assert_bug(const char *msg, ...)
434 vsnprintf(str, sizeof(str), msg, ap);
437 priv->bugs.push_back(new bug_message(str));
439 if (isfeasibleprefix()) {
447 * @brief Assert a bug in the executing program, asserted by a user thread
448 * @see ModelChecker::assert_bug
449 * @param msg Descriptive message for the bug (do not include newline char)
451 void ModelChecker::assert_user_bug(const char *msg)
453 /* If feasible bug, bail out now */
455 switch_to_master(NULL);
458 /** @return True, if any bugs have been reported for this execution */
459 bool ModelChecker::have_bug_reports() const
461 return priv->bugs.size() != 0;
464 /** @brief Print bug report listing for this execution (if any bugs exist) */
465 void ModelChecker::print_bugs() const
467 if (have_bug_reports()) {
468 model_print("Bug report: %zu bug%s detected\n",
470 priv->bugs.size() > 1 ? "s" : "");
471 for (unsigned int i = 0; i < priv->bugs.size(); i++)
472 priv->bugs[i]->print();
477 * @brief Record end-of-execution stats
479 * Must be run when exiting an execution. Records various stats.
480 * @see struct execution_stats
482 void ModelChecker::record_stats()
485 if (!isfeasibleprefix())
486 stats.num_infeasible++;
487 else if (have_bug_reports())
488 stats.num_buggy_executions++;
489 else if (is_complete_execution())
490 stats.num_complete++;
492 stats.num_redundant++;
495 * @todo We can violate this ASSERT() when fairness/sleep sets
496 * conflict to cause an execution to terminate, e.g. with:
497 * Scheduler: [0: disabled][1: disabled][2: sleep][3: current, enabled]
499 //ASSERT(scheduler->all_threads_sleeping());
503 /** @brief Print execution stats */
504 void ModelChecker::print_stats() const
506 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
507 model_print("Number of redundant executions: %d\n", stats.num_redundant);
508 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
509 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
510 model_print("Total executions: %d\n", stats.num_total);
511 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
515 * @brief End-of-exeuction print
516 * @param printbugs Should any existing bugs be printed?
518 void ModelChecker::print_execution(bool printbugs) const
520 print_program_output();
522 if (params.verbose) {
523 model_print("Earliest divergence point since last feasible execution:\n");
524 if (earliest_diverge)
525 earliest_diverge->print();
527 model_print("(Not set)\n");
533 /* Don't print invalid bugs */
542 * Queries the model-checker for more executions to explore and, if one
543 * exists, resets the model-checker state to execute a new execution.
545 * @return If there are more executions to explore, return true. Otherwise,
548 bool ModelChecker::next_execution()
551 /* Is this execution a feasible execution that's worth bug-checking? */
552 bool complete = isfeasibleprefix() && (is_complete_execution() ||
555 /* End-of-execution bug checks */
558 assert_bug("Deadlock detected");
561 run_trace_analyses();
567 if (params.verbose || (complete && have_bug_reports()))
568 print_execution(complete);
570 clear_program_output();
573 earliest_diverge = NULL;
575 if ((diverge = get_next_backtrack()) == NULL)
579 model_print("Next execution will diverge at:\n");
583 reset_to_initial_state();
588 * @brief Run trace analyses on complete trace. */
590 void ModelChecker::run_trace_analyses() {
591 for(unsigned int i=0; i < trace_analyses->size(); i++) {
592 (*trace_analyses)[i]->analyze(action_trace);
597 * @brief Find the last fence-related backtracking conflict for a ModelAction
599 * This function performs the search for the most recent conflicting action
600 * against which we should perform backtracking, as affected by fence
601 * operations. This includes pairs of potentially-synchronizing actions which
602 * occur due to fence-acquire or fence-release, and hence should be explored in
603 * the opposite execution order.
605 * @param act The current action
606 * @return The most recent action which conflicts with act due to fences
608 ModelAction * ModelChecker::get_last_fence_conflict(ModelAction *act) const
610 /* Only perform release/acquire fence backtracking for stores */
611 if (!act->is_write())
614 /* Find a fence-release (or, act is a release) */
615 ModelAction *last_release;
616 if (act->is_release())
619 last_release = get_last_fence_release(act->get_tid());
623 /* Skip past the release */
624 action_list_t *list = action_trace;
625 action_list_t::reverse_iterator rit;
626 for (rit = list->rbegin(); rit != list->rend(); rit++)
627 if (*rit == last_release)
629 ASSERT(rit != list->rend());
634 * load --sb-> fence-acquire */
635 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
636 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
637 bool found_acquire_fences = false;
638 for ( ; rit != list->rend(); rit++) {
639 ModelAction *prev = *rit;
640 if (act->same_thread(prev))
643 int tid = id_to_int(prev->get_tid());
645 if (prev->is_read() && act->same_var(prev)) {
646 if (prev->is_acquire()) {
647 /* Found most recent load-acquire, don't need
648 * to search for more fences */
649 if (!found_acquire_fences)
652 prior_loads[tid] = prev;
655 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
656 found_acquire_fences = true;
657 acquire_fences[tid] = prev;
661 ModelAction *latest_backtrack = NULL;
662 for (unsigned int i = 0; i < acquire_fences.size(); i++)
663 if (acquire_fences[i] && prior_loads[i])
664 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
665 latest_backtrack = acquire_fences[i];
666 return latest_backtrack;
670 * @brief Find the last backtracking conflict for a ModelAction
672 * This function performs the search for the most recent conflicting action
673 * against which we should perform backtracking. This primary includes pairs of
674 * synchronizing actions which should be explored in the opposite execution
677 * @param act The current action
678 * @return The most recent action which conflicts with act
680 ModelAction * ModelChecker::get_last_conflict(ModelAction *act) const
682 switch (act->get_type()) {
683 /* case ATOMIC_FENCE: fences don't directly cause backtracking */
687 ModelAction *ret = NULL;
689 /* linear search: from most recent to oldest */
690 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
691 action_list_t::reverse_iterator rit;
692 for (rit = list->rbegin(); rit != list->rend(); rit++) {
693 ModelAction *prev = *rit;
694 if (prev->could_synchronize_with(act)) {
700 ModelAction *ret2 = get_last_fence_conflict(act);
710 case ATOMIC_TRYLOCK: {
711 /* linear search: from most recent to oldest */
712 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
713 action_list_t::reverse_iterator rit;
714 for (rit = list->rbegin(); rit != list->rend(); rit++) {
715 ModelAction *prev = *rit;
716 if (act->is_conflicting_lock(prev))
721 case ATOMIC_UNLOCK: {
722 /* linear search: from most recent to oldest */
723 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
724 action_list_t::reverse_iterator rit;
725 for (rit = list->rbegin(); rit != list->rend(); rit++) {
726 ModelAction *prev = *rit;
727 if (!act->same_thread(prev) && prev->is_failed_trylock())
733 /* linear search: from most recent to oldest */
734 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
735 action_list_t::reverse_iterator rit;
736 for (rit = list->rbegin(); rit != list->rend(); rit++) {
737 ModelAction *prev = *rit;
738 if (!act->same_thread(prev) && prev->is_failed_trylock())
740 if (!act->same_thread(prev) && prev->is_notify())
746 case ATOMIC_NOTIFY_ALL:
747 case ATOMIC_NOTIFY_ONE: {
748 /* linear search: from most recent to oldest */
749 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
750 action_list_t::reverse_iterator rit;
751 for (rit = list->rbegin(); rit != list->rend(); rit++) {
752 ModelAction *prev = *rit;
753 if (!act->same_thread(prev) && prev->is_wait())
764 /** This method finds backtracking points where we should try to
765 * reorder the parameter ModelAction against.
767 * @param the ModelAction to find backtracking points for.
769 void ModelChecker::set_backtracking(ModelAction *act)
771 Thread *t = get_thread(act);
772 ModelAction *prev = get_last_conflict(act);
776 Node *node = prev->get_node()->get_parent();
778 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
779 int low_tid, high_tid;
780 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
781 low_tid = id_to_int(act->get_tid());
782 high_tid = low_tid + 1;
785 high_tid = get_num_threads();
788 for (int i = low_tid; i < high_tid; i++) {
789 thread_id_t tid = int_to_id(i);
791 /* Make sure this thread can be enabled here. */
792 if (i >= node->get_num_threads())
795 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
796 /* Don't backtrack into a point where the thread is disabled or sleeping. */
797 if (node->enabled_status(tid) != THREAD_ENABLED)
800 /* Check if this has been explored already */
801 if (node->has_been_explored(tid))
804 /* See if fairness allows */
805 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
807 for (int t = 0; t < node->get_num_threads(); t++) {
808 thread_id_t tother = int_to_id(t);
809 if (node->is_enabled(tother) && node->has_priority(tother)) {
818 /* See if CHESS-like yield fairness allows */
819 if (model->params.yieldon) {
821 for (int t = 0; t < node->get_num_threads(); t++) {
822 thread_id_t tother = int_to_id(t);
823 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
832 /* Cache the latest backtracking point */
833 set_latest_backtrack(prev);
835 /* If this is a new backtracking point, mark the tree */
836 if (!node->set_backtrack(tid))
838 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
839 id_to_int(prev->get_tid()),
840 id_to_int(t->get_id()));
849 * @brief Cache the a backtracking point as the "most recent", if eligible
851 * Note that this does not prepare the NodeStack for this backtracking
852 * operation, it only caches the action on a per-execution basis
854 * @param act The operation at which we should explore a different next action
855 * (i.e., backtracking point)
856 * @return True, if this action is now the most recent backtracking point;
859 bool ModelChecker::set_latest_backtrack(ModelAction *act)
861 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
862 priv->next_backtrack = act;
869 * Returns last backtracking point. The model checker will explore a different
870 * path for this point in the next execution.
871 * @return The ModelAction at which the next execution should diverge.
873 ModelAction * ModelChecker::get_next_backtrack()
875 ModelAction *next = priv->next_backtrack;
876 priv->next_backtrack = NULL;
881 * Processes a read model action.
882 * @param curr is the read model action to process.
883 * @return True if processing this read updates the mo_graph.
885 bool ModelChecker::process_read(ModelAction *curr)
887 Node *node = curr->get_node();
889 bool updated = false;
890 switch (node->get_read_from_status()) {
891 case READ_FROM_PAST: {
892 const ModelAction *rf = node->get_read_from_past();
895 mo_graph->startChanges();
897 ASSERT(!is_infeasible());
898 if (!check_recency(curr, rf)) {
899 if (node->increment_read_from()) {
900 mo_graph->rollbackChanges();
903 priv->too_many_reads = true;
907 updated = r_modification_order(curr, rf);
909 mo_graph->commitChanges();
910 mo_check_promises(curr, true);
913 case READ_FROM_PROMISE: {
914 Promise *promise = curr->get_node()->get_read_from_promise();
915 if (promise->add_reader(curr))
916 priv->failed_promise = true;
917 curr->set_read_from_promise(promise);
918 mo_graph->startChanges();
919 if (!check_recency(curr, promise))
920 priv->too_many_reads = true;
921 updated = r_modification_order(curr, promise);
922 mo_graph->commitChanges();
925 case READ_FROM_FUTURE: {
926 /* Read from future value */
927 struct future_value fv = node->get_future_value();
928 Promise *promise = new Promise(curr, fv);
929 curr->set_read_from_promise(promise);
930 promises->push_back(promise);
931 mo_graph->startChanges();
932 updated = r_modification_order(curr, promise);
933 mo_graph->commitChanges();
939 get_thread(curr)->set_return_value(curr->get_return_value());
945 * Processes a lock, trylock, or unlock model action. @param curr is
946 * the read model action to process.
948 * The try lock operation checks whether the lock is taken. If not,
949 * it falls to the normal lock operation case. If so, it returns
952 * The lock operation has already been checked that it is enabled, so
953 * it just grabs the lock and synchronizes with the previous unlock.
955 * The unlock operation has to re-enable all of the threads that are
956 * waiting on the lock.
958 * @return True if synchronization was updated; false otherwise
960 bool ModelChecker::process_mutex(ModelAction *curr)
962 std::mutex *mutex = curr->get_mutex();
963 struct std::mutex_state *state = NULL;
966 state = mutex->get_state();
968 switch (curr->get_type()) {
969 case ATOMIC_TRYLOCK: {
970 bool success = !state->locked;
971 curr->set_try_lock(success);
973 get_thread(curr)->set_return_value(0);
976 get_thread(curr)->set_return_value(1);
978 //otherwise fall into the lock case
980 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
981 assert_bug("Lock access before initialization");
982 state->locked = get_thread(curr);
983 ModelAction *unlock = get_last_unlock(curr);
984 //synchronize with the previous unlock statement
985 if (unlock != NULL) {
986 curr->synchronize_with(unlock);
992 case ATOMIC_UNLOCK: {
993 /* wake up the other threads */
994 for (unsigned int i = 0; i < get_num_threads(); i++) {
995 Thread *t = get_thread(int_to_id(i));
996 Thread *curr_thrd = get_thread(curr);
997 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
1001 /* unlock the lock - after checking who was waiting on it */
1002 state->locked = NULL;
1004 if (!curr->is_wait())
1005 break; /* The rest is only for ATOMIC_WAIT */
1007 /* Should we go to sleep? (simulate spurious failures) */
1008 if (curr->get_node()->get_misc() == 0) {
1009 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
1011 scheduler->sleep(get_thread(curr));
1015 case ATOMIC_NOTIFY_ALL: {
1016 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1017 //activate all the waiting threads
1018 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
1019 scheduler->wake(get_thread(*rit));
1024 case ATOMIC_NOTIFY_ONE: {
1025 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1026 int wakeupthread = curr->get_node()->get_misc();
1027 action_list_t::iterator it = waiters->begin();
1028 advance(it, wakeupthread);
1029 scheduler->wake(get_thread(*it));
1041 * @brief Check if the current pending promises allow a future value to be sent
1043 * If one of the following is true:
1044 * (a) there are no pending promises
1045 * (b) the reader and writer do not cross any promises
1046 * Then, it is safe to pass a future value back now.
1048 * Otherwise, we must save the pending future value until (a) or (b) is true
1050 * @param writer The operation which sends the future value. Must be a write.
1051 * @param reader The operation which will observe the value. Must be a read.
1052 * @return True if the future value can be sent now; false if it must wait.
1054 bool ModelChecker::promises_may_allow(const ModelAction *writer,
1055 const ModelAction *reader) const
1057 if (promises->empty())
1059 for(int i=promises->size()-1;i>=0;i--) {
1060 ModelAction *pr=(*promises)[i]->get_reader(0);
1061 //reader is after promise...doesn't cross any promise
1064 //writer is after promise, reader before...bad...
1072 * @brief Add a future value to a reader
1074 * This function performs a few additional checks to ensure that the future
1075 * value can be feasibly observed by the reader
1077 * @param writer The operation whose value is sent. Must be a write.
1078 * @param reader The read operation which may read the future value. Must be a read.
1080 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
1082 /* Do more ambitious checks now that mo is more complete */
1083 if (!mo_may_allow(writer, reader))
1086 Node *node = reader->get_node();
1088 /* Find an ancestor thread which exists at the time of the reader */
1089 Thread *write_thread = get_thread(writer);
1090 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
1091 write_thread = write_thread->get_parent();
1093 struct future_value fv = {
1094 writer->get_write_value(),
1095 writer->get_seq_number() + params.maxfuturedelay,
1096 write_thread->get_id(),
1098 if (node->add_future_value(fv))
1099 set_latest_backtrack(reader);
1103 * Process a write ModelAction
1104 * @param curr The ModelAction to process
1105 * @return True if the mo_graph was updated or promises were resolved
1107 bool ModelChecker::process_write(ModelAction *curr)
1109 /* Readers to which we may send our future value */
1110 ModelVector<ModelAction *> send_fv;
1112 const ModelAction *earliest_promise_reader;
1113 bool updated_promises = false;
1115 bool updated_mod_order = w_modification_order(curr, &send_fv);
1116 Promise *promise = pop_promise_to_resolve(curr);
1119 earliest_promise_reader = promise->get_reader(0);
1120 updated_promises = resolve_promise(curr, promise);
1122 earliest_promise_reader = NULL;
1124 for (unsigned int i = 0; i < send_fv.size(); i++) {
1125 ModelAction *read = send_fv[i];
1127 /* Don't send future values to reads after the Promise we resolve */
1128 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
1129 /* Check if future value can be sent immediately */
1130 if (promises_may_allow(curr, read)) {
1131 add_future_value(curr, read);
1133 futurevalues->push_back(PendingFutureValue(curr, read));
1138 /* Check the pending future values */
1139 for (int i = (int)futurevalues->size() - 1; i >= 0; i--) {
1140 struct PendingFutureValue pfv = (*futurevalues)[i];
1141 if (promises_may_allow(pfv.writer, pfv.reader)) {
1142 add_future_value(pfv.writer, pfv.reader);
1143 futurevalues->erase(futurevalues->begin() + i);
1147 mo_graph->commitChanges();
1148 mo_check_promises(curr, false);
1150 get_thread(curr)->set_return_value(VALUE_NONE);
1151 return updated_mod_order || updated_promises;
1155 * Process a fence ModelAction
1156 * @param curr The ModelAction to process
1157 * @return True if synchronization was updated
1159 bool ModelChecker::process_fence(ModelAction *curr)
1162 * fence-relaxed: no-op
1163 * fence-release: only log the occurence (not in this function), for
1164 * use in later synchronization
1165 * fence-acquire (this function): search for hypothetical release
1167 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
1169 bool updated = false;
1170 if (curr->is_acquire()) {
1171 action_list_t *list = action_trace;
1172 action_list_t::reverse_iterator rit;
1173 /* Find X : is_read(X) && X --sb-> curr */
1174 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1175 ModelAction *act = *rit;
1178 if (act->get_tid() != curr->get_tid())
1180 /* Stop at the beginning of the thread */
1181 if (act->is_thread_start())
1183 /* Stop once we reach a prior fence-acquire */
1184 if (act->is_fence() && act->is_acquire())
1186 if (!act->is_read())
1188 /* read-acquire will find its own release sequences */
1189 if (act->is_acquire())
1192 /* Establish hypothetical release sequences */
1193 rel_heads_list_t release_heads;
1194 get_release_seq_heads(curr, act, &release_heads);
1195 for (unsigned int i = 0; i < release_heads.size(); i++)
1196 if (!curr->synchronize_with(release_heads[i]))
1197 set_bad_synchronization();
1198 if (release_heads.size() != 0)
1206 * @brief Process the current action for thread-related activity
1208 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
1209 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
1210 * synchronization, etc. This function is a no-op for non-THREAD actions
1211 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
1213 * @param curr The current action
1214 * @return True if synchronization was updated or a thread completed
1216 bool ModelChecker::process_thread_action(ModelAction *curr)
1218 bool updated = false;
1220 switch (curr->get_type()) {
1221 case THREAD_CREATE: {
1222 thrd_t *thrd = (thrd_t *)curr->get_location();
1223 struct thread_params *params = (struct thread_params *)curr->get_value();
1224 Thread *th = new Thread(thrd, params->func, params->arg, get_thread(curr));
1226 th->set_creation(curr);
1227 /* Promises can be satisfied by children */
1228 for (unsigned int i = 0; i < promises->size(); i++) {
1229 Promise *promise = (*promises)[i];
1230 if (promise->thread_is_available(curr->get_tid()))
1231 promise->add_thread(th->get_id());
1236 Thread *blocking = curr->get_thread_operand();
1237 ModelAction *act = get_last_action(blocking->get_id());
1238 curr->synchronize_with(act);
1239 updated = true; /* trigger rel-seq checks */
1242 case THREAD_FINISH: {
1243 Thread *th = get_thread(curr);
1244 /* Wake up any joining threads */
1245 for (unsigned int i = 0; i < get_num_threads(); i++) {
1246 Thread *waiting = get_thread(int_to_id(i));
1247 if (waiting->waiting_on() == th &&
1248 waiting->get_pending()->is_thread_join())
1249 scheduler->wake(waiting);
1252 /* Completed thread can't satisfy promises */
1253 for (unsigned int i = 0; i < promises->size(); i++) {
1254 Promise *promise = (*promises)[i];
1255 if (promise->thread_is_available(th->get_id()))
1256 if (promise->eliminate_thread(th->get_id()))
1257 priv->failed_promise = true;
1259 updated = true; /* trigger rel-seq checks */
1262 case THREAD_START: {
1263 check_promises(curr->get_tid(), NULL, curr->get_cv());
1274 * @brief Process the current action for release sequence fixup activity
1276 * Performs model-checker release sequence fixups for the current action,
1277 * forcing a single pending release sequence to break (with a given, potential
1278 * "loose" write) or to complete (i.e., synchronize). If a pending release
1279 * sequence forms a complete release sequence, then we must perform the fixup
1280 * synchronization, mo_graph additions, etc.
1282 * @param curr The current action; must be a release sequence fixup action
1283 * @param work_queue The work queue to which to add work items as they are
1286 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1288 const ModelAction *write = curr->get_node()->get_relseq_break();
1289 struct release_seq *sequence = pending_rel_seqs->back();
1290 pending_rel_seqs->pop_back();
1292 ModelAction *acquire = sequence->acquire;
1293 const ModelAction *rf = sequence->rf;
1294 const ModelAction *release = sequence->release;
1298 ASSERT(release->same_thread(rf));
1300 if (write == NULL) {
1302 * @todo Forcing a synchronization requires that we set
1303 * modification order constraints. For instance, we can't allow
1304 * a fixup sequence in which two separate read-acquire
1305 * operations read from the same sequence, where the first one
1306 * synchronizes and the other doesn't. Essentially, we can't
1307 * allow any writes to insert themselves between 'release' and
1311 /* Must synchronize */
1312 if (!acquire->synchronize_with(release)) {
1313 set_bad_synchronization();
1316 /* Re-check all pending release sequences */
1317 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1318 /* Re-check act for mo_graph edges */
1319 work_queue->push_back(MOEdgeWorkEntry(acquire));
1321 /* propagate synchronization to later actions */
1322 action_list_t::reverse_iterator rit = action_trace->rbegin();
1323 for (; (*rit) != acquire; rit++) {
1324 ModelAction *propagate = *rit;
1325 if (acquire->happens_before(propagate)) {
1326 propagate->synchronize_with(acquire);
1327 /* Re-check 'propagate' for mo_graph edges */
1328 work_queue->push_back(MOEdgeWorkEntry(propagate));
1332 /* Break release sequence with new edges:
1333 * release --mo--> write --mo--> rf */
1334 mo_graph->addEdge(release, write);
1335 mo_graph->addEdge(write, rf);
1338 /* See if we have realized a data race */
1343 * Initialize the current action by performing one or more of the following
1344 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1345 * in the NodeStack, manipulating backtracking sets, allocating and
1346 * initializing clock vectors, and computing the promises to fulfill.
1348 * @param curr The current action, as passed from the user context; may be
1349 * freed/invalidated after the execution of this function, with a different
1350 * action "returned" its place (pass-by-reference)
1351 * @return True if curr is a newly-explored action; false otherwise
1353 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1355 ModelAction *newcurr;
1357 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1358 newcurr = process_rmw(*curr);
1361 if (newcurr->is_rmw())
1362 compute_promises(newcurr);
1368 (*curr)->set_seq_number(get_next_seq_num());
1370 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1372 /* First restore type and order in case of RMW operation */
1373 if ((*curr)->is_rmwr())
1374 newcurr->copy_typeandorder(*curr);
1376 ASSERT((*curr)->get_location() == newcurr->get_location());
1377 newcurr->copy_from_new(*curr);
1379 /* Discard duplicate ModelAction; use action from NodeStack */
1382 /* Always compute new clock vector */
1383 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1386 return false; /* Action was explored previously */
1390 /* Always compute new clock vector */
1391 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1393 /* Assign most recent release fence */
1394 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1397 * Perform one-time actions when pushing new ModelAction onto
1400 if (newcurr->is_write())
1401 compute_promises(newcurr);
1402 else if (newcurr->is_relseq_fixup())
1403 compute_relseq_breakwrites(newcurr);
1404 else if (newcurr->is_wait())
1405 newcurr->get_node()->set_misc_max(2);
1406 else if (newcurr->is_notify_one()) {
1407 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1409 return true; /* This was a new ModelAction */
1414 * @brief Establish reads-from relation between two actions
1416 * Perform basic operations involved with establishing a concrete rf relation,
1417 * including setting the ModelAction data and checking for release sequences.
1419 * @param act The action that is reading (must be a read)
1420 * @param rf The action from which we are reading (must be a write)
1422 * @return True if this read established synchronization
1424 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1427 ASSERT(rf->is_write());
1429 act->set_read_from(rf);
1430 if (act->is_acquire()) {
1431 rel_heads_list_t release_heads;
1432 get_release_seq_heads(act, act, &release_heads);
1433 int num_heads = release_heads.size();
1434 for (unsigned int i = 0; i < release_heads.size(); i++)
1435 if (!act->synchronize_with(release_heads[i])) {
1436 set_bad_synchronization();
1439 return num_heads > 0;
1445 * Check promises and eliminate potentially-satisfying threads when a thread is
1446 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1447 * no longer satisfy a promise generated from that thread.
1449 * @param blocker The thread on which a thread is waiting
1450 * @param waiting The waiting thread
1452 void ModelChecker::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1454 for (unsigned int i = 0; i < promises->size(); i++) {
1455 Promise *promise = (*promises)[i];
1456 if (!promise->thread_is_available(waiting->get_id()))
1458 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1459 ModelAction *reader = promise->get_reader(j);
1460 if (reader->get_tid() != blocker->get_id())
1462 if (promise->eliminate_thread(waiting->get_id())) {
1463 /* Promise has failed */
1464 priv->failed_promise = true;
1466 /* Only eliminate the 'waiting' thread once */
1474 * @brief Check whether a model action is enabled.
1476 * Checks whether a lock or join operation would be successful (i.e., is the
1477 * lock already locked, or is the joined thread already complete). If not, put
1478 * the action in a waiter list.
1480 * @param curr is the ModelAction to check whether it is enabled.
1481 * @return a bool that indicates whether the action is enabled.
1483 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1484 if (curr->is_lock()) {
1485 std::mutex *lock = curr->get_mutex();
1486 struct std::mutex_state *state = lock->get_state();
1489 } else if (curr->is_thread_join()) {
1490 Thread *blocking = curr->get_thread_operand();
1491 if (!blocking->is_complete()) {
1492 thread_blocking_check_promises(blocking, get_thread(curr));
1501 * This is the heart of the model checker routine. It performs model-checking
1502 * actions corresponding to a given "current action." Among other processes, it
1503 * calculates reads-from relationships, updates synchronization clock vectors,
1504 * forms a memory_order constraints graph, and handles replay/backtrack
1505 * execution when running permutations of previously-observed executions.
1507 * @param curr The current action to process
1508 * @return The ModelAction that is actually executed; may be different than
1509 * curr; may be NULL, if the current action is not enabled to run
1511 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1514 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1515 bool newly_explored = initialize_curr_action(&curr);
1519 wake_up_sleeping_actions(curr);
1521 /* Compute fairness information for CHESS yield algorithm */
1522 if (model->params.yieldon) {
1523 curr->get_node()->update_yield(scheduler);
1526 /* Add the action to lists before any other model-checking tasks */
1527 if (!second_part_of_rmw)
1528 add_action_to_lists(curr);
1530 /* Build may_read_from set for newly-created actions */
1531 if (newly_explored && curr->is_read())
1532 build_may_read_from(curr);
1534 /* Initialize work_queue with the "current action" work */
1535 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1536 while (!work_queue.empty() && !has_asserted()) {
1537 WorkQueueEntry work = work_queue.front();
1538 work_queue.pop_front();
1540 switch (work.type) {
1541 case WORK_CHECK_CURR_ACTION: {
1542 ModelAction *act = work.action;
1543 bool update = false; /* update this location's release seq's */
1544 bool update_all = false; /* update all release seq's */
1546 if (process_thread_action(curr))
1549 if (act->is_read() && !second_part_of_rmw && process_read(act))
1552 if (act->is_write() && process_write(act))
1555 if (act->is_fence() && process_fence(act))
1558 if (act->is_mutex_op() && process_mutex(act))
1561 if (act->is_relseq_fixup())
1562 process_relseq_fixup(curr, &work_queue);
1565 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1567 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1570 case WORK_CHECK_RELEASE_SEQ:
1571 resolve_release_sequences(work.location, &work_queue);
1573 case WORK_CHECK_MO_EDGES: {
1574 /** @todo Complete verification of work_queue */
1575 ModelAction *act = work.action;
1576 bool updated = false;
1578 if (act->is_read()) {
1579 const ModelAction *rf = act->get_reads_from();
1580 const Promise *promise = act->get_reads_from_promise();
1582 if (r_modification_order(act, rf))
1584 } else if (promise) {
1585 if (r_modification_order(act, promise))
1589 if (act->is_write()) {
1590 if (w_modification_order(act, NULL))
1593 mo_graph->commitChanges();
1596 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1605 check_curr_backtracking(curr);
1606 set_backtracking(curr);
1610 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1612 Node *currnode = curr->get_node();
1613 Node *parnode = currnode->get_parent();
1615 if ((parnode && !parnode->backtrack_empty()) ||
1616 !currnode->misc_empty() ||
1617 !currnode->read_from_empty() ||
1618 !currnode->promise_empty() ||
1619 !currnode->relseq_break_empty()) {
1620 set_latest_backtrack(curr);
1624 bool ModelChecker::promises_expired() const
1626 for (unsigned int i = 0; i < promises->size(); i++) {
1627 Promise *promise = (*promises)[i];
1628 if (promise->get_expiration() < priv->used_sequence_numbers)
1635 * This is the strongest feasibility check available.
1636 * @return whether the current trace (partial or complete) must be a prefix of
1639 bool ModelChecker::isfeasibleprefix() const
1641 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1645 * Print disagnostic information about an infeasible execution
1646 * @param prefix A string to prefix the output with; if NULL, then a default
1647 * message prefix will be provided
1649 void ModelChecker::print_infeasibility(const char *prefix) const
1653 if (mo_graph->checkForCycles())
1654 ptr += sprintf(ptr, "[mo cycle]");
1655 if (priv->failed_promise)
1656 ptr += sprintf(ptr, "[failed promise]");
1657 if (priv->too_many_reads)
1658 ptr += sprintf(ptr, "[too many reads]");
1659 if (priv->no_valid_reads)
1660 ptr += sprintf(ptr, "[no valid reads-from]");
1661 if (priv->bad_synchronization)
1662 ptr += sprintf(ptr, "[bad sw ordering]");
1663 if (promises_expired())
1664 ptr += sprintf(ptr, "[promise expired]");
1665 if (promises->size() != 0)
1666 ptr += sprintf(ptr, "[unresolved promise]");
1668 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1672 * Returns whether the current completed trace is feasible, except for pending
1673 * release sequences.
1675 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1677 return !is_infeasible() && promises->size() == 0;
1681 * Check if the current partial trace is infeasible. Does not check any
1682 * end-of-execution flags, which might rule out the execution. Thus, this is
1683 * useful only for ruling an execution as infeasible.
1684 * @return whether the current partial trace is infeasible.
1686 bool ModelChecker::is_infeasible() const
1688 return mo_graph->checkForCycles() ||
1689 priv->no_valid_reads ||
1690 priv->failed_promise ||
1691 priv->too_many_reads ||
1692 priv->bad_synchronization ||
1696 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1697 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1698 ModelAction *lastread = get_last_action(act->get_tid());
1699 lastread->process_rmw(act);
1700 if (act->is_rmw()) {
1701 if (lastread->get_reads_from())
1702 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1704 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1705 mo_graph->commitChanges();
1711 * A helper function for ModelChecker::check_recency, to check if the current
1712 * thread is able to read from a different write/promise for 'params.maxreads'
1713 * number of steps and if that write/promise should become visible (i.e., is
1714 * ordered later in the modification order). This helps model memory liveness.
1716 * @param curr The current action. Must be a read.
1717 * @param rf The write/promise from which we plan to read
1718 * @param other_rf The write/promise from which we may read
1719 * @return True if we were able to read from other_rf for params.maxreads steps
1721 template <typename T, typename U>
1722 bool ModelChecker::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1724 /* Need a different write/promise */
1725 if (other_rf->equals(rf))
1728 /* Only look for "newer" writes/promises */
1729 if (!mo_graph->checkReachable(rf, other_rf))
1732 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1733 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1734 action_list_t::reverse_iterator rit = list->rbegin();
1735 ASSERT((*rit) == curr);
1736 /* Skip past curr */
1739 /* Does this write/promise work for everyone? */
1740 for (int i = 0; i < params.maxreads; i++, rit++) {
1741 ModelAction *act = *rit;
1742 if (!act->may_read_from(other_rf))
1749 * Checks whether a thread has read from the same write or Promise for too many
1750 * times without seeing the effects of a later write/Promise.
1753 * 1) there must a different write/promise that we could read from,
1754 * 2) we must have read from the same write/promise in excess of maxreads times,
1755 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1756 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1758 * If so, we decide that the execution is no longer feasible.
1760 * @param curr The current action. Must be a read.
1761 * @param rf The ModelAction/Promise from which we might read.
1762 * @return True if the read should succeed; false otherwise
1764 template <typename T>
1765 bool ModelChecker::check_recency(ModelAction *curr, const T *rf) const
1767 if (!params.maxreads)
1770 //NOTE: Next check is just optimization, not really necessary....
1771 if (curr->get_node()->get_read_from_past_size() +
1772 curr->get_node()->get_read_from_promise_size() <= 1)
1775 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1776 int tid = id_to_int(curr->get_tid());
1777 ASSERT(tid < (int)thrd_lists->size());
1778 action_list_t *list = &(*thrd_lists)[tid];
1779 action_list_t::reverse_iterator rit = list->rbegin();
1780 ASSERT((*rit) == curr);
1781 /* Skip past curr */
1784 action_list_t::reverse_iterator ritcopy = rit;
1785 /* See if we have enough reads from the same value */
1786 for (int count = 0; count < params.maxreads; ritcopy++, count++) {
1787 if (ritcopy == list->rend())
1789 ModelAction *act = *ritcopy;
1790 if (!act->is_read())
1792 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1794 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1796 if (act->get_node()->get_read_from_past_size() +
1797 act->get_node()->get_read_from_promise_size() <= 1)
1800 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1801 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1802 if (should_read_instead(curr, rf, write))
1803 return false; /* liveness failure */
1805 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1806 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1807 if (should_read_instead(curr, rf, promise))
1808 return false; /* liveness failure */
1814 * @brief Updates the mo_graph with the constraints imposed from the current
1817 * Basic idea is the following: Go through each other thread and find
1818 * the last action that happened before our read. Two cases:
1820 * -# The action is a write: that write must either occur before
1821 * the write we read from or be the write we read from.
1822 * -# The action is a read: the write that that action read from
1823 * must occur before the write we read from or be the same write.
1825 * @param curr The current action. Must be a read.
1826 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1827 * @return True if modification order edges were added; false otherwise
1829 template <typename rf_type>
1830 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1832 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1835 ASSERT(curr->is_read());
1837 /* Last SC fence in the current thread */
1838 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1839 ModelAction *last_sc_write = NULL;
1840 if (curr->is_seqcst())
1841 last_sc_write = get_last_seq_cst_write(curr);
1843 /* Iterate over all threads */
1844 for (i = 0; i < thrd_lists->size(); i++) {
1845 /* Last SC fence in thread i */
1846 ModelAction *last_sc_fence_thread_local = NULL;
1847 if (int_to_id((int)i) != curr->get_tid())
1848 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1850 /* Last SC fence in thread i, before last SC fence in current thread */
1851 ModelAction *last_sc_fence_thread_before = NULL;
1852 if (last_sc_fence_local)
1853 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1855 /* Iterate over actions in thread, starting from most recent */
1856 action_list_t *list = &(*thrd_lists)[i];
1857 action_list_t::reverse_iterator rit;
1858 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1859 ModelAction *act = *rit;
1864 /* Don't want to add reflexive edges on 'rf' */
1865 if (act->equals(rf)) {
1866 if (act->happens_before(curr))
1872 if (act->is_write()) {
1873 /* C++, Section 29.3 statement 5 */
1874 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1875 *act < *last_sc_fence_thread_local) {
1876 added = mo_graph->addEdge(act, rf) || added;
1879 /* C++, Section 29.3 statement 4 */
1880 else if (act->is_seqcst() && last_sc_fence_local &&
1881 *act < *last_sc_fence_local) {
1882 added = mo_graph->addEdge(act, rf) || added;
1885 /* C++, Section 29.3 statement 6 */
1886 else if (last_sc_fence_thread_before &&
1887 *act < *last_sc_fence_thread_before) {
1888 added = mo_graph->addEdge(act, rf) || added;
1893 /* C++, Section 29.3 statement 3 (second subpoint) */
1894 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1895 added = mo_graph->addEdge(act, rf) || added;
1900 * Include at most one act per-thread that "happens
1903 if (act->happens_before(curr)) {
1904 if (act->is_write()) {
1905 added = mo_graph->addEdge(act, rf) || added;
1907 const ModelAction *prevrf = act->get_reads_from();
1908 const Promise *prevrf_promise = act->get_reads_from_promise();
1910 if (!prevrf->equals(rf))
1911 added = mo_graph->addEdge(prevrf, rf) || added;
1912 } else if (!prevrf_promise->equals(rf)) {
1913 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1922 * All compatible, thread-exclusive promises must be ordered after any
1923 * concrete loads from the same thread
1925 for (unsigned int i = 0; i < promises->size(); i++)
1926 if ((*promises)[i]->is_compatible_exclusive(curr))
1927 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1933 * Updates the mo_graph with the constraints imposed from the current write.
1935 * Basic idea is the following: Go through each other thread and find
1936 * the lastest action that happened before our write. Two cases:
1938 * (1) The action is a write => that write must occur before
1941 * (2) The action is a read => the write that that action read from
1942 * must occur before the current write.
1944 * This method also handles two other issues:
1946 * (I) Sequential Consistency: Making sure that if the current write is
1947 * seq_cst, that it occurs after the previous seq_cst write.
1949 * (II) Sending the write back to non-synchronizing reads.
1951 * @param curr The current action. Must be a write.
1952 * @param send_fv A vector for stashing reads to which we may pass our future
1953 * value. If NULL, then don't record any future values.
1954 * @return True if modification order edges were added; false otherwise
1956 bool ModelChecker::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1958 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1961 ASSERT(curr->is_write());
1963 if (curr->is_seqcst()) {
1964 /* We have to at least see the last sequentially consistent write,
1965 so we are initialized. */
1966 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1967 if (last_seq_cst != NULL) {
1968 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1972 /* Last SC fence in the current thread */
1973 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1975 /* Iterate over all threads */
1976 for (i = 0; i < thrd_lists->size(); i++) {
1977 /* Last SC fence in thread i, before last SC fence in current thread */
1978 ModelAction *last_sc_fence_thread_before = NULL;
1979 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1980 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1982 /* Iterate over actions in thread, starting from most recent */
1983 action_list_t *list = &(*thrd_lists)[i];
1984 action_list_t::reverse_iterator rit;
1985 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1986 ModelAction *act = *rit;
1989 * 1) If RMW and it actually read from something, then we
1990 * already have all relevant edges, so just skip to next
1993 * 2) If RMW and it didn't read from anything, we should
1994 * whatever edge we can get to speed up convergence.
1996 * 3) If normal write, we need to look at earlier actions, so
1997 * continue processing list.
1999 if (curr->is_rmw()) {
2000 if (curr->get_reads_from() != NULL)
2008 /* C++, Section 29.3 statement 7 */
2009 if (last_sc_fence_thread_before && act->is_write() &&
2010 *act < *last_sc_fence_thread_before) {
2011 added = mo_graph->addEdge(act, curr) || added;
2016 * Include at most one act per-thread that "happens
2019 if (act->happens_before(curr)) {
2021 * Note: if act is RMW, just add edge:
2023 * The following edge should be handled elsewhere:
2024 * readfrom(act) --mo--> act
2026 if (act->is_write())
2027 added = mo_graph->addEdge(act, curr) || added;
2028 else if (act->is_read()) {
2029 //if previous read accessed a null, just keep going
2030 if (act->get_reads_from() == NULL)
2032 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
2035 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
2036 !act->same_thread(curr)) {
2037 /* We have an action that:
2038 (1) did not happen before us
2039 (2) is a read and we are a write
2040 (3) cannot synchronize with us
2041 (4) is in a different thread
2043 that read could potentially read from our write. Note that
2044 these checks are overly conservative at this point, we'll
2045 do more checks before actually removing the
2049 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
2050 if (!is_infeasible())
2051 send_fv->push_back(act);
2052 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
2053 add_future_value(curr, act);
2060 * All compatible, thread-exclusive promises must be ordered after any
2061 * concrete stores to the same thread, or else they can be merged with
2064 for (unsigned int i = 0; i < promises->size(); i++)
2065 if ((*promises)[i]->is_compatible_exclusive(curr))
2066 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
2071 /** Arbitrary reads from the future are not allowed. Section 29.3
2072 * part 9 places some constraints. This method checks one result of constraint
2073 * constraint. Others require compiler support. */
2074 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
2076 if (!writer->is_rmw())
2079 if (!reader->is_rmw())
2082 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
2083 if (search == reader)
2085 if (search->get_tid() == reader->get_tid() &&
2086 search->happens_before(reader))
2094 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
2095 * some constraints. This method checks one the following constraint (others
2096 * require compiler support):
2098 * If X --hb-> Y --mo-> Z, then X should not read from Z.
2100 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
2102 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
2104 /* Iterate over all threads */
2105 for (i = 0; i < thrd_lists->size(); i++) {
2106 const ModelAction *write_after_read = NULL;
2108 /* Iterate over actions in thread, starting from most recent */
2109 action_list_t *list = &(*thrd_lists)[i];
2110 action_list_t::reverse_iterator rit;
2111 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2112 ModelAction *act = *rit;
2114 /* Don't disallow due to act == reader */
2115 if (!reader->happens_before(act) || reader == act)
2117 else if (act->is_write())
2118 write_after_read = act;
2119 else if (act->is_read() && act->get_reads_from() != NULL)
2120 write_after_read = act->get_reads_from();
2123 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
2130 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
2131 * The ModelAction under consideration is expected to be taking part in
2132 * release/acquire synchronization as an object of the "reads from" relation.
2133 * Note that this can only provide release sequence support for RMW chains
2134 * which do not read from the future, as those actions cannot be traced until
2135 * their "promise" is fulfilled. Similarly, we may not even establish the
2136 * presence of a release sequence with certainty, as some modification order
2137 * constraints may be decided further in the future. Thus, this function
2138 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
2139 * and a boolean representing certainty.
2141 * @param rf The action that might be part of a release sequence. Must be a
2143 * @param release_heads A pass-by-reference style return parameter. After
2144 * execution of this function, release_heads will contain the heads of all the
2145 * relevant release sequences, if any exists with certainty
2146 * @param pending A pass-by-reference style return parameter which is only used
2147 * when returning false (i.e., uncertain). Returns most information regarding
2148 * an uncertain release sequence, including any write operations that might
2149 * break the sequence.
2150 * @return true, if the ModelChecker is certain that release_heads is complete;
2153 bool ModelChecker::release_seq_heads(const ModelAction *rf,
2154 rel_heads_list_t *release_heads,
2155 struct release_seq *pending) const
2157 /* Only check for release sequences if there are no cycles */
2158 if (mo_graph->checkForCycles())
2161 for ( ; rf != NULL; rf = rf->get_reads_from()) {
2162 ASSERT(rf->is_write());
2164 if (rf->is_release())
2165 release_heads->push_back(rf);
2166 else if (rf->get_last_fence_release())
2167 release_heads->push_back(rf->get_last_fence_release());
2169 break; /* End of RMW chain */
2171 /** @todo Need to be smarter here... In the linux lock
2172 * example, this will run to the beginning of the program for
2174 /** @todo The way to be smarter here is to keep going until 1
2175 * thread has a release preceded by an acquire and you've seen
2178 /* acq_rel RMW is a sufficient stopping condition */
2179 if (rf->is_acquire() && rf->is_release())
2180 return true; /* complete */
2183 /* read from future: need to settle this later */
2185 return false; /* incomplete */
2188 if (rf->is_release())
2189 return true; /* complete */
2191 /* else relaxed write
2192 * - check for fence-release in the same thread (29.8, stmt. 3)
2193 * - check modification order for contiguous subsequence
2194 * -> rf must be same thread as release */
2196 const ModelAction *fence_release = rf->get_last_fence_release();
2197 /* Synchronize with a fence-release unconditionally; we don't need to
2198 * find any more "contiguous subsequence..." for it */
2200 release_heads->push_back(fence_release);
2202 int tid = id_to_int(rf->get_tid());
2203 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
2204 action_list_t *list = &(*thrd_lists)[tid];
2205 action_list_t::const_reverse_iterator rit;
2207 /* Find rf in the thread list */
2208 rit = std::find(list->rbegin(), list->rend(), rf);
2209 ASSERT(rit != list->rend());
2211 /* Find the last {write,fence}-release */
2212 for (; rit != list->rend(); rit++) {
2213 if (fence_release && *(*rit) < *fence_release)
2215 if ((*rit)->is_release())
2218 if (rit == list->rend()) {
2219 /* No write-release in this thread */
2220 return true; /* complete */
2221 } else if (fence_release && *(*rit) < *fence_release) {
2222 /* The fence-release is more recent (and so, "stronger") than
2223 * the most recent write-release */
2224 return true; /* complete */
2225 } /* else, need to establish contiguous release sequence */
2226 ModelAction *release = *rit;
2228 ASSERT(rf->same_thread(release));
2230 pending->writes.clear();
2232 bool certain = true;
2233 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2234 if (id_to_int(rf->get_tid()) == (int)i)
2236 list = &(*thrd_lists)[i];
2238 /* Can we ensure no future writes from this thread may break
2239 * the release seq? */
2240 bool future_ordered = false;
2242 ModelAction *last = get_last_action(int_to_id(i));
2243 Thread *th = get_thread(int_to_id(i));
2244 if ((last && rf->happens_before(last)) ||
2247 future_ordered = true;
2249 ASSERT(!th->is_model_thread() || future_ordered);
2251 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2252 const ModelAction *act = *rit;
2253 /* Reach synchronization -> this thread is complete */
2254 if (act->happens_before(release))
2256 if (rf->happens_before(act)) {
2257 future_ordered = true;
2261 /* Only non-RMW writes can break release sequences */
2262 if (!act->is_write() || act->is_rmw())
2265 /* Check modification order */
2266 if (mo_graph->checkReachable(rf, act)) {
2267 /* rf --mo--> act */
2268 future_ordered = true;
2271 if (mo_graph->checkReachable(act, release))
2272 /* act --mo--> release */
2274 if (mo_graph->checkReachable(release, act) &&
2275 mo_graph->checkReachable(act, rf)) {
2276 /* release --mo-> act --mo--> rf */
2277 return true; /* complete */
2279 /* act may break release sequence */
2280 pending->writes.push_back(act);
2283 if (!future_ordered)
2284 certain = false; /* This thread is uncertain */
2288 release_heads->push_back(release);
2289 pending->writes.clear();
2291 pending->release = release;
2298 * An interface for getting the release sequence head(s) with which a
2299 * given ModelAction must synchronize. This function only returns a non-empty
2300 * result when it can locate a release sequence head with certainty. Otherwise,
2301 * it may mark the internal state of the ModelChecker so that it will handle
2302 * the release sequence at a later time, causing @a acquire to update its
2303 * synchronization at some later point in execution.
2305 * @param acquire The 'acquire' action that may synchronize with a release
2307 * @param read The read action that may read from a release sequence; this may
2308 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2309 * when 'acquire' is a fence-acquire)
2310 * @param release_heads A pass-by-reference return parameter. Will be filled
2311 * with the head(s) of the release sequence(s), if they exists with certainty.
2312 * @see ModelChecker::release_seq_heads
2314 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2315 ModelAction *read, rel_heads_list_t *release_heads)
2317 const ModelAction *rf = read->get_reads_from();
2318 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2319 sequence->acquire = acquire;
2320 sequence->read = read;
2322 if (!release_seq_heads(rf, release_heads, sequence)) {
2323 /* add act to 'lazy checking' list */
2324 pending_rel_seqs->push_back(sequence);
2326 snapshot_free(sequence);
2331 * Attempt to resolve all stashed operations that might synchronize with a
2332 * release sequence for a given location. This implements the "lazy" portion of
2333 * determining whether or not a release sequence was contiguous, since not all
2334 * modification order information is present at the time an action occurs.
2336 * @param location The location/object that should be checked for release
2337 * sequence resolutions. A NULL value means to check all locations.
2338 * @param work_queue The work queue to which to add work items as they are
2340 * @return True if any updates occurred (new synchronization, new mo_graph
2343 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2345 bool updated = false;
2346 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs->begin();
2347 while (it != pending_rel_seqs->end()) {
2348 struct release_seq *pending = *it;
2349 ModelAction *acquire = pending->acquire;
2350 const ModelAction *read = pending->read;
2352 /* Only resolve sequences on the given location, if provided */
2353 if (location && read->get_location() != location) {
2358 const ModelAction *rf = read->get_reads_from();
2359 rel_heads_list_t release_heads;
2361 complete = release_seq_heads(rf, &release_heads, pending);
2362 for (unsigned int i = 0; i < release_heads.size(); i++) {
2363 if (!acquire->has_synchronized_with(release_heads[i])) {
2364 if (acquire->synchronize_with(release_heads[i]))
2367 set_bad_synchronization();
2372 /* Re-check all pending release sequences */
2373 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2374 /* Re-check read-acquire for mo_graph edges */
2375 if (acquire->is_read())
2376 work_queue->push_back(MOEdgeWorkEntry(acquire));
2378 /* propagate synchronization to later actions */
2379 action_list_t::reverse_iterator rit = action_trace->rbegin();
2380 for (; (*rit) != acquire; rit++) {
2381 ModelAction *propagate = *rit;
2382 if (acquire->happens_before(propagate)) {
2383 propagate->synchronize_with(acquire);
2384 /* Re-check 'propagate' for mo_graph edges */
2385 work_queue->push_back(MOEdgeWorkEntry(propagate));
2390 it = pending_rel_seqs->erase(it);
2391 snapshot_free(pending);
2397 // If we resolved promises or data races, see if we have realized a data race.
2404 * Performs various bookkeeping operations for the current ModelAction. For
2405 * instance, adds action to the per-object, per-thread action vector and to the
2406 * action trace list of all thread actions.
2408 * @param act is the ModelAction to add.
2410 void ModelChecker::add_action_to_lists(ModelAction *act)
2412 int tid = id_to_int(act->get_tid());
2413 ModelAction *uninit = NULL;
2415 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2416 if (list->empty() && act->is_atomic_var()) {
2417 uninit = get_uninitialized_action(act);
2418 uninit_id = id_to_int(uninit->get_tid());
2419 list->push_front(uninit);
2421 list->push_back(act);
2423 action_trace->push_back(act);
2425 action_trace->push_front(uninit);
2427 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2428 if (tid >= (int)vec->size())
2429 vec->resize(priv->next_thread_id);
2430 (*vec)[tid].push_back(act);
2432 (*vec)[uninit_id].push_front(uninit);
2434 if ((int)thrd_last_action->size() <= tid)
2435 thrd_last_action->resize(get_num_threads());
2436 (*thrd_last_action)[tid] = act;
2438 (*thrd_last_action)[uninit_id] = uninit;
2440 if (act->is_fence() && act->is_release()) {
2441 if ((int)thrd_last_fence_release->size() <= tid)
2442 thrd_last_fence_release->resize(get_num_threads());
2443 (*thrd_last_fence_release)[tid] = act;
2446 if (act->is_wait()) {
2447 void *mutex_loc = (void *) act->get_value();
2448 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2450 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2451 if (tid >= (int)vec->size())
2452 vec->resize(priv->next_thread_id);
2453 (*vec)[tid].push_back(act);
2458 * @brief Get the last action performed by a particular Thread
2459 * @param tid The thread ID of the Thread in question
2460 * @return The last action in the thread
2462 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2464 int threadid = id_to_int(tid);
2465 if (threadid < (int)thrd_last_action->size())
2466 return (*thrd_last_action)[id_to_int(tid)];
2472 * @brief Get the last fence release performed by a particular Thread
2473 * @param tid The thread ID of the Thread in question
2474 * @return The last fence release in the thread, if one exists; NULL otherwise
2476 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2478 int threadid = id_to_int(tid);
2479 if (threadid < (int)thrd_last_fence_release->size())
2480 return (*thrd_last_fence_release)[id_to_int(tid)];
2486 * Gets the last memory_order_seq_cst write (in the total global sequence)
2487 * performed on a particular object (i.e., memory location), not including the
2489 * @param curr The current ModelAction; also denotes the object location to
2491 * @return The last seq_cst write
2493 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2495 void *location = curr->get_location();
2496 action_list_t *list = get_safe_ptr_action(obj_map, location);
2497 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2498 action_list_t::reverse_iterator rit;
2499 for (rit = list->rbegin(); (*rit) != curr; rit++)
2501 rit++; /* Skip past curr */
2502 for ( ; rit != list->rend(); rit++)
2503 if ((*rit)->is_write() && (*rit)->is_seqcst())
2509 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2510 * performed in a particular thread, prior to a particular fence.
2511 * @param tid The ID of the thread to check
2512 * @param before_fence The fence from which to begin the search; if NULL, then
2513 * search for the most recent fence in the thread.
2514 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2516 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2518 /* All fences should have NULL location */
2519 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2520 action_list_t::reverse_iterator rit = list->rbegin();
2523 for (; rit != list->rend(); rit++)
2524 if (*rit == before_fence)
2527 ASSERT(*rit == before_fence);
2531 for (; rit != list->rend(); rit++)
2532 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2538 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2539 * location). This function identifies the mutex according to the current
2540 * action, which is presumed to perform on the same mutex.
2541 * @param curr The current ModelAction; also denotes the object location to
2543 * @return The last unlock operation
2545 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2547 void *location = curr->get_location();
2548 action_list_t *list = get_safe_ptr_action(obj_map, location);
2549 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2550 action_list_t::reverse_iterator rit;
2551 for (rit = list->rbegin(); rit != list->rend(); rit++)
2552 if ((*rit)->is_unlock() || (*rit)->is_wait())
2557 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2559 ModelAction *parent = get_last_action(tid);
2561 parent = get_thread(tid)->get_creation();
2566 * Returns the clock vector for a given thread.
2567 * @param tid The thread whose clock vector we want
2568 * @return Desired clock vector
2570 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2572 return get_parent_action(tid)->get_cv();
2576 * @brief Find the promise (if any) to resolve for the current action and
2577 * remove it from the pending promise vector
2578 * @param curr The current ModelAction. Should be a write.
2579 * @return The Promise to resolve, if any; otherwise NULL
2581 Promise * ModelChecker::pop_promise_to_resolve(const ModelAction *curr)
2583 for (unsigned int i = 0; i < promises->size(); i++)
2584 if (curr->get_node()->get_promise(i)) {
2585 Promise *ret = (*promises)[i];
2586 promises->erase(promises->begin() + i);
2593 * Resolve a Promise with a current write.
2594 * @param write The ModelAction that is fulfilling Promises
2595 * @param promise The Promise to resolve
2596 * @return True if the Promise was successfully resolved; false otherwise
2598 bool ModelChecker::resolve_promise(ModelAction *write, Promise *promise)
2600 ModelVector<ModelAction *> actions_to_check;
2602 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2603 ModelAction *read = promise->get_reader(i);
2604 read_from(read, write);
2605 actions_to_check.push_back(read);
2607 /* Make sure the promise's value matches the write's value */
2608 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2609 if (!mo_graph->resolvePromise(promise, write))
2610 priv->failed_promise = true;
2613 * @todo It is possible to end up in an inconsistent state, where a
2614 * "resolved" promise may still be referenced if
2615 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2617 * Note that the inconsistency only matters when dumping mo_graph to
2623 //Check whether reading these writes has made threads unable to
2625 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2626 ModelAction *read = actions_to_check[i];
2627 mo_check_promises(read, true);
2634 * Compute the set of promises that could potentially be satisfied by this
2635 * action. Note that the set computation actually appears in the Node, not in
2637 * @param curr The ModelAction that may satisfy promises
2639 void ModelChecker::compute_promises(ModelAction *curr)
2641 for (unsigned int i = 0; i < promises->size(); i++) {
2642 Promise *promise = (*promises)[i];
2643 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2646 bool satisfy = true;
2647 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2648 const ModelAction *act = promise->get_reader(j);
2649 if (act->happens_before(curr) ||
2650 act->could_synchronize_with(curr)) {
2656 curr->get_node()->set_promise(i);
2660 /** Checks promises in response to change in ClockVector Threads. */
2661 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2663 for (unsigned int i = 0; i < promises->size(); i++) {
2664 Promise *promise = (*promises)[i];
2665 if (!promise->thread_is_available(tid))
2667 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2668 const ModelAction *act = promise->get_reader(j);
2669 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2670 merge_cv->synchronized_since(act)) {
2671 if (promise->eliminate_thread(tid)) {
2672 /* Promise has failed */
2673 priv->failed_promise = true;
2681 void ModelChecker::check_promises_thread_disabled()
2683 for (unsigned int i = 0; i < promises->size(); i++) {
2684 Promise *promise = (*promises)[i];
2685 if (promise->has_failed()) {
2686 priv->failed_promise = true;
2693 * @brief Checks promises in response to addition to modification order for
2696 * We test whether threads are still available for satisfying promises after an
2697 * addition to our modification order constraints. Those that are unavailable
2698 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2699 * that promise has failed.
2701 * @param act The ModelAction which updated the modification order
2702 * @param is_read_check Should be true if act is a read and we must check for
2703 * updates to the store from which it read (there is a distinction here for
2704 * RMW's, which are both a load and a store)
2706 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2708 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2710 for (unsigned int i = 0; i < promises->size(); i++) {
2711 Promise *promise = (*promises)[i];
2713 // Is this promise on the same location?
2714 if (!promise->same_location(write))
2717 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2718 const ModelAction *pread = promise->get_reader(j);
2719 if (!pread->happens_before(act))
2721 if (mo_graph->checkPromise(write, promise)) {
2722 priv->failed_promise = true;
2728 // Don't do any lookups twice for the same thread
2729 if (!promise->thread_is_available(act->get_tid()))
2732 if (mo_graph->checkReachable(promise, write)) {
2733 if (mo_graph->checkPromise(write, promise)) {
2734 priv->failed_promise = true;
2742 * Compute the set of writes that may break the current pending release
2743 * sequence. This information is extracted from previou release sequence
2746 * @param curr The current ModelAction. Must be a release sequence fixup
2749 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2751 if (pending_rel_seqs->empty())
2754 struct release_seq *pending = pending_rel_seqs->back();
2755 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2756 const ModelAction *write = pending->writes[i];
2757 curr->get_node()->add_relseq_break(write);
2760 /* NULL means don't break the sequence; just synchronize */
2761 curr->get_node()->add_relseq_break(NULL);
2765 * Build up an initial set of all past writes that this 'read' action may read
2766 * from, as well as any previously-observed future values that must still be valid.
2768 * @param curr is the current ModelAction that we are exploring; it must be a
2771 void ModelChecker::build_may_read_from(ModelAction *curr)
2773 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2775 ASSERT(curr->is_read());
2777 ModelAction *last_sc_write = NULL;
2779 if (curr->is_seqcst())
2780 last_sc_write = get_last_seq_cst_write(curr);
2782 /* Iterate over all threads */
2783 for (i = 0; i < thrd_lists->size(); i++) {
2784 /* Iterate over actions in thread, starting from most recent */
2785 action_list_t *list = &(*thrd_lists)[i];
2786 action_list_t::reverse_iterator rit;
2787 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2788 ModelAction *act = *rit;
2790 /* Only consider 'write' actions */
2791 if (!act->is_write() || act == curr)
2794 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2795 bool allow_read = true;
2797 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2799 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2803 /* Only add feasible reads */
2804 mo_graph->startChanges();
2805 r_modification_order(curr, act);
2806 if (!is_infeasible())
2807 curr->get_node()->add_read_from_past(act);
2808 mo_graph->rollbackChanges();
2811 /* Include at most one act per-thread that "happens before" curr */
2812 if (act->happens_before(curr))
2817 /* Inherit existing, promised future values */
2818 for (i = 0; i < promises->size(); i++) {
2819 const Promise *promise = (*promises)[i];
2820 const ModelAction *promise_read = promise->get_reader(0);
2821 if (promise_read->same_var(curr)) {
2822 /* Only add feasible future-values */
2823 mo_graph->startChanges();
2824 r_modification_order(curr, promise);
2825 if (!is_infeasible())
2826 curr->get_node()->add_read_from_promise(promise_read);
2827 mo_graph->rollbackChanges();
2831 /* We may find no valid may-read-from only if the execution is doomed */
2832 if (!curr->get_node()->read_from_size()) {
2833 priv->no_valid_reads = true;
2837 if (DBG_ENABLED()) {
2838 model_print("Reached read action:\n");
2840 model_print("Printing read_from_past\n");
2841 curr->get_node()->print_read_from_past();
2842 model_print("End printing read_from_past\n");
2846 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2848 for ( ; write != NULL; write = write->get_reads_from()) {
2849 /* UNINIT actions don't have a Node, and they never sleep */
2850 if (write->is_uninitialized())
2852 Node *prevnode = write->get_node()->get_parent();
2854 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2855 if (write->is_release() && thread_sleep)
2857 if (!write->is_rmw())
2864 * @brief Get an action representing an uninitialized atomic
2866 * This function may create a new one or try to retrieve one from the NodeStack
2868 * @param curr The current action, which prompts the creation of an UNINIT action
2869 * @return A pointer to the UNINIT ModelAction
2871 ModelAction * ModelChecker::get_uninitialized_action(const ModelAction *curr) const
2873 Node *node = curr->get_node();
2874 ModelAction *act = node->get_uninit_action();
2876 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), model->params.uninitvalue, model_thread);
2877 node->set_uninit_action(act);
2879 act->create_cv(NULL);
2883 static void print_list(action_list_t *list)
2885 action_list_t::iterator it;
2887 model_print("---------------------------------------------------------------------\n");
2889 unsigned int hash = 0;
2891 for (it = list->begin(); it != list->end(); it++) {
2892 const ModelAction *act = *it;
2893 if (act->get_seq_number() > 0)
2895 hash = hash^(hash<<3)^((*it)->hash());
2897 model_print("HASH %u\n", hash);
2898 model_print("---------------------------------------------------------------------\n");
2901 #if SUPPORT_MOD_ORDER_DUMP
2902 void ModelChecker::dumpGraph(char *filename) const
2905 sprintf(buffer, "%s.dot", filename);
2906 FILE *file = fopen(buffer, "w");
2907 fprintf(file, "digraph %s {\n", filename);
2908 mo_graph->dumpNodes(file);
2909 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2911 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2912 ModelAction *act = *it;
2913 if (act->is_read()) {
2914 mo_graph->dot_print_node(file, act);
2915 if (act->get_reads_from())
2916 mo_graph->dot_print_edge(file,
2917 act->get_reads_from(),
2919 "label=\"rf\", color=red, weight=2");
2921 mo_graph->dot_print_edge(file,
2922 act->get_reads_from_promise(),
2924 "label=\"rf\", color=red");
2926 if (thread_array[act->get_tid()]) {
2927 mo_graph->dot_print_edge(file,
2928 thread_array[id_to_int(act->get_tid())],
2930 "label=\"sb\", color=blue, weight=400");
2933 thread_array[act->get_tid()] = act;
2935 fprintf(file, "}\n");
2936 model_free(thread_array);
2941 /** @brief Prints an execution trace summary. */
2942 void ModelChecker::print_summary() const
2944 #if SUPPORT_MOD_ORDER_DUMP
2945 char buffername[100];
2946 sprintf(buffername, "exec%04u", stats.num_total);
2947 mo_graph->dumpGraphToFile(buffername);
2948 sprintf(buffername, "graph%04u", stats.num_total);
2949 dumpGraph(buffername);
2952 model_print("Execution %d:", stats.num_total);
2953 if (isfeasibleprefix()) {
2954 if (scheduler->all_threads_sleeping())
2955 model_print(" SLEEP-SET REDUNDANT");
2958 print_infeasibility(" INFEASIBLE");
2959 print_list(action_trace);
2961 if (!promises->empty()) {
2962 model_print("Pending promises:\n");
2963 for (unsigned int i = 0; i < promises->size(); i++) {
2964 model_print(" [P%u] ", i);
2965 (*promises)[i]->print();
2972 * Add a Thread to the system for the first time. Should only be called once
2974 * @param t The Thread to add
2976 void ModelChecker::add_thread(Thread *t)
2978 thread_map->put(id_to_int(t->get_id()), t);
2979 scheduler->add_thread(t);
2983 * @brief Get a Thread reference by its ID
2984 * @param tid The Thread's ID
2985 * @return A Thread reference
2987 Thread * ModelChecker::get_thread(thread_id_t tid) const
2989 return thread_map->get(id_to_int(tid));
2993 * @brief Get a reference to the Thread in which a ModelAction was executed
2994 * @param act The ModelAction
2995 * @return A Thread reference
2997 Thread * ModelChecker::get_thread(const ModelAction *act) const
2999 return get_thread(act->get_tid());
3003 * @brief Get a Promise's "promise number"
3005 * A "promise number" is an index number that is unique to a promise, valid
3006 * only for a specific snapshot of an execution trace. Promises may come and go
3007 * as they are generated an resolved, so an index only retains meaning for the
3010 * @param promise The Promise to check
3011 * @return The promise index, if the promise still is valid; otherwise -1
3013 int ModelChecker::get_promise_number(const Promise *promise) const
3015 for (unsigned int i = 0; i < promises->size(); i++)
3016 if ((*promises)[i] == promise)
3023 * @brief Check if a Thread is currently enabled
3024 * @param t The Thread to check
3025 * @return True if the Thread is currently enabled
3027 bool ModelChecker::is_enabled(Thread *t) const
3029 return scheduler->is_enabled(t);
3033 * @brief Check if a Thread is currently enabled
3034 * @param tid The ID of the Thread to check
3035 * @return True if the Thread is currently enabled
3037 bool ModelChecker::is_enabled(thread_id_t tid) const
3039 return scheduler->is_enabled(tid);
3043 * Switch from a model-checker context to a user-thread context. This is the
3044 * complement of ModelChecker::switch_to_master and must be called from the
3045 * model-checker context
3047 * @param thread The user-thread to switch to
3049 void ModelChecker::switch_from_master(Thread *thread)
3051 scheduler->set_current_thread(thread);
3052 Thread::swap(&system_context, thread);
3056 * Switch from a user-context to the "master thread" context (a.k.a. system
3057 * context). This switch is made with the intention of exploring a particular
3058 * model-checking action (described by a ModelAction object). Must be called
3059 * from a user-thread context.
3061 * @param act The current action that will be explored. May be NULL only if
3062 * trace is exiting via an assertion (see ModelChecker::set_assert and
3063 * ModelChecker::has_asserted).
3064 * @return Return the value returned by the current action
3066 uint64_t ModelChecker::switch_to_master(ModelAction *act)
3069 Thread *old = thread_current();
3070 scheduler->set_current_thread(NULL);
3071 ASSERT(!old->get_pending());
3072 old->set_pending(act);
3073 if (Thread::swap(old, &system_context) < 0) {
3074 perror("swap threads");
3077 return old->get_return_value();
3081 * Takes the next step in the execution, if possible.
3082 * @param curr The current step to take
3083 * @return Returns the next Thread to run, if any; NULL if this execution
3086 Thread * ModelChecker::take_step(ModelAction *curr)
3088 Thread *curr_thrd = get_thread(curr);
3089 ASSERT(curr_thrd->get_state() == THREAD_READY);
3091 ASSERT(check_action_enabled(curr)); /* May have side effects? */
3092 curr = check_current_action(curr);
3095 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
3096 scheduler->remove_thread(curr_thrd);
3098 return action_select_next_thread(curr);
3101 /** Wrapper to run the user's main function, with appropriate arguments */
3102 void user_main_wrapper(void *)
3104 user_main(model->params.argc, model->params.argv);
3107 bool ModelChecker::should_terminate_execution()
3109 /* Infeasible -> don't take any more steps */
3110 if (is_infeasible())
3112 else if (isfeasibleprefix() && have_bug_reports()) {
3117 if (params.bound != 0 && priv->used_sequence_numbers > params.bound)
3122 /** @brief Run ModelChecker for the user program */
3123 void ModelChecker::run()
3127 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL, NULL);
3132 * Stash next pending action(s) for thread(s). There
3133 * should only need to stash one thread's action--the
3134 * thread which just took a step--plus the first step
3135 * for any newly-created thread
3137 for (unsigned int i = 0; i < get_num_threads(); i++) {
3138 thread_id_t tid = int_to_id(i);
3139 Thread *thr = get_thread(tid);
3140 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
3141 switch_from_master(thr);
3142 if (thr->is_waiting_on(thr))
3143 assert_bug("Deadlock detected (thread %u)", i);
3147 /* Don't schedule threads which should be disabled */
3148 for (unsigned int i = 0; i < get_num_threads(); i++) {
3149 Thread *th = get_thread(int_to_id(i));
3150 ModelAction *act = th->get_pending();
3151 if (act && is_enabled(th) && !check_action_enabled(act)) {
3152 scheduler->sleep(th);
3156 /* Catch assertions from prior take_step or from
3157 * between-ModelAction bugs (e.g., data races) */
3162 t = get_next_thread();
3163 if (!t || t->is_model_thread())
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 (!should_terminate_execution());
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");