11 #include "snapshot-interface.h"
13 #include "clockvector.h"
14 #include "cyclegraph.h"
17 #include "threads-model.h"
19 #include "traceanalysis.h"
20 #include "bugmessage.h"
22 #define INITIAL_THREAD_ID 0
27 * Structure for holding small ModelChecker members that should be snapshotted
29 struct model_snapshot_members {
30 model_snapshot_members() :
31 /* First thread created will have id INITIAL_THREAD_ID */
32 next_thread_id(INITIAL_THREAD_ID),
33 used_sequence_numbers(0),
37 failed_promise(false),
38 too_many_reads(false),
39 no_valid_reads(false),
40 bad_synchronization(false),
44 ~model_snapshot_members() {
45 for (unsigned int i = 0; i < bugs.size(); i++)
50 unsigned int next_thread_id;
51 modelclock_t used_sequence_numbers;
52 ModelAction *next_backtrack;
53 SnapVector<bug_message *> bugs;
54 struct execution_stats stats;
58 /** @brief Incorrectly-ordered synchronization was made */
59 bool bad_synchronization;
65 /** @brief Constructor */
66 ModelChecker::ModelChecker(struct model_params params) :
67 /* Initialize default scheduler */
69 scheduler(new Scheduler()),
71 earliest_diverge(NULL),
72 action_trace(new action_list_t()),
73 thread_map(new HashTable<int, Thread *, int>()),
74 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
75 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
76 obj_thrd_map(new HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4 >()),
77 promises(new SnapVector<Promise *>()),
78 futurevalues(new SnapVector<struct PendingFutureValue>()),
79 pending_rel_seqs(new SnapVector<struct release_seq *>()),
80 thrd_last_action(new SnapVector<ModelAction *>(1)),
81 thrd_last_fence_release(new SnapVector<ModelAction *>()),
82 node_stack(new NodeStack()),
83 trace_analyses(new ModelVector<TraceAnalysis *>()),
84 priv(new struct model_snapshot_members()),
85 mo_graph(new CycleGraph())
87 /* Initialize a model-checker thread, for special ModelActions */
88 model_thread = new Thread(get_next_id());
89 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
92 /** @brief Destructor */
93 ModelChecker::~ModelChecker()
95 for (unsigned int i = 0; i < get_num_threads(); i++)
96 delete thread_map->get(i);
101 delete condvar_waiters_map;
104 for (unsigned int i = 0; i < promises->size(); i++)
105 delete (*promises)[i];
108 delete pending_rel_seqs;
110 delete thrd_last_action;
111 delete thrd_last_fence_release;
113 for (unsigned int i = 0; i < trace_analyses->size(); i++)
114 delete (*trace_analyses)[i];
115 delete trace_analyses;
121 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
123 action_list_t *tmp = hash->get(ptr);
125 tmp = new action_list_t();
131 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
133 SnapVector<action_list_t> *tmp = hash->get(ptr);
135 tmp = new SnapVector<action_list_t>();
141 action_list_t * ModelChecker::get_actions_on_obj(void * obj, thread_id_t tid) {
142 SnapVector<action_list_t> *wrv=obj_thrd_map->get(obj);
145 unsigned int thread=id_to_int(tid);
146 if (thread < wrv->size())
147 return &(*wrv)[thread];
154 * Restores user program to initial state and resets all model-checker data
157 void ModelChecker::reset_to_initial_state()
159 DEBUG("+++ Resetting to initial state +++\n");
160 node_stack->reset_execution();
163 * FIXME: if we utilize partial rollback, we will need to free only
164 * those pending actions which were NOT pending before the rollback
167 for (unsigned int i = 0; i < get_num_threads(); i++)
168 delete get_thread(int_to_id(i))->get_pending();
170 snapshot_backtrack_before(0);
173 /** @return a thread ID for a new Thread */
174 thread_id_t ModelChecker::get_next_id()
176 return priv->next_thread_id++;
179 /** @return the number of user threads created during this execution */
180 unsigned int ModelChecker::get_num_threads() const
182 return priv->next_thread_id;
186 * Must be called from user-thread context (e.g., through the global
187 * thread_current() interface)
189 * @return The currently executing Thread.
191 Thread * ModelChecker::get_current_thread() const
193 return scheduler->get_current_thread();
196 /** @return a sequence number for a new ModelAction */
197 modelclock_t ModelChecker::get_next_seq_num()
199 return ++priv->used_sequence_numbers;
202 Node * ModelChecker::get_curr_node() const
204 return node_stack->get_head();
208 * @brief Select the next thread to execute based on the curren action
210 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
211 * actions should be followed by the execution of their child thread. In either
212 * case, the current action should determine the next thread schedule.
214 * @param curr The current action
215 * @return The next thread to run, if the current action will determine this
216 * selection; otherwise NULL
218 Thread * ModelChecker::action_select_next_thread(const ModelAction *curr) const
220 /* Do not split atomic RMW */
222 return get_thread(curr);
223 /* Follow CREATE with the created thread */
224 if (curr->get_type() == THREAD_CREATE)
225 return curr->get_thread_operand();
230 * @brief Choose the next thread to execute.
232 * This function chooses the next thread that should execute. It can enforce
233 * execution replay/backtracking or, if the model-checker has no preference
234 * regarding the next thread (i.e., when exploring a new execution ordering),
235 * we defer to the scheduler.
237 * @return The next chosen thread to run, if any exist. Or else if the current
238 * execution should terminate, return NULL.
240 Thread * ModelChecker::get_next_thread()
245 * Have we completed exploring the preselected path? Then let the
249 return scheduler->select_next_thread();
251 /* Else, we are trying to replay an execution */
252 ModelAction *next = node_stack->get_next()->get_action();
254 if (next == diverge) {
255 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
256 earliest_diverge = diverge;
258 Node *nextnode = next->get_node();
259 Node *prevnode = nextnode->get_parent();
260 scheduler->update_sleep_set(prevnode);
262 /* Reached divergence point */
263 if (nextnode->increment_behaviors()) {
264 /* Execute the same thread with a new behavior */
265 tid = next->get_tid();
266 node_stack->pop_restofstack(2);
269 /* Make a different thread execute for next step */
270 scheduler->add_sleep(get_thread(next->get_tid()));
271 tid = prevnode->get_next_backtrack();
272 /* Make sure the backtracked thread isn't sleeping. */
273 node_stack->pop_restofstack(1);
274 if (diverge == earliest_diverge) {
275 earliest_diverge = prevnode->get_action();
278 /* Start the round robin scheduler from this thread id */
279 scheduler->set_scheduler_thread(tid);
280 /* The correct sleep set is in the parent node. */
283 DEBUG("*** Divergence point ***\n");
287 tid = next->get_tid();
289 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
290 ASSERT(tid != THREAD_ID_T_NONE);
291 return get_thread(id_to_int(tid));
295 * We need to know what the next actions of all threads in the sleep
296 * set will be. This method computes them and stores the actions at
297 * the corresponding thread object's pending action.
300 void ModelChecker::execute_sleep_set()
302 for (unsigned int i = 0; i < get_num_threads(); i++) {
303 thread_id_t tid = int_to_id(i);
304 Thread *thr = get_thread(tid);
305 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
306 thr->get_pending()->set_sleep_flag();
312 * @brief Should the current action wake up a given thread?
314 * @param curr The current action
315 * @param thread The thread that we might wake up
316 * @return True, if we should wake up the sleeping thread; false otherwise
318 bool ModelChecker::should_wake_up(const ModelAction *curr, const Thread *thread) const
320 const ModelAction *asleep = thread->get_pending();
321 /* Don't allow partial RMW to wake anyone up */
324 /* Synchronizing actions may have been backtracked */
325 if (asleep->could_synchronize_with(curr))
327 /* All acquire/release fences and fence-acquire/store-release */
328 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
330 /* Fence-release + store can awake load-acquire on the same location */
331 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
332 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
333 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
339 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
341 for (unsigned int i = 0; i < get_num_threads(); i++) {
342 Thread *thr = get_thread(int_to_id(i));
343 if (scheduler->is_sleep_set(thr)) {
344 if (should_wake_up(curr, thr))
345 /* Remove this thread from sleep set */
346 scheduler->remove_sleep(thr);
351 /** @brief Alert the model-checker that an incorrectly-ordered
352 * synchronization was made */
353 void ModelChecker::set_bad_synchronization()
355 priv->bad_synchronization = true;
359 * Check whether the current trace has triggered an assertion which should halt
362 * @return True, if the execution should be aborted; false otherwise
364 bool ModelChecker::has_asserted() const
366 return priv->asserted;
370 * Trigger a trace assertion which should cause this execution to be halted.
371 * This can be due to a detected bug or due to an infeasibility that should
374 void ModelChecker::set_assert()
376 priv->asserted = true;
380 * Check if we are in a deadlock. Should only be called at the end of an
381 * execution, although it should not give false positives in the middle of an
382 * execution (there should be some ENABLED thread).
384 * @return True if program is in a deadlock; false otherwise
386 bool ModelChecker::is_deadlocked() const
388 bool blocking_threads = false;
389 for (unsigned int i = 0; i < get_num_threads(); i++) {
390 thread_id_t tid = int_to_id(i);
393 Thread *t = get_thread(tid);
394 if (!t->is_model_thread() && t->get_pending())
395 blocking_threads = true;
397 return blocking_threads;
401 * Check if this is a complete execution. That is, have all thread completed
402 * execution (rather than exiting because sleep sets have forced a redundant
405 * @return True if the execution is complete.
407 bool ModelChecker::is_complete_execution() const
409 for (unsigned int i = 0; i < get_num_threads(); i++)
410 if (is_enabled(int_to_id(i)))
416 * @brief Assert a bug in the executing program.
418 * Use this function to assert any sort of bug in the user program. If the
419 * current trace is feasible (actually, a prefix of some feasible execution),
420 * then this execution will be aborted, printing the appropriate message. If
421 * the current trace is not yet feasible, the error message will be stashed and
422 * printed if the execution ever becomes feasible.
424 * @param msg Descriptive message for the bug (do not include newline char)
425 * @return True if bug is immediately-feasible
427 bool ModelChecker::assert_bug(const char *msg, ...)
433 vsnprintf(str, sizeof(str), msg, ap);
436 priv->bugs.push_back(new bug_message(str));
438 if (isfeasibleprefix()) {
446 * @brief Assert a bug in the executing program, asserted by a user thread
447 * @see ModelChecker::assert_bug
448 * @param msg Descriptive message for the bug (do not include newline char)
450 void ModelChecker::assert_user_bug(const char *msg)
452 /* If feasible bug, bail out now */
454 switch_to_master(NULL);
457 /** @return True, if any bugs have been reported for this execution */
458 bool ModelChecker::have_bug_reports() const
460 return priv->bugs.size() != 0;
463 /** @brief Print bug report listing for this execution (if any bugs exist) */
464 void ModelChecker::print_bugs() const
466 if (have_bug_reports()) {
467 model_print("Bug report: %zu bug%s detected\n",
469 priv->bugs.size() > 1 ? "s" : "");
470 for (unsigned int i = 0; i < priv->bugs.size(); i++)
471 priv->bugs[i]->print();
476 * @brief Record end-of-execution stats
478 * Must be run when exiting an execution. Records various stats.
479 * @see struct execution_stats
481 void ModelChecker::record_stats()
484 if (!isfeasibleprefix())
485 stats.num_infeasible++;
486 else if (have_bug_reports())
487 stats.num_buggy_executions++;
488 else if (is_complete_execution())
489 stats.num_complete++;
491 stats.num_redundant++;
494 * @todo We can violate this ASSERT() when fairness/sleep sets
495 * conflict to cause an execution to terminate, e.g. with:
496 * Scheduler: [0: disabled][1: disabled][2: sleep][3: current, enabled]
498 //ASSERT(scheduler->all_threads_sleeping());
502 /** @brief Print execution stats */
503 void ModelChecker::print_stats() const
505 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
506 model_print("Number of redundant executions: %d\n", stats.num_redundant);
507 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
508 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
509 model_print("Total executions: %d\n", stats.num_total);
510 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
514 * @brief End-of-exeuction print
515 * @param printbugs Should any existing bugs be printed?
517 void ModelChecker::print_execution(bool printbugs) const
519 print_program_output();
521 if (params.verbose) {
522 model_print("Earliest divergence point since last feasible execution:\n");
523 if (earliest_diverge)
524 earliest_diverge->print();
526 model_print("(Not set)\n");
532 /* Don't print invalid bugs */
541 * Queries the model-checker for more executions to explore and, if one
542 * exists, resets the model-checker state to execute a new execution.
544 * @return If there are more executions to explore, return true. Otherwise,
547 bool ModelChecker::next_execution()
550 /* Is this execution a feasible execution that's worth bug-checking? */
551 bool complete = isfeasibleprefix() && (is_complete_execution() ||
554 /* End-of-execution bug checks */
557 assert_bug("Deadlock detected");
560 run_trace_analyses();
566 if (params.verbose || (complete && have_bug_reports()))
567 print_execution(complete);
569 clear_program_output();
572 earliest_diverge = NULL;
574 if ((diverge = get_next_backtrack()) == NULL)
578 model_print("Next execution will diverge at:\n");
582 reset_to_initial_state();
586 /** @brief Run trace analyses on complete trace */
587 void ModelChecker::run_trace_analyses() {
588 for (unsigned int i = 0; i < trace_analyses->size(); i++)
589 (*trace_analyses)[i]->analyze(action_trace);
593 * @brief Find the last fence-related backtracking conflict for a ModelAction
595 * This function performs the search for the most recent conflicting action
596 * against which we should perform backtracking, as affected by fence
597 * operations. This includes pairs of potentially-synchronizing actions which
598 * occur due to fence-acquire or fence-release, and hence should be explored in
599 * the opposite execution order.
601 * @param act The current action
602 * @return The most recent action which conflicts with act due to fences
604 ModelAction * ModelChecker::get_last_fence_conflict(ModelAction *act) const
606 /* Only perform release/acquire fence backtracking for stores */
607 if (!act->is_write())
610 /* Find a fence-release (or, act is a release) */
611 ModelAction *last_release;
612 if (act->is_release())
615 last_release = get_last_fence_release(act->get_tid());
619 /* Skip past the release */
620 action_list_t *list = action_trace;
621 action_list_t::reverse_iterator rit;
622 for (rit = list->rbegin(); rit != list->rend(); rit++)
623 if (*rit == last_release)
625 ASSERT(rit != list->rend());
630 * load --sb-> fence-acquire */
631 ModelVector<ModelAction *> acquire_fences(get_num_threads(), NULL);
632 ModelVector<ModelAction *> prior_loads(get_num_threads(), NULL);
633 bool found_acquire_fences = false;
634 for ( ; rit != list->rend(); rit++) {
635 ModelAction *prev = *rit;
636 if (act->same_thread(prev))
639 int tid = id_to_int(prev->get_tid());
641 if (prev->is_read() && act->same_var(prev)) {
642 if (prev->is_acquire()) {
643 /* Found most recent load-acquire, don't need
644 * to search for more fences */
645 if (!found_acquire_fences)
648 prior_loads[tid] = prev;
651 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
652 found_acquire_fences = true;
653 acquire_fences[tid] = prev;
657 ModelAction *latest_backtrack = NULL;
658 for (unsigned int i = 0; i < acquire_fences.size(); i++)
659 if (acquire_fences[i] && prior_loads[i])
660 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
661 latest_backtrack = acquire_fences[i];
662 return latest_backtrack;
666 * @brief Find the last backtracking conflict for a ModelAction
668 * This function performs the search for the most recent conflicting action
669 * against which we should perform backtracking. This primary includes pairs of
670 * synchronizing actions which should be explored in the opposite execution
673 * @param act The current action
674 * @return The most recent action which conflicts with act
676 ModelAction * ModelChecker::get_last_conflict(ModelAction *act) const
678 switch (act->get_type()) {
679 /* case ATOMIC_FENCE: fences don't directly cause backtracking */
683 ModelAction *ret = NULL;
685 /* linear search: from most recent to oldest */
686 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
687 action_list_t::reverse_iterator rit;
688 for (rit = list->rbegin(); rit != list->rend(); rit++) {
689 ModelAction *prev = *rit;
690 if (prev->could_synchronize_with(act)) {
696 ModelAction *ret2 = get_last_fence_conflict(act);
706 case ATOMIC_TRYLOCK: {
707 /* linear search: from most recent to oldest */
708 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
709 action_list_t::reverse_iterator rit;
710 for (rit = list->rbegin(); rit != list->rend(); rit++) {
711 ModelAction *prev = *rit;
712 if (act->is_conflicting_lock(prev))
717 case ATOMIC_UNLOCK: {
718 /* linear search: from most recent to oldest */
719 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
720 action_list_t::reverse_iterator rit;
721 for (rit = list->rbegin(); rit != list->rend(); rit++) {
722 ModelAction *prev = *rit;
723 if (!act->same_thread(prev) && prev->is_failed_trylock())
729 /* linear search: from most recent to oldest */
730 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
731 action_list_t::reverse_iterator rit;
732 for (rit = list->rbegin(); rit != list->rend(); rit++) {
733 ModelAction *prev = *rit;
734 if (!act->same_thread(prev) && prev->is_failed_trylock())
736 if (!act->same_thread(prev) && prev->is_notify())
742 case ATOMIC_NOTIFY_ALL:
743 case ATOMIC_NOTIFY_ONE: {
744 /* linear search: from most recent to oldest */
745 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
746 action_list_t::reverse_iterator rit;
747 for (rit = list->rbegin(); rit != list->rend(); rit++) {
748 ModelAction *prev = *rit;
749 if (!act->same_thread(prev) && prev->is_wait())
760 /** This method finds backtracking points where we should try to
761 * reorder the parameter ModelAction against.
763 * @param the ModelAction to find backtracking points for.
765 void ModelChecker::set_backtracking(ModelAction *act)
767 Thread *t = get_thread(act);
768 ModelAction *prev = get_last_conflict(act);
772 Node *node = prev->get_node()->get_parent();
774 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
775 int low_tid, high_tid;
776 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
777 low_tid = id_to_int(act->get_tid());
778 high_tid = low_tid + 1;
781 high_tid = get_num_threads();
784 for (int i = low_tid; i < high_tid; i++) {
785 thread_id_t tid = int_to_id(i);
787 /* Make sure this thread can be enabled here. */
788 if (i >= node->get_num_threads())
791 /* See Dynamic Partial Order Reduction (addendum), POPL '05 */
792 /* Don't backtrack into a point where the thread is disabled or sleeping. */
793 if (node->enabled_status(tid) != THREAD_ENABLED)
796 /* Check if this has been explored already */
797 if (node->has_been_explored(tid))
800 /* See if fairness allows */
801 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
803 for (int t = 0; t < node->get_num_threads(); t++) {
804 thread_id_t tother = int_to_id(t);
805 if (node->is_enabled(tother) && node->has_priority(tother)) {
814 /* See if CHESS-like yield fairness allows */
815 if (model->params.yieldon) {
817 for (int t = 0; t < node->get_num_threads(); t++) {
818 thread_id_t tother = int_to_id(t);
819 if (node->is_enabled(tother) && node->has_priority_over(tid, tother)) {
828 /* Cache the latest backtracking point */
829 set_latest_backtrack(prev);
831 /* If this is a new backtracking point, mark the tree */
832 if (!node->set_backtrack(tid))
834 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
835 id_to_int(prev->get_tid()),
836 id_to_int(t->get_id()));
845 * @brief Cache the a backtracking point as the "most recent", if eligible
847 * Note that this does not prepare the NodeStack for this backtracking
848 * operation, it only caches the action on a per-execution basis
850 * @param act The operation at which we should explore a different next action
851 * (i.e., backtracking point)
852 * @return True, if this action is now the most recent backtracking point;
855 bool ModelChecker::set_latest_backtrack(ModelAction *act)
857 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
858 priv->next_backtrack = act;
865 * Returns last backtracking point. The model checker will explore a different
866 * path for this point in the next execution.
867 * @return The ModelAction at which the next execution should diverge.
869 ModelAction * ModelChecker::get_next_backtrack()
871 ModelAction *next = priv->next_backtrack;
872 priv->next_backtrack = NULL;
877 * Processes a read model action.
878 * @param curr is the read model action to process.
879 * @return True if processing this read updates the mo_graph.
881 bool ModelChecker::process_read(ModelAction *curr)
883 Node *node = curr->get_node();
885 bool updated = false;
886 switch (node->get_read_from_status()) {
887 case READ_FROM_PAST: {
888 const ModelAction *rf = node->get_read_from_past();
891 mo_graph->startChanges();
893 ASSERT(!is_infeasible());
894 if (!check_recency(curr, rf)) {
895 if (node->increment_read_from()) {
896 mo_graph->rollbackChanges();
899 priv->too_many_reads = true;
903 updated = r_modification_order(curr, rf);
905 mo_graph->commitChanges();
906 mo_check_promises(curr, true);
909 case READ_FROM_PROMISE: {
910 Promise *promise = curr->get_node()->get_read_from_promise();
911 if (promise->add_reader(curr))
912 priv->failed_promise = true;
913 curr->set_read_from_promise(promise);
914 mo_graph->startChanges();
915 if (!check_recency(curr, promise))
916 priv->too_many_reads = true;
917 updated = r_modification_order(curr, promise);
918 mo_graph->commitChanges();
921 case READ_FROM_FUTURE: {
922 /* Read from future value */
923 struct future_value fv = node->get_future_value();
924 Promise *promise = new Promise(curr, fv);
925 curr->set_read_from_promise(promise);
926 promises->push_back(promise);
927 mo_graph->startChanges();
928 updated = r_modification_order(curr, promise);
929 mo_graph->commitChanges();
935 get_thread(curr)->set_return_value(curr->get_return_value());
941 * Processes a lock, trylock, or unlock model action. @param curr is
942 * the read model action to process.
944 * The try lock operation checks whether the lock is taken. If not,
945 * it falls to the normal lock operation case. If so, it returns
948 * The lock operation has already been checked that it is enabled, so
949 * it just grabs the lock and synchronizes with the previous unlock.
951 * The unlock operation has to re-enable all of the threads that are
952 * waiting on the lock.
954 * @return True if synchronization was updated; false otherwise
956 bool ModelChecker::process_mutex(ModelAction *curr)
958 std::mutex *mutex = curr->get_mutex();
959 struct std::mutex_state *state = NULL;
962 state = mutex->get_state();
964 switch (curr->get_type()) {
965 case ATOMIC_TRYLOCK: {
966 bool success = !state->locked;
967 curr->set_try_lock(success);
969 get_thread(curr)->set_return_value(0);
972 get_thread(curr)->set_return_value(1);
974 //otherwise fall into the lock case
976 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
977 assert_bug("Lock access before initialization");
978 state->locked = get_thread(curr);
979 ModelAction *unlock = get_last_unlock(curr);
980 //synchronize with the previous unlock statement
981 if (unlock != NULL) {
982 synchronize(unlock, curr);
988 case ATOMIC_UNLOCK: {
989 /* wake up the other threads */
990 for (unsigned int i = 0; i < get_num_threads(); i++) {
991 Thread *t = get_thread(int_to_id(i));
992 Thread *curr_thrd = get_thread(curr);
993 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
997 /* unlock the lock - after checking who was waiting on it */
998 state->locked = NULL;
1000 if (!curr->is_wait())
1001 break; /* The rest is only for ATOMIC_WAIT */
1003 /* Should we go to sleep? (simulate spurious failures) */
1004 if (curr->get_node()->get_misc() == 0) {
1005 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
1007 scheduler->sleep(get_thread(curr));
1011 case ATOMIC_NOTIFY_ALL: {
1012 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1013 //activate all the waiting threads
1014 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
1015 scheduler->wake(get_thread(*rit));
1020 case ATOMIC_NOTIFY_ONE: {
1021 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1022 int wakeupthread = curr->get_node()->get_misc();
1023 action_list_t::iterator it = waiters->begin();
1024 advance(it, wakeupthread);
1025 scheduler->wake(get_thread(*it));
1037 * @brief Check if the current pending promises allow a future value to be sent
1039 * If one of the following is true:
1040 * (a) there are no pending promises
1041 * (b) the reader and writer do not cross any promises
1042 * Then, it is safe to pass a future value back now.
1044 * Otherwise, we must save the pending future value until (a) or (b) is true
1046 * @param writer The operation which sends the future value. Must be a write.
1047 * @param reader The operation which will observe the value. Must be a read.
1048 * @return True if the future value can be sent now; false if it must wait.
1050 bool ModelChecker::promises_may_allow(const ModelAction *writer,
1051 const ModelAction *reader) const
1053 if (promises->empty())
1055 for(int i=promises->size()-1;i>=0;i--) {
1056 ModelAction *pr=(*promises)[i]->get_reader(0);
1057 //reader is after promise...doesn't cross any promise
1060 //writer is after promise, reader before...bad...
1068 * @brief Add a future value to a reader
1070 * This function performs a few additional checks to ensure that the future
1071 * value can be feasibly observed by the reader
1073 * @param writer The operation whose value is sent. Must be a write.
1074 * @param reader The read operation which may read the future value. Must be a read.
1076 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
1078 /* Do more ambitious checks now that mo is more complete */
1079 if (!mo_may_allow(writer, reader))
1082 Node *node = reader->get_node();
1084 /* Find an ancestor thread which exists at the time of the reader */
1085 Thread *write_thread = get_thread(writer);
1086 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
1087 write_thread = write_thread->get_parent();
1089 struct future_value fv = {
1090 writer->get_write_value(),
1091 writer->get_seq_number() + params.maxfuturedelay,
1092 write_thread->get_id(),
1094 if (node->add_future_value(fv))
1095 set_latest_backtrack(reader);
1099 * Process a write ModelAction
1100 * @param curr The ModelAction to process
1101 * @return True if the mo_graph was updated or promises were resolved
1103 bool ModelChecker::process_write(ModelAction *curr)
1105 /* Readers to which we may send our future value */
1106 ModelVector<ModelAction *> send_fv;
1108 const ModelAction *earliest_promise_reader;
1109 bool updated_promises = false;
1111 bool updated_mod_order = w_modification_order(curr, &send_fv);
1112 Promise *promise = pop_promise_to_resolve(curr);
1115 earliest_promise_reader = promise->get_reader(0);
1116 updated_promises = resolve_promise(curr, promise);
1118 earliest_promise_reader = NULL;
1120 for (unsigned int i = 0; i < send_fv.size(); i++) {
1121 ModelAction *read = send_fv[i];
1123 /* Don't send future values to reads after the Promise we resolve */
1124 if (!earliest_promise_reader || *read < *earliest_promise_reader) {
1125 /* Check if future value can be sent immediately */
1126 if (promises_may_allow(curr, read)) {
1127 add_future_value(curr, read);
1129 futurevalues->push_back(PendingFutureValue(curr, read));
1134 /* Check the pending future values */
1135 for (int i = (int)futurevalues->size() - 1; i >= 0; i--) {
1136 struct PendingFutureValue pfv = (*futurevalues)[i];
1137 if (promises_may_allow(pfv.writer, pfv.reader)) {
1138 add_future_value(pfv.writer, pfv.reader);
1139 futurevalues->erase(futurevalues->begin() + i);
1143 mo_graph->commitChanges();
1144 mo_check_promises(curr, false);
1146 get_thread(curr)->set_return_value(VALUE_NONE);
1147 return updated_mod_order || updated_promises;
1151 * Process a fence ModelAction
1152 * @param curr The ModelAction to process
1153 * @return True if synchronization was updated
1155 bool ModelChecker::process_fence(ModelAction *curr)
1158 * fence-relaxed: no-op
1159 * fence-release: only log the occurence (not in this function), for
1160 * use in later synchronization
1161 * fence-acquire (this function): search for hypothetical release
1163 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
1165 bool updated = false;
1166 if (curr->is_acquire()) {
1167 action_list_t *list = action_trace;
1168 action_list_t::reverse_iterator rit;
1169 /* Find X : is_read(X) && X --sb-> curr */
1170 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1171 ModelAction *act = *rit;
1174 if (act->get_tid() != curr->get_tid())
1176 /* Stop at the beginning of the thread */
1177 if (act->is_thread_start())
1179 /* Stop once we reach a prior fence-acquire */
1180 if (act->is_fence() && act->is_acquire())
1182 if (!act->is_read())
1184 /* read-acquire will find its own release sequences */
1185 if (act->is_acquire())
1188 /* Establish hypothetical release sequences */
1189 rel_heads_list_t release_heads;
1190 get_release_seq_heads(curr, act, &release_heads);
1191 for (unsigned int i = 0; i < release_heads.size(); i++)
1192 synchronize(release_heads[i], curr);
1193 if (release_heads.size() != 0)
1201 * @brief Process the current action for thread-related activity
1203 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
1204 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
1205 * synchronization, etc. This function is a no-op for non-THREAD actions
1206 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
1208 * @param curr The current action
1209 * @return True if synchronization was updated or a thread completed
1211 bool ModelChecker::process_thread_action(ModelAction *curr)
1213 bool updated = false;
1215 switch (curr->get_type()) {
1216 case THREAD_CREATE: {
1217 thrd_t *thrd = (thrd_t *)curr->get_location();
1218 struct thread_params *params = (struct thread_params *)curr->get_value();
1219 Thread *th = new Thread(thrd, params->func, params->arg, get_thread(curr));
1221 th->set_creation(curr);
1222 /* Promises can be satisfied by children */
1223 for (unsigned int i = 0; i < promises->size(); i++) {
1224 Promise *promise = (*promises)[i];
1225 if (promise->thread_is_available(curr->get_tid()))
1226 promise->add_thread(th->get_id());
1231 Thread *blocking = curr->get_thread_operand();
1232 ModelAction *act = get_last_action(blocking->get_id());
1233 synchronize(act, curr);
1234 updated = true; /* trigger rel-seq checks */
1237 case THREAD_FINISH: {
1238 Thread *th = get_thread(curr);
1239 /* Wake up any joining threads */
1240 for (unsigned int i = 0; i < get_num_threads(); i++) {
1241 Thread *waiting = get_thread(int_to_id(i));
1242 if (waiting->waiting_on() == th &&
1243 waiting->get_pending()->is_thread_join())
1244 scheduler->wake(waiting);
1247 /* Completed thread can't satisfy promises */
1248 for (unsigned int i = 0; i < promises->size(); i++) {
1249 Promise *promise = (*promises)[i];
1250 if (promise->thread_is_available(th->get_id()))
1251 if (promise->eliminate_thread(th->get_id()))
1252 priv->failed_promise = true;
1254 updated = true; /* trigger rel-seq checks */
1257 case THREAD_START: {
1258 check_promises(curr->get_tid(), NULL, curr->get_cv());
1269 * @brief Process the current action for release sequence fixup activity
1271 * Performs model-checker release sequence fixups for the current action,
1272 * forcing a single pending release sequence to break (with a given, potential
1273 * "loose" write) or to complete (i.e., synchronize). If a pending release
1274 * sequence forms a complete release sequence, then we must perform the fixup
1275 * synchronization, mo_graph additions, etc.
1277 * @param curr The current action; must be a release sequence fixup action
1278 * @param work_queue The work queue to which to add work items as they are
1281 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1283 const ModelAction *write = curr->get_node()->get_relseq_break();
1284 struct release_seq *sequence = pending_rel_seqs->back();
1285 pending_rel_seqs->pop_back();
1287 ModelAction *acquire = sequence->acquire;
1288 const ModelAction *rf = sequence->rf;
1289 const ModelAction *release = sequence->release;
1293 ASSERT(release->same_thread(rf));
1295 if (write == NULL) {
1297 * @todo Forcing a synchronization requires that we set
1298 * modification order constraints. For instance, we can't allow
1299 * a fixup sequence in which two separate read-acquire
1300 * operations read from the same sequence, where the first one
1301 * synchronizes and the other doesn't. Essentially, we can't
1302 * allow any writes to insert themselves between 'release' and
1306 /* Must synchronize */
1307 if (!synchronize(release, acquire))
1309 /* Re-check all pending release sequences */
1310 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1311 /* Re-check act for mo_graph edges */
1312 work_queue->push_back(MOEdgeWorkEntry(acquire));
1314 /* propagate synchronization to later actions */
1315 action_list_t::reverse_iterator rit = action_trace->rbegin();
1316 for (; (*rit) != acquire; rit++) {
1317 ModelAction *propagate = *rit;
1318 if (acquire->happens_before(propagate)) {
1319 synchronize(acquire, propagate);
1320 /* Re-check 'propagate' for mo_graph edges */
1321 work_queue->push_back(MOEdgeWorkEntry(propagate));
1325 /* Break release sequence with new edges:
1326 * release --mo--> write --mo--> rf */
1327 mo_graph->addEdge(release, write);
1328 mo_graph->addEdge(write, rf);
1331 /* See if we have realized a data race */
1336 * Initialize the current action by performing one or more of the following
1337 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1338 * in the NodeStack, manipulating backtracking sets, allocating and
1339 * initializing clock vectors, and computing the promises to fulfill.
1341 * @param curr The current action, as passed from the user context; may be
1342 * freed/invalidated after the execution of this function, with a different
1343 * action "returned" its place (pass-by-reference)
1344 * @return True if curr is a newly-explored action; false otherwise
1346 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1348 ModelAction *newcurr;
1350 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1351 newcurr = process_rmw(*curr);
1354 if (newcurr->is_rmw())
1355 compute_promises(newcurr);
1361 (*curr)->set_seq_number(get_next_seq_num());
1363 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1365 /* First restore type and order in case of RMW operation */
1366 if ((*curr)->is_rmwr())
1367 newcurr->copy_typeandorder(*curr);
1369 ASSERT((*curr)->get_location() == newcurr->get_location());
1370 newcurr->copy_from_new(*curr);
1372 /* Discard duplicate ModelAction; use action from NodeStack */
1375 /* Always compute new clock vector */
1376 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1379 return false; /* Action was explored previously */
1383 /* Always compute new clock vector */
1384 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1386 /* Assign most recent release fence */
1387 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1390 * Perform one-time actions when pushing new ModelAction onto
1393 if (newcurr->is_write())
1394 compute_promises(newcurr);
1395 else if (newcurr->is_relseq_fixup())
1396 compute_relseq_breakwrites(newcurr);
1397 else if (newcurr->is_wait())
1398 newcurr->get_node()->set_misc_max(2);
1399 else if (newcurr->is_notify_one()) {
1400 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1402 return true; /* This was a new ModelAction */
1407 * @brief Establish reads-from relation between two actions
1409 * Perform basic operations involved with establishing a concrete rf relation,
1410 * including setting the ModelAction data and checking for release sequences.
1412 * @param act The action that is reading (must be a read)
1413 * @param rf The action from which we are reading (must be a write)
1415 * @return True if this read established synchronization
1417 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1420 ASSERT(rf->is_write());
1422 act->set_read_from(rf);
1423 if (act->is_acquire()) {
1424 rel_heads_list_t release_heads;
1425 get_release_seq_heads(act, act, &release_heads);
1426 int num_heads = release_heads.size();
1427 for (unsigned int i = 0; i < release_heads.size(); i++)
1428 if (!synchronize(release_heads[i], act))
1430 return num_heads > 0;
1436 * @brief Synchronizes two actions
1438 * When A synchronizes with B (or A --sw-> B), B inherits A's clock vector.
1439 * This function performs the synchronization as well as providing other hooks
1440 * for other checks along with synchronization.
1442 * @param first The left-hand side of the synchronizes-with relation
1443 * @param second The right-hand side of the synchronizes-with relation
1444 * @return True if the synchronization was successful (i.e., was consistent
1445 * with the execution order); false otherwise
1447 bool ModelChecker::synchronize(const ModelAction *first, ModelAction *second)
1449 if (*second < *first) {
1450 set_bad_synchronization();
1453 check_promises(first->get_tid(), second->get_cv(), first->get_cv());
1454 return second->synchronize_with(first);
1458 * Check promises and eliminate potentially-satisfying threads when a thread is
1459 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1460 * no longer satisfy a promise generated from that thread.
1462 * @param blocker The thread on which a thread is waiting
1463 * @param waiting The waiting thread
1465 void ModelChecker::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1467 for (unsigned int i = 0; i < promises->size(); i++) {
1468 Promise *promise = (*promises)[i];
1469 if (!promise->thread_is_available(waiting->get_id()))
1471 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1472 ModelAction *reader = promise->get_reader(j);
1473 if (reader->get_tid() != blocker->get_id())
1475 if (promise->eliminate_thread(waiting->get_id())) {
1476 /* Promise has failed */
1477 priv->failed_promise = true;
1479 /* Only eliminate the 'waiting' thread once */
1487 * @brief Check whether a model action is enabled.
1489 * Checks whether a lock or join operation would be successful (i.e., is the
1490 * lock already locked, or is the joined thread already complete). If not, put
1491 * the action in a waiter list.
1493 * @param curr is the ModelAction to check whether it is enabled.
1494 * @return a bool that indicates whether the action is enabled.
1496 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1497 if (curr->is_lock()) {
1498 std::mutex *lock = curr->get_mutex();
1499 struct std::mutex_state *state = lock->get_state();
1502 } else if (curr->is_thread_join()) {
1503 Thread *blocking = curr->get_thread_operand();
1504 if (!blocking->is_complete()) {
1505 thread_blocking_check_promises(blocking, get_thread(curr));
1514 * This is the heart of the model checker routine. It performs model-checking
1515 * actions corresponding to a given "current action." Among other processes, it
1516 * calculates reads-from relationships, updates synchronization clock vectors,
1517 * forms a memory_order constraints graph, and handles replay/backtrack
1518 * execution when running permutations of previously-observed executions.
1520 * @param curr The current action to process
1521 * @return The ModelAction that is actually executed; may be different than
1522 * curr; may be NULL, if the current action is not enabled to run
1524 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1527 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1528 bool newly_explored = initialize_curr_action(&curr);
1532 wake_up_sleeping_actions(curr);
1534 /* Compute fairness information for CHESS yield algorithm */
1535 if (model->params.yieldon) {
1536 curr->get_node()->update_yield(scheduler);
1539 /* Add the action to lists before any other model-checking tasks */
1540 if (!second_part_of_rmw)
1541 add_action_to_lists(curr);
1543 /* Build may_read_from set for newly-created actions */
1544 if (newly_explored && curr->is_read())
1545 build_may_read_from(curr);
1547 /* Initialize work_queue with the "current action" work */
1548 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1549 while (!work_queue.empty() && !has_asserted()) {
1550 WorkQueueEntry work = work_queue.front();
1551 work_queue.pop_front();
1553 switch (work.type) {
1554 case WORK_CHECK_CURR_ACTION: {
1555 ModelAction *act = work.action;
1556 bool update = false; /* update this location's release seq's */
1557 bool update_all = false; /* update all release seq's */
1559 if (process_thread_action(curr))
1562 if (act->is_read() && !second_part_of_rmw && process_read(act))
1565 if (act->is_write() && process_write(act))
1568 if (act->is_fence() && process_fence(act))
1571 if (act->is_mutex_op() && process_mutex(act))
1574 if (act->is_relseq_fixup())
1575 process_relseq_fixup(curr, &work_queue);
1578 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1580 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1583 case WORK_CHECK_RELEASE_SEQ:
1584 resolve_release_sequences(work.location, &work_queue);
1586 case WORK_CHECK_MO_EDGES: {
1587 /** @todo Complete verification of work_queue */
1588 ModelAction *act = work.action;
1589 bool updated = false;
1591 if (act->is_read()) {
1592 const ModelAction *rf = act->get_reads_from();
1593 const Promise *promise = act->get_reads_from_promise();
1595 if (r_modification_order(act, rf))
1597 } else if (promise) {
1598 if (r_modification_order(act, promise))
1602 if (act->is_write()) {
1603 if (w_modification_order(act, NULL))
1606 mo_graph->commitChanges();
1609 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1618 check_curr_backtracking(curr);
1619 set_backtracking(curr);
1623 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1625 Node *currnode = curr->get_node();
1626 Node *parnode = currnode->get_parent();
1628 if ((parnode && !parnode->backtrack_empty()) ||
1629 !currnode->misc_empty() ||
1630 !currnode->read_from_empty() ||
1631 !currnode->promise_empty() ||
1632 !currnode->relseq_break_empty()) {
1633 set_latest_backtrack(curr);
1637 bool ModelChecker::promises_expired() const
1639 for (unsigned int i = 0; i < promises->size(); i++) {
1640 Promise *promise = (*promises)[i];
1641 if (promise->get_expiration() < priv->used_sequence_numbers)
1648 * This is the strongest feasibility check available.
1649 * @return whether the current trace (partial or complete) must be a prefix of
1652 bool ModelChecker::isfeasibleprefix() const
1654 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1658 * Print disagnostic information about an infeasible execution
1659 * @param prefix A string to prefix the output with; if NULL, then a default
1660 * message prefix will be provided
1662 void ModelChecker::print_infeasibility(const char *prefix) const
1666 if (mo_graph->checkForCycles())
1667 ptr += sprintf(ptr, "[mo cycle]");
1668 if (priv->failed_promise)
1669 ptr += sprintf(ptr, "[failed promise]");
1670 if (priv->too_many_reads)
1671 ptr += sprintf(ptr, "[too many reads]");
1672 if (priv->no_valid_reads)
1673 ptr += sprintf(ptr, "[no valid reads-from]");
1674 if (priv->bad_synchronization)
1675 ptr += sprintf(ptr, "[bad sw ordering]");
1676 if (promises_expired())
1677 ptr += sprintf(ptr, "[promise expired]");
1678 if (promises->size() != 0)
1679 ptr += sprintf(ptr, "[unresolved promise]");
1681 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1685 * Returns whether the current completed trace is feasible, except for pending
1686 * release sequences.
1688 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1690 return !is_infeasible() && promises->size() == 0;
1694 * Check if the current partial trace is infeasible. Does not check any
1695 * end-of-execution flags, which might rule out the execution. Thus, this is
1696 * useful only for ruling an execution as infeasible.
1697 * @return whether the current partial trace is infeasible.
1699 bool ModelChecker::is_infeasible() const
1701 return mo_graph->checkForCycles() ||
1702 priv->no_valid_reads ||
1703 priv->failed_promise ||
1704 priv->too_many_reads ||
1705 priv->bad_synchronization ||
1709 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1710 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1711 ModelAction *lastread = get_last_action(act->get_tid());
1712 lastread->process_rmw(act);
1713 if (act->is_rmw()) {
1714 if (lastread->get_reads_from())
1715 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1717 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1718 mo_graph->commitChanges();
1724 * A helper function for ModelChecker::check_recency, to check if the current
1725 * thread is able to read from a different write/promise for 'params.maxreads'
1726 * number of steps and if that write/promise should become visible (i.e., is
1727 * ordered later in the modification order). This helps model memory liveness.
1729 * @param curr The current action. Must be a read.
1730 * @param rf The write/promise from which we plan to read
1731 * @param other_rf The write/promise from which we may read
1732 * @return True if we were able to read from other_rf for params.maxreads steps
1734 template <typename T, typename U>
1735 bool ModelChecker::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1737 /* Need a different write/promise */
1738 if (other_rf->equals(rf))
1741 /* Only look for "newer" writes/promises */
1742 if (!mo_graph->checkReachable(rf, other_rf))
1745 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1746 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1747 action_list_t::reverse_iterator rit = list->rbegin();
1748 ASSERT((*rit) == curr);
1749 /* Skip past curr */
1752 /* Does this write/promise work for everyone? */
1753 for (int i = 0; i < params.maxreads; i++, rit++) {
1754 ModelAction *act = *rit;
1755 if (!act->may_read_from(other_rf))
1762 * Checks whether a thread has read from the same write or Promise for too many
1763 * times without seeing the effects of a later write/Promise.
1766 * 1) there must a different write/promise that we could read from,
1767 * 2) we must have read from the same write/promise in excess of maxreads times,
1768 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1769 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1771 * If so, we decide that the execution is no longer feasible.
1773 * @param curr The current action. Must be a read.
1774 * @param rf The ModelAction/Promise from which we might read.
1775 * @return True if the read should succeed; false otherwise
1777 template <typename T>
1778 bool ModelChecker::check_recency(ModelAction *curr, const T *rf) const
1780 if (!params.maxreads)
1783 //NOTE: Next check is just optimization, not really necessary....
1784 if (curr->get_node()->get_read_from_past_size() +
1785 curr->get_node()->get_read_from_promise_size() <= 1)
1788 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1789 int tid = id_to_int(curr->get_tid());
1790 ASSERT(tid < (int)thrd_lists->size());
1791 action_list_t *list = &(*thrd_lists)[tid];
1792 action_list_t::reverse_iterator rit = list->rbegin();
1793 ASSERT((*rit) == curr);
1794 /* Skip past curr */
1797 action_list_t::reverse_iterator ritcopy = rit;
1798 /* See if we have enough reads from the same value */
1799 for (int count = 0; count < params.maxreads; ritcopy++, count++) {
1800 if (ritcopy == list->rend())
1802 ModelAction *act = *ritcopy;
1803 if (!act->is_read())
1805 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1807 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1809 if (act->get_node()->get_read_from_past_size() +
1810 act->get_node()->get_read_from_promise_size() <= 1)
1813 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1814 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1815 if (should_read_instead(curr, rf, write))
1816 return false; /* liveness failure */
1818 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1819 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1820 if (should_read_instead(curr, rf, promise))
1821 return false; /* liveness failure */
1827 * @brief Updates the mo_graph with the constraints imposed from the current
1830 * Basic idea is the following: Go through each other thread and find
1831 * the last action that happened before our read. Two cases:
1833 * -# The action is a write: that write must either occur before
1834 * the write we read from or be the write we read from.
1835 * -# The action is a read: the write that that action read from
1836 * must occur before the write we read from or be the same write.
1838 * @param curr The current action. Must be a read.
1839 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1840 * @return True if modification order edges were added; false otherwise
1842 template <typename rf_type>
1843 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1845 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1848 ASSERT(curr->is_read());
1850 /* Last SC fence in the current thread */
1851 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1852 ModelAction *last_sc_write = NULL;
1853 if (curr->is_seqcst())
1854 last_sc_write = get_last_seq_cst_write(curr);
1856 /* Iterate over all threads */
1857 for (i = 0; i < thrd_lists->size(); i++) {
1858 /* Last SC fence in thread i */
1859 ModelAction *last_sc_fence_thread_local = NULL;
1860 if (int_to_id((int)i) != curr->get_tid())
1861 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1863 /* Last SC fence in thread i, before last SC fence in current thread */
1864 ModelAction *last_sc_fence_thread_before = NULL;
1865 if (last_sc_fence_local)
1866 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1868 /* Iterate over actions in thread, starting from most recent */
1869 action_list_t *list = &(*thrd_lists)[i];
1870 action_list_t::reverse_iterator rit;
1871 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1872 ModelAction *act = *rit;
1877 /* Don't want to add reflexive edges on 'rf' */
1878 if (act->equals(rf)) {
1879 if (act->happens_before(curr))
1885 if (act->is_write()) {
1886 /* C++, Section 29.3 statement 5 */
1887 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1888 *act < *last_sc_fence_thread_local) {
1889 added = mo_graph->addEdge(act, rf) || added;
1892 /* C++, Section 29.3 statement 4 */
1893 else if (act->is_seqcst() && last_sc_fence_local &&
1894 *act < *last_sc_fence_local) {
1895 added = mo_graph->addEdge(act, rf) || added;
1898 /* C++, Section 29.3 statement 6 */
1899 else if (last_sc_fence_thread_before &&
1900 *act < *last_sc_fence_thread_before) {
1901 added = mo_graph->addEdge(act, rf) || added;
1906 /* C++, Section 29.3 statement 3 (second subpoint) */
1907 if (curr->is_seqcst() && last_sc_write && act == last_sc_write) {
1908 added = mo_graph->addEdge(act, rf) || added;
1913 * Include at most one act per-thread that "happens
1916 if (act->happens_before(curr)) {
1917 if (act->is_write()) {
1918 added = mo_graph->addEdge(act, rf) || added;
1920 const ModelAction *prevrf = act->get_reads_from();
1921 const Promise *prevrf_promise = act->get_reads_from_promise();
1923 if (!prevrf->equals(rf))
1924 added = mo_graph->addEdge(prevrf, rf) || added;
1925 } else if (!prevrf_promise->equals(rf)) {
1926 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1935 * All compatible, thread-exclusive promises must be ordered after any
1936 * concrete loads from the same thread
1938 for (unsigned int i = 0; i < promises->size(); i++)
1939 if ((*promises)[i]->is_compatible_exclusive(curr))
1940 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1946 * Updates the mo_graph with the constraints imposed from the current write.
1948 * Basic idea is the following: Go through each other thread and find
1949 * the lastest action that happened before our write. Two cases:
1951 * (1) The action is a write => that write must occur before
1954 * (2) The action is a read => the write that that action read from
1955 * must occur before the current write.
1957 * This method also handles two other issues:
1959 * (I) Sequential Consistency: Making sure that if the current write is
1960 * seq_cst, that it occurs after the previous seq_cst write.
1962 * (II) Sending the write back to non-synchronizing reads.
1964 * @param curr The current action. Must be a write.
1965 * @param send_fv A vector for stashing reads to which we may pass our future
1966 * value. If NULL, then don't record any future values.
1967 * @return True if modification order edges were added; false otherwise
1969 bool ModelChecker::w_modification_order(ModelAction *curr, ModelVector<ModelAction *> *send_fv)
1971 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1974 ASSERT(curr->is_write());
1976 if (curr->is_seqcst()) {
1977 /* We have to at least see the last sequentially consistent write,
1978 so we are initialized. */
1979 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1980 if (last_seq_cst != NULL) {
1981 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1985 /* Last SC fence in the current thread */
1986 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1988 /* Iterate over all threads */
1989 for (i = 0; i < thrd_lists->size(); i++) {
1990 /* Last SC fence in thread i, before last SC fence in current thread */
1991 ModelAction *last_sc_fence_thread_before = NULL;
1992 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1993 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1995 /* Iterate over actions in thread, starting from most recent */
1996 action_list_t *list = &(*thrd_lists)[i];
1997 action_list_t::reverse_iterator rit;
1998 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1999 ModelAction *act = *rit;
2002 * 1) If RMW and it actually read from something, then we
2003 * already have all relevant edges, so just skip to next
2006 * 2) If RMW and it didn't read from anything, we should
2007 * whatever edge we can get to speed up convergence.
2009 * 3) If normal write, we need to look at earlier actions, so
2010 * continue processing list.
2012 if (curr->is_rmw()) {
2013 if (curr->get_reads_from() != NULL)
2021 /* C++, Section 29.3 statement 7 */
2022 if (last_sc_fence_thread_before && act->is_write() &&
2023 *act < *last_sc_fence_thread_before) {
2024 added = mo_graph->addEdge(act, curr) || added;
2029 * Include at most one act per-thread that "happens
2032 if (act->happens_before(curr)) {
2034 * Note: if act is RMW, just add edge:
2036 * The following edge should be handled elsewhere:
2037 * readfrom(act) --mo--> act
2039 if (act->is_write())
2040 added = mo_graph->addEdge(act, curr) || added;
2041 else if (act->is_read()) {
2042 //if previous read accessed a null, just keep going
2043 if (act->get_reads_from() == NULL)
2045 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
2048 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
2049 !act->same_thread(curr)) {
2050 /* We have an action that:
2051 (1) did not happen before us
2052 (2) is a read and we are a write
2053 (3) cannot synchronize with us
2054 (4) is in a different thread
2056 that read could potentially read from our write. Note that
2057 these checks are overly conservative at this point, we'll
2058 do more checks before actually removing the
2062 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
2063 if (!is_infeasible())
2064 send_fv->push_back(act);
2065 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
2066 add_future_value(curr, act);
2073 * All compatible, thread-exclusive promises must be ordered after any
2074 * concrete stores to the same thread, or else they can be merged with
2077 for (unsigned int i = 0; i < promises->size(); i++)
2078 if ((*promises)[i]->is_compatible_exclusive(curr))
2079 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
2084 /** Arbitrary reads from the future are not allowed. Section 29.3
2085 * part 9 places some constraints. This method checks one result of constraint
2086 * constraint. Others require compiler support. */
2087 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader) const
2089 if (!writer->is_rmw())
2092 if (!reader->is_rmw())
2095 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
2096 if (search == reader)
2098 if (search->get_tid() == reader->get_tid() &&
2099 search->happens_before(reader))
2107 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
2108 * some constraints. This method checks one the following constraint (others
2109 * require compiler support):
2111 * If X --hb-> Y --mo-> Z, then X should not read from Z.
2113 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
2115 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
2117 /* Iterate over all threads */
2118 for (i = 0; i < thrd_lists->size(); i++) {
2119 const ModelAction *write_after_read = NULL;
2121 /* Iterate over actions in thread, starting from most recent */
2122 action_list_t *list = &(*thrd_lists)[i];
2123 action_list_t::reverse_iterator rit;
2124 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2125 ModelAction *act = *rit;
2127 /* Don't disallow due to act == reader */
2128 if (!reader->happens_before(act) || reader == act)
2130 else if (act->is_write())
2131 write_after_read = act;
2132 else if (act->is_read() && act->get_reads_from() != NULL)
2133 write_after_read = act->get_reads_from();
2136 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
2143 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
2144 * The ModelAction under consideration is expected to be taking part in
2145 * release/acquire synchronization as an object of the "reads from" relation.
2146 * Note that this can only provide release sequence support for RMW chains
2147 * which do not read from the future, as those actions cannot be traced until
2148 * their "promise" is fulfilled. Similarly, we may not even establish the
2149 * presence of a release sequence with certainty, as some modification order
2150 * constraints may be decided further in the future. Thus, this function
2151 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
2152 * and a boolean representing certainty.
2154 * @param rf The action that might be part of a release sequence. Must be a
2156 * @param release_heads A pass-by-reference style return parameter. After
2157 * execution of this function, release_heads will contain the heads of all the
2158 * relevant release sequences, if any exists with certainty
2159 * @param pending A pass-by-reference style return parameter which is only used
2160 * when returning false (i.e., uncertain). Returns most information regarding
2161 * an uncertain release sequence, including any write operations that might
2162 * break the sequence.
2163 * @return true, if the ModelChecker is certain that release_heads is complete;
2166 bool ModelChecker::release_seq_heads(const ModelAction *rf,
2167 rel_heads_list_t *release_heads,
2168 struct release_seq *pending) const
2170 /* Only check for release sequences if there are no cycles */
2171 if (mo_graph->checkForCycles())
2174 for ( ; rf != NULL; rf = rf->get_reads_from()) {
2175 ASSERT(rf->is_write());
2177 if (rf->is_release())
2178 release_heads->push_back(rf);
2179 else if (rf->get_last_fence_release())
2180 release_heads->push_back(rf->get_last_fence_release());
2182 break; /* End of RMW chain */
2184 /** @todo Need to be smarter here... In the linux lock
2185 * example, this will run to the beginning of the program for
2187 /** @todo The way to be smarter here is to keep going until 1
2188 * thread has a release preceded by an acquire and you've seen
2191 /* acq_rel RMW is a sufficient stopping condition */
2192 if (rf->is_acquire() && rf->is_release())
2193 return true; /* complete */
2196 /* read from future: need to settle this later */
2198 return false; /* incomplete */
2201 if (rf->is_release())
2202 return true; /* complete */
2204 /* else relaxed write
2205 * - check for fence-release in the same thread (29.8, stmt. 3)
2206 * - check modification order for contiguous subsequence
2207 * -> rf must be same thread as release */
2209 const ModelAction *fence_release = rf->get_last_fence_release();
2210 /* Synchronize with a fence-release unconditionally; we don't need to
2211 * find any more "contiguous subsequence..." for it */
2213 release_heads->push_back(fence_release);
2215 int tid = id_to_int(rf->get_tid());
2216 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
2217 action_list_t *list = &(*thrd_lists)[tid];
2218 action_list_t::const_reverse_iterator rit;
2220 /* Find rf in the thread list */
2221 rit = std::find(list->rbegin(), list->rend(), rf);
2222 ASSERT(rit != list->rend());
2224 /* Find the last {write,fence}-release */
2225 for (; rit != list->rend(); rit++) {
2226 if (fence_release && *(*rit) < *fence_release)
2228 if ((*rit)->is_release())
2231 if (rit == list->rend()) {
2232 /* No write-release in this thread */
2233 return true; /* complete */
2234 } else if (fence_release && *(*rit) < *fence_release) {
2235 /* The fence-release is more recent (and so, "stronger") than
2236 * the most recent write-release */
2237 return true; /* complete */
2238 } /* else, need to establish contiguous release sequence */
2239 ModelAction *release = *rit;
2241 ASSERT(rf->same_thread(release));
2243 pending->writes.clear();
2245 bool certain = true;
2246 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2247 if (id_to_int(rf->get_tid()) == (int)i)
2249 list = &(*thrd_lists)[i];
2251 /* Can we ensure no future writes from this thread may break
2252 * the release seq? */
2253 bool future_ordered = false;
2255 ModelAction *last = get_last_action(int_to_id(i));
2256 Thread *th = get_thread(int_to_id(i));
2257 if ((last && rf->happens_before(last)) ||
2260 future_ordered = true;
2262 ASSERT(!th->is_model_thread() || future_ordered);
2264 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2265 const ModelAction *act = *rit;
2266 /* Reach synchronization -> this thread is complete */
2267 if (act->happens_before(release))
2269 if (rf->happens_before(act)) {
2270 future_ordered = true;
2274 /* Only non-RMW writes can break release sequences */
2275 if (!act->is_write() || act->is_rmw())
2278 /* Check modification order */
2279 if (mo_graph->checkReachable(rf, act)) {
2280 /* rf --mo--> act */
2281 future_ordered = true;
2284 if (mo_graph->checkReachable(act, release))
2285 /* act --mo--> release */
2287 if (mo_graph->checkReachable(release, act) &&
2288 mo_graph->checkReachable(act, rf)) {
2289 /* release --mo-> act --mo--> rf */
2290 return true; /* complete */
2292 /* act may break release sequence */
2293 pending->writes.push_back(act);
2296 if (!future_ordered)
2297 certain = false; /* This thread is uncertain */
2301 release_heads->push_back(release);
2302 pending->writes.clear();
2304 pending->release = release;
2311 * An interface for getting the release sequence head(s) with which a
2312 * given ModelAction must synchronize. This function only returns a non-empty
2313 * result when it can locate a release sequence head with certainty. Otherwise,
2314 * it may mark the internal state of the ModelChecker so that it will handle
2315 * the release sequence at a later time, causing @a acquire to update its
2316 * synchronization at some later point in execution.
2318 * @param acquire The 'acquire' action that may synchronize with a release
2320 * @param read The read action that may read from a release sequence; this may
2321 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2322 * when 'acquire' is a fence-acquire)
2323 * @param release_heads A pass-by-reference return parameter. Will be filled
2324 * with the head(s) of the release sequence(s), if they exists with certainty.
2325 * @see ModelChecker::release_seq_heads
2327 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2328 ModelAction *read, rel_heads_list_t *release_heads)
2330 const ModelAction *rf = read->get_reads_from();
2331 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2332 sequence->acquire = acquire;
2333 sequence->read = read;
2335 if (!release_seq_heads(rf, release_heads, sequence)) {
2336 /* add act to 'lazy checking' list */
2337 pending_rel_seqs->push_back(sequence);
2339 snapshot_free(sequence);
2344 * Attempt to resolve all stashed operations that might synchronize with a
2345 * release sequence for a given location. This implements the "lazy" portion of
2346 * determining whether or not a release sequence was contiguous, since not all
2347 * modification order information is present at the time an action occurs.
2349 * @param location The location/object that should be checked for release
2350 * sequence resolutions. A NULL value means to check all locations.
2351 * @param work_queue The work queue to which to add work items as they are
2353 * @return True if any updates occurred (new synchronization, new mo_graph
2356 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2358 bool updated = false;
2359 SnapVector<struct release_seq *>::iterator it = pending_rel_seqs->begin();
2360 while (it != pending_rel_seqs->end()) {
2361 struct release_seq *pending = *it;
2362 ModelAction *acquire = pending->acquire;
2363 const ModelAction *read = pending->read;
2365 /* Only resolve sequences on the given location, if provided */
2366 if (location && read->get_location() != location) {
2371 const ModelAction *rf = read->get_reads_from();
2372 rel_heads_list_t release_heads;
2374 complete = release_seq_heads(rf, &release_heads, pending);
2375 for (unsigned int i = 0; i < release_heads.size(); i++)
2376 if (!acquire->has_synchronized_with(release_heads[i]))
2377 if (synchronize(release_heads[i], acquire))
2381 /* Re-check all pending release sequences */
2382 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2383 /* Re-check read-acquire for mo_graph edges */
2384 if (acquire->is_read())
2385 work_queue->push_back(MOEdgeWorkEntry(acquire));
2387 /* propagate synchronization to later actions */
2388 action_list_t::reverse_iterator rit = action_trace->rbegin();
2389 for (; (*rit) != acquire; rit++) {
2390 ModelAction *propagate = *rit;
2391 if (acquire->happens_before(propagate)) {
2392 synchronize(acquire, propagate);
2393 /* Re-check 'propagate' for mo_graph edges */
2394 work_queue->push_back(MOEdgeWorkEntry(propagate));
2399 it = pending_rel_seqs->erase(it);
2400 snapshot_free(pending);
2406 // If we resolved promises or data races, see if we have realized a data race.
2413 * Performs various bookkeeping operations for the current ModelAction. For
2414 * instance, adds action to the per-object, per-thread action vector and to the
2415 * action trace list of all thread actions.
2417 * @param act is the ModelAction to add.
2419 void ModelChecker::add_action_to_lists(ModelAction *act)
2421 int tid = id_to_int(act->get_tid());
2422 ModelAction *uninit = NULL;
2424 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2425 if (list->empty() && act->is_atomic_var()) {
2426 uninit = get_uninitialized_action(act);
2427 uninit_id = id_to_int(uninit->get_tid());
2428 list->push_front(uninit);
2430 list->push_back(act);
2432 action_trace->push_back(act);
2434 action_trace->push_front(uninit);
2436 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2437 if (tid >= (int)vec->size())
2438 vec->resize(priv->next_thread_id);
2439 (*vec)[tid].push_back(act);
2441 (*vec)[uninit_id].push_front(uninit);
2443 if ((int)thrd_last_action->size() <= tid)
2444 thrd_last_action->resize(get_num_threads());
2445 (*thrd_last_action)[tid] = act;
2447 (*thrd_last_action)[uninit_id] = uninit;
2449 if (act->is_fence() && act->is_release()) {
2450 if ((int)thrd_last_fence_release->size() <= tid)
2451 thrd_last_fence_release->resize(get_num_threads());
2452 (*thrd_last_fence_release)[tid] = act;
2455 if (act->is_wait()) {
2456 void *mutex_loc = (void *) act->get_value();
2457 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2459 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2460 if (tid >= (int)vec->size())
2461 vec->resize(priv->next_thread_id);
2462 (*vec)[tid].push_back(act);
2467 * @brief Get the last action performed by a particular Thread
2468 * @param tid The thread ID of the Thread in question
2469 * @return The last action in the thread
2471 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2473 int threadid = id_to_int(tid);
2474 if (threadid < (int)thrd_last_action->size())
2475 return (*thrd_last_action)[id_to_int(tid)];
2481 * @brief Get the last fence release performed by a particular Thread
2482 * @param tid The thread ID of the Thread in question
2483 * @return The last fence release in the thread, if one exists; NULL otherwise
2485 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2487 int threadid = id_to_int(tid);
2488 if (threadid < (int)thrd_last_fence_release->size())
2489 return (*thrd_last_fence_release)[id_to_int(tid)];
2495 * Gets the last memory_order_seq_cst write (in the total global sequence)
2496 * performed on a particular object (i.e., memory location), not including the
2498 * @param curr The current ModelAction; also denotes the object location to
2500 * @return The last seq_cst write
2502 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2504 void *location = curr->get_location();
2505 action_list_t *list = get_safe_ptr_action(obj_map, location);
2506 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2507 action_list_t::reverse_iterator rit;
2508 for (rit = list->rbegin(); (*rit) != curr; rit++)
2510 rit++; /* Skip past curr */
2511 for ( ; rit != list->rend(); rit++)
2512 if ((*rit)->is_write() && (*rit)->is_seqcst())
2518 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2519 * performed in a particular thread, prior to a particular fence.
2520 * @param tid The ID of the thread to check
2521 * @param before_fence The fence from which to begin the search; if NULL, then
2522 * search for the most recent fence in the thread.
2523 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2525 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2527 /* All fences should have NULL location */
2528 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2529 action_list_t::reverse_iterator rit = list->rbegin();
2532 for (; rit != list->rend(); rit++)
2533 if (*rit == before_fence)
2536 ASSERT(*rit == before_fence);
2540 for (; rit != list->rend(); rit++)
2541 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2547 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2548 * location). This function identifies the mutex according to the current
2549 * action, which is presumed to perform on the same mutex.
2550 * @param curr The current ModelAction; also denotes the object location to
2552 * @return The last unlock operation
2554 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2556 void *location = curr->get_location();
2557 action_list_t *list = get_safe_ptr_action(obj_map, location);
2558 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2559 action_list_t::reverse_iterator rit;
2560 for (rit = list->rbegin(); rit != list->rend(); rit++)
2561 if ((*rit)->is_unlock() || (*rit)->is_wait())
2566 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2568 ModelAction *parent = get_last_action(tid);
2570 parent = get_thread(tid)->get_creation();
2575 * Returns the clock vector for a given thread.
2576 * @param tid The thread whose clock vector we want
2577 * @return Desired clock vector
2579 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2581 return get_parent_action(tid)->get_cv();
2585 * @brief Find the promise (if any) to resolve for the current action and
2586 * remove it from the pending promise vector
2587 * @param curr The current ModelAction. Should be a write.
2588 * @return The Promise to resolve, if any; otherwise NULL
2590 Promise * ModelChecker::pop_promise_to_resolve(const ModelAction *curr)
2592 for (unsigned int i = 0; i < promises->size(); i++)
2593 if (curr->get_node()->get_promise(i)) {
2594 Promise *ret = (*promises)[i];
2595 promises->erase(promises->begin() + i);
2602 * Resolve a Promise with a current write.
2603 * @param write The ModelAction that is fulfilling Promises
2604 * @param promise The Promise to resolve
2605 * @return True if the Promise was successfully resolved; false otherwise
2607 bool ModelChecker::resolve_promise(ModelAction *write, Promise *promise)
2609 ModelVector<ModelAction *> actions_to_check;
2611 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2612 ModelAction *read = promise->get_reader(i);
2613 read_from(read, write);
2614 actions_to_check.push_back(read);
2616 /* Make sure the promise's value matches the write's value */
2617 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2618 if (!mo_graph->resolvePromise(promise, write))
2619 priv->failed_promise = true;
2622 * @todo It is possible to end up in an inconsistent state, where a
2623 * "resolved" promise may still be referenced if
2624 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2626 * Note that the inconsistency only matters when dumping mo_graph to
2632 //Check whether reading these writes has made threads unable to
2634 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2635 ModelAction *read = actions_to_check[i];
2636 mo_check_promises(read, true);
2643 * Compute the set of promises that could potentially be satisfied by this
2644 * action. Note that the set computation actually appears in the Node, not in
2646 * @param curr The ModelAction that may satisfy promises
2648 void ModelChecker::compute_promises(ModelAction *curr)
2650 for (unsigned int i = 0; i < promises->size(); i++) {
2651 Promise *promise = (*promises)[i];
2652 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2655 bool satisfy = true;
2656 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2657 const ModelAction *act = promise->get_reader(j);
2658 if (act->happens_before(curr) ||
2659 act->could_synchronize_with(curr)) {
2665 curr->get_node()->set_promise(i);
2669 /** Checks promises in response to change in ClockVector Threads. */
2670 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2672 for (unsigned int i = 0; i < promises->size(); i++) {
2673 Promise *promise = (*promises)[i];
2674 if (!promise->thread_is_available(tid))
2676 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2677 const ModelAction *act = promise->get_reader(j);
2678 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2679 merge_cv->synchronized_since(act)) {
2680 if (promise->eliminate_thread(tid)) {
2681 /* Promise has failed */
2682 priv->failed_promise = true;
2690 void ModelChecker::check_promises_thread_disabled()
2692 for (unsigned int i = 0; i < promises->size(); i++) {
2693 Promise *promise = (*promises)[i];
2694 if (promise->has_failed()) {
2695 priv->failed_promise = true;
2702 * @brief Checks promises in response to addition to modification order for
2705 * We test whether threads are still available for satisfying promises after an
2706 * addition to our modification order constraints. Those that are unavailable
2707 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2708 * that promise has failed.
2710 * @param act The ModelAction which updated the modification order
2711 * @param is_read_check Should be true if act is a read and we must check for
2712 * updates to the store from which it read (there is a distinction here for
2713 * RMW's, which are both a load and a store)
2715 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2717 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2719 for (unsigned int i = 0; i < promises->size(); i++) {
2720 Promise *promise = (*promises)[i];
2722 // Is this promise on the same location?
2723 if (!promise->same_location(write))
2726 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2727 const ModelAction *pread = promise->get_reader(j);
2728 if (!pread->happens_before(act))
2730 if (mo_graph->checkPromise(write, promise)) {
2731 priv->failed_promise = true;
2737 // Don't do any lookups twice for the same thread
2738 if (!promise->thread_is_available(act->get_tid()))
2741 if (mo_graph->checkReachable(promise, write)) {
2742 if (mo_graph->checkPromise(write, promise)) {
2743 priv->failed_promise = true;
2751 * Compute the set of writes that may break the current pending release
2752 * sequence. This information is extracted from previou release sequence
2755 * @param curr The current ModelAction. Must be a release sequence fixup
2758 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2760 if (pending_rel_seqs->empty())
2763 struct release_seq *pending = pending_rel_seqs->back();
2764 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2765 const ModelAction *write = pending->writes[i];
2766 curr->get_node()->add_relseq_break(write);
2769 /* NULL means don't break the sequence; just synchronize */
2770 curr->get_node()->add_relseq_break(NULL);
2774 * Build up an initial set of all past writes that this 'read' action may read
2775 * from, as well as any previously-observed future values that must still be valid.
2777 * @param curr is the current ModelAction that we are exploring; it must be a
2780 void ModelChecker::build_may_read_from(ModelAction *curr)
2782 SnapVector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2784 ASSERT(curr->is_read());
2786 ModelAction *last_sc_write = NULL;
2788 if (curr->is_seqcst())
2789 last_sc_write = get_last_seq_cst_write(curr);
2791 /* Iterate over all threads */
2792 for (i = 0; i < thrd_lists->size(); i++) {
2793 /* Iterate over actions in thread, starting from most recent */
2794 action_list_t *list = &(*thrd_lists)[i];
2795 action_list_t::reverse_iterator rit;
2796 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2797 ModelAction *act = *rit;
2799 /* Only consider 'write' actions */
2800 if (!act->is_write() || act == curr)
2803 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2804 bool allow_read = true;
2806 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2808 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2812 /* Only add feasible reads */
2813 mo_graph->startChanges();
2814 r_modification_order(curr, act);
2815 if (!is_infeasible())
2816 curr->get_node()->add_read_from_past(act);
2817 mo_graph->rollbackChanges();
2820 /* Include at most one act per-thread that "happens before" curr */
2821 if (act->happens_before(curr))
2826 /* Inherit existing, promised future values */
2827 for (i = 0; i < promises->size(); i++) {
2828 const Promise *promise = (*promises)[i];
2829 const ModelAction *promise_read = promise->get_reader(0);
2830 if (promise_read->same_var(curr)) {
2831 /* Only add feasible future-values */
2832 mo_graph->startChanges();
2833 r_modification_order(curr, promise);
2834 if (!is_infeasible())
2835 curr->get_node()->add_read_from_promise(promise_read);
2836 mo_graph->rollbackChanges();
2840 /* We may find no valid may-read-from only if the execution is doomed */
2841 if (!curr->get_node()->read_from_size()) {
2842 priv->no_valid_reads = true;
2846 if (DBG_ENABLED()) {
2847 model_print("Reached read action:\n");
2849 model_print("Printing read_from_past\n");
2850 curr->get_node()->print_read_from_past();
2851 model_print("End printing read_from_past\n");
2855 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2857 for ( ; write != NULL; write = write->get_reads_from()) {
2858 /* UNINIT actions don't have a Node, and they never sleep */
2859 if (write->is_uninitialized())
2861 Node *prevnode = write->get_node()->get_parent();
2863 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2864 if (write->is_release() && thread_sleep)
2866 if (!write->is_rmw())
2873 * @brief Get an action representing an uninitialized atomic
2875 * This function may create a new one or try to retrieve one from the NodeStack
2877 * @param curr The current action, which prompts the creation of an UNINIT action
2878 * @return A pointer to the UNINIT ModelAction
2880 ModelAction * ModelChecker::get_uninitialized_action(const ModelAction *curr) const
2882 Node *node = curr->get_node();
2883 ModelAction *act = node->get_uninit_action();
2885 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), model->params.uninitvalue, model_thread);
2886 node->set_uninit_action(act);
2888 act->create_cv(NULL);
2892 static void print_list(action_list_t *list)
2894 action_list_t::iterator it;
2896 model_print("---------------------------------------------------------------------\n");
2898 unsigned int hash = 0;
2900 for (it = list->begin(); it != list->end(); it++) {
2901 const ModelAction *act = *it;
2902 if (act->get_seq_number() > 0)
2904 hash = hash^(hash<<3)^((*it)->hash());
2906 model_print("HASH %u\n", hash);
2907 model_print("---------------------------------------------------------------------\n");
2910 #if SUPPORT_MOD_ORDER_DUMP
2911 void ModelChecker::dumpGraph(char *filename) const
2914 sprintf(buffer, "%s.dot", filename);
2915 FILE *file = fopen(buffer, "w");
2916 fprintf(file, "digraph %s {\n", filename);
2917 mo_graph->dumpNodes(file);
2918 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2920 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2921 ModelAction *act = *it;
2922 if (act->is_read()) {
2923 mo_graph->dot_print_node(file, act);
2924 if (act->get_reads_from())
2925 mo_graph->dot_print_edge(file,
2926 act->get_reads_from(),
2928 "label=\"rf\", color=red, weight=2");
2930 mo_graph->dot_print_edge(file,
2931 act->get_reads_from_promise(),
2933 "label=\"rf\", color=red");
2935 if (thread_array[act->get_tid()]) {
2936 mo_graph->dot_print_edge(file,
2937 thread_array[id_to_int(act->get_tid())],
2939 "label=\"sb\", color=blue, weight=400");
2942 thread_array[act->get_tid()] = act;
2944 fprintf(file, "}\n");
2945 model_free(thread_array);
2950 /** @brief Prints an execution trace summary. */
2951 void ModelChecker::print_summary() const
2953 #if SUPPORT_MOD_ORDER_DUMP
2954 char buffername[100];
2955 sprintf(buffername, "exec%04u", stats.num_total);
2956 mo_graph->dumpGraphToFile(buffername);
2957 sprintf(buffername, "graph%04u", stats.num_total);
2958 dumpGraph(buffername);
2961 model_print("Execution %d:", stats.num_total);
2962 if (isfeasibleprefix()) {
2963 if (scheduler->all_threads_sleeping())
2964 model_print(" SLEEP-SET REDUNDANT");
2967 print_infeasibility(" INFEASIBLE");
2968 print_list(action_trace);
2970 if (!promises->empty()) {
2971 model_print("Pending promises:\n");
2972 for (unsigned int i = 0; i < promises->size(); i++) {
2973 model_print(" [P%u] ", i);
2974 (*promises)[i]->print();
2981 * Add a Thread to the system for the first time. Should only be called once
2983 * @param t The Thread to add
2985 void ModelChecker::add_thread(Thread *t)
2987 thread_map->put(id_to_int(t->get_id()), t);
2988 scheduler->add_thread(t);
2992 * @brief Get a Thread reference by its ID
2993 * @param tid The Thread's ID
2994 * @return A Thread reference
2996 Thread * ModelChecker::get_thread(thread_id_t tid) const
2998 return thread_map->get(id_to_int(tid));
3002 * @brief Get a reference to the Thread in which a ModelAction was executed
3003 * @param act The ModelAction
3004 * @return A Thread reference
3006 Thread * ModelChecker::get_thread(const ModelAction *act) const
3008 return get_thread(act->get_tid());
3012 * @brief Get a Promise's "promise number"
3014 * A "promise number" is an index number that is unique to a promise, valid
3015 * only for a specific snapshot of an execution trace. Promises may come and go
3016 * as they are generated an resolved, so an index only retains meaning for the
3019 * @param promise The Promise to check
3020 * @return The promise index, if the promise still is valid; otherwise -1
3022 int ModelChecker::get_promise_number(const Promise *promise) const
3024 for (unsigned int i = 0; i < promises->size(); i++)
3025 if ((*promises)[i] == promise)
3032 * @brief Check if a Thread is currently enabled
3033 * @param t The Thread to check
3034 * @return True if the Thread is currently enabled
3036 bool ModelChecker::is_enabled(Thread *t) const
3038 return scheduler->is_enabled(t);
3042 * @brief Check if a Thread is currently enabled
3043 * @param tid The ID of the Thread to check
3044 * @return True if the Thread is currently enabled
3046 bool ModelChecker::is_enabled(thread_id_t tid) const
3048 return scheduler->is_enabled(tid);
3052 * Switch from a model-checker context to a user-thread context. This is the
3053 * complement of ModelChecker::switch_to_master and must be called from the
3054 * model-checker context
3056 * @param thread The user-thread to switch to
3058 void ModelChecker::switch_from_master(Thread *thread)
3060 scheduler->set_current_thread(thread);
3061 Thread::swap(&system_context, thread);
3065 * Switch from a user-context to the "master thread" context (a.k.a. system
3066 * context). This switch is made with the intention of exploring a particular
3067 * model-checking action (described by a ModelAction object). Must be called
3068 * from a user-thread context.
3070 * @param act The current action that will be explored. May be NULL only if
3071 * trace is exiting via an assertion (see ModelChecker::set_assert and
3072 * ModelChecker::has_asserted).
3073 * @return Return the value returned by the current action
3075 uint64_t ModelChecker::switch_to_master(ModelAction *act)
3078 Thread *old = thread_current();
3079 scheduler->set_current_thread(NULL);
3080 ASSERT(!old->get_pending());
3081 old->set_pending(act);
3082 if (Thread::swap(old, &system_context) < 0) {
3083 perror("swap threads");
3086 return old->get_return_value();
3090 * Takes the next step in the execution, if possible.
3091 * @param curr The current step to take
3092 * @return Returns the next Thread to run, if any; NULL if this execution
3095 Thread * ModelChecker::take_step(ModelAction *curr)
3097 Thread *curr_thrd = get_thread(curr);
3098 ASSERT(curr_thrd->get_state() == THREAD_READY);
3100 ASSERT(check_action_enabled(curr)); /* May have side effects? */
3101 curr = check_current_action(curr);
3104 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
3105 scheduler->remove_thread(curr_thrd);
3107 return action_select_next_thread(curr);
3110 /** Wrapper to run the user's main function, with appropriate arguments */
3111 void user_main_wrapper(void *)
3113 user_main(model->params.argc, model->params.argv);
3116 /** @return True if the execution has taken too many steps */
3117 bool ModelChecker::too_many_steps() const
3119 return params.bound != 0 && priv->used_sequence_numbers > params.bound;
3122 bool ModelChecker::should_terminate_execution()
3124 /* Infeasible -> don't take any more steps */
3125 if (is_infeasible())
3127 else if (isfeasibleprefix() && have_bug_reports()) {
3132 if (too_many_steps())
3137 /** @brief Run ModelChecker for the user program */
3138 void ModelChecker::run()
3142 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL, NULL);
3147 * Stash next pending action(s) for thread(s). There
3148 * should only need to stash one thread's action--the
3149 * thread which just took a step--plus the first step
3150 * for any newly-created thread
3152 for (unsigned int i = 0; i < get_num_threads(); i++) {
3153 thread_id_t tid = int_to_id(i);
3154 Thread *thr = get_thread(tid);
3155 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
3156 switch_from_master(thr);
3157 if (thr->is_waiting_on(thr))
3158 assert_bug("Deadlock detected (thread %u)", i);
3162 /* Don't schedule threads which should be disabled */
3163 for (unsigned int i = 0; i < get_num_threads(); i++) {
3164 Thread *th = get_thread(int_to_id(i));
3165 ModelAction *act = th->get_pending();
3166 if (act && is_enabled(th) && !check_action_enabled(act)) {
3167 scheduler->sleep(th);
3171 /* Catch assertions from prior take_step or from
3172 * between-ModelAction bugs (e.g., data races) */
3177 t = get_next_thread();
3178 if (!t || t->is_model_thread())
3181 /* Consume the next action for a Thread */
3182 ModelAction *curr = t->get_pending();
3183 t->set_pending(NULL);
3184 t = take_step(curr);
3185 } while (!should_terminate_execution());
3188 * Launch end-of-execution release sequence fixups only when
3189 * the execution is otherwise feasible AND there are:
3191 * (1) pending release sequences
3192 * (2) pending assertions that could be invalidated by a change
3193 * in clock vectors (i.e., data races)
3194 * (3) no pending promises
3196 while (!pending_rel_seqs->empty() &&
3197 is_feasible_prefix_ignore_relseq() &&
3198 !unrealizedraces.empty()) {
3199 model_print("*** WARNING: release sequence fixup action "
3200 "(%zu pending release seuqence(s)) ***\n",
3201 pending_rel_seqs->size());
3202 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
3203 std::memory_order_seq_cst, NULL, VALUE_NONE,
3207 } while (next_execution());
3209 model_print("******* Model-checking complete: *******\n");