9 #include "snapshot-interface.h"
11 #include "clockvector.h"
12 #include "cyclegraph.h"
15 #include "threads-model.h"
18 #define INITIAL_THREAD_ID 0
23 bug_message(const char *str) {
24 const char *fmt = " [BUG] %s\n";
25 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
26 sprintf(msg, fmt, str);
28 ~bug_message() { if (msg) snapshot_free(msg); }
31 void print() { model_print("%s", msg); }
37 * Structure for holding small ModelChecker members that should be snapshotted
39 struct model_snapshot_members {
40 ModelAction *current_action;
41 unsigned int next_thread_id;
42 modelclock_t used_sequence_numbers;
44 ModelAction *next_backtrack;
45 std::vector< bug_message *, SnapshotAlloc<bug_message *> > bugs;
46 struct execution_stats stats;
49 /** @brief Constructor */
50 ModelChecker::ModelChecker(struct model_params params) :
51 /* Initialize default scheduler */
53 scheduler(new Scheduler()),
55 earliest_diverge(NULL),
56 action_trace(new action_list_t()),
57 thread_map(new HashTable<int, Thread *, int>()),
58 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
59 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
60 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
61 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
62 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
63 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
64 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
65 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
66 node_stack(new NodeStack()),
67 mo_graph(new CycleGraph()),
68 failed_promise(false),
69 too_many_reads(false),
71 bad_synchronization(false)
73 /* Allocate this "size" on the snapshotting heap */
74 priv = (struct model_snapshot_members *)snapshot_calloc(1, sizeof(*priv));
75 /* First thread created will have id INITIAL_THREAD_ID */
76 priv->next_thread_id = INITIAL_THREAD_ID;
78 /* Initialize a model-checker thread, for special ModelActions */
79 model_thread = new Thread(get_next_id());
80 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
83 /** @brief Destructor */
84 ModelChecker::~ModelChecker()
86 for (unsigned int i = 0; i < get_num_threads(); i++)
87 delete thread_map->get(i);
92 delete lock_waiters_map;
93 delete condvar_waiters_map;
96 for (unsigned int i = 0; i < promises->size(); i++)
97 delete (*promises)[i];
100 delete pending_rel_seqs;
102 delete thrd_last_action;
107 for (unsigned int i = 0; i < priv->bugs.size(); i++)
108 delete priv->bugs[i];
113 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr) {
114 action_list_t * tmp=hash->get(ptr);
116 tmp=new action_list_t();
122 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr) {
123 std::vector<action_list_t> * tmp=hash->get(ptr);
125 tmp=new std::vector<action_list_t>();
132 * Restores user program to initial state and resets all model-checker data
135 void ModelChecker::reset_to_initial_state()
137 DEBUG("+++ Resetting to initial state +++\n");
138 node_stack->reset_execution();
139 failed_promise = false;
140 too_many_reads = false;
141 bad_synchronization = false;
144 /* Print all model-checker output before rollback */
147 snapshotObject->backTrackBeforeStep(0);
150 /** @return a thread ID for a new Thread */
151 thread_id_t ModelChecker::get_next_id()
153 return priv->next_thread_id++;
156 /** @return the number of user threads created during this execution */
157 unsigned int ModelChecker::get_num_threads() const
159 return priv->next_thread_id;
162 /** @return The currently executing Thread. */
163 Thread * ModelChecker::get_current_thread()
165 return scheduler->get_current_thread();
168 /** @return a sequence number for a new ModelAction */
169 modelclock_t ModelChecker::get_next_seq_num()
171 return ++priv->used_sequence_numbers;
174 Node * ModelChecker::get_curr_node() {
175 return node_stack->get_head();
179 * @brief Choose the next thread to execute.
181 * This function chooses the next thread that should execute. It can force the
182 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
183 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
184 * The model-checker may have no preference regarding the next thread (i.e.,
185 * when exploring a new execution ordering), in which case this will return
187 * @param curr The current ModelAction. This action might guide the choice of
189 * @return The next thread to run. If the model-checker has no preference, NULL.
191 Thread * ModelChecker::get_next_thread(ModelAction *curr)
196 /* Do not split atomic actions. */
198 return thread_current();
199 /* The THREAD_CREATE action points to the created Thread */
200 else if (curr->get_type() == THREAD_CREATE)
201 return (Thread *)curr->get_location();
204 /* Have we completed exploring the preselected path? */
208 /* Else, we are trying to replay an execution */
209 ModelAction *next = node_stack->get_next()->get_action();
211 if (next == diverge) {
212 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
213 earliest_diverge=diverge;
215 Node *nextnode = next->get_node();
216 Node *prevnode = nextnode->get_parent();
217 scheduler->update_sleep_set(prevnode);
219 /* Reached divergence point */
220 if (nextnode->increment_misc()) {
221 /* The next node will try to satisfy a different misc_index values. */
222 tid = next->get_tid();
223 node_stack->pop_restofstack(2);
224 } else if (nextnode->increment_promise()) {
225 /* The next node will try to satisfy a different set of promises. */
226 tid = next->get_tid();
227 node_stack->pop_restofstack(2);
228 } else if (nextnode->increment_read_from()) {
229 /* The next node will read from a different value. */
230 tid = next->get_tid();
231 node_stack->pop_restofstack(2);
232 } else if (nextnode->increment_future_value()) {
233 /* The next node will try to read from a different future value. */
234 tid = next->get_tid();
235 node_stack->pop_restofstack(2);
236 } else if (nextnode->increment_relseq_break()) {
237 /* The next node will try to resolve a release sequence differently */
238 tid = next->get_tid();
239 node_stack->pop_restofstack(2);
241 /* Make a different thread execute for next step */
242 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
243 tid = prevnode->get_next_backtrack();
244 /* Make sure the backtracked thread isn't sleeping. */
245 node_stack->pop_restofstack(1);
246 if (diverge==earliest_diverge) {
247 earliest_diverge=prevnode->get_action();
250 /* The correct sleep set is in the parent node. */
253 DEBUG("*** Divergence point ***\n");
257 tid = next->get_tid();
259 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
260 ASSERT(tid != THREAD_ID_T_NONE);
261 return thread_map->get(id_to_int(tid));
265 * We need to know what the next actions of all threads in the sleep
266 * set will be. This method computes them and stores the actions at
267 * the corresponding thread object's pending action.
270 void ModelChecker::execute_sleep_set() {
271 for(unsigned int i=0;i<get_num_threads();i++) {
272 thread_id_t tid=int_to_id(i);
273 Thread *thr=get_thread(tid);
274 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET &&
275 thr->get_pending() == NULL ) {
276 thr->set_state(THREAD_RUNNING);
277 scheduler->next_thread(thr);
278 Thread::swap(&system_context, thr);
279 priv->current_action->set_sleep_flag();
280 thr->set_pending(priv->current_action);
283 priv->current_action = NULL;
286 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
287 for(unsigned int i=0;i<get_num_threads();i++) {
288 thread_id_t tid=int_to_id(i);
289 Thread *thr=get_thread(tid);
290 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
291 ModelAction *pending_act=thr->get_pending();
292 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
293 //Remove this thread from sleep set
294 scheduler->remove_sleep(thr);
301 * Check if we are in a deadlock. Should only be called at the end of an
302 * execution, although it should not give false positives in the middle of an
303 * execution (there should be some ENABLED thread).
305 * @return True if program is in a deadlock; false otherwise
307 bool ModelChecker::is_deadlocked() const
309 bool blocking_threads = false;
310 for (unsigned int i = 0; i < get_num_threads(); i++) {
311 thread_id_t tid = int_to_id(i);
314 Thread *t = get_thread(tid);
315 if (!t->is_model_thread() && t->get_pending())
316 blocking_threads = true;
318 return blocking_threads;
322 * Check if this is a complete execution. That is, have all thread completed
323 * execution (rather than exiting because sleep sets have forced a redundant
326 * @return True if the execution is complete.
328 bool ModelChecker::is_complete_execution() const
330 for (unsigned int i = 0; i < get_num_threads(); i++)
331 if (is_enabled(int_to_id(i)))
337 * @brief Assert a bug in the executing program.
339 * Use this function to assert any sort of bug in the user program. If the
340 * current trace is feasible (actually, a prefix of some feasible execution),
341 * then this execution will be aborted, printing the appropriate message. If
342 * the current trace is not yet feasible, the error message will be stashed and
343 * printed if the execution ever becomes feasible.
345 * @param msg Descriptive message for the bug (do not include newline char)
346 * @return True if bug is immediately-feasible
348 bool ModelChecker::assert_bug(const char *msg)
350 priv->bugs.push_back(new bug_message(msg));
352 if (isfeasibleprefix()) {
360 * @brief Assert a bug in the executing program, asserted by a user thread
361 * @see ModelChecker::assert_bug
362 * @param msg Descriptive message for the bug (do not include newline char)
364 void ModelChecker::assert_user_bug(const char *msg)
366 /* If feasible bug, bail out now */
368 switch_to_master(NULL);
371 /** @return True, if any bugs have been reported for this execution */
372 bool ModelChecker::have_bug_reports() const
374 return priv->bugs.size() != 0;
377 /** @brief Print bug report listing for this execution (if any bugs exist) */
378 void ModelChecker::print_bugs() const
380 if (have_bug_reports()) {
381 model_print("Bug report: %zu bug%s detected\n",
383 priv->bugs.size() > 1 ? "s" : "");
384 for (unsigned int i = 0; i < priv->bugs.size(); i++)
385 priv->bugs[i]->print();
390 * @brief Record end-of-execution stats
392 * Must be run when exiting an execution. Records various stats.
393 * @see struct execution_stats
395 void ModelChecker::record_stats()
398 if (!isfinalfeasible())
399 stats.num_infeasible++;
400 else if (have_bug_reports())
401 stats.num_buggy_executions++;
402 else if (is_complete_execution())
403 stats.num_complete++;
406 /** @brief Print execution stats */
407 void ModelChecker::print_stats() const
409 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
410 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
411 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
412 model_print("Total executions: %d\n", stats.num_total);
413 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
417 * Queries the model-checker for more executions to explore and, if one
418 * exists, resets the model-checker state to execute a new execution.
420 * @return If there are more executions to explore, return true. Otherwise,
423 bool ModelChecker::next_execution()
427 if (isfinalfeasible() && (is_complete_execution() || have_bug_reports())) {
429 assert_bug("Deadlock detected");
433 if (DBG_ENABLED() || params.verbose || have_bug_reports()) {
434 print_program_output();
436 if (DBG_ENABLED() || params.verbose) {
437 model_print("Earliest divergence point since last feasible execution:\n");
438 if (earliest_diverge)
439 earliest_diverge->print();
441 model_print("(Not set)\n");
451 clear_program_output();
453 earliest_diverge = NULL;
454 } else if (DBG_ENABLED()) {
455 print_program_output();
460 clear_program_output();
465 if ((diverge = get_next_backtrack()) == NULL)
469 model_print("Next execution will diverge at:\n");
473 reset_to_initial_state();
477 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
479 switch (act->get_type()) {
483 /* linear search: from most recent to oldest */
484 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
485 action_list_t::reverse_iterator rit;
486 for (rit = list->rbegin(); rit != list->rend(); rit++) {
487 ModelAction *prev = *rit;
488 if (prev->could_synchronize_with(act))
494 case ATOMIC_TRYLOCK: {
495 /* linear search: from most recent to oldest */
496 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
497 action_list_t::reverse_iterator rit;
498 for (rit = list->rbegin(); rit != list->rend(); rit++) {
499 ModelAction *prev = *rit;
500 if (act->is_conflicting_lock(prev))
505 case ATOMIC_UNLOCK: {
506 /* linear search: from most recent to oldest */
507 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
508 action_list_t::reverse_iterator rit;
509 for (rit = list->rbegin(); rit != list->rend(); rit++) {
510 ModelAction *prev = *rit;
511 if (!act->same_thread(prev)&&prev->is_failed_trylock())
517 /* linear search: from most recent to oldest */
518 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
519 action_list_t::reverse_iterator rit;
520 for (rit = list->rbegin(); rit != list->rend(); rit++) {
521 ModelAction *prev = *rit;
522 if (!act->same_thread(prev)&&prev->is_failed_trylock())
524 if (!act->same_thread(prev)&&prev->is_notify())
530 case ATOMIC_NOTIFY_ALL:
531 case ATOMIC_NOTIFY_ONE: {
532 /* linear search: from most recent to oldest */
533 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
534 action_list_t::reverse_iterator rit;
535 for (rit = list->rbegin(); rit != list->rend(); rit++) {
536 ModelAction *prev = *rit;
537 if (!act->same_thread(prev)&&prev->is_wait())
548 /** This method finds backtracking points where we should try to
549 * reorder the parameter ModelAction against.
551 * @param the ModelAction to find backtracking points for.
553 void ModelChecker::set_backtracking(ModelAction *act)
555 Thread *t = get_thread(act);
556 ModelAction * prev = get_last_conflict(act);
560 Node * node = prev->get_node()->get_parent();
562 int low_tid, high_tid;
563 if (node->is_enabled(t)) {
564 low_tid = id_to_int(act->get_tid());
565 high_tid = low_tid+1;
568 high_tid = get_num_threads();
571 for(int i = low_tid; i < high_tid; i++) {
572 thread_id_t tid = int_to_id(i);
574 /* Make sure this thread can be enabled here. */
575 if (i >= node->get_num_threads())
578 /* Don't backtrack into a point where the thread is disabled or sleeping. */
579 if (node->enabled_status(tid)!=THREAD_ENABLED)
582 /* Check if this has been explored already */
583 if (node->has_been_explored(tid))
586 /* See if fairness allows */
587 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
589 for(int t=0;t<node->get_num_threads();t++) {
590 thread_id_t tother=int_to_id(t);
591 if (node->is_enabled(tother) && node->has_priority(tother)) {
599 /* Cache the latest backtracking point */
600 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
601 priv->next_backtrack = prev;
603 /* If this is a new backtracking point, mark the tree */
604 if (!node->set_backtrack(tid))
606 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
607 id_to_int(prev->get_tid()),
608 id_to_int(t->get_id()));
617 * Returns last backtracking point. The model checker will explore a different
618 * path for this point in the next execution.
619 * @return The ModelAction at which the next execution should diverge.
621 ModelAction * ModelChecker::get_next_backtrack()
623 ModelAction *next = priv->next_backtrack;
624 priv->next_backtrack = NULL;
629 * Processes a read or rmw model action.
630 * @param curr is the read model action to process.
631 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
632 * @return True if processing this read updates the mo_graph.
634 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
636 uint64_t value = VALUE_NONE;
637 bool updated = false;
639 const ModelAction *reads_from = curr->get_node()->get_read_from();
640 if (reads_from != NULL) {
641 mo_graph->startChanges();
643 value = reads_from->get_value();
644 bool r_status = false;
646 if (!second_part_of_rmw) {
647 check_recency(curr, reads_from);
648 r_status = r_modification_order(curr, reads_from);
652 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
653 mo_graph->rollbackChanges();
654 too_many_reads = false;
658 curr->read_from(reads_from);
659 mo_graph->commitChanges();
660 mo_check_promises(curr->get_tid(), reads_from);
663 } else if (!second_part_of_rmw) {
664 /* Read from future value */
665 value = curr->get_node()->get_future_value();
666 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
667 curr->read_from(NULL);
668 Promise *valuepromise = new Promise(curr, value, expiration);
669 promises->push_back(valuepromise);
671 get_thread(curr)->set_return_value(value);
677 * Processes a lock, trylock, or unlock model action. @param curr is
678 * the read model action to process.
680 * The try lock operation checks whether the lock is taken. If not,
681 * it falls to the normal lock operation case. If so, it returns
684 * The lock operation has already been checked that it is enabled, so
685 * it just grabs the lock and synchronizes with the previous unlock.
687 * The unlock operation has to re-enable all of the threads that are
688 * waiting on the lock.
690 * @return True if synchronization was updated; false otherwise
692 bool ModelChecker::process_mutex(ModelAction *curr) {
693 std::mutex *mutex=NULL;
694 struct std::mutex_state *state=NULL;
696 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
697 mutex = (std::mutex *)curr->get_location();
698 state = mutex->get_state();
699 } else if(curr->is_wait()) {
700 mutex = (std::mutex *)curr->get_value();
701 state = mutex->get_state();
704 switch (curr->get_type()) {
705 case ATOMIC_TRYLOCK: {
706 bool success = !state->islocked;
707 curr->set_try_lock(success);
709 get_thread(curr)->set_return_value(0);
712 get_thread(curr)->set_return_value(1);
714 //otherwise fall into the lock case
716 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
717 assert_bug("Lock access before initialization");
718 state->islocked = true;
719 ModelAction *unlock = get_last_unlock(curr);
720 //synchronize with the previous unlock statement
721 if (unlock != NULL) {
722 curr->synchronize_with(unlock);
727 case ATOMIC_UNLOCK: {
729 state->islocked = false;
730 //wake up the other threads
731 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
732 //activate all the waiting threads
733 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
734 scheduler->wake(get_thread(*rit));
741 state->islocked = false;
742 //wake up the other threads
743 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
744 //activate all the waiting threads
745 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
746 scheduler->wake(get_thread(*rit));
749 //check whether we should go to sleep or not...simulate spurious failures
750 if (curr->get_node()->get_misc()==0) {
751 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
753 scheduler->sleep(get_current_thread());
757 case ATOMIC_NOTIFY_ALL: {
758 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
759 //activate all the waiting threads
760 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
761 scheduler->wake(get_thread(*rit));
766 case ATOMIC_NOTIFY_ONE: {
767 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
768 int wakeupthread=curr->get_node()->get_misc();
769 action_list_t::iterator it = waiters->begin();
770 advance(it, wakeupthread);
771 scheduler->wake(get_thread(*it));
783 * Process a write ModelAction
784 * @param curr The ModelAction to process
785 * @return True if the mo_graph was updated or promises were resolved
787 bool ModelChecker::process_write(ModelAction *curr)
789 bool updated_mod_order = w_modification_order(curr);
790 bool updated_promises = resolve_promises(curr);
792 if (promises->size() == 0) {
793 for (unsigned int i = 0; i < futurevalues->size(); i++) {
794 struct PendingFutureValue pfv = (*futurevalues)[i];
795 //Do more ambitious checks now that mo is more complete
796 if (mo_may_allow(pfv.writer, pfv.act)&&
797 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
798 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
799 priv->next_backtrack = pfv.act;
801 futurevalues->resize(0);
804 mo_graph->commitChanges();
805 mo_check_promises(curr->get_tid(), curr);
807 get_thread(curr)->set_return_value(VALUE_NONE);
808 return updated_mod_order || updated_promises;
812 * @brief Process the current action for thread-related activity
814 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
815 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
816 * synchronization, etc. This function is a no-op for non-THREAD actions
817 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
819 * @param curr The current action
820 * @return True if synchronization was updated or a thread completed
822 bool ModelChecker::process_thread_action(ModelAction *curr)
824 bool updated = false;
826 switch (curr->get_type()) {
827 case THREAD_CREATE: {
828 Thread *th = (Thread *)curr->get_location();
829 th->set_creation(curr);
833 Thread *blocking = (Thread *)curr->get_location();
834 ModelAction *act = get_last_action(blocking->get_id());
835 curr->synchronize_with(act);
836 updated = true; /* trigger rel-seq checks */
839 case THREAD_FINISH: {
840 Thread *th = get_thread(curr);
841 while (!th->wait_list_empty()) {
842 ModelAction *act = th->pop_wait_list();
843 scheduler->wake(get_thread(act));
846 updated = true; /* trigger rel-seq checks */
850 check_promises(curr->get_tid(), NULL, curr->get_cv());
861 * @brief Process the current action for release sequence fixup activity
863 * Performs model-checker release sequence fixups for the current action,
864 * forcing a single pending release sequence to break (with a given, potential
865 * "loose" write) or to complete (i.e., synchronize). If a pending release
866 * sequence forms a complete release sequence, then we must perform the fixup
867 * synchronization, mo_graph additions, etc.
869 * @param curr The current action; must be a release sequence fixup action
870 * @param work_queue The work queue to which to add work items as they are
873 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
875 const ModelAction *write = curr->get_node()->get_relseq_break();
876 struct release_seq *sequence = pending_rel_seqs->back();
877 pending_rel_seqs->pop_back();
879 ModelAction *acquire = sequence->acquire;
880 const ModelAction *rf = sequence->rf;
881 const ModelAction *release = sequence->release;
885 ASSERT(release->same_thread(rf));
889 * @todo Forcing a synchronization requires that we set
890 * modification order constraints. For instance, we can't allow
891 * a fixup sequence in which two separate read-acquire
892 * operations read from the same sequence, where the first one
893 * synchronizes and the other doesn't. Essentially, we can't
894 * allow any writes to insert themselves between 'release' and
898 /* Must synchronize */
899 if (!acquire->synchronize_with(release)) {
900 set_bad_synchronization();
903 /* Re-check all pending release sequences */
904 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
905 /* Re-check act for mo_graph edges */
906 work_queue->push_back(MOEdgeWorkEntry(acquire));
908 /* propagate synchronization to later actions */
909 action_list_t::reverse_iterator rit = action_trace->rbegin();
910 for (; (*rit) != acquire; rit++) {
911 ModelAction *propagate = *rit;
912 if (acquire->happens_before(propagate)) {
913 propagate->synchronize_with(acquire);
914 /* Re-check 'propagate' for mo_graph edges */
915 work_queue->push_back(MOEdgeWorkEntry(propagate));
919 /* Break release sequence with new edges:
920 * release --mo--> write --mo--> rf */
921 mo_graph->addEdge(release, write);
922 mo_graph->addEdge(write, rf);
925 /* See if we have realized a data race */
930 * Initialize the current action by performing one or more of the following
931 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
932 * in the NodeStack, manipulating backtracking sets, allocating and
933 * initializing clock vectors, and computing the promises to fulfill.
935 * @param curr The current action, as passed from the user context; may be
936 * freed/invalidated after the execution of this function, with a different
937 * action "returned" its place (pass-by-reference)
938 * @return True if curr is a newly-explored action; false otherwise
940 bool ModelChecker::initialize_curr_action(ModelAction **curr)
942 ModelAction *newcurr;
944 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
945 newcurr = process_rmw(*curr);
948 if (newcurr->is_rmw())
949 compute_promises(newcurr);
955 (*curr)->set_seq_number(get_next_seq_num());
957 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled());
959 /* First restore type and order in case of RMW operation */
960 if ((*curr)->is_rmwr())
961 newcurr->copy_typeandorder(*curr);
963 ASSERT((*curr)->get_location() == newcurr->get_location());
964 newcurr->copy_from_new(*curr);
966 /* Discard duplicate ModelAction; use action from NodeStack */
969 /* Always compute new clock vector */
970 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
973 return false; /* Action was explored previously */
977 /* Always compute new clock vector */
978 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
980 * Perform one-time actions when pushing new ModelAction onto
983 if (newcurr->is_write())
984 compute_promises(newcurr);
985 else if (newcurr->is_relseq_fixup())
986 compute_relseq_breakwrites(newcurr);
987 else if (newcurr->is_wait())
988 newcurr->get_node()->set_misc_max(2);
989 else if (newcurr->is_notify_one()) {
990 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
992 return true; /* This was a new ModelAction */
997 * @brief Check whether a model action is enabled.
999 * Checks whether a lock or join operation would be successful (i.e., is the
1000 * lock already locked, or is the joined thread already complete). If not, put
1001 * the action in a waiter list.
1003 * @param curr is the ModelAction to check whether it is enabled.
1004 * @return a bool that indicates whether the action is enabled.
1006 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1007 if (curr->is_lock()) {
1008 std::mutex * lock = (std::mutex *)curr->get_location();
1009 struct std::mutex_state * state = lock->get_state();
1010 if (state->islocked) {
1011 //Stick the action in the appropriate waiting queue
1012 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
1015 } else if (curr->get_type() == THREAD_JOIN) {
1016 Thread *blocking = (Thread *)curr->get_location();
1017 if (!blocking->is_complete()) {
1018 blocking->push_wait_list(curr);
1027 * Stores the ModelAction for the current thread action. Call this
1028 * immediately before switching from user- to system-context to pass
1029 * data between them.
1030 * @param act The ModelAction created by the user-thread action
1032 void ModelChecker::set_current_action(ModelAction *act) {
1033 priv->current_action = act;
1037 * This is the heart of the model checker routine. It performs model-checking
1038 * actions corresponding to a given "current action." Among other processes, it
1039 * calculates reads-from relationships, updates synchronization clock vectors,
1040 * forms a memory_order constraints graph, and handles replay/backtrack
1041 * execution when running permutations of previously-observed executions.
1043 * @param curr The current action to process
1044 * @return The next Thread that must be executed. May be NULL if ModelChecker
1045 * makes no choice (e.g., according to replay execution, combining RMW actions,
1048 Thread * ModelChecker::check_current_action(ModelAction *curr)
1051 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1053 if (!check_action_enabled(curr)) {
1054 /* Make the execution look like we chose to run this action
1055 * much later, when a lock/join can succeed */
1056 get_current_thread()->set_pending(curr);
1057 scheduler->sleep(get_current_thread());
1058 return get_next_thread(NULL);
1061 bool newly_explored = initialize_curr_action(&curr);
1063 wake_up_sleeping_actions(curr);
1065 /* Add the action to lists before any other model-checking tasks */
1066 if (!second_part_of_rmw)
1067 add_action_to_lists(curr);
1069 /* Build may_read_from set for newly-created actions */
1070 if (newly_explored && curr->is_read())
1071 build_reads_from_past(curr);
1073 /* Initialize work_queue with the "current action" work */
1074 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1075 while (!work_queue.empty() && !has_asserted()) {
1076 WorkQueueEntry work = work_queue.front();
1077 work_queue.pop_front();
1079 switch (work.type) {
1080 case WORK_CHECK_CURR_ACTION: {
1081 ModelAction *act = work.action;
1082 bool update = false; /* update this location's release seq's */
1083 bool update_all = false; /* update all release seq's */
1085 if (process_thread_action(curr))
1088 if (act->is_read() && process_read(act, second_part_of_rmw))
1091 if (act->is_write() && process_write(act))
1094 if (act->is_mutex_op() && process_mutex(act))
1097 if (act->is_relseq_fixup())
1098 process_relseq_fixup(curr, &work_queue);
1101 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1103 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1106 case WORK_CHECK_RELEASE_SEQ:
1107 resolve_release_sequences(work.location, &work_queue);
1109 case WORK_CHECK_MO_EDGES: {
1110 /** @todo Complete verification of work_queue */
1111 ModelAction *act = work.action;
1112 bool updated = false;
1114 if (act->is_read()) {
1115 const ModelAction *rf = act->get_reads_from();
1116 if (rf != NULL && r_modification_order(act, rf))
1119 if (act->is_write()) {
1120 if (w_modification_order(act))
1123 mo_graph->commitChanges();
1126 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1135 check_curr_backtracking(curr);
1136 set_backtracking(curr);
1137 return get_next_thread(curr);
1140 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
1141 Node *currnode = curr->get_node();
1142 Node *parnode = currnode->get_parent();
1144 if ((!parnode->backtrack_empty() ||
1145 !currnode->misc_empty() ||
1146 !currnode->read_from_empty() ||
1147 !currnode->future_value_empty() ||
1148 !currnode->promise_empty() ||
1149 !currnode->relseq_break_empty())
1150 && (!priv->next_backtrack ||
1151 *curr > *priv->next_backtrack)) {
1152 priv->next_backtrack = curr;
1156 bool ModelChecker::promises_expired() const
1158 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
1159 Promise *promise = (*promises)[promise_index];
1160 if (promise->get_expiration()<priv->used_sequence_numbers) {
1167 /** @return whether the current partial trace must be a prefix of a
1168 * feasible trace. */
1169 bool ModelChecker::isfeasibleprefix() const
1171 return promises->size() == 0 && pending_rel_seqs->size() == 0 && isfeasible();
1174 /** @return whether the current partial trace is feasible. */
1175 bool ModelChecker::isfeasible() const
1177 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
1178 DEBUG("Infeasible: RMW violation\n");
1180 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
1183 /** @return whether the current partial trace is feasible other than
1184 * multiple RMW reading from the same store. */
1185 bool ModelChecker::isfeasibleotherthanRMW() const
1187 if (DBG_ENABLED()) {
1188 if (mo_graph->checkForCycles())
1189 DEBUG("Infeasible: modification order cycles\n");
1191 DEBUG("Infeasible: failed promise\n");
1193 DEBUG("Infeasible: too many reads\n");
1194 if (bad_synchronization)
1195 DEBUG("Infeasible: bad synchronization ordering\n");
1196 if (promises_expired())
1197 DEBUG("Infeasible: promises expired\n");
1199 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
1202 /** Returns whether the current completed trace is feasible. */
1203 bool ModelChecker::isfinalfeasible() const
1205 if (DBG_ENABLED() && promises->size() != 0)
1206 DEBUG("Infeasible: unrevolved promises\n");
1208 return isfeasible() && promises->size() == 0;
1211 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1212 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1213 ModelAction *lastread = get_last_action(act->get_tid());
1214 lastread->process_rmw(act);
1215 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1216 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1217 mo_graph->commitChanges();
1223 * Checks whether a thread has read from the same write for too many times
1224 * without seeing the effects of a later write.
1227 * 1) there must a different write that we could read from that would satisfy the modification order,
1228 * 2) we must have read from the same value in excess of maxreads times, and
1229 * 3) that other write must have been in the reads_from set for maxreads times.
1231 * If so, we decide that the execution is no longer feasible.
1233 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1234 if (params.maxreads != 0) {
1236 if (curr->get_node()->get_read_from_size() <= 1)
1238 //Must make sure that execution is currently feasible... We could
1239 //accidentally clear by rolling back
1242 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1243 int tid = id_to_int(curr->get_tid());
1246 if ((int)thrd_lists->size() <= tid)
1248 action_list_t *list = &(*thrd_lists)[tid];
1250 action_list_t::reverse_iterator rit = list->rbegin();
1251 /* Skip past curr */
1252 for (; (*rit) != curr; rit++)
1254 /* go past curr now */
1257 action_list_t::reverse_iterator ritcopy = rit;
1258 //See if we have enough reads from the same value
1260 for (; count < params.maxreads; rit++,count++) {
1261 if (rit==list->rend())
1263 ModelAction *act = *rit;
1264 if (!act->is_read())
1267 if (act->get_reads_from() != rf)
1269 if (act->get_node()->get_read_from_size() <= 1)
1272 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1274 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1276 //Need a different write
1280 /* Test to see whether this is a feasible write to read from*/
1281 mo_graph->startChanges();
1282 r_modification_order(curr, write);
1283 bool feasiblereadfrom = isfeasible();
1284 mo_graph->rollbackChanges();
1286 if (!feasiblereadfrom)
1290 bool feasiblewrite = true;
1291 //new we need to see if this write works for everyone
1293 for (int loop = count; loop>0; loop--,rit++) {
1294 ModelAction *act=*rit;
1295 bool foundvalue = false;
1296 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1297 if (act->get_node()->get_read_from_at(j)==write) {
1303 feasiblewrite = false;
1307 if (feasiblewrite) {
1308 too_many_reads = true;
1316 * Updates the mo_graph with the constraints imposed from the current
1319 * Basic idea is the following: Go through each other thread and find
1320 * the lastest action that happened before our read. Two cases:
1322 * (1) The action is a write => that write must either occur before
1323 * the write we read from or be the write we read from.
1325 * (2) The action is a read => the write that that action read from
1326 * must occur before the write we read from or be the same write.
1328 * @param curr The current action. Must be a read.
1329 * @param rf The action that curr reads from. Must be a write.
1330 * @return True if modification order edges were added; false otherwise
1332 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1334 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1337 ASSERT(curr->is_read());
1339 /* Iterate over all threads */
1340 for (i = 0; i < thrd_lists->size(); i++) {
1341 /* Iterate over actions in thread, starting from most recent */
1342 action_list_t *list = &(*thrd_lists)[i];
1343 action_list_t::reverse_iterator rit;
1344 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1345 ModelAction *act = *rit;
1348 * Include at most one act per-thread that "happens
1349 * before" curr. Don't consider reflexively.
1351 if (act->happens_before(curr) && act != curr) {
1352 if (act->is_write()) {
1354 mo_graph->addEdge(act, rf);
1358 const ModelAction *prevreadfrom = act->get_reads_from();
1359 //if the previous read is unresolved, keep going...
1360 if (prevreadfrom == NULL)
1363 if (rf != prevreadfrom) {
1364 mo_graph->addEdge(prevreadfrom, rf);
1376 /** This method fixes up the modification order when we resolve a
1377 * promises. The basic problem is that actions that occur after the
1378 * read curr could not property add items to the modification order
1381 * So for each thread, we find the earliest item that happens after
1382 * the read curr. This is the item we have to fix up with additional
1383 * constraints. If that action is write, we add a MO edge between
1384 * the Action rf and that action. If the action is a read, we add a
1385 * MO edge between the Action rf, and whatever the read accessed.
1387 * @param curr is the read ModelAction that we are fixing up MO edges for.
1388 * @param rf is the write ModelAction that curr reads from.
1391 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1393 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1395 ASSERT(curr->is_read());
1397 /* Iterate over all threads */
1398 for (i = 0; i < thrd_lists->size(); i++) {
1399 /* Iterate over actions in thread, starting from most recent */
1400 action_list_t *list = &(*thrd_lists)[i];
1401 action_list_t::reverse_iterator rit;
1402 ModelAction *lastact = NULL;
1404 /* Find last action that happens after curr that is either not curr or a rmw */
1405 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1406 ModelAction *act = *rit;
1407 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1413 /* Include at most one act per-thread that "happens before" curr */
1414 if (lastact != NULL) {
1415 if (lastact==curr) {
1416 //Case 1: The resolved read is a RMW, and we need to make sure
1417 //that the write portion of the RMW mod order after rf
1419 mo_graph->addEdge(rf, lastact);
1420 } else if (lastact->is_read()) {
1421 //Case 2: The resolved read is a normal read and the next
1422 //operation is a read, and we need to make sure the value read
1423 //is mod ordered after rf
1425 const ModelAction *postreadfrom = lastact->get_reads_from();
1426 if (postreadfrom != NULL&&rf != postreadfrom)
1427 mo_graph->addEdge(rf, postreadfrom);
1429 //Case 3: The resolved read is a normal read and the next
1430 //operation is a write, and we need to make sure that the
1431 //write is mod ordered after rf
1433 mo_graph->addEdge(rf, lastact);
1441 * Updates the mo_graph with the constraints imposed from the current write.
1443 * Basic idea is the following: Go through each other thread and find
1444 * the lastest action that happened before our write. Two cases:
1446 * (1) The action is a write => that write must occur before
1449 * (2) The action is a read => the write that that action read from
1450 * must occur before the current write.
1452 * This method also handles two other issues:
1454 * (I) Sequential Consistency: Making sure that if the current write is
1455 * seq_cst, that it occurs after the previous seq_cst write.
1457 * (II) Sending the write back to non-synchronizing reads.
1459 * @param curr The current action. Must be a write.
1460 * @return True if modification order edges were added; false otherwise
1462 bool ModelChecker::w_modification_order(ModelAction *curr)
1464 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1467 ASSERT(curr->is_write());
1469 if (curr->is_seqcst()) {
1470 /* We have to at least see the last sequentially consistent write,
1471 so we are initialized. */
1472 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1473 if (last_seq_cst != NULL) {
1474 mo_graph->addEdge(last_seq_cst, curr);
1479 /* Iterate over all threads */
1480 for (i = 0; i < thrd_lists->size(); i++) {
1481 /* Iterate over actions in thread, starting from most recent */
1482 action_list_t *list = &(*thrd_lists)[i];
1483 action_list_t::reverse_iterator rit;
1484 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1485 ModelAction *act = *rit;
1488 * 1) If RMW and it actually read from something, then we
1489 * already have all relevant edges, so just skip to next
1492 * 2) If RMW and it didn't read from anything, we should
1493 * whatever edge we can get to speed up convergence.
1495 * 3) If normal write, we need to look at earlier actions, so
1496 * continue processing list.
1498 if (curr->is_rmw()) {
1499 if (curr->get_reads_from()!=NULL)
1508 * Include at most one act per-thread that "happens
1511 if (act->happens_before(curr)) {
1513 * Note: if act is RMW, just add edge:
1515 * The following edge should be handled elsewhere:
1516 * readfrom(act) --mo--> act
1518 if (act->is_write())
1519 mo_graph->addEdge(act, curr);
1520 else if (act->is_read()) {
1521 //if previous read accessed a null, just keep going
1522 if (act->get_reads_from() == NULL)
1524 mo_graph->addEdge(act->get_reads_from(), curr);
1528 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1529 !act->same_thread(curr)) {
1530 /* We have an action that:
1531 (1) did not happen before us
1532 (2) is a read and we are a write
1533 (3) cannot synchronize with us
1534 (4) is in a different thread
1536 that read could potentially read from our write. Note that
1537 these checks are overly conservative at this point, we'll
1538 do more checks before actually removing the
1542 if (thin_air_constraint_may_allow(curr, act)) {
1544 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1545 struct PendingFutureValue pfv = {curr,act};
1546 futurevalues->push_back(pfv);
1556 /** Arbitrary reads from the future are not allowed. Section 29.3
1557 * part 9 places some constraints. This method checks one result of constraint
1558 * constraint. Others require compiler support. */
1559 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1560 if (!writer->is_rmw())
1563 if (!reader->is_rmw())
1566 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1567 if (search == reader)
1569 if (search->get_tid() == reader->get_tid() &&
1570 search->happens_before(reader))
1578 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1579 * some constraints. This method checks one the following constraint (others
1580 * require compiler support):
1582 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1584 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1586 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1588 /* Iterate over all threads */
1589 for (i = 0; i < thrd_lists->size(); i++) {
1590 const ModelAction *write_after_read = NULL;
1592 /* Iterate over actions in thread, starting from most recent */
1593 action_list_t *list = &(*thrd_lists)[i];
1594 action_list_t::reverse_iterator rit;
1595 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1596 ModelAction *act = *rit;
1598 if (!reader->happens_before(act))
1600 else if (act->is_write())
1601 write_after_read = act;
1602 else if (act->is_read() && act->get_reads_from() != NULL && act != reader) {
1603 write_after_read = act->get_reads_from();
1607 if (write_after_read && write_after_read!=writer && mo_graph->checkReachable(write_after_read, writer))
1614 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1615 * The ModelAction under consideration is expected to be taking part in
1616 * release/acquire synchronization as an object of the "reads from" relation.
1617 * Note that this can only provide release sequence support for RMW chains
1618 * which do not read from the future, as those actions cannot be traced until
1619 * their "promise" is fulfilled. Similarly, we may not even establish the
1620 * presence of a release sequence with certainty, as some modification order
1621 * constraints may be decided further in the future. Thus, this function
1622 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1623 * and a boolean representing certainty.
1625 * @param rf The action that might be part of a release sequence. Must be a
1627 * @param release_heads A pass-by-reference style return parameter. After
1628 * execution of this function, release_heads will contain the heads of all the
1629 * relevant release sequences, if any exists with certainty
1630 * @param pending A pass-by-reference style return parameter which is only used
1631 * when returning false (i.e., uncertain). Returns most information regarding
1632 * an uncertain release sequence, including any write operations that might
1633 * break the sequence.
1634 * @return true, if the ModelChecker is certain that release_heads is complete;
1637 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1638 rel_heads_list_t *release_heads,
1639 struct release_seq *pending) const
1641 /* Only check for release sequences if there are no cycles */
1642 if (mo_graph->checkForCycles())
1646 ASSERT(rf->is_write());
1648 if (rf->is_release())
1649 release_heads->push_back(rf);
1651 break; /* End of RMW chain */
1653 /** @todo Need to be smarter here... In the linux lock
1654 * example, this will run to the beginning of the program for
1656 /** @todo The way to be smarter here is to keep going until 1
1657 * thread has a release preceded by an acquire and you've seen
1660 /* acq_rel RMW is a sufficient stopping condition */
1661 if (rf->is_acquire() && rf->is_release())
1662 return true; /* complete */
1664 rf = rf->get_reads_from();
1667 /* read from future: need to settle this later */
1669 return false; /* incomplete */
1672 if (rf->is_release())
1673 return true; /* complete */
1675 /* else relaxed write; check modification order for contiguous subsequence
1676 * -> rf must be same thread as release */
1677 int tid = id_to_int(rf->get_tid());
1678 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1679 action_list_t *list = &(*thrd_lists)[tid];
1680 action_list_t::const_reverse_iterator rit;
1682 /* Find rf in the thread list */
1683 rit = std::find(list->rbegin(), list->rend(), rf);
1684 ASSERT(rit != list->rend());
1686 /* Find the last write/release */
1687 for (; rit != list->rend(); rit++)
1688 if ((*rit)->is_release())
1690 if (rit == list->rend()) {
1691 /* No write-release in this thread */
1692 return true; /* complete */
1694 ModelAction *release = *rit;
1696 ASSERT(rf->same_thread(release));
1698 pending->writes.clear();
1700 bool certain = true;
1701 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1702 if (id_to_int(rf->get_tid()) == (int)i)
1704 list = &(*thrd_lists)[i];
1706 /* Can we ensure no future writes from this thread may break
1707 * the release seq? */
1708 bool future_ordered = false;
1710 ModelAction *last = get_last_action(int_to_id(i));
1711 Thread *th = get_thread(int_to_id(i));
1712 if ((last && rf->happens_before(last)) ||
1715 future_ordered = true;
1717 ASSERT(!th->is_model_thread() || future_ordered);
1719 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1720 const ModelAction *act = *rit;
1721 /* Reach synchronization -> this thread is complete */
1722 if (act->happens_before(release))
1724 if (rf->happens_before(act)) {
1725 future_ordered = true;
1729 /* Only non-RMW writes can break release sequences */
1730 if (!act->is_write() || act->is_rmw())
1733 /* Check modification order */
1734 if (mo_graph->checkReachable(rf, act)) {
1735 /* rf --mo--> act */
1736 future_ordered = true;
1739 if (mo_graph->checkReachable(act, release))
1740 /* act --mo--> release */
1742 if (mo_graph->checkReachable(release, act) &&
1743 mo_graph->checkReachable(act, rf)) {
1744 /* release --mo-> act --mo--> rf */
1745 return true; /* complete */
1747 /* act may break release sequence */
1748 pending->writes.push_back(act);
1751 if (!future_ordered)
1752 certain = false; /* This thread is uncertain */
1756 release_heads->push_back(release);
1757 pending->writes.clear();
1759 pending->release = release;
1766 * A public interface for getting the release sequence head(s) with which a
1767 * given ModelAction must synchronize. This function only returns a non-empty
1768 * result when it can locate a release sequence head with certainty. Otherwise,
1769 * it may mark the internal state of the ModelChecker so that it will handle
1770 * the release sequence at a later time, causing @a act to update its
1771 * synchronization at some later point in execution.
1772 * @param act The 'acquire' action that may read from a release sequence
1773 * @param release_heads A pass-by-reference return parameter. Will be filled
1774 * with the head(s) of the release sequence(s), if they exists with certainty.
1775 * @see ModelChecker::release_seq_heads
1777 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1779 const ModelAction *rf = act->get_reads_from();
1780 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1781 sequence->acquire = act;
1783 if (!release_seq_heads(rf, release_heads, sequence)) {
1784 /* add act to 'lazy checking' list */
1785 pending_rel_seqs->push_back(sequence);
1787 snapshot_free(sequence);
1792 * Attempt to resolve all stashed operations that might synchronize with a
1793 * release sequence for a given location. This implements the "lazy" portion of
1794 * determining whether or not a release sequence was contiguous, since not all
1795 * modification order information is present at the time an action occurs.
1797 * @param location The location/object that should be checked for release
1798 * sequence resolutions. A NULL value means to check all locations.
1799 * @param work_queue The work queue to which to add work items as they are
1801 * @return True if any updates occurred (new synchronization, new mo_graph
1804 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1806 bool updated = false;
1807 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1808 while (it != pending_rel_seqs->end()) {
1809 struct release_seq *pending = *it;
1810 ModelAction *act = pending->acquire;
1812 /* Only resolve sequences on the given location, if provided */
1813 if (location && act->get_location() != location) {
1818 const ModelAction *rf = act->get_reads_from();
1819 rel_heads_list_t release_heads;
1821 complete = release_seq_heads(rf, &release_heads, pending);
1822 for (unsigned int i = 0; i < release_heads.size(); i++) {
1823 if (!act->has_synchronized_with(release_heads[i])) {
1824 if (act->synchronize_with(release_heads[i]))
1827 set_bad_synchronization();
1832 /* Re-check all pending release sequences */
1833 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1834 /* Re-check act for mo_graph edges */
1835 work_queue->push_back(MOEdgeWorkEntry(act));
1837 /* propagate synchronization to later actions */
1838 action_list_t::reverse_iterator rit = action_trace->rbegin();
1839 for (; (*rit) != act; rit++) {
1840 ModelAction *propagate = *rit;
1841 if (act->happens_before(propagate)) {
1842 propagate->synchronize_with(act);
1843 /* Re-check 'propagate' for mo_graph edges */
1844 work_queue->push_back(MOEdgeWorkEntry(propagate));
1849 it = pending_rel_seqs->erase(it);
1850 snapshot_free(pending);
1856 // If we resolved promises or data races, see if we have realized a data race.
1863 * Performs various bookkeeping operations for the current ModelAction. For
1864 * instance, adds action to the per-object, per-thread action vector and to the
1865 * action trace list of all thread actions.
1867 * @param act is the ModelAction to add.
1869 void ModelChecker::add_action_to_lists(ModelAction *act)
1871 int tid = id_to_int(act->get_tid());
1872 action_trace->push_back(act);
1874 get_safe_ptr_action(obj_map, act->get_location())->push_back(act);
1876 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
1877 if (tid >= (int)vec->size())
1878 vec->resize(priv->next_thread_id);
1879 (*vec)[tid].push_back(act);
1881 if ((int)thrd_last_action->size() <= tid)
1882 thrd_last_action->resize(get_num_threads());
1883 (*thrd_last_action)[tid] = act;
1885 if (act->is_wait()) {
1886 void *mutex_loc=(void *) act->get_value();
1887 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
1889 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
1890 if (tid >= (int)vec->size())
1891 vec->resize(priv->next_thread_id);
1892 (*vec)[tid].push_back(act);
1894 if ((int)thrd_last_action->size() <= tid)
1895 thrd_last_action->resize(get_num_threads());
1896 (*thrd_last_action)[tid] = act;
1901 * @brief Get the last action performed by a particular Thread
1902 * @param tid The thread ID of the Thread in question
1903 * @return The last action in the thread
1905 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1907 int threadid = id_to_int(tid);
1908 if (threadid < (int)thrd_last_action->size())
1909 return (*thrd_last_action)[id_to_int(tid)];
1915 * Gets the last memory_order_seq_cst write (in the total global sequence)
1916 * performed on a particular object (i.e., memory location), not including the
1918 * @param curr The current ModelAction; also denotes the object location to
1920 * @return The last seq_cst write
1922 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1924 void *location = curr->get_location();
1925 action_list_t *list = get_safe_ptr_action(obj_map, location);
1926 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1927 action_list_t::reverse_iterator rit;
1928 for (rit = list->rbegin(); rit != list->rend(); rit++)
1929 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1935 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1936 * location). This function identifies the mutex according to the current
1937 * action, which is presumed to perform on the same mutex.
1938 * @param curr The current ModelAction; also denotes the object location to
1940 * @return The last unlock operation
1942 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1944 void *location = curr->get_location();
1945 action_list_t *list = get_safe_ptr_action(obj_map, location);
1946 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1947 action_list_t::reverse_iterator rit;
1948 for (rit = list->rbegin(); rit != list->rend(); rit++)
1949 if ((*rit)->is_unlock() || (*rit)->is_wait())
1954 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1956 ModelAction *parent = get_last_action(tid);
1958 parent = get_thread(tid)->get_creation();
1963 * Returns the clock vector for a given thread.
1964 * @param tid The thread whose clock vector we want
1965 * @return Desired clock vector
1967 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1969 return get_parent_action(tid)->get_cv();
1973 * Resolve a set of Promises with a current write. The set is provided in the
1974 * Node corresponding to @a write.
1975 * @param write The ModelAction that is fulfilling Promises
1976 * @return True if promises were resolved; false otherwise
1978 bool ModelChecker::resolve_promises(ModelAction *write)
1980 bool resolved = false;
1981 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1983 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1984 Promise *promise = (*promises)[promise_index];
1985 if (write->get_node()->get_promise(i)) {
1986 ModelAction *read = promise->get_action();
1987 if (read->is_rmw()) {
1988 mo_graph->addRMWEdge(write, read);
1990 read->read_from(write);
1991 //First fix up the modification order for actions that happened
1993 r_modification_order(read, write);
1994 //Next fix up the modification order for actions that happened
1996 post_r_modification_order(read, write);
1997 //Make sure the promise's value matches the write's value
1998 ASSERT(promise->get_value() == write->get_value());
2001 promises->erase(promises->begin() + promise_index);
2002 threads_to_check.push_back(read->get_tid());
2009 //Check whether reading these writes has made threads unable to
2012 for(unsigned int i=0;i<threads_to_check.size();i++)
2013 mo_check_promises(threads_to_check[i], write);
2019 * Compute the set of promises that could potentially be satisfied by this
2020 * action. Note that the set computation actually appears in the Node, not in
2022 * @param curr The ModelAction that may satisfy promises
2024 void ModelChecker::compute_promises(ModelAction *curr)
2026 for (unsigned int i = 0; i < promises->size(); i++) {
2027 Promise *promise = (*promises)[i];
2028 const ModelAction *act = promise->get_action();
2029 if (!act->happens_before(curr) &&
2031 !act->could_synchronize_with(curr) &&
2032 !act->same_thread(curr) &&
2033 act->get_location() == curr->get_location() &&
2034 promise->get_value() == curr->get_value()) {
2035 curr->get_node()->set_promise(i, act->is_rmw());
2040 /** Checks promises in response to change in ClockVector Threads. */
2041 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2043 for (unsigned int i = 0; i < promises->size(); i++) {
2044 Promise *promise = (*promises)[i];
2045 const ModelAction *act = promise->get_action();
2046 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
2047 merge_cv->synchronized_since(act)) {
2048 if (promise->increment_threads(tid)) {
2049 //Promise has failed
2050 failed_promise = true;
2057 void ModelChecker::check_promises_thread_disabled() {
2058 for (unsigned int i = 0; i < promises->size(); i++) {
2059 Promise *promise = (*promises)[i];
2060 if (promise->check_promise()) {
2061 failed_promise = true;
2067 /** Checks promises in response to addition to modification order for threads.
2069 * pthread is the thread that performed the read that created the promise
2071 * pread is the read that created the promise
2073 * pwrite is either the first write to same location as pread by
2074 * pthread that is sequenced after pread or the value read by the
2075 * first read to the same lcoation as pread by pthread that is
2076 * sequenced after pread..
2078 * 1. If tid=pthread, then we check what other threads are reachable
2079 * through the mode order starting with pwrite. Those threads cannot
2080 * perform a write that will resolve the promise due to modification
2081 * order constraints.
2083 * 2. If the tid is not pthread, we check whether pwrite can reach the
2084 * action write through the modification order. If so, that thread
2085 * cannot perform a future write that will resolve the promise due to
2086 * modificatin order constraints.
2088 * @parem tid The thread that either read from the model action
2089 * write, or actually did the model action write.
2091 * @parem write The ModelAction representing the relevant write.
2094 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
2095 void * location = write->get_location();
2096 for (unsigned int i = 0; i < promises->size(); i++) {
2097 Promise *promise = (*promises)[i];
2098 const ModelAction *act = promise->get_action();
2100 //Is this promise on the same location?
2101 if ( act->get_location() != location )
2104 //same thread as the promise
2105 if ( act->get_tid()==tid ) {
2107 //do we have a pwrite for the promise, if not, set it
2108 if (promise->get_write() == NULL ) {
2109 promise->set_write(write);
2110 //The pwrite cannot happen before the promise
2111 if (write->happens_before(act) && (write != act)) {
2112 failed_promise = true;
2116 if (mo_graph->checkPromise(write, promise)) {
2117 failed_promise = true;
2122 //Don't do any lookups twice for the same thread
2123 if (promise->has_sync_thread(tid))
2126 if (promise->get_write()&&mo_graph->checkReachable(promise->get_write(), write)) {
2127 if (promise->increment_threads(tid)) {
2128 failed_promise = true;
2136 * Compute the set of writes that may break the current pending release
2137 * sequence. This information is extracted from previou release sequence
2140 * @param curr The current ModelAction. Must be a release sequence fixup
2143 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2145 if (pending_rel_seqs->empty())
2148 struct release_seq *pending = pending_rel_seqs->back();
2149 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2150 const ModelAction *write = pending->writes[i];
2151 curr->get_node()->add_relseq_break(write);
2154 /* NULL means don't break the sequence; just synchronize */
2155 curr->get_node()->add_relseq_break(NULL);
2159 * Build up an initial set of all past writes that this 'read' action may read
2160 * from. This set is determined by the clock vector's "happens before"
2162 * @param curr is the current ModelAction that we are exploring; it must be a
2165 void ModelChecker::build_reads_from_past(ModelAction *curr)
2167 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2169 ASSERT(curr->is_read());
2171 ModelAction *last_seq_cst = NULL;
2173 /* Track whether this object has been initialized */
2174 bool initialized = false;
2176 if (curr->is_seqcst()) {
2177 last_seq_cst = get_last_seq_cst(curr);
2178 /* We have to at least see the last sequentially consistent write,
2179 so we are initialized. */
2180 if (last_seq_cst != NULL)
2184 /* Iterate over all threads */
2185 for (i = 0; i < thrd_lists->size(); i++) {
2186 /* Iterate over actions in thread, starting from most recent */
2187 action_list_t *list = &(*thrd_lists)[i];
2188 action_list_t::reverse_iterator rit;
2189 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2190 ModelAction *act = *rit;
2192 /* Only consider 'write' actions */
2193 if (!act->is_write() || act == curr)
2196 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2197 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
2198 if (!curr->get_sleep_flag() || curr->is_seqcst() || sleep_can_read_from(curr, act)) {
2199 DEBUG("Adding action to may_read_from:\n");
2200 if (DBG_ENABLED()) {
2204 curr->get_node()->add_read_from(act);
2208 /* Include at most one act per-thread that "happens before" curr */
2209 if (act->happens_before(curr)) {
2217 assert_bug("May read from uninitialized atomic");
2219 if (DBG_ENABLED() || !initialized) {
2220 model_print("Reached read action:\n");
2222 model_print("Printing may_read_from\n");
2223 curr->get_node()->print_may_read_from();
2224 model_print("End printing may_read_from\n");
2228 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2230 Node *prevnode=write->get_node()->get_parent();
2232 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2233 if (write->is_release()&&thread_sleep)
2235 if (!write->is_rmw()) {
2238 if (write->get_reads_from()==NULL)
2240 write=write->get_reads_from();
2244 static void print_list(action_list_t *list, int exec_num = -1)
2246 action_list_t::iterator it;
2248 model_print("---------------------------------------------------------------------\n");
2250 model_print("Execution %d:\n", exec_num);
2252 unsigned int hash=0;
2254 for (it = list->begin(); it != list->end(); it++) {
2256 hash=hash^(hash<<3)^((*it)->hash());
2258 model_print("HASH %u\n", hash);
2259 model_print("---------------------------------------------------------------------\n");
2262 #if SUPPORT_MOD_ORDER_DUMP
2263 void ModelChecker::dumpGraph(char *filename) {
2265 sprintf(buffer, "%s.dot",filename);
2266 FILE *file=fopen(buffer, "w");
2267 fprintf(file, "digraph %s {\n",filename);
2268 mo_graph->dumpNodes(file);
2269 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2271 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2272 ModelAction *action=*it;
2273 if (action->is_read()) {
2274 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2275 if (action->get_reads_from()!=NULL)
2276 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2278 if (thread_array[action->get_tid()] != NULL) {
2279 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2282 thread_array[action->get_tid()]=action;
2284 fprintf(file,"}\n");
2285 model_free(thread_array);
2290 void ModelChecker::print_summary()
2292 #if SUPPORT_MOD_ORDER_DUMP
2294 char buffername[100];
2295 sprintf(buffername, "exec%04u", stats.num_total);
2296 mo_graph->dumpGraphToFile(buffername);
2297 sprintf(buffername, "graph%04u", stats.num_total);
2298 dumpGraph(buffername);
2301 if (!isfinalfeasible())
2302 model_print("INFEASIBLE EXECUTION!\n");
2303 print_list(action_trace, stats.num_total);
2308 * Add a Thread to the system for the first time. Should only be called once
2310 * @param t The Thread to add
2312 void ModelChecker::add_thread(Thread *t)
2314 thread_map->put(id_to_int(t->get_id()), t);
2315 scheduler->add_thread(t);
2319 * Removes a thread from the scheduler.
2320 * @param the thread to remove.
2322 void ModelChecker::remove_thread(Thread *t)
2324 scheduler->remove_thread(t);
2328 * @brief Get a Thread reference by its ID
2329 * @param tid The Thread's ID
2330 * @return A Thread reference
2332 Thread * ModelChecker::get_thread(thread_id_t tid) const
2334 return thread_map->get(id_to_int(tid));
2338 * @brief Get a reference to the Thread in which a ModelAction was executed
2339 * @param act The ModelAction
2340 * @return A Thread reference
2342 Thread * ModelChecker::get_thread(ModelAction *act) const
2344 return get_thread(act->get_tid());
2348 * @brief Check if a Thread is currently enabled
2349 * @param t The Thread to check
2350 * @return True if the Thread is currently enabled
2352 bool ModelChecker::is_enabled(Thread *t) const
2354 return scheduler->is_enabled(t);
2358 * @brief Check if a Thread is currently enabled
2359 * @param tid The ID of the Thread to check
2360 * @return True if the Thread is currently enabled
2362 bool ModelChecker::is_enabled(thread_id_t tid) const
2364 return scheduler->is_enabled(tid);
2368 * Switch from a user-context to the "master thread" context (a.k.a. system
2369 * context). This switch is made with the intention of exploring a particular
2370 * model-checking action (described by a ModelAction object). Must be called
2371 * from a user-thread context.
2373 * @param act The current action that will be explored. May be NULL only if
2374 * trace is exiting via an assertion (see ModelChecker::set_assert and
2375 * ModelChecker::has_asserted).
2376 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2378 int ModelChecker::switch_to_master(ModelAction *act)
2381 Thread *old = thread_current();
2382 set_current_action(act);
2383 old->set_state(THREAD_READY);
2384 return Thread::swap(old, &system_context);
2388 * Takes the next step in the execution, if possible.
2389 * @return Returns true (success) if a step was taken and false otherwise.
2391 bool ModelChecker::take_step() {
2395 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2397 if (curr->get_state() == THREAD_READY) {
2398 ASSERT(priv->current_action);
2400 priv->nextThread = check_current_action(priv->current_action);
2401 priv->current_action = NULL;
2403 if (curr->is_blocked() || curr->is_complete())
2404 scheduler->remove_thread(curr);
2409 Thread *next = scheduler->next_thread(priv->nextThread);
2411 /* Infeasible -> don't take any more steps */
2414 else if (isfeasibleprefix() && have_bug_reports()) {
2419 if (params.bound != 0) {
2420 if (priv->used_sequence_numbers > params.bound) {
2425 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2426 next ? id_to_int(next->get_id()) : -1);
2429 * Launch end-of-execution release sequence fixups only when there are:
2431 * (1) no more user threads to run (or when execution replay chooses
2432 * the 'model_thread')
2433 * (2) pending release sequences
2434 * (3) pending assertions (i.e., data races)
2435 * (4) no pending promises
2437 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2438 isfinalfeasible() && !unrealizedraces.empty()) {
2439 model_print("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2440 pending_rel_seqs->size());
2441 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2442 std::memory_order_seq_cst, NULL, VALUE_NONE,
2444 set_current_action(fixup);
2448 /* next == NULL -> don't take any more steps */
2452 next->set_state(THREAD_RUNNING);
2454 if (next->get_pending() != NULL) {
2455 /* restart a pending action */
2456 set_current_action(next->get_pending());
2457 next->set_pending(NULL);
2458 next->set_state(THREAD_READY);
2462 /* Return false only if swap fails with an error */
2463 return (Thread::swap(&system_context, next) == 0);
2466 /** Runs the current execution until threre are no more steps to take. */
2467 void ModelChecker::finish_execution() {
2470 while (take_step());