9 #include "snapshot-interface.h"
11 #include "clockvector.h"
12 #include "cyclegraph.h"
15 #include "threads-model.h"
17 #define INITIAL_THREAD_ID 0
21 /** @brief Constructor */
22 ModelChecker::ModelChecker(struct model_params params) :
23 /* Initialize default scheduler */
25 scheduler(new Scheduler()),
27 num_feasible_executions(0),
29 earliest_diverge(NULL),
30 action_trace(new action_list_t()),
31 thread_map(new HashTable<int, Thread *, int>()),
32 obj_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
33 lock_waiters_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
34 condvar_waiters_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
35 obj_thrd_map(new HashTable<void *, std::vector<action_list_t>, uintptr_t, 4 >()),
36 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
37 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
38 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
39 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
40 node_stack(new NodeStack()),
41 mo_graph(new CycleGraph()),
42 failed_promise(false),
43 too_many_reads(false),
45 bad_synchronization(false)
47 /* Allocate this "size" on the snapshotting heap */
48 priv = (struct model_snapshot_members *)calloc(1, sizeof(*priv));
49 /* First thread created will have id INITIAL_THREAD_ID */
50 priv->next_thread_id = INITIAL_THREAD_ID;
52 /* Initialize a model-checker thread, for special ModelActions */
53 model_thread = new Thread(get_next_id());
54 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
57 /** @brief Destructor */
58 ModelChecker::~ModelChecker()
60 for (unsigned int i = 0; i < get_num_threads(); i++)
61 delete thread_map->get(i);
66 delete lock_waiters_map;
67 delete condvar_waiters_map;
70 for (unsigned int i = 0; i < promises->size(); i++)
71 delete (*promises)[i];
74 delete pending_rel_seqs;
76 delete thrd_last_action;
83 * Restores user program to initial state and resets all model-checker data
86 void ModelChecker::reset_to_initial_state()
88 DEBUG("+++ Resetting to initial state +++\n");
89 node_stack->reset_execution();
90 failed_promise = false;
91 too_many_reads = false;
92 bad_synchronization = false;
94 snapshotObject->backTrackBeforeStep(0);
97 /** @return a thread ID for a new Thread */
98 thread_id_t ModelChecker::get_next_id()
100 return priv->next_thread_id++;
103 /** @return the number of user threads created during this execution */
104 unsigned int ModelChecker::get_num_threads()
106 return priv->next_thread_id;
109 /** @return The currently executing Thread. */
110 Thread * ModelChecker::get_current_thread()
112 return scheduler->get_current_thread();
115 /** @return a sequence number for a new ModelAction */
116 modelclock_t ModelChecker::get_next_seq_num()
118 return ++priv->used_sequence_numbers;
121 Node * ModelChecker::get_curr_node() {
122 return node_stack->get_head();
126 * @brief Choose the next thread to execute.
128 * This function chooses the next thread that should execute. It can force the
129 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
130 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
131 * The model-checker may have no preference regarding the next thread (i.e.,
132 * when exploring a new execution ordering), in which case this will return
134 * @param curr The current ModelAction. This action might guide the choice of
136 * @return The next thread to run. If the model-checker has no preference, NULL.
138 Thread * ModelChecker::get_next_thread(ModelAction *curr)
143 /* Do not split atomic actions. */
145 return thread_current();
146 /* The THREAD_CREATE action points to the created Thread */
147 else if (curr->get_type() == THREAD_CREATE)
148 return (Thread *)curr->get_location();
151 /* Have we completed exploring the preselected path? */
155 /* Else, we are trying to replay an execution */
156 ModelAction *next = node_stack->get_next()->get_action();
158 if (next == diverge) {
159 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
160 earliest_diverge=diverge;
162 Node *nextnode = next->get_node();
163 Node *prevnode = nextnode->get_parent();
164 scheduler->update_sleep_set(prevnode);
166 /* Reached divergence point */
167 if (nextnode->increment_misc()) {
168 /* The next node will try to satisfy a different misc_index values. */
169 tid = next->get_tid();
170 node_stack->pop_restofstack(2);
171 } else if (nextnode->increment_promise()) {
172 /* The next node will try to satisfy a different set of promises. */
173 tid = next->get_tid();
174 node_stack->pop_restofstack(2);
175 } else if (nextnode->increment_read_from()) {
176 /* The next node will read from a different value. */
177 tid = next->get_tid();
178 node_stack->pop_restofstack(2);
179 } else if (nextnode->increment_future_value()) {
180 /* The next node will try to read from a different future value. */
181 tid = next->get_tid();
182 node_stack->pop_restofstack(2);
183 } else if (nextnode->increment_relseq_break()) {
184 /* The next node will try to resolve a release sequence differently */
185 tid = next->get_tid();
186 node_stack->pop_restofstack(2);
188 /* Make a different thread execute for next step */
189 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
190 tid = prevnode->get_next_backtrack();
191 /* Make sure the backtracked thread isn't sleeping. */
192 node_stack->pop_restofstack(1);
193 if (diverge==earliest_diverge) {
194 earliest_diverge=prevnode->get_action();
197 /* The correct sleep set is in the parent node. */
200 DEBUG("*** Divergence point ***\n");
204 tid = next->get_tid();
206 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
207 ASSERT(tid != THREAD_ID_T_NONE);
208 return thread_map->get(id_to_int(tid));
212 * We need to know what the next actions of all threads in the sleep
213 * set will be. This method computes them and stores the actions at
214 * the corresponding thread object's pending action.
217 void ModelChecker::execute_sleep_set() {
218 for(unsigned int i=0;i<get_num_threads();i++) {
219 thread_id_t tid=int_to_id(i);
220 Thread *thr=get_thread(tid);
221 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET &&
222 thr->get_pending() == NULL ) {
223 thr->set_state(THREAD_RUNNING);
224 scheduler->next_thread(thr);
225 Thread::swap(&system_context, thr);
226 priv->current_action->set_sleep_flag();
227 thr->set_pending(priv->current_action);
230 priv->current_action = NULL;
233 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
234 for(unsigned int i=0;i<get_num_threads();i++) {
235 thread_id_t tid=int_to_id(i);
236 Thread *thr=get_thread(tid);
237 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
238 ModelAction *pending_act=thr->get_pending();
239 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
240 //Remove this thread from sleep set
241 scheduler->remove_sleep(thr);
248 * Queries the model-checker for more executions to explore and, if one
249 * exists, resets the model-checker state to execute a new execution.
251 * @return If there are more executions to explore, return true. Otherwise,
254 bool ModelChecker::next_execution()
260 if (isfinalfeasible()) {
261 printf("Earliest divergence point since last feasible execution:\n");
262 if (earliest_diverge)
263 earliest_diverge->print();
265 printf("(Not set)\n");
267 earliest_diverge = NULL;
268 num_feasible_executions++;
271 DEBUG("Number of acquires waiting on pending release sequences: %zu\n",
272 pending_rel_seqs->size());
275 if (isfinalfeasible() || (params.bound != 0 && priv->used_sequence_numbers > params.bound ) || DBG_ENABLED() )
278 if ((diverge = get_next_backtrack()) == NULL)
282 printf("Next execution will diverge at:\n");
286 reset_to_initial_state();
290 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
292 switch (act->get_type()) {
296 /* linear search: from most recent to oldest */
297 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
298 action_list_t::reverse_iterator rit;
299 for (rit = list->rbegin(); rit != list->rend(); rit++) {
300 ModelAction *prev = *rit;
301 if (prev->could_synchronize_with(act))
307 case ATOMIC_TRYLOCK: {
308 /* linear search: from most recent to oldest */
309 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
310 action_list_t::reverse_iterator rit;
311 for (rit = list->rbegin(); rit != list->rend(); rit++) {
312 ModelAction *prev = *rit;
313 if (act->is_conflicting_lock(prev))
318 case ATOMIC_UNLOCK: {
319 /* linear search: from most recent to oldest */
320 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
321 action_list_t::reverse_iterator rit;
322 for (rit = list->rbegin(); rit != list->rend(); rit++) {
323 ModelAction *prev = *rit;
324 if (!act->same_thread(prev)&&prev->is_failed_trylock())
330 /* linear search: from most recent to oldest */
331 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
332 action_list_t::reverse_iterator rit;
333 for (rit = list->rbegin(); rit != list->rend(); rit++) {
334 ModelAction *prev = *rit;
335 if (!act->same_thread(prev)&&prev->is_failed_trylock())
337 if (!act->same_thread(prev)&&prev->is_notify())
343 case ATOMIC_NOTIFY_ALL:
344 case ATOMIC_NOTIFY_ONE: {
345 /* linear search: from most recent to oldest */
346 action_list_t *list = obj_map->get_safe_ptr(act->get_location());
347 action_list_t::reverse_iterator rit;
348 for (rit = list->rbegin(); rit != list->rend(); rit++) {
349 ModelAction *prev = *rit;
350 if (!act->same_thread(prev)&&prev->is_wait())
361 /** This method finds backtracking points where we should try to
362 * reorder the parameter ModelAction against.
364 * @param the ModelAction to find backtracking points for.
366 void ModelChecker::set_backtracking(ModelAction *act)
368 Thread *t = get_thread(act);
369 ModelAction * prev = get_last_conflict(act);
373 Node * node = prev->get_node()->get_parent();
375 int low_tid, high_tid;
376 if (node->is_enabled(t)) {
377 low_tid = id_to_int(act->get_tid());
378 high_tid = low_tid+1;
381 high_tid = get_num_threads();
384 for(int i = low_tid; i < high_tid; i++) {
385 thread_id_t tid = int_to_id(i);
387 /* Make sure this thread can be enabled here. */
388 if (i >= node->get_num_threads())
391 /* Don't backtrack into a point where the thread is disabled or sleeping. */
392 if (node->enabled_status(tid)!=THREAD_ENABLED)
395 /* Check if this has been explored already */
396 if (node->has_been_explored(tid))
399 /* See if fairness allows */
400 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
402 for(int t=0;t<node->get_num_threads();t++) {
403 thread_id_t tother=int_to_id(t);
404 if (node->is_enabled(tother) && node->has_priority(tother)) {
412 /* Cache the latest backtracking point */
413 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
414 priv->next_backtrack = prev;
416 /* If this is a new backtracking point, mark the tree */
417 if (!node->set_backtrack(tid))
419 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
420 id_to_int(prev->get_tid()),
421 id_to_int(t->get_id()));
430 * Returns last backtracking point. The model checker will explore a different
431 * path for this point in the next execution.
432 * @return The ModelAction at which the next execution should diverge.
434 ModelAction * ModelChecker::get_next_backtrack()
436 ModelAction *next = priv->next_backtrack;
437 priv->next_backtrack = NULL;
442 * Processes a read or rmw model action.
443 * @param curr is the read model action to process.
444 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
445 * @return True if processing this read updates the mo_graph.
447 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
449 uint64_t value = VALUE_NONE;
450 bool updated = false;
452 const ModelAction *reads_from = curr->get_node()->get_read_from();
453 if (reads_from != NULL) {
454 mo_graph->startChanges();
456 value = reads_from->get_value();
457 bool r_status = false;
459 if (!second_part_of_rmw) {
460 check_recency(curr, reads_from);
461 r_status = r_modification_order(curr, reads_from);
465 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
466 mo_graph->rollbackChanges();
467 too_many_reads = false;
471 curr->read_from(reads_from);
472 mo_graph->commitChanges();
473 mo_check_promises(curr->get_tid(), reads_from);
476 } else if (!second_part_of_rmw) {
477 /* Read from future value */
478 value = curr->get_node()->get_future_value();
479 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
480 curr->read_from(NULL);
481 Promise *valuepromise = new Promise(curr, value, expiration);
482 promises->push_back(valuepromise);
484 get_thread(curr)->set_return_value(value);
490 * Processes a lock, trylock, or unlock model action. @param curr is
491 * the read model action to process.
493 * The try lock operation checks whether the lock is taken. If not,
494 * it falls to the normal lock operation case. If so, it returns
497 * The lock operation has already been checked that it is enabled, so
498 * it just grabs the lock and synchronizes with the previous unlock.
500 * The unlock operation has to re-enable all of the threads that are
501 * waiting on the lock.
503 * @return True if synchronization was updated; false otherwise
505 bool ModelChecker::process_mutex(ModelAction *curr) {
506 std::mutex *mutex=NULL;
507 struct std::mutex_state *state=NULL;
509 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
510 mutex = (std::mutex *)curr->get_location();
511 state = mutex->get_state();
512 } else if(curr->is_wait()) {
513 mutex = (std::mutex *)curr->get_value();
514 state = mutex->get_state();
517 switch (curr->get_type()) {
518 case ATOMIC_TRYLOCK: {
519 bool success = !state->islocked;
520 curr->set_try_lock(success);
522 get_thread(curr)->set_return_value(0);
525 get_thread(curr)->set_return_value(1);
527 //otherwise fall into the lock case
529 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock) {
530 printf("Lock access before initialization\n");
533 state->islocked = true;
534 ModelAction *unlock = get_last_unlock(curr);
535 //synchronize with the previous unlock statement
536 if (unlock != NULL) {
537 curr->synchronize_with(unlock);
542 case ATOMIC_UNLOCK: {
544 state->islocked = false;
545 //wake up the other threads
546 action_list_t *waiters = lock_waiters_map->get_safe_ptr(curr->get_location());
547 //activate all the waiting threads
548 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
549 scheduler->wake(get_thread(*rit));
556 state->islocked = false;
557 //wake up the other threads
558 action_list_t *waiters = lock_waiters_map->get_safe_ptr((void *) curr->get_value());
559 //activate all the waiting threads
560 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
561 scheduler->wake(get_thread(*rit));
564 //check whether we should go to sleep or not...simulate spurious failures
565 if (curr->get_node()->get_misc()==0) {
566 condvar_waiters_map->get_safe_ptr(curr->get_location())->push_back(curr);
568 scheduler->sleep(get_current_thread());
572 case ATOMIC_NOTIFY_ALL: {
573 action_list_t *waiters = condvar_waiters_map->get_safe_ptr(curr->get_location());
574 //activate all the waiting threads
575 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
576 scheduler->wake(get_thread(*rit));
581 case ATOMIC_NOTIFY_ONE: {
582 action_list_t *waiters = condvar_waiters_map->get_safe_ptr(curr->get_location());
583 int wakeupthread=curr->get_node()->get_misc();
584 action_list_t::iterator it = waiters->begin();
585 advance(it, wakeupthread);
586 scheduler->wake(get_thread(*it));
598 * Process a write ModelAction
599 * @param curr The ModelAction to process
600 * @return True if the mo_graph was updated or promises were resolved
602 bool ModelChecker::process_write(ModelAction *curr)
604 bool updated_mod_order = w_modification_order(curr);
605 bool updated_promises = resolve_promises(curr);
607 if (promises->size() == 0) {
608 for (unsigned int i = 0; i < futurevalues->size(); i++) {
609 struct PendingFutureValue pfv = (*futurevalues)[i];
610 //Do more ambitious checks now that mo is more complete
611 if (mo_may_allow(pfv.writer, pfv.act)&&
612 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
613 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
614 priv->next_backtrack = pfv.act;
616 futurevalues->resize(0);
619 mo_graph->commitChanges();
620 mo_check_promises(curr->get_tid(), curr);
622 get_thread(curr)->set_return_value(VALUE_NONE);
623 return updated_mod_order || updated_promises;
627 * @brief Process the current action for thread-related activity
629 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
630 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
631 * synchronization, etc. This function is a no-op for non-THREAD actions
632 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
634 * @param curr The current action
635 * @return True if synchronization was updated or a thread completed
637 bool ModelChecker::process_thread_action(ModelAction *curr)
639 bool updated = false;
641 switch (curr->get_type()) {
642 case THREAD_CREATE: {
643 Thread *th = (Thread *)curr->get_location();
644 th->set_creation(curr);
648 Thread *blocking = (Thread *)curr->get_location();
649 ModelAction *act = get_last_action(blocking->get_id());
650 curr->synchronize_with(act);
651 updated = true; /* trigger rel-seq checks */
654 case THREAD_FINISH: {
655 Thread *th = get_thread(curr);
656 while (!th->wait_list_empty()) {
657 ModelAction *act = th->pop_wait_list();
658 scheduler->wake(get_thread(act));
661 updated = true; /* trigger rel-seq checks */
665 check_promises(curr->get_tid(), NULL, curr->get_cv());
676 * @brief Process the current action for release sequence fixup activity
678 * Performs model-checker release sequence fixups for the current action,
679 * forcing a single pending release sequence to break (with a given, potential
680 * "loose" write) or to complete (i.e., synchronize). If a pending release
681 * sequence forms a complete release sequence, then we must perform the fixup
682 * synchronization, mo_graph additions, etc.
684 * @param curr The current action; must be a release sequence fixup action
685 * @param work_queue The work queue to which to add work items as they are
688 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
690 const ModelAction *write = curr->get_node()->get_relseq_break();
691 struct release_seq *sequence = pending_rel_seqs->back();
692 pending_rel_seqs->pop_back();
694 ModelAction *acquire = sequence->acquire;
695 const ModelAction *rf = sequence->rf;
696 const ModelAction *release = sequence->release;
700 ASSERT(release->same_thread(rf));
704 * @todo Forcing a synchronization requires that we set
705 * modification order constraints. For instance, we can't allow
706 * a fixup sequence in which two separate read-acquire
707 * operations read from the same sequence, where the first one
708 * synchronizes and the other doesn't. Essentially, we can't
709 * allow any writes to insert themselves between 'release' and
713 /* Must synchronize */
714 if (!acquire->synchronize_with(release)) {
715 set_bad_synchronization();
718 /* Re-check all pending release sequences */
719 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
720 /* Re-check act for mo_graph edges */
721 work_queue->push_back(MOEdgeWorkEntry(acquire));
723 /* propagate synchronization to later actions */
724 action_list_t::reverse_iterator rit = action_trace->rbegin();
725 for (; (*rit) != acquire; rit++) {
726 ModelAction *propagate = *rit;
727 if (acquire->happens_before(propagate)) {
728 propagate->synchronize_with(acquire);
729 /* Re-check 'propagate' for mo_graph edges */
730 work_queue->push_back(MOEdgeWorkEntry(propagate));
734 /* Break release sequence with new edges:
735 * release --mo--> write --mo--> rf */
736 mo_graph->addEdge(release, write);
737 mo_graph->addEdge(write, rf);
740 /* See if we have realized a data race */
741 if (checkDataRaces())
746 * Initialize the current action by performing one or more of the following
747 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
748 * in the NodeStack, manipulating backtracking sets, allocating and
749 * initializing clock vectors, and computing the promises to fulfill.
751 * @param curr The current action, as passed from the user context; may be
752 * freed/invalidated after the execution of this function, with a different
753 * action "returned" its place (pass-by-reference)
754 * @return True if curr is a newly-explored action; false otherwise
756 bool ModelChecker::initialize_curr_action(ModelAction **curr)
758 ModelAction *newcurr;
760 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
761 newcurr = process_rmw(*curr);
764 if (newcurr->is_rmw())
765 compute_promises(newcurr);
771 (*curr)->set_seq_number(get_next_seq_num());
773 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled());
775 /* First restore type and order in case of RMW operation */
776 if ((*curr)->is_rmwr())
777 newcurr->copy_typeandorder(*curr);
779 ASSERT((*curr)->get_location() == newcurr->get_location());
780 newcurr->copy_from_new(*curr);
782 /* Discard duplicate ModelAction; use action from NodeStack */
785 /* Always compute new clock vector */
786 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
789 return false; /* Action was explored previously */
793 /* Always compute new clock vector */
794 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
796 * Perform one-time actions when pushing new ModelAction onto
799 if (newcurr->is_write())
800 compute_promises(newcurr);
801 else if (newcurr->is_relseq_fixup())
802 compute_relseq_breakwrites(newcurr);
803 else if (newcurr->is_wait())
804 newcurr->get_node()->set_misc_max(2);
805 else if (newcurr->is_notify_one()) {
806 newcurr->get_node()->set_misc_max(condvar_waiters_map->get_safe_ptr(newcurr->get_location())->size());
808 return true; /* This was a new ModelAction */
813 * @brief Check whether a model action is enabled.
815 * Checks whether a lock or join operation would be successful (i.e., is the
816 * lock already locked, or is the joined thread already complete). If not, put
817 * the action in a waiter list.
819 * @param curr is the ModelAction to check whether it is enabled.
820 * @return a bool that indicates whether the action is enabled.
822 bool ModelChecker::check_action_enabled(ModelAction *curr) {
823 if (curr->is_lock()) {
824 std::mutex * lock = (std::mutex *)curr->get_location();
825 struct std::mutex_state * state = lock->get_state();
826 if (state->islocked) {
827 //Stick the action in the appropriate waiting queue
828 lock_waiters_map->get_safe_ptr(curr->get_location())->push_back(curr);
831 } else if (curr->get_type() == THREAD_JOIN) {
832 Thread *blocking = (Thread *)curr->get_location();
833 if (!blocking->is_complete()) {
834 blocking->push_wait_list(curr);
843 * This is the heart of the model checker routine. It performs model-checking
844 * actions corresponding to a given "current action." Among other processes, it
845 * calculates reads-from relationships, updates synchronization clock vectors,
846 * forms a memory_order constraints graph, and handles replay/backtrack
847 * execution when running permutations of previously-observed executions.
849 * @param curr The current action to process
850 * @return The next Thread that must be executed. May be NULL if ModelChecker
851 * makes no choice (e.g., according to replay execution, combining RMW actions,
854 Thread * ModelChecker::check_current_action(ModelAction *curr)
857 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
859 if (!check_action_enabled(curr)) {
860 /* Make the execution look like we chose to run this action
861 * much later, when a lock/join can succeed */
862 get_current_thread()->set_pending(curr);
863 scheduler->sleep(get_current_thread());
864 return get_next_thread(NULL);
867 bool newly_explored = initialize_curr_action(&curr);
869 wake_up_sleeping_actions(curr);
871 /* Add the action to lists before any other model-checking tasks */
872 if (!second_part_of_rmw)
873 add_action_to_lists(curr);
875 /* Build may_read_from set for newly-created actions */
876 if (newly_explored && curr->is_read())
877 build_reads_from_past(curr);
879 /* Initialize work_queue with the "current action" work */
880 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
881 while (!work_queue.empty()) {
882 WorkQueueEntry work = work_queue.front();
883 work_queue.pop_front();
886 case WORK_CHECK_CURR_ACTION: {
887 ModelAction *act = work.action;
888 bool update = false; /* update this location's release seq's */
889 bool update_all = false; /* update all release seq's */
891 if (process_thread_action(curr))
894 if (act->is_read() && process_read(act, second_part_of_rmw))
897 if (act->is_write() && process_write(act))
900 if (act->is_mutex_op() && process_mutex(act))
903 if (act->is_relseq_fixup())
904 process_relseq_fixup(curr, &work_queue);
907 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
909 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
912 case WORK_CHECK_RELEASE_SEQ:
913 resolve_release_sequences(work.location, &work_queue);
915 case WORK_CHECK_MO_EDGES: {
916 /** @todo Complete verification of work_queue */
917 ModelAction *act = work.action;
918 bool updated = false;
920 if (act->is_read()) {
921 const ModelAction *rf = act->get_reads_from();
922 if (rf != NULL && r_modification_order(act, rf))
925 if (act->is_write()) {
926 if (w_modification_order(act))
929 mo_graph->commitChanges();
932 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
941 check_curr_backtracking(curr);
942 set_backtracking(curr);
943 return get_next_thread(curr);
946 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
947 Node *currnode = curr->get_node();
948 Node *parnode = currnode->get_parent();
950 if ((!parnode->backtrack_empty() ||
951 !currnode->misc_empty() ||
952 !currnode->read_from_empty() ||
953 !currnode->future_value_empty() ||
954 !currnode->promise_empty() ||
955 !currnode->relseq_break_empty())
956 && (!priv->next_backtrack ||
957 *curr > *priv->next_backtrack)) {
958 priv->next_backtrack = curr;
962 bool ModelChecker::promises_expired() {
963 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
964 Promise *promise = (*promises)[promise_index];
965 if (promise->get_expiration()<priv->used_sequence_numbers) {
972 /** @return whether the current partial trace must be a prefix of a
974 bool ModelChecker::isfeasibleprefix() {
975 return promises->size() == 0 && pending_rel_seqs->size() == 0 && isfeasible();
978 /** @return whether the current partial trace is feasible. */
979 bool ModelChecker::isfeasible() {
980 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
981 DEBUG("Infeasible: RMW violation\n");
983 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
986 /** @return whether the current partial trace is feasible other than
987 * multiple RMW reading from the same store. */
988 bool ModelChecker::isfeasibleotherthanRMW() {
990 if (mo_graph->checkForCycles())
991 DEBUG("Infeasible: modification order cycles\n");
993 DEBUG("Infeasible: failed promise\n");
995 DEBUG("Infeasible: too many reads\n");
996 if (bad_synchronization)
997 DEBUG("Infeasible: bad synchronization ordering\n");
998 if (promises_expired())
999 DEBUG("Infeasible: promises expired\n");
1001 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
1004 /** Returns whether the current completed trace is feasible. */
1005 bool ModelChecker::isfinalfeasible() {
1006 if (DBG_ENABLED() && promises->size() != 0)
1007 DEBUG("Infeasible: unrevolved promises\n");
1009 return isfeasible() && promises->size() == 0;
1012 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1013 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1014 ModelAction *lastread = get_last_action(act->get_tid());
1015 lastread->process_rmw(act);
1016 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1017 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1018 mo_graph->commitChanges();
1024 * Checks whether a thread has read from the same write for too many times
1025 * without seeing the effects of a later write.
1028 * 1) there must a different write that we could read from that would satisfy the modification order,
1029 * 2) we must have read from the same value in excess of maxreads times, and
1030 * 3) that other write must have been in the reads_from set for maxreads times.
1032 * If so, we decide that the execution is no longer feasible.
1034 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1035 if (params.maxreads != 0) {
1037 if (curr->get_node()->get_read_from_size() <= 1)
1039 //Must make sure that execution is currently feasible... We could
1040 //accidentally clear by rolling back
1043 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1044 int tid = id_to_int(curr->get_tid());
1047 if ((int)thrd_lists->size() <= tid)
1049 action_list_t *list = &(*thrd_lists)[tid];
1051 action_list_t::reverse_iterator rit = list->rbegin();
1052 /* Skip past curr */
1053 for (; (*rit) != curr; rit++)
1055 /* go past curr now */
1058 action_list_t::reverse_iterator ritcopy = rit;
1059 //See if we have enough reads from the same value
1061 for (; count < params.maxreads; rit++,count++) {
1062 if (rit==list->rend())
1064 ModelAction *act = *rit;
1065 if (!act->is_read())
1068 if (act->get_reads_from() != rf)
1070 if (act->get_node()->get_read_from_size() <= 1)
1073 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1075 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1077 //Need a different write
1081 /* Test to see whether this is a feasible write to read from*/
1082 mo_graph->startChanges();
1083 r_modification_order(curr, write);
1084 bool feasiblereadfrom = isfeasible();
1085 mo_graph->rollbackChanges();
1087 if (!feasiblereadfrom)
1091 bool feasiblewrite = true;
1092 //new we need to see if this write works for everyone
1094 for (int loop = count; loop>0; loop--,rit++) {
1095 ModelAction *act=*rit;
1096 bool foundvalue = false;
1097 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1098 if (act->get_node()->get_read_from_at(j)==write) {
1104 feasiblewrite = false;
1108 if (feasiblewrite) {
1109 too_many_reads = true;
1117 * Updates the mo_graph with the constraints imposed from the current
1120 * Basic idea is the following: Go through each other thread and find
1121 * the lastest action that happened before our read. Two cases:
1123 * (1) The action is a write => that write must either occur before
1124 * the write we read from or be the write we read from.
1126 * (2) The action is a read => the write that that action read from
1127 * must occur before the write we read from or be the same write.
1129 * @param curr The current action. Must be a read.
1130 * @param rf The action that curr reads from. Must be a write.
1131 * @return True if modification order edges were added; false otherwise
1133 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1135 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1138 ASSERT(curr->is_read());
1140 /* Iterate over all threads */
1141 for (i = 0; i < thrd_lists->size(); i++) {
1142 /* Iterate over actions in thread, starting from most recent */
1143 action_list_t *list = &(*thrd_lists)[i];
1144 action_list_t::reverse_iterator rit;
1145 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1146 ModelAction *act = *rit;
1149 * Include at most one act per-thread that "happens
1150 * before" curr. Don't consider reflexively.
1152 if (act->happens_before(curr) && act != curr) {
1153 if (act->is_write()) {
1155 mo_graph->addEdge(act, rf);
1159 const ModelAction *prevreadfrom = act->get_reads_from();
1160 //if the previous read is unresolved, keep going...
1161 if (prevreadfrom == NULL)
1164 if (rf != prevreadfrom) {
1165 mo_graph->addEdge(prevreadfrom, rf);
1177 /** This method fixes up the modification order when we resolve a
1178 * promises. The basic problem is that actions that occur after the
1179 * read curr could not property add items to the modification order
1182 * So for each thread, we find the earliest item that happens after
1183 * the read curr. This is the item we have to fix up with additional
1184 * constraints. If that action is write, we add a MO edge between
1185 * the Action rf and that action. If the action is a read, we add a
1186 * MO edge between the Action rf, and whatever the read accessed.
1188 * @param curr is the read ModelAction that we are fixing up MO edges for.
1189 * @param rf is the write ModelAction that curr reads from.
1192 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1194 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1196 ASSERT(curr->is_read());
1198 /* Iterate over all threads */
1199 for (i = 0; i < thrd_lists->size(); i++) {
1200 /* Iterate over actions in thread, starting from most recent */
1201 action_list_t *list = &(*thrd_lists)[i];
1202 action_list_t::reverse_iterator rit;
1203 ModelAction *lastact = NULL;
1205 /* Find last action that happens after curr that is either not curr or a rmw */
1206 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1207 ModelAction *act = *rit;
1208 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1214 /* Include at most one act per-thread that "happens before" curr */
1215 if (lastact != NULL) {
1216 if (lastact==curr) {
1217 //Case 1: The resolved read is a RMW, and we need to make sure
1218 //that the write portion of the RMW mod order after rf
1220 mo_graph->addEdge(rf, lastact);
1221 } else if (lastact->is_read()) {
1222 //Case 2: The resolved read is a normal read and the next
1223 //operation is a read, and we need to make sure the value read
1224 //is mod ordered after rf
1226 const ModelAction *postreadfrom = lastact->get_reads_from();
1227 if (postreadfrom != NULL&&rf != postreadfrom)
1228 mo_graph->addEdge(rf, postreadfrom);
1230 //Case 3: The resolved read is a normal read and the next
1231 //operation is a write, and we need to make sure that the
1232 //write is mod ordered after rf
1234 mo_graph->addEdge(rf, lastact);
1242 * Updates the mo_graph with the constraints imposed from the current write.
1244 * Basic idea is the following: Go through each other thread and find
1245 * the lastest action that happened before our write. Two cases:
1247 * (1) The action is a write => that write must occur before
1250 * (2) The action is a read => the write that that action read from
1251 * must occur before the current write.
1253 * This method also handles two other issues:
1255 * (I) Sequential Consistency: Making sure that if the current write is
1256 * seq_cst, that it occurs after the previous seq_cst write.
1258 * (II) Sending the write back to non-synchronizing reads.
1260 * @param curr The current action. Must be a write.
1261 * @return True if modification order edges were added; false otherwise
1263 bool ModelChecker::w_modification_order(ModelAction *curr)
1265 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1268 ASSERT(curr->is_write());
1270 if (curr->is_seqcst()) {
1271 /* We have to at least see the last sequentially consistent write,
1272 so we are initialized. */
1273 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1274 if (last_seq_cst != NULL) {
1275 mo_graph->addEdge(last_seq_cst, curr);
1280 /* Iterate over all threads */
1281 for (i = 0; i < thrd_lists->size(); i++) {
1282 /* Iterate over actions in thread, starting from most recent */
1283 action_list_t *list = &(*thrd_lists)[i];
1284 action_list_t::reverse_iterator rit;
1285 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1286 ModelAction *act = *rit;
1289 * 1) If RMW and it actually read from something, then we
1290 * already have all relevant edges, so just skip to next
1293 * 2) If RMW and it didn't read from anything, we should
1294 * whatever edge we can get to speed up convergence.
1296 * 3) If normal write, we need to look at earlier actions, so
1297 * continue processing list.
1299 if (curr->is_rmw()) {
1300 if (curr->get_reads_from()!=NULL)
1309 * Include at most one act per-thread that "happens
1312 if (act->happens_before(curr)) {
1314 * Note: if act is RMW, just add edge:
1316 * The following edge should be handled elsewhere:
1317 * readfrom(act) --mo--> act
1319 if (act->is_write())
1320 mo_graph->addEdge(act, curr);
1321 else if (act->is_read()) {
1322 //if previous read accessed a null, just keep going
1323 if (act->get_reads_from() == NULL)
1325 mo_graph->addEdge(act->get_reads_from(), curr);
1329 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1330 !act->same_thread(curr)) {
1331 /* We have an action that:
1332 (1) did not happen before us
1333 (2) is a read and we are a write
1334 (3) cannot synchronize with us
1335 (4) is in a different thread
1337 that read could potentially read from our write. Note that
1338 these checks are overly conservative at this point, we'll
1339 do more checks before actually removing the
1343 if (thin_air_constraint_may_allow(curr, act)) {
1345 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1346 struct PendingFutureValue pfv = {curr,act};
1347 futurevalues->push_back(pfv);
1357 /** Arbitrary reads from the future are not allowed. Section 29.3
1358 * part 9 places some constraints. This method checks one result of constraint
1359 * constraint. Others require compiler support. */
1360 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1361 if (!writer->is_rmw())
1364 if (!reader->is_rmw())
1367 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1368 if (search == reader)
1370 if (search->get_tid() == reader->get_tid() &&
1371 search->happens_before(reader))
1379 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1380 * some constraints. This method checks one the following constraint (others
1381 * require compiler support):
1383 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1385 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1387 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(reader->get_location());
1390 /* Iterate over all threads */
1391 for (i = 0; i < thrd_lists->size(); i++) {
1392 ModelAction *write_after_read = NULL;
1394 /* Iterate over actions in thread, starting from most recent */
1395 action_list_t *list = &(*thrd_lists)[i];
1396 action_list_t::reverse_iterator rit;
1397 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1398 ModelAction *act = *rit;
1400 if (!reader->happens_before(act))
1402 else if (act->is_write())
1403 write_after_read = act;
1406 if (write_after_read && mo_graph->checkReachable(write_after_read, writer))
1414 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1415 * The ModelAction under consideration is expected to be taking part in
1416 * release/acquire synchronization as an object of the "reads from" relation.
1417 * Note that this can only provide release sequence support for RMW chains
1418 * which do not read from the future, as those actions cannot be traced until
1419 * their "promise" is fulfilled. Similarly, we may not even establish the
1420 * presence of a release sequence with certainty, as some modification order
1421 * constraints may be decided further in the future. Thus, this function
1422 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1423 * and a boolean representing certainty.
1425 * @param rf The action that might be part of a release sequence. Must be a
1427 * @param release_heads A pass-by-reference style return parameter. After
1428 * execution of this function, release_heads will contain the heads of all the
1429 * relevant release sequences, if any exists with certainty
1430 * @param pending A pass-by-reference style return parameter which is only used
1431 * when returning false (i.e., uncertain). Returns most information regarding
1432 * an uncertain release sequence, including any write operations that might
1433 * break the sequence.
1434 * @return true, if the ModelChecker is certain that release_heads is complete;
1437 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1438 rel_heads_list_t *release_heads,
1439 struct release_seq *pending) const
1441 /* Only check for release sequences if there are no cycles */
1442 if (mo_graph->checkForCycles())
1446 ASSERT(rf->is_write());
1448 if (rf->is_release())
1449 release_heads->push_back(rf);
1451 break; /* End of RMW chain */
1453 /** @todo Need to be smarter here... In the linux lock
1454 * example, this will run to the beginning of the program for
1456 /** @todo The way to be smarter here is to keep going until 1
1457 * thread has a release preceded by an acquire and you've seen
1460 /* acq_rel RMW is a sufficient stopping condition */
1461 if (rf->is_acquire() && rf->is_release())
1462 return true; /* complete */
1464 rf = rf->get_reads_from();
1467 /* read from future: need to settle this later */
1469 return false; /* incomplete */
1472 if (rf->is_release())
1473 return true; /* complete */
1475 /* else relaxed write; check modification order for contiguous subsequence
1476 * -> rf must be same thread as release */
1477 int tid = id_to_int(rf->get_tid());
1478 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(rf->get_location());
1479 action_list_t *list = &(*thrd_lists)[tid];
1480 action_list_t::const_reverse_iterator rit;
1482 /* Find rf in the thread list */
1483 rit = std::find(list->rbegin(), list->rend(), rf);
1484 ASSERT(rit != list->rend());
1486 /* Find the last write/release */
1487 for (; rit != list->rend(); rit++)
1488 if ((*rit)->is_release())
1490 if (rit == list->rend()) {
1491 /* No write-release in this thread */
1492 return true; /* complete */
1494 ModelAction *release = *rit;
1496 ASSERT(rf->same_thread(release));
1498 pending->writes.clear();
1500 bool certain = true;
1501 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1502 if (id_to_int(rf->get_tid()) == (int)i)
1504 list = &(*thrd_lists)[i];
1506 /* Can we ensure no future writes from this thread may break
1507 * the release seq? */
1508 bool future_ordered = false;
1510 ModelAction *last = get_last_action(int_to_id(i));
1511 Thread *th = get_thread(int_to_id(i));
1512 if ((last && rf->happens_before(last)) ||
1513 !scheduler->is_enabled(th) ||
1515 future_ordered = true;
1517 ASSERT(!th->is_model_thread() || future_ordered);
1519 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1520 const ModelAction *act = *rit;
1521 /* Reach synchronization -> this thread is complete */
1522 if (act->happens_before(release))
1524 if (rf->happens_before(act)) {
1525 future_ordered = true;
1529 /* Only non-RMW writes can break release sequences */
1530 if (!act->is_write() || act->is_rmw())
1533 /* Check modification order */
1534 if (mo_graph->checkReachable(rf, act)) {
1535 /* rf --mo--> act */
1536 future_ordered = true;
1539 if (mo_graph->checkReachable(act, release))
1540 /* act --mo--> release */
1542 if (mo_graph->checkReachable(release, act) &&
1543 mo_graph->checkReachable(act, rf)) {
1544 /* release --mo-> act --mo--> rf */
1545 return true; /* complete */
1547 /* act may break release sequence */
1548 pending->writes.push_back(act);
1551 if (!future_ordered)
1552 certain = false; /* This thread is uncertain */
1556 release_heads->push_back(release);
1557 pending->writes.clear();
1559 pending->release = release;
1566 * A public interface for getting the release sequence head(s) with which a
1567 * given ModelAction must synchronize. This function only returns a non-empty
1568 * result when it can locate a release sequence head with certainty. Otherwise,
1569 * it may mark the internal state of the ModelChecker so that it will handle
1570 * the release sequence at a later time, causing @a act to update its
1571 * synchronization at some later point in execution.
1572 * @param act The 'acquire' action that may read from a release sequence
1573 * @param release_heads A pass-by-reference return parameter. Will be filled
1574 * with the head(s) of the release sequence(s), if they exists with certainty.
1575 * @see ModelChecker::release_seq_heads
1577 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1579 const ModelAction *rf = act->get_reads_from();
1580 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1581 sequence->acquire = act;
1583 if (!release_seq_heads(rf, release_heads, sequence)) {
1584 /* add act to 'lazy checking' list */
1585 pending_rel_seqs->push_back(sequence);
1587 snapshot_free(sequence);
1592 * Attempt to resolve all stashed operations that might synchronize with a
1593 * release sequence for a given location. This implements the "lazy" portion of
1594 * determining whether or not a release sequence was contiguous, since not all
1595 * modification order information is present at the time an action occurs.
1597 * @param location The location/object that should be checked for release
1598 * sequence resolutions. A NULL value means to check all locations.
1599 * @param work_queue The work queue to which to add work items as they are
1601 * @return True if any updates occurred (new synchronization, new mo_graph
1604 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1606 bool updated = false;
1607 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1608 while (it != pending_rel_seqs->end()) {
1609 struct release_seq *pending = *it;
1610 ModelAction *act = pending->acquire;
1612 /* Only resolve sequences on the given location, if provided */
1613 if (location && act->get_location() != location) {
1618 const ModelAction *rf = act->get_reads_from();
1619 rel_heads_list_t release_heads;
1621 complete = release_seq_heads(rf, &release_heads, pending);
1622 for (unsigned int i = 0; i < release_heads.size(); i++) {
1623 if (!act->has_synchronized_with(release_heads[i])) {
1624 if (act->synchronize_with(release_heads[i]))
1627 set_bad_synchronization();
1632 /* Re-check all pending release sequences */
1633 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1634 /* Re-check act for mo_graph edges */
1635 work_queue->push_back(MOEdgeWorkEntry(act));
1637 /* propagate synchronization to later actions */
1638 action_list_t::reverse_iterator rit = action_trace->rbegin();
1639 for (; (*rit) != act; rit++) {
1640 ModelAction *propagate = *rit;
1641 if (act->happens_before(propagate)) {
1642 propagate->synchronize_with(act);
1643 /* Re-check 'propagate' for mo_graph edges */
1644 work_queue->push_back(MOEdgeWorkEntry(propagate));
1649 it = pending_rel_seqs->erase(it);
1650 snapshot_free(pending);
1656 // If we resolved promises or data races, see if we have realized a data race.
1657 if (checkDataRaces()) {
1665 * Performs various bookkeeping operations for the current ModelAction. For
1666 * instance, adds action to the per-object, per-thread action vector and to the
1667 * action trace list of all thread actions.
1669 * @param act is the ModelAction to add.
1671 void ModelChecker::add_action_to_lists(ModelAction *act)
1673 int tid = id_to_int(act->get_tid());
1674 action_trace->push_back(act);
1676 obj_map->get_safe_ptr(act->get_location())->push_back(act);
1678 std::vector<action_list_t> *vec = obj_thrd_map->get_safe_ptr(act->get_location());
1679 if (tid >= (int)vec->size())
1680 vec->resize(priv->next_thread_id);
1681 (*vec)[tid].push_back(act);
1683 if ((int)thrd_last_action->size() <= tid)
1684 thrd_last_action->resize(get_num_threads());
1685 (*thrd_last_action)[tid] = act;
1687 if (act->is_wait()) {
1688 void *mutex_loc=(void *) act->get_value();
1689 obj_map->get_safe_ptr(mutex_loc)->push_back(act);
1691 std::vector<action_list_t> *vec = obj_thrd_map->get_safe_ptr(mutex_loc);
1692 if (tid >= (int)vec->size())
1693 vec->resize(priv->next_thread_id);
1694 (*vec)[tid].push_back(act);
1696 if ((int)thrd_last_action->size() <= tid)
1697 thrd_last_action->resize(get_num_threads());
1698 (*thrd_last_action)[tid] = act;
1703 * @brief Get the last action performed by a particular Thread
1704 * @param tid The thread ID of the Thread in question
1705 * @return The last action in the thread
1707 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1709 int threadid = id_to_int(tid);
1710 if (threadid < (int)thrd_last_action->size())
1711 return (*thrd_last_action)[id_to_int(tid)];
1717 * Gets the last memory_order_seq_cst write (in the total global sequence)
1718 * performed on a particular object (i.e., memory location), not including the
1720 * @param curr The current ModelAction; also denotes the object location to
1722 * @return The last seq_cst write
1724 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1726 void *location = curr->get_location();
1727 action_list_t *list = obj_map->get_safe_ptr(location);
1728 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1729 action_list_t::reverse_iterator rit;
1730 for (rit = list->rbegin(); rit != list->rend(); rit++)
1731 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1737 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1738 * location). This function identifies the mutex according to the current
1739 * action, which is presumed to perform on the same mutex.
1740 * @param curr The current ModelAction; also denotes the object location to
1742 * @return The last unlock operation
1744 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1746 void *location = curr->get_location();
1747 action_list_t *list = obj_map->get_safe_ptr(location);
1748 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1749 action_list_t::reverse_iterator rit;
1750 for (rit = list->rbegin(); rit != list->rend(); rit++)
1751 if ((*rit)->is_unlock() || (*rit)->is_wait())
1756 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1758 ModelAction *parent = get_last_action(tid);
1760 parent = get_thread(tid)->get_creation();
1765 * Returns the clock vector for a given thread.
1766 * @param tid The thread whose clock vector we want
1767 * @return Desired clock vector
1769 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1771 return get_parent_action(tid)->get_cv();
1775 * Resolve a set of Promises with a current write. The set is provided in the
1776 * Node corresponding to @a write.
1777 * @param write The ModelAction that is fulfilling Promises
1778 * @return True if promises were resolved; false otherwise
1780 bool ModelChecker::resolve_promises(ModelAction *write)
1782 bool resolved = false;
1783 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1785 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1786 Promise *promise = (*promises)[promise_index];
1787 if (write->get_node()->get_promise(i)) {
1788 ModelAction *read = promise->get_action();
1789 if (read->is_rmw()) {
1790 mo_graph->addRMWEdge(write, read);
1792 read->read_from(write);
1793 //First fix up the modification order for actions that happened
1795 r_modification_order(read, write);
1796 //Next fix up the modification order for actions that happened
1798 post_r_modification_order(read, write);
1799 //Make sure the promise's value matches the write's value
1800 ASSERT(promise->get_value() == write->get_value());
1803 promises->erase(promises->begin() + promise_index);
1804 threads_to_check.push_back(read->get_tid());
1811 //Check whether reading these writes has made threads unable to
1814 for(unsigned int i=0;i<threads_to_check.size();i++)
1815 mo_check_promises(threads_to_check[i], write);
1821 * Compute the set of promises that could potentially be satisfied by this
1822 * action. Note that the set computation actually appears in the Node, not in
1824 * @param curr The ModelAction that may satisfy promises
1826 void ModelChecker::compute_promises(ModelAction *curr)
1828 for (unsigned int i = 0; i < promises->size(); i++) {
1829 Promise *promise = (*promises)[i];
1830 const ModelAction *act = promise->get_action();
1831 if (!act->happens_before(curr) &&
1833 !act->could_synchronize_with(curr) &&
1834 !act->same_thread(curr) &&
1835 act->get_location() == curr->get_location() &&
1836 promise->get_value() == curr->get_value()) {
1837 curr->get_node()->set_promise(i);
1842 /** Checks promises in response to change in ClockVector Threads. */
1843 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
1845 for (unsigned int i = 0; i < promises->size(); i++) {
1846 Promise *promise = (*promises)[i];
1847 const ModelAction *act = promise->get_action();
1848 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
1849 merge_cv->synchronized_since(act)) {
1850 if (promise->increment_threads(tid)) {
1851 //Promise has failed
1852 failed_promise = true;
1859 /** Checks promises in response to addition to modification order for threads.
1861 * pthread is the thread that performed the read that created the promise
1863 * pread is the read that created the promise
1865 * pwrite is either the first write to same location as pread by
1866 * pthread that is sequenced after pread or the value read by the
1867 * first read to the same lcoation as pread by pthread that is
1868 * sequenced after pread..
1870 * 1. If tid=pthread, then we check what other threads are reachable
1871 * through the mode order starting with pwrite. Those threads cannot
1872 * perform a write that will resolve the promise due to modification
1873 * order constraints.
1875 * 2. If the tid is not pthread, we check whether pwrite can reach the
1876 * action write through the modification order. If so, that thread
1877 * cannot perform a future write that will resolve the promise due to
1878 * modificatin order constraints.
1880 * @parem tid The thread that either read from the model action
1881 * write, or actually did the model action write.
1883 * @parem write The ModelAction representing the relevant write.
1886 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
1887 void * location = write->get_location();
1888 for (unsigned int i = 0; i < promises->size(); i++) {
1889 Promise *promise = (*promises)[i];
1890 const ModelAction *act = promise->get_action();
1892 //Is this promise on the same location?
1893 if ( act->get_location() != location )
1896 //same thread as the promise
1897 if ( act->get_tid()==tid ) {
1899 //do we have a pwrite for the promise, if not, set it
1900 if (promise->get_write() == NULL ) {
1901 promise->set_write(write);
1902 //The pwrite cannot happen before the promise
1903 if (write->happens_before(act) && (write != act)) {
1904 failed_promise = true;
1908 if (mo_graph->checkPromise(write, promise)) {
1909 failed_promise = true;
1914 //Don't do any lookups twice for the same thread
1915 if (promise->has_sync_thread(tid))
1918 if (mo_graph->checkReachable(promise->get_write(), write)) {
1919 if (promise->increment_threads(tid)) {
1920 failed_promise = true;
1928 * Compute the set of writes that may break the current pending release
1929 * sequence. This information is extracted from previou release sequence
1932 * @param curr The current ModelAction. Must be a release sequence fixup
1935 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
1937 if (pending_rel_seqs->empty())
1940 struct release_seq *pending = pending_rel_seqs->back();
1941 for (unsigned int i = 0; i < pending->writes.size(); i++) {
1942 const ModelAction *write = pending->writes[i];
1943 curr->get_node()->add_relseq_break(write);
1946 /* NULL means don't break the sequence; just synchronize */
1947 curr->get_node()->add_relseq_break(NULL);
1951 * Build up an initial set of all past writes that this 'read' action may read
1952 * from. This set is determined by the clock vector's "happens before"
1954 * @param curr is the current ModelAction that we are exploring; it must be a
1957 void ModelChecker::build_reads_from_past(ModelAction *curr)
1959 std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
1961 ASSERT(curr->is_read());
1963 ModelAction *last_seq_cst = NULL;
1965 /* Track whether this object has been initialized */
1966 bool initialized = false;
1968 if (curr->is_seqcst()) {
1969 last_seq_cst = get_last_seq_cst(curr);
1970 /* We have to at least see the last sequentially consistent write,
1971 so we are initialized. */
1972 if (last_seq_cst != NULL)
1976 /* Iterate over all threads */
1977 for (i = 0; i < thrd_lists->size(); i++) {
1978 /* Iterate over actions in thread, starting from most recent */
1979 action_list_t *list = &(*thrd_lists)[i];
1980 action_list_t::reverse_iterator rit;
1981 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1982 ModelAction *act = *rit;
1984 /* Only consider 'write' actions */
1985 if (!act->is_write() || act == curr)
1988 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
1989 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
1990 DEBUG("Adding action to may_read_from:\n");
1991 if (DBG_ENABLED()) {
1996 if (curr->get_sleep_flag() && ! curr->is_seqcst()) {
1997 if (sleep_can_read_from(curr, act))
1998 curr->get_node()->add_read_from(act);
2000 curr->get_node()->add_read_from(act);
2003 /* Include at most one act per-thread that "happens before" curr */
2004 if (act->happens_before(curr)) {
2012 /** @todo Need a more informative way of reporting errors. */
2013 printf("ERROR: may read from uninitialized atomic\n");
2017 if (DBG_ENABLED() || !initialized) {
2018 printf("Reached read action:\n");
2020 printf("Printing may_read_from\n");
2021 curr->get_node()->print_may_read_from();
2022 printf("End printing may_read_from\n");
2026 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2028 Node *prevnode=write->get_node()->get_parent();
2030 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2031 if (write->is_release()&&thread_sleep)
2033 if (!write->is_rmw()) {
2036 if (write->get_reads_from()==NULL)
2038 write=write->get_reads_from();
2042 static void print_list(action_list_t *list)
2044 action_list_t::iterator it;
2046 printf("---------------------------------------------------------------------\n");
2048 unsigned int hash=0;
2050 for (it = list->begin(); it != list->end(); it++) {
2052 hash=hash^(hash<<3)^((*it)->hash());
2054 printf("HASH %u\n", hash);
2055 printf("---------------------------------------------------------------------\n");
2058 #if SUPPORT_MOD_ORDER_DUMP
2059 void ModelChecker::dumpGraph(char *filename) {
2061 sprintf(buffer, "%s.dot",filename);
2062 FILE *file=fopen(buffer, "w");
2063 fprintf(file, "digraph %s {\n",filename);
2064 mo_graph->dumpNodes(file);
2065 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2067 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2068 ModelAction *action=*it;
2069 if (action->is_read()) {
2070 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2071 if (action->get_reads_from()!=NULL)
2072 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2074 if (thread_array[action->get_tid()] != NULL) {
2075 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2078 thread_array[action->get_tid()]=action;
2080 fprintf(file,"}\n");
2081 model_free(thread_array);
2086 void ModelChecker::print_summary()
2089 printf("Number of executions: %d\n", num_executions);
2090 printf("Number of feasible executions: %d\n", num_feasible_executions);
2091 printf("Total nodes created: %d\n", node_stack->get_total_nodes());
2093 #if SUPPORT_MOD_ORDER_DUMP
2095 char buffername[100];
2096 sprintf(buffername, "exec%04u", num_executions);
2097 mo_graph->dumpGraphToFile(buffername);
2098 sprintf(buffername, "graph%04u", num_executions);
2099 dumpGraph(buffername);
2102 if (!isfinalfeasible())
2103 printf("INFEASIBLE EXECUTION!\n");
2104 print_list(action_trace);
2109 * Add a Thread to the system for the first time. Should only be called once
2111 * @param t The Thread to add
2113 void ModelChecker::add_thread(Thread *t)
2115 thread_map->put(id_to_int(t->get_id()), t);
2116 scheduler->add_thread(t);
2120 * Removes a thread from the scheduler.
2121 * @param the thread to remove.
2123 void ModelChecker::remove_thread(Thread *t)
2125 scheduler->remove_thread(t);
2129 * @brief Get a Thread reference by its ID
2130 * @param tid The Thread's ID
2131 * @return A Thread reference
2133 Thread * ModelChecker::get_thread(thread_id_t tid) const
2135 return thread_map->get(id_to_int(tid));
2139 * @brief Get a reference to the Thread in which a ModelAction was executed
2140 * @param act The ModelAction
2141 * @return A Thread reference
2143 Thread * ModelChecker::get_thread(ModelAction *act) const
2145 return get_thread(act->get_tid());
2149 * Switch from a user-context to the "master thread" context (a.k.a. system
2150 * context). This switch is made with the intention of exploring a particular
2151 * model-checking action (described by a ModelAction object). Must be called
2152 * from a user-thread context.
2154 * @param act The current action that will be explored. May be NULL only if
2155 * trace is exiting via an assertion (see ModelChecker::set_assert and
2156 * ModelChecker::has_asserted).
2157 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2159 int ModelChecker::switch_to_master(ModelAction *act)
2162 Thread *old = thread_current();
2163 set_current_action(act);
2164 old->set_state(THREAD_READY);
2165 return Thread::swap(old, &system_context);
2169 * Takes the next step in the execution, if possible.
2170 * @return Returns true (success) if a step was taken and false otherwise.
2172 bool ModelChecker::take_step() {
2176 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2178 if (curr->get_state() == THREAD_READY) {
2179 ASSERT(priv->current_action);
2181 priv->nextThread = check_current_action(priv->current_action);
2182 priv->current_action = NULL;
2184 if (curr->is_blocked() || curr->is_complete())
2185 scheduler->remove_thread(curr);
2190 Thread *next = scheduler->next_thread(priv->nextThread);
2192 /* Infeasible -> don't take any more steps */
2196 if (params.bound != 0) {
2197 if (priv->used_sequence_numbers > params.bound) {
2202 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2203 next ? id_to_int(next->get_id()) : -1);
2206 * Launch end-of-execution release sequence fixups only when there are:
2208 * (1) no more user threads to run (or when execution replay chooses
2209 * the 'model_thread')
2210 * (2) pending release sequences
2211 * (3) pending assertions (i.e., data races)
2212 * (4) no pending promises
2214 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2215 isfinalfeasible() && !unrealizedraces.empty()) {
2216 printf("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2217 pending_rel_seqs->size());
2218 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2219 std::memory_order_seq_cst, NULL, VALUE_NONE,
2221 set_current_action(fixup);
2225 /* next == NULL -> don't take any more steps */
2229 next->set_state(THREAD_RUNNING);
2231 if (next->get_pending() != NULL) {
2232 /* restart a pending action */
2233 set_current_action(next->get_pending());
2234 next->set_pending(NULL);
2235 next->set_state(THREAD_READY);
2239 /* Return false only if swap fails with an error */
2240 return (Thread::swap(&system_context, next) == 0);
2243 /** Runs the current execution until threre are no more steps to take. */
2244 void ModelChecker::finish_execution() {
2247 while (take_step());