9 #include "snapshot-interface.h"
11 #include "clockvector.h"
12 #include "cyclegraph.h"
15 #include "threads-model.h"
17 #define INITIAL_THREAD_ID 0
22 * Structure for holding small ModelChecker members that should be snapshotted
24 struct model_snapshot_members {
25 ModelAction *current_action;
26 unsigned int next_thread_id;
27 modelclock_t used_sequence_numbers;
29 ModelAction *next_backtrack;
32 /** @brief Constructor */
33 ModelChecker::ModelChecker(struct model_params params) :
34 /* Initialize default scheduler */
36 scheduler(new Scheduler()),
38 num_feasible_executions(0),
40 earliest_diverge(NULL),
41 action_trace(new action_list_t()),
42 thread_map(new HashTable<int, Thread *, int>()),
43 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
44 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
45 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
46 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
47 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
48 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
49 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
50 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
51 node_stack(new NodeStack()),
52 mo_graph(new CycleGraph()),
53 failed_promise(false),
54 too_many_reads(false),
56 bad_synchronization(false)
58 /* Allocate this "size" on the snapshotting heap */
59 priv = (struct model_snapshot_members *)snapshot_calloc(1, sizeof(*priv));
60 /* First thread created will have id INITIAL_THREAD_ID */
61 priv->next_thread_id = INITIAL_THREAD_ID;
63 /* Initialize a model-checker thread, for special ModelActions */
64 model_thread = new Thread(get_next_id());
65 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
68 /** @brief Destructor */
69 ModelChecker::~ModelChecker()
71 for (unsigned int i = 0; i < get_num_threads(); i++)
72 delete thread_map->get(i);
77 delete lock_waiters_map;
78 delete condvar_waiters_map;
81 for (unsigned int i = 0; i < promises->size(); i++)
82 delete (*promises)[i];
85 delete pending_rel_seqs;
87 delete thrd_last_action;
95 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr) {
96 action_list_t * tmp=hash->get(ptr);
98 tmp=new action_list_t();
104 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr) {
105 std::vector<action_list_t> * tmp=hash->get(ptr);
107 tmp=new std::vector<action_list_t>();
114 * Restores user program to initial state and resets all model-checker data
117 void ModelChecker::reset_to_initial_state()
119 DEBUG("+++ Resetting to initial state +++\n");
120 node_stack->reset_execution();
121 failed_promise = false;
122 too_many_reads = false;
123 bad_synchronization = false;
125 snapshotObject->backTrackBeforeStep(0);
128 /** @return a thread ID for a new Thread */
129 thread_id_t ModelChecker::get_next_id()
131 return priv->next_thread_id++;
134 /** @return the number of user threads created during this execution */
135 unsigned int ModelChecker::get_num_threads() const
137 return priv->next_thread_id;
140 /** @return The currently executing Thread. */
141 Thread * ModelChecker::get_current_thread()
143 return scheduler->get_current_thread();
146 /** @return a sequence number for a new ModelAction */
147 modelclock_t ModelChecker::get_next_seq_num()
149 return ++priv->used_sequence_numbers;
152 Node * ModelChecker::get_curr_node() {
153 return node_stack->get_head();
157 * @brief Choose the next thread to execute.
159 * This function chooses the next thread that should execute. It can force the
160 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
161 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
162 * The model-checker may have no preference regarding the next thread (i.e.,
163 * when exploring a new execution ordering), in which case this will return
165 * @param curr The current ModelAction. This action might guide the choice of
167 * @return The next thread to run. If the model-checker has no preference, NULL.
169 Thread * ModelChecker::get_next_thread(ModelAction *curr)
174 /* Do not split atomic actions. */
176 return thread_current();
177 /* The THREAD_CREATE action points to the created Thread */
178 else if (curr->get_type() == THREAD_CREATE)
179 return (Thread *)curr->get_location();
182 /* Have we completed exploring the preselected path? */
186 /* Else, we are trying to replay an execution */
187 ModelAction *next = node_stack->get_next()->get_action();
189 if (next == diverge) {
190 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
191 earliest_diverge=diverge;
193 Node *nextnode = next->get_node();
194 Node *prevnode = nextnode->get_parent();
195 scheduler->update_sleep_set(prevnode);
197 /* Reached divergence point */
198 if (nextnode->increment_misc()) {
199 /* The next node will try to satisfy a different misc_index values. */
200 tid = next->get_tid();
201 node_stack->pop_restofstack(2);
202 } else if (nextnode->increment_promise()) {
203 /* The next node will try to satisfy a different set of promises. */
204 tid = next->get_tid();
205 node_stack->pop_restofstack(2);
206 } else if (nextnode->increment_read_from()) {
207 /* The next node will read from a different value. */
208 tid = next->get_tid();
209 node_stack->pop_restofstack(2);
210 } else if (nextnode->increment_future_value()) {
211 /* The next node will try to read from a different future value. */
212 tid = next->get_tid();
213 node_stack->pop_restofstack(2);
214 } else if (nextnode->increment_relseq_break()) {
215 /* The next node will try to resolve a release sequence differently */
216 tid = next->get_tid();
217 node_stack->pop_restofstack(2);
219 /* Make a different thread execute for next step */
220 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
221 tid = prevnode->get_next_backtrack();
222 /* Make sure the backtracked thread isn't sleeping. */
223 node_stack->pop_restofstack(1);
224 if (diverge==earliest_diverge) {
225 earliest_diverge=prevnode->get_action();
228 /* The correct sleep set is in the parent node. */
231 DEBUG("*** Divergence point ***\n");
235 tid = next->get_tid();
237 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
238 ASSERT(tid != THREAD_ID_T_NONE);
239 return thread_map->get(id_to_int(tid));
243 * We need to know what the next actions of all threads in the sleep
244 * set will be. This method computes them and stores the actions at
245 * the corresponding thread object's pending action.
248 void ModelChecker::execute_sleep_set() {
249 for(unsigned int i=0;i<get_num_threads();i++) {
250 thread_id_t tid=int_to_id(i);
251 Thread *thr=get_thread(tid);
252 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET &&
253 thr->get_pending() == NULL ) {
254 thr->set_state(THREAD_RUNNING);
255 scheduler->next_thread(thr);
256 Thread::swap(&system_context, thr);
257 priv->current_action->set_sleep_flag();
258 thr->set_pending(priv->current_action);
261 priv->current_action = NULL;
264 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
265 for(unsigned int i=0;i<get_num_threads();i++) {
266 thread_id_t tid=int_to_id(i);
267 Thread *thr=get_thread(tid);
268 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
269 ModelAction *pending_act=thr->get_pending();
270 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
271 //Remove this thread from sleep set
272 scheduler->remove_sleep(thr);
279 * Check if we are in a deadlock. Should only be called at the end of an
280 * execution, although it should not give false positives in the middle of an
281 * execution (there should be some ENABLED thread).
283 * @return True if program is in a deadlock; false otherwise
285 bool ModelChecker::is_deadlocked() const
287 bool blocking_threads = false;
288 for (unsigned int i = 0; i < get_num_threads(); i++) {
289 thread_id_t tid = int_to_id(i);
292 Thread *t = get_thread(tid);
293 if (!t->is_model_thread() && t->get_pending())
294 blocking_threads = true;
296 return blocking_threads;
300 * Check if this is a complete execution. That is, have all thread completed
301 * execution (rather than exiting because sleep sets have forced a redundant
304 * @return True if the execution is complete.
306 bool ModelChecker::is_complete_execution() const
308 for (unsigned int i = 0; i < get_num_threads(); i++)
309 if (is_enabled(int_to_id(i)))
315 * Queries the model-checker for more executions to explore and, if one
316 * exists, resets the model-checker state to execute a new execution.
318 * @return If there are more executions to explore, return true. Otherwise,
321 bool ModelChecker::next_execution()
328 printf("ERROR: DEADLOCK\n");
329 if (isfinalfeasible()) {
330 printf("Earliest divergence point since last feasible execution:\n");
331 if (earliest_diverge)
332 earliest_diverge->print();
334 printf("(Not set)\n");
336 earliest_diverge = NULL;
337 num_feasible_executions++;
341 if (isfinalfeasible() || DBG_ENABLED()) {
346 if ((diverge = get_next_backtrack()) == NULL)
350 printf("Next execution will diverge at:\n");
354 reset_to_initial_state();
358 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
360 switch (act->get_type()) {
364 /* linear search: from most recent to oldest */
365 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
366 action_list_t::reverse_iterator rit;
367 for (rit = list->rbegin(); rit != list->rend(); rit++) {
368 ModelAction *prev = *rit;
369 if (prev->could_synchronize_with(act))
375 case ATOMIC_TRYLOCK: {
376 /* linear search: from most recent to oldest */
377 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
378 action_list_t::reverse_iterator rit;
379 for (rit = list->rbegin(); rit != list->rend(); rit++) {
380 ModelAction *prev = *rit;
381 if (act->is_conflicting_lock(prev))
386 case ATOMIC_UNLOCK: {
387 /* linear search: from most recent to oldest */
388 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
389 action_list_t::reverse_iterator rit;
390 for (rit = list->rbegin(); rit != list->rend(); rit++) {
391 ModelAction *prev = *rit;
392 if (!act->same_thread(prev)&&prev->is_failed_trylock())
398 /* linear search: from most recent to oldest */
399 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
400 action_list_t::reverse_iterator rit;
401 for (rit = list->rbegin(); rit != list->rend(); rit++) {
402 ModelAction *prev = *rit;
403 if (!act->same_thread(prev)&&prev->is_failed_trylock())
405 if (!act->same_thread(prev)&&prev->is_notify())
411 case ATOMIC_NOTIFY_ALL:
412 case ATOMIC_NOTIFY_ONE: {
413 /* linear search: from most recent to oldest */
414 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
415 action_list_t::reverse_iterator rit;
416 for (rit = list->rbegin(); rit != list->rend(); rit++) {
417 ModelAction *prev = *rit;
418 if (!act->same_thread(prev)&&prev->is_wait())
429 /** This method finds backtracking points where we should try to
430 * reorder the parameter ModelAction against.
432 * @param the ModelAction to find backtracking points for.
434 void ModelChecker::set_backtracking(ModelAction *act)
436 Thread *t = get_thread(act);
437 ModelAction * prev = get_last_conflict(act);
441 Node * node = prev->get_node()->get_parent();
443 int low_tid, high_tid;
444 if (node->is_enabled(t)) {
445 low_tid = id_to_int(act->get_tid());
446 high_tid = low_tid+1;
449 high_tid = get_num_threads();
452 for(int i = low_tid; i < high_tid; i++) {
453 thread_id_t tid = int_to_id(i);
455 /* Make sure this thread can be enabled here. */
456 if (i >= node->get_num_threads())
459 /* Don't backtrack into a point where the thread is disabled or sleeping. */
460 if (node->enabled_status(tid)!=THREAD_ENABLED)
463 /* Check if this has been explored already */
464 if (node->has_been_explored(tid))
467 /* See if fairness allows */
468 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
470 for(int t=0;t<node->get_num_threads();t++) {
471 thread_id_t tother=int_to_id(t);
472 if (node->is_enabled(tother) && node->has_priority(tother)) {
480 /* Cache the latest backtracking point */
481 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
482 priv->next_backtrack = prev;
484 /* If this is a new backtracking point, mark the tree */
485 if (!node->set_backtrack(tid))
487 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
488 id_to_int(prev->get_tid()),
489 id_to_int(t->get_id()));
498 * Returns last backtracking point. The model checker will explore a different
499 * path for this point in the next execution.
500 * @return The ModelAction at which the next execution should diverge.
502 ModelAction * ModelChecker::get_next_backtrack()
504 ModelAction *next = priv->next_backtrack;
505 priv->next_backtrack = NULL;
510 * Processes a read or rmw model action.
511 * @param curr is the read model action to process.
512 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
513 * @return True if processing this read updates the mo_graph.
515 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
517 uint64_t value = VALUE_NONE;
518 bool updated = false;
520 const ModelAction *reads_from = curr->get_node()->get_read_from();
521 if (reads_from != NULL) {
522 mo_graph->startChanges();
524 value = reads_from->get_value();
525 bool r_status = false;
527 if (!second_part_of_rmw) {
528 check_recency(curr, reads_from);
529 r_status = r_modification_order(curr, reads_from);
533 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
534 mo_graph->rollbackChanges();
535 too_many_reads = false;
539 curr->read_from(reads_from);
540 mo_graph->commitChanges();
541 mo_check_promises(curr->get_tid(), reads_from);
544 } else if (!second_part_of_rmw) {
545 /* Read from future value */
546 value = curr->get_node()->get_future_value();
547 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
548 curr->read_from(NULL);
549 Promise *valuepromise = new Promise(curr, value, expiration);
550 promises->push_back(valuepromise);
552 get_thread(curr)->set_return_value(value);
558 * Processes a lock, trylock, or unlock model action. @param curr is
559 * the read model action to process.
561 * The try lock operation checks whether the lock is taken. If not,
562 * it falls to the normal lock operation case. If so, it returns
565 * The lock operation has already been checked that it is enabled, so
566 * it just grabs the lock and synchronizes with the previous unlock.
568 * The unlock operation has to re-enable all of the threads that are
569 * waiting on the lock.
571 * @return True if synchronization was updated; false otherwise
573 bool ModelChecker::process_mutex(ModelAction *curr) {
574 std::mutex *mutex=NULL;
575 struct std::mutex_state *state=NULL;
577 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
578 mutex = (std::mutex *)curr->get_location();
579 state = mutex->get_state();
580 } else if(curr->is_wait()) {
581 mutex = (std::mutex *)curr->get_value();
582 state = mutex->get_state();
585 switch (curr->get_type()) {
586 case ATOMIC_TRYLOCK: {
587 bool success = !state->islocked;
588 curr->set_try_lock(success);
590 get_thread(curr)->set_return_value(0);
593 get_thread(curr)->set_return_value(1);
595 //otherwise fall into the lock case
597 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock) {
598 printf("Lock access before initialization\n");
601 state->islocked = true;
602 ModelAction *unlock = get_last_unlock(curr);
603 //synchronize with the previous unlock statement
604 if (unlock != NULL) {
605 curr->synchronize_with(unlock);
610 case ATOMIC_UNLOCK: {
612 state->islocked = false;
613 //wake up the other threads
614 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
615 //activate all the waiting threads
616 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
617 scheduler->wake(get_thread(*rit));
624 state->islocked = false;
625 //wake up the other threads
626 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
627 //activate all the waiting threads
628 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
629 scheduler->wake(get_thread(*rit));
632 //check whether we should go to sleep or not...simulate spurious failures
633 if (curr->get_node()->get_misc()==0) {
634 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
636 scheduler->sleep(get_current_thread());
640 case ATOMIC_NOTIFY_ALL: {
641 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
642 //activate all the waiting threads
643 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
644 scheduler->wake(get_thread(*rit));
649 case ATOMIC_NOTIFY_ONE: {
650 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
651 int wakeupthread=curr->get_node()->get_misc();
652 action_list_t::iterator it = waiters->begin();
653 advance(it, wakeupthread);
654 scheduler->wake(get_thread(*it));
666 * Process a write ModelAction
667 * @param curr The ModelAction to process
668 * @return True if the mo_graph was updated or promises were resolved
670 bool ModelChecker::process_write(ModelAction *curr)
672 bool updated_mod_order = w_modification_order(curr);
673 bool updated_promises = resolve_promises(curr);
675 if (promises->size() == 0) {
676 for (unsigned int i = 0; i < futurevalues->size(); i++) {
677 struct PendingFutureValue pfv = (*futurevalues)[i];
678 //Do more ambitious checks now that mo is more complete
679 if (mo_may_allow(pfv.writer, pfv.act)&&
680 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
681 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
682 priv->next_backtrack = pfv.act;
684 futurevalues->resize(0);
687 mo_graph->commitChanges();
688 mo_check_promises(curr->get_tid(), curr);
690 get_thread(curr)->set_return_value(VALUE_NONE);
691 return updated_mod_order || updated_promises;
695 * @brief Process the current action for thread-related activity
697 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
698 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
699 * synchronization, etc. This function is a no-op for non-THREAD actions
700 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
702 * @param curr The current action
703 * @return True if synchronization was updated or a thread completed
705 bool ModelChecker::process_thread_action(ModelAction *curr)
707 bool updated = false;
709 switch (curr->get_type()) {
710 case THREAD_CREATE: {
711 Thread *th = (Thread *)curr->get_location();
712 th->set_creation(curr);
716 Thread *blocking = (Thread *)curr->get_location();
717 ModelAction *act = get_last_action(blocking->get_id());
718 curr->synchronize_with(act);
719 updated = true; /* trigger rel-seq checks */
722 case THREAD_FINISH: {
723 Thread *th = get_thread(curr);
724 while (!th->wait_list_empty()) {
725 ModelAction *act = th->pop_wait_list();
726 scheduler->wake(get_thread(act));
729 updated = true; /* trigger rel-seq checks */
733 check_promises(curr->get_tid(), NULL, curr->get_cv());
744 * @brief Process the current action for release sequence fixup activity
746 * Performs model-checker release sequence fixups for the current action,
747 * forcing a single pending release sequence to break (with a given, potential
748 * "loose" write) or to complete (i.e., synchronize). If a pending release
749 * sequence forms a complete release sequence, then we must perform the fixup
750 * synchronization, mo_graph additions, etc.
752 * @param curr The current action; must be a release sequence fixup action
753 * @param work_queue The work queue to which to add work items as they are
756 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
758 const ModelAction *write = curr->get_node()->get_relseq_break();
759 struct release_seq *sequence = pending_rel_seqs->back();
760 pending_rel_seqs->pop_back();
762 ModelAction *acquire = sequence->acquire;
763 const ModelAction *rf = sequence->rf;
764 const ModelAction *release = sequence->release;
768 ASSERT(release->same_thread(rf));
772 * @todo Forcing a synchronization requires that we set
773 * modification order constraints. For instance, we can't allow
774 * a fixup sequence in which two separate read-acquire
775 * operations read from the same sequence, where the first one
776 * synchronizes and the other doesn't. Essentially, we can't
777 * allow any writes to insert themselves between 'release' and
781 /* Must synchronize */
782 if (!acquire->synchronize_with(release)) {
783 set_bad_synchronization();
786 /* Re-check all pending release sequences */
787 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
788 /* Re-check act for mo_graph edges */
789 work_queue->push_back(MOEdgeWorkEntry(acquire));
791 /* propagate synchronization to later actions */
792 action_list_t::reverse_iterator rit = action_trace->rbegin();
793 for (; (*rit) != acquire; rit++) {
794 ModelAction *propagate = *rit;
795 if (acquire->happens_before(propagate)) {
796 propagate->synchronize_with(acquire);
797 /* Re-check 'propagate' for mo_graph edges */
798 work_queue->push_back(MOEdgeWorkEntry(propagate));
802 /* Break release sequence with new edges:
803 * release --mo--> write --mo--> rf */
804 mo_graph->addEdge(release, write);
805 mo_graph->addEdge(write, rf);
808 /* See if we have realized a data race */
809 if (checkDataRaces())
814 * Initialize the current action by performing one or more of the following
815 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
816 * in the NodeStack, manipulating backtracking sets, allocating and
817 * initializing clock vectors, and computing the promises to fulfill.
819 * @param curr The current action, as passed from the user context; may be
820 * freed/invalidated after the execution of this function, with a different
821 * action "returned" its place (pass-by-reference)
822 * @return True if curr is a newly-explored action; false otherwise
824 bool ModelChecker::initialize_curr_action(ModelAction **curr)
826 ModelAction *newcurr;
828 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
829 newcurr = process_rmw(*curr);
832 if (newcurr->is_rmw())
833 compute_promises(newcurr);
839 (*curr)->set_seq_number(get_next_seq_num());
841 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled());
843 /* First restore type and order in case of RMW operation */
844 if ((*curr)->is_rmwr())
845 newcurr->copy_typeandorder(*curr);
847 ASSERT((*curr)->get_location() == newcurr->get_location());
848 newcurr->copy_from_new(*curr);
850 /* Discard duplicate ModelAction; use action from NodeStack */
853 /* Always compute new clock vector */
854 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
857 return false; /* Action was explored previously */
861 /* Always compute new clock vector */
862 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
864 * Perform one-time actions when pushing new ModelAction onto
867 if (newcurr->is_write())
868 compute_promises(newcurr);
869 else if (newcurr->is_relseq_fixup())
870 compute_relseq_breakwrites(newcurr);
871 else if (newcurr->is_wait())
872 newcurr->get_node()->set_misc_max(2);
873 else if (newcurr->is_notify_one()) {
874 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
876 return true; /* This was a new ModelAction */
881 * @brief Check whether a model action is enabled.
883 * Checks whether a lock or join operation would be successful (i.e., is the
884 * lock already locked, or is the joined thread already complete). If not, put
885 * the action in a waiter list.
887 * @param curr is the ModelAction to check whether it is enabled.
888 * @return a bool that indicates whether the action is enabled.
890 bool ModelChecker::check_action_enabled(ModelAction *curr) {
891 if (curr->is_lock()) {
892 std::mutex * lock = (std::mutex *)curr->get_location();
893 struct std::mutex_state * state = lock->get_state();
894 if (state->islocked) {
895 //Stick the action in the appropriate waiting queue
896 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
899 } else if (curr->get_type() == THREAD_JOIN) {
900 Thread *blocking = (Thread *)curr->get_location();
901 if (!blocking->is_complete()) {
902 blocking->push_wait_list(curr);
911 * Stores the ModelAction for the current thread action. Call this
912 * immediately before switching from user- to system-context to pass
914 * @param act The ModelAction created by the user-thread action
916 void ModelChecker::set_current_action(ModelAction *act) {
917 priv->current_action = act;
921 * This is the heart of the model checker routine. It performs model-checking
922 * actions corresponding to a given "current action." Among other processes, it
923 * calculates reads-from relationships, updates synchronization clock vectors,
924 * forms a memory_order constraints graph, and handles replay/backtrack
925 * execution when running permutations of previously-observed executions.
927 * @param curr The current action to process
928 * @return The next Thread that must be executed. May be NULL if ModelChecker
929 * makes no choice (e.g., according to replay execution, combining RMW actions,
932 Thread * ModelChecker::check_current_action(ModelAction *curr)
935 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
937 if (!check_action_enabled(curr)) {
938 /* Make the execution look like we chose to run this action
939 * much later, when a lock/join can succeed */
940 get_current_thread()->set_pending(curr);
941 scheduler->sleep(get_current_thread());
942 return get_next_thread(NULL);
945 bool newly_explored = initialize_curr_action(&curr);
947 wake_up_sleeping_actions(curr);
949 /* Add the action to lists before any other model-checking tasks */
950 if (!second_part_of_rmw)
951 add_action_to_lists(curr);
953 /* Build may_read_from set for newly-created actions */
954 if (newly_explored && curr->is_read())
955 build_reads_from_past(curr);
957 /* Initialize work_queue with the "current action" work */
958 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
959 while (!work_queue.empty()) {
960 WorkQueueEntry work = work_queue.front();
961 work_queue.pop_front();
964 case WORK_CHECK_CURR_ACTION: {
965 ModelAction *act = work.action;
966 bool update = false; /* update this location's release seq's */
967 bool update_all = false; /* update all release seq's */
969 if (process_thread_action(curr))
972 if (act->is_read() && process_read(act, second_part_of_rmw))
975 if (act->is_write() && process_write(act))
978 if (act->is_mutex_op() && process_mutex(act))
981 if (act->is_relseq_fixup())
982 process_relseq_fixup(curr, &work_queue);
985 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
987 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
990 case WORK_CHECK_RELEASE_SEQ:
991 resolve_release_sequences(work.location, &work_queue);
993 case WORK_CHECK_MO_EDGES: {
994 /** @todo Complete verification of work_queue */
995 ModelAction *act = work.action;
996 bool updated = false;
998 if (act->is_read()) {
999 const ModelAction *rf = act->get_reads_from();
1000 if (rf != NULL && r_modification_order(act, rf))
1003 if (act->is_write()) {
1004 if (w_modification_order(act))
1007 mo_graph->commitChanges();
1010 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1019 check_curr_backtracking(curr);
1020 set_backtracking(curr);
1021 return get_next_thread(curr);
1024 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
1025 Node *currnode = curr->get_node();
1026 Node *parnode = currnode->get_parent();
1028 if ((!parnode->backtrack_empty() ||
1029 !currnode->misc_empty() ||
1030 !currnode->read_from_empty() ||
1031 !currnode->future_value_empty() ||
1032 !currnode->promise_empty() ||
1033 !currnode->relseq_break_empty())
1034 && (!priv->next_backtrack ||
1035 *curr > *priv->next_backtrack)) {
1036 priv->next_backtrack = curr;
1040 bool ModelChecker::promises_expired() const
1042 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
1043 Promise *promise = (*promises)[promise_index];
1044 if (promise->get_expiration()<priv->used_sequence_numbers) {
1051 /** @return whether the current partial trace must be a prefix of a
1052 * feasible trace. */
1053 bool ModelChecker::isfeasibleprefix() const
1055 return promises->size() == 0 && pending_rel_seqs->size() == 0 && isfeasible();
1058 /** @return whether the current partial trace is feasible. */
1059 bool ModelChecker::isfeasible() const
1061 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
1062 DEBUG("Infeasible: RMW violation\n");
1064 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
1067 /** @return whether the current partial trace is feasible other than
1068 * multiple RMW reading from the same store. */
1069 bool ModelChecker::isfeasibleotherthanRMW() const
1071 if (DBG_ENABLED()) {
1072 if (mo_graph->checkForCycles())
1073 DEBUG("Infeasible: modification order cycles\n");
1075 DEBUG("Infeasible: failed promise\n");
1077 DEBUG("Infeasible: too many reads\n");
1078 if (bad_synchronization)
1079 DEBUG("Infeasible: bad synchronization ordering\n");
1080 if (promises_expired())
1081 DEBUG("Infeasible: promises expired\n");
1083 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
1086 /** Returns whether the current completed trace is feasible. */
1087 bool ModelChecker::isfinalfeasible() const
1089 if (DBG_ENABLED() && promises->size() != 0)
1090 DEBUG("Infeasible: unrevolved promises\n");
1092 return isfeasible() && promises->size() == 0;
1095 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1096 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1097 ModelAction *lastread = get_last_action(act->get_tid());
1098 lastread->process_rmw(act);
1099 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1100 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1101 mo_graph->commitChanges();
1107 * Checks whether a thread has read from the same write for too many times
1108 * without seeing the effects of a later write.
1111 * 1) there must a different write that we could read from that would satisfy the modification order,
1112 * 2) we must have read from the same value in excess of maxreads times, and
1113 * 3) that other write must have been in the reads_from set for maxreads times.
1115 * If so, we decide that the execution is no longer feasible.
1117 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1118 if (params.maxreads != 0) {
1120 if (curr->get_node()->get_read_from_size() <= 1)
1122 //Must make sure that execution is currently feasible... We could
1123 //accidentally clear by rolling back
1126 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1127 int tid = id_to_int(curr->get_tid());
1130 if ((int)thrd_lists->size() <= tid)
1132 action_list_t *list = &(*thrd_lists)[tid];
1134 action_list_t::reverse_iterator rit = list->rbegin();
1135 /* Skip past curr */
1136 for (; (*rit) != curr; rit++)
1138 /* go past curr now */
1141 action_list_t::reverse_iterator ritcopy = rit;
1142 //See if we have enough reads from the same value
1144 for (; count < params.maxreads; rit++,count++) {
1145 if (rit==list->rend())
1147 ModelAction *act = *rit;
1148 if (!act->is_read())
1151 if (act->get_reads_from() != rf)
1153 if (act->get_node()->get_read_from_size() <= 1)
1156 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1158 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1160 //Need a different write
1164 /* Test to see whether this is a feasible write to read from*/
1165 mo_graph->startChanges();
1166 r_modification_order(curr, write);
1167 bool feasiblereadfrom = isfeasible();
1168 mo_graph->rollbackChanges();
1170 if (!feasiblereadfrom)
1174 bool feasiblewrite = true;
1175 //new we need to see if this write works for everyone
1177 for (int loop = count; loop>0; loop--,rit++) {
1178 ModelAction *act=*rit;
1179 bool foundvalue = false;
1180 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1181 if (act->get_node()->get_read_from_at(j)==write) {
1187 feasiblewrite = false;
1191 if (feasiblewrite) {
1192 too_many_reads = true;
1200 * Updates the mo_graph with the constraints imposed from the current
1203 * Basic idea is the following: Go through each other thread and find
1204 * the lastest action that happened before our read. Two cases:
1206 * (1) The action is a write => that write must either occur before
1207 * the write we read from or be the write we read from.
1209 * (2) The action is a read => the write that that action read from
1210 * must occur before the write we read from or be the same write.
1212 * @param curr The current action. Must be a read.
1213 * @param rf The action that curr reads from. Must be a write.
1214 * @return True if modification order edges were added; false otherwise
1216 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1218 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1221 ASSERT(curr->is_read());
1223 /* Iterate over all threads */
1224 for (i = 0; i < thrd_lists->size(); i++) {
1225 /* Iterate over actions in thread, starting from most recent */
1226 action_list_t *list = &(*thrd_lists)[i];
1227 action_list_t::reverse_iterator rit;
1228 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1229 ModelAction *act = *rit;
1232 * Include at most one act per-thread that "happens
1233 * before" curr. Don't consider reflexively.
1235 if (act->happens_before(curr) && act != curr) {
1236 if (act->is_write()) {
1238 mo_graph->addEdge(act, rf);
1242 const ModelAction *prevreadfrom = act->get_reads_from();
1243 //if the previous read is unresolved, keep going...
1244 if (prevreadfrom == NULL)
1247 if (rf != prevreadfrom) {
1248 mo_graph->addEdge(prevreadfrom, rf);
1260 /** This method fixes up the modification order when we resolve a
1261 * promises. The basic problem is that actions that occur after the
1262 * read curr could not property add items to the modification order
1265 * So for each thread, we find the earliest item that happens after
1266 * the read curr. This is the item we have to fix up with additional
1267 * constraints. If that action is write, we add a MO edge between
1268 * the Action rf and that action. If the action is a read, we add a
1269 * MO edge between the Action rf, and whatever the read accessed.
1271 * @param curr is the read ModelAction that we are fixing up MO edges for.
1272 * @param rf is the write ModelAction that curr reads from.
1275 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1277 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1279 ASSERT(curr->is_read());
1281 /* Iterate over all threads */
1282 for (i = 0; i < thrd_lists->size(); i++) {
1283 /* Iterate over actions in thread, starting from most recent */
1284 action_list_t *list = &(*thrd_lists)[i];
1285 action_list_t::reverse_iterator rit;
1286 ModelAction *lastact = NULL;
1288 /* Find last action that happens after curr that is either not curr or a rmw */
1289 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1290 ModelAction *act = *rit;
1291 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1297 /* Include at most one act per-thread that "happens before" curr */
1298 if (lastact != NULL) {
1299 if (lastact==curr) {
1300 //Case 1: The resolved read is a RMW, and we need to make sure
1301 //that the write portion of the RMW mod order after rf
1303 mo_graph->addEdge(rf, lastact);
1304 } else if (lastact->is_read()) {
1305 //Case 2: The resolved read is a normal read and the next
1306 //operation is a read, and we need to make sure the value read
1307 //is mod ordered after rf
1309 const ModelAction *postreadfrom = lastact->get_reads_from();
1310 if (postreadfrom != NULL&&rf != postreadfrom)
1311 mo_graph->addEdge(rf, postreadfrom);
1313 //Case 3: The resolved read is a normal read and the next
1314 //operation is a write, and we need to make sure that the
1315 //write is mod ordered after rf
1317 mo_graph->addEdge(rf, lastact);
1325 * Updates the mo_graph with the constraints imposed from the current write.
1327 * Basic idea is the following: Go through each other thread and find
1328 * the lastest action that happened before our write. Two cases:
1330 * (1) The action is a write => that write must occur before
1333 * (2) The action is a read => the write that that action read from
1334 * must occur before the current write.
1336 * This method also handles two other issues:
1338 * (I) Sequential Consistency: Making sure that if the current write is
1339 * seq_cst, that it occurs after the previous seq_cst write.
1341 * (II) Sending the write back to non-synchronizing reads.
1343 * @param curr The current action. Must be a write.
1344 * @return True if modification order edges were added; false otherwise
1346 bool ModelChecker::w_modification_order(ModelAction *curr)
1348 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1351 ASSERT(curr->is_write());
1353 if (curr->is_seqcst()) {
1354 /* We have to at least see the last sequentially consistent write,
1355 so we are initialized. */
1356 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1357 if (last_seq_cst != NULL) {
1358 mo_graph->addEdge(last_seq_cst, curr);
1363 /* Iterate over all threads */
1364 for (i = 0; i < thrd_lists->size(); i++) {
1365 /* Iterate over actions in thread, starting from most recent */
1366 action_list_t *list = &(*thrd_lists)[i];
1367 action_list_t::reverse_iterator rit;
1368 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1369 ModelAction *act = *rit;
1372 * 1) If RMW and it actually read from something, then we
1373 * already have all relevant edges, so just skip to next
1376 * 2) If RMW and it didn't read from anything, we should
1377 * whatever edge we can get to speed up convergence.
1379 * 3) If normal write, we need to look at earlier actions, so
1380 * continue processing list.
1382 if (curr->is_rmw()) {
1383 if (curr->get_reads_from()!=NULL)
1392 * Include at most one act per-thread that "happens
1395 if (act->happens_before(curr)) {
1397 * Note: if act is RMW, just add edge:
1399 * The following edge should be handled elsewhere:
1400 * readfrom(act) --mo--> act
1402 if (act->is_write())
1403 mo_graph->addEdge(act, curr);
1404 else if (act->is_read()) {
1405 //if previous read accessed a null, just keep going
1406 if (act->get_reads_from() == NULL)
1408 mo_graph->addEdge(act->get_reads_from(), curr);
1412 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1413 !act->same_thread(curr)) {
1414 /* We have an action that:
1415 (1) did not happen before us
1416 (2) is a read and we are a write
1417 (3) cannot synchronize with us
1418 (4) is in a different thread
1420 that read could potentially read from our write. Note that
1421 these checks are overly conservative at this point, we'll
1422 do more checks before actually removing the
1426 if (thin_air_constraint_may_allow(curr, act)) {
1428 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1429 struct PendingFutureValue pfv = {curr,act};
1430 futurevalues->push_back(pfv);
1440 /** Arbitrary reads from the future are not allowed. Section 29.3
1441 * part 9 places some constraints. This method checks one result of constraint
1442 * constraint. Others require compiler support. */
1443 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1444 if (!writer->is_rmw())
1447 if (!reader->is_rmw())
1450 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1451 if (search == reader)
1453 if (search->get_tid() == reader->get_tid() &&
1454 search->happens_before(reader))
1462 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1463 * some constraints. This method checks one the following constraint (others
1464 * require compiler support):
1466 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1468 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1470 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1472 /* Iterate over all threads */
1473 for (i = 0; i < thrd_lists->size(); i++) {
1474 const ModelAction *write_after_read = NULL;
1476 /* Iterate over actions in thread, starting from most recent */
1477 action_list_t *list = &(*thrd_lists)[i];
1478 action_list_t::reverse_iterator rit;
1479 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1480 ModelAction *act = *rit;
1482 if (!reader->happens_before(act))
1484 else if (act->is_write())
1485 write_after_read = act;
1486 else if (act->is_read() && act->get_reads_from() != NULL && act != reader) {
1487 write_after_read = act->get_reads_from();
1491 if (write_after_read && write_after_read!=writer && mo_graph->checkReachable(write_after_read, writer))
1498 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1499 * The ModelAction under consideration is expected to be taking part in
1500 * release/acquire synchronization as an object of the "reads from" relation.
1501 * Note that this can only provide release sequence support for RMW chains
1502 * which do not read from the future, as those actions cannot be traced until
1503 * their "promise" is fulfilled. Similarly, we may not even establish the
1504 * presence of a release sequence with certainty, as some modification order
1505 * constraints may be decided further in the future. Thus, this function
1506 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1507 * and a boolean representing certainty.
1509 * @param rf The action that might be part of a release sequence. Must be a
1511 * @param release_heads A pass-by-reference style return parameter. After
1512 * execution of this function, release_heads will contain the heads of all the
1513 * relevant release sequences, if any exists with certainty
1514 * @param pending A pass-by-reference style return parameter which is only used
1515 * when returning false (i.e., uncertain). Returns most information regarding
1516 * an uncertain release sequence, including any write operations that might
1517 * break the sequence.
1518 * @return true, if the ModelChecker is certain that release_heads is complete;
1521 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1522 rel_heads_list_t *release_heads,
1523 struct release_seq *pending) const
1525 /* Only check for release sequences if there are no cycles */
1526 if (mo_graph->checkForCycles())
1530 ASSERT(rf->is_write());
1532 if (rf->is_release())
1533 release_heads->push_back(rf);
1535 break; /* End of RMW chain */
1537 /** @todo Need to be smarter here... In the linux lock
1538 * example, this will run to the beginning of the program for
1540 /** @todo The way to be smarter here is to keep going until 1
1541 * thread has a release preceded by an acquire and you've seen
1544 /* acq_rel RMW is a sufficient stopping condition */
1545 if (rf->is_acquire() && rf->is_release())
1546 return true; /* complete */
1548 rf = rf->get_reads_from();
1551 /* read from future: need to settle this later */
1553 return false; /* incomplete */
1556 if (rf->is_release())
1557 return true; /* complete */
1559 /* else relaxed write; check modification order for contiguous subsequence
1560 * -> rf must be same thread as release */
1561 int tid = id_to_int(rf->get_tid());
1562 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1563 action_list_t *list = &(*thrd_lists)[tid];
1564 action_list_t::const_reverse_iterator rit;
1566 /* Find rf in the thread list */
1567 rit = std::find(list->rbegin(), list->rend(), rf);
1568 ASSERT(rit != list->rend());
1570 /* Find the last write/release */
1571 for (; rit != list->rend(); rit++)
1572 if ((*rit)->is_release())
1574 if (rit == list->rend()) {
1575 /* No write-release in this thread */
1576 return true; /* complete */
1578 ModelAction *release = *rit;
1580 ASSERT(rf->same_thread(release));
1582 pending->writes.clear();
1584 bool certain = true;
1585 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1586 if (id_to_int(rf->get_tid()) == (int)i)
1588 list = &(*thrd_lists)[i];
1590 /* Can we ensure no future writes from this thread may break
1591 * the release seq? */
1592 bool future_ordered = false;
1594 ModelAction *last = get_last_action(int_to_id(i));
1595 Thread *th = get_thread(int_to_id(i));
1596 if ((last && rf->happens_before(last)) ||
1599 future_ordered = true;
1601 ASSERT(!th->is_model_thread() || future_ordered);
1603 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1604 const ModelAction *act = *rit;
1605 /* Reach synchronization -> this thread is complete */
1606 if (act->happens_before(release))
1608 if (rf->happens_before(act)) {
1609 future_ordered = true;
1613 /* Only non-RMW writes can break release sequences */
1614 if (!act->is_write() || act->is_rmw())
1617 /* Check modification order */
1618 if (mo_graph->checkReachable(rf, act)) {
1619 /* rf --mo--> act */
1620 future_ordered = true;
1623 if (mo_graph->checkReachable(act, release))
1624 /* act --mo--> release */
1626 if (mo_graph->checkReachable(release, act) &&
1627 mo_graph->checkReachable(act, rf)) {
1628 /* release --mo-> act --mo--> rf */
1629 return true; /* complete */
1631 /* act may break release sequence */
1632 pending->writes.push_back(act);
1635 if (!future_ordered)
1636 certain = false; /* This thread is uncertain */
1640 release_heads->push_back(release);
1641 pending->writes.clear();
1643 pending->release = release;
1650 * A public interface for getting the release sequence head(s) with which a
1651 * given ModelAction must synchronize. This function only returns a non-empty
1652 * result when it can locate a release sequence head with certainty. Otherwise,
1653 * it may mark the internal state of the ModelChecker so that it will handle
1654 * the release sequence at a later time, causing @a act to update its
1655 * synchronization at some later point in execution.
1656 * @param act The 'acquire' action that may read from a release sequence
1657 * @param release_heads A pass-by-reference return parameter. Will be filled
1658 * with the head(s) of the release sequence(s), if they exists with certainty.
1659 * @see ModelChecker::release_seq_heads
1661 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1663 const ModelAction *rf = act->get_reads_from();
1664 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1665 sequence->acquire = act;
1667 if (!release_seq_heads(rf, release_heads, sequence)) {
1668 /* add act to 'lazy checking' list */
1669 pending_rel_seqs->push_back(sequence);
1671 snapshot_free(sequence);
1676 * Attempt to resolve all stashed operations that might synchronize with a
1677 * release sequence for a given location. This implements the "lazy" portion of
1678 * determining whether or not a release sequence was contiguous, since not all
1679 * modification order information is present at the time an action occurs.
1681 * @param location The location/object that should be checked for release
1682 * sequence resolutions. A NULL value means to check all locations.
1683 * @param work_queue The work queue to which to add work items as they are
1685 * @return True if any updates occurred (new synchronization, new mo_graph
1688 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1690 bool updated = false;
1691 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1692 while (it != pending_rel_seqs->end()) {
1693 struct release_seq *pending = *it;
1694 ModelAction *act = pending->acquire;
1696 /* Only resolve sequences on the given location, if provided */
1697 if (location && act->get_location() != location) {
1702 const ModelAction *rf = act->get_reads_from();
1703 rel_heads_list_t release_heads;
1705 complete = release_seq_heads(rf, &release_heads, pending);
1706 for (unsigned int i = 0; i < release_heads.size(); i++) {
1707 if (!act->has_synchronized_with(release_heads[i])) {
1708 if (act->synchronize_with(release_heads[i]))
1711 set_bad_synchronization();
1716 /* Re-check all pending release sequences */
1717 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1718 /* Re-check act for mo_graph edges */
1719 work_queue->push_back(MOEdgeWorkEntry(act));
1721 /* propagate synchronization to later actions */
1722 action_list_t::reverse_iterator rit = action_trace->rbegin();
1723 for (; (*rit) != act; rit++) {
1724 ModelAction *propagate = *rit;
1725 if (act->happens_before(propagate)) {
1726 propagate->synchronize_with(act);
1727 /* Re-check 'propagate' for mo_graph edges */
1728 work_queue->push_back(MOEdgeWorkEntry(propagate));
1733 it = pending_rel_seqs->erase(it);
1734 snapshot_free(pending);
1740 // If we resolved promises or data races, see if we have realized a data race.
1741 if (checkDataRaces()) {
1749 * Performs various bookkeeping operations for the current ModelAction. For
1750 * instance, adds action to the per-object, per-thread action vector and to the
1751 * action trace list of all thread actions.
1753 * @param act is the ModelAction to add.
1755 void ModelChecker::add_action_to_lists(ModelAction *act)
1757 int tid = id_to_int(act->get_tid());
1758 action_trace->push_back(act);
1760 get_safe_ptr_action(obj_map, act->get_location())->push_back(act);
1762 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
1763 if (tid >= (int)vec->size())
1764 vec->resize(priv->next_thread_id);
1765 (*vec)[tid].push_back(act);
1767 if ((int)thrd_last_action->size() <= tid)
1768 thrd_last_action->resize(get_num_threads());
1769 (*thrd_last_action)[tid] = act;
1771 if (act->is_wait()) {
1772 void *mutex_loc=(void *) act->get_value();
1773 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
1775 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
1776 if (tid >= (int)vec->size())
1777 vec->resize(priv->next_thread_id);
1778 (*vec)[tid].push_back(act);
1780 if ((int)thrd_last_action->size() <= tid)
1781 thrd_last_action->resize(get_num_threads());
1782 (*thrd_last_action)[tid] = act;
1787 * @brief Get the last action performed by a particular Thread
1788 * @param tid The thread ID of the Thread in question
1789 * @return The last action in the thread
1791 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1793 int threadid = id_to_int(tid);
1794 if (threadid < (int)thrd_last_action->size())
1795 return (*thrd_last_action)[id_to_int(tid)];
1801 * Gets the last memory_order_seq_cst write (in the total global sequence)
1802 * performed on a particular object (i.e., memory location), not including the
1804 * @param curr The current ModelAction; also denotes the object location to
1806 * @return The last seq_cst write
1808 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1810 void *location = curr->get_location();
1811 action_list_t *list = get_safe_ptr_action(obj_map, location);
1812 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1813 action_list_t::reverse_iterator rit;
1814 for (rit = list->rbegin(); rit != list->rend(); rit++)
1815 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1821 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1822 * location). This function identifies the mutex according to the current
1823 * action, which is presumed to perform on the same mutex.
1824 * @param curr The current ModelAction; also denotes the object location to
1826 * @return The last unlock operation
1828 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1830 void *location = curr->get_location();
1831 action_list_t *list = get_safe_ptr_action(obj_map, location);
1832 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1833 action_list_t::reverse_iterator rit;
1834 for (rit = list->rbegin(); rit != list->rend(); rit++)
1835 if ((*rit)->is_unlock() || (*rit)->is_wait())
1840 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1842 ModelAction *parent = get_last_action(tid);
1844 parent = get_thread(tid)->get_creation();
1849 * Returns the clock vector for a given thread.
1850 * @param tid The thread whose clock vector we want
1851 * @return Desired clock vector
1853 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1855 return get_parent_action(tid)->get_cv();
1859 * Resolve a set of Promises with a current write. The set is provided in the
1860 * Node corresponding to @a write.
1861 * @param write The ModelAction that is fulfilling Promises
1862 * @return True if promises were resolved; false otherwise
1864 bool ModelChecker::resolve_promises(ModelAction *write)
1866 bool resolved = false;
1867 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1869 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1870 Promise *promise = (*promises)[promise_index];
1871 if (write->get_node()->get_promise(i)) {
1872 ModelAction *read = promise->get_action();
1873 if (read->is_rmw()) {
1874 mo_graph->addRMWEdge(write, read);
1876 read->read_from(write);
1877 //First fix up the modification order for actions that happened
1879 r_modification_order(read, write);
1880 //Next fix up the modification order for actions that happened
1882 post_r_modification_order(read, write);
1883 //Make sure the promise's value matches the write's value
1884 ASSERT(promise->get_value() == write->get_value());
1887 promises->erase(promises->begin() + promise_index);
1888 threads_to_check.push_back(read->get_tid());
1895 //Check whether reading these writes has made threads unable to
1898 for(unsigned int i=0;i<threads_to_check.size();i++)
1899 mo_check_promises(threads_to_check[i], write);
1905 * Compute the set of promises that could potentially be satisfied by this
1906 * action. Note that the set computation actually appears in the Node, not in
1908 * @param curr The ModelAction that may satisfy promises
1910 void ModelChecker::compute_promises(ModelAction *curr)
1912 for (unsigned int i = 0; i < promises->size(); i++) {
1913 Promise *promise = (*promises)[i];
1914 const ModelAction *act = promise->get_action();
1915 if (!act->happens_before(curr) &&
1917 !act->could_synchronize_with(curr) &&
1918 !act->same_thread(curr) &&
1919 act->get_location() == curr->get_location() &&
1920 promise->get_value() == curr->get_value()) {
1921 curr->get_node()->set_promise(i, act->is_rmw());
1926 /** Checks promises in response to change in ClockVector Threads. */
1927 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
1929 for (unsigned int i = 0; i < promises->size(); i++) {
1930 Promise *promise = (*promises)[i];
1931 const ModelAction *act = promise->get_action();
1932 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
1933 merge_cv->synchronized_since(act)) {
1934 if (promise->increment_threads(tid)) {
1935 //Promise has failed
1936 failed_promise = true;
1943 void ModelChecker::check_promises_thread_disabled() {
1944 for (unsigned int i = 0; i < promises->size(); i++) {
1945 Promise *promise = (*promises)[i];
1946 if (promise->check_promise()) {
1947 failed_promise = true;
1953 /** Checks promises in response to addition to modification order for threads.
1955 * pthread is the thread that performed the read that created the promise
1957 * pread is the read that created the promise
1959 * pwrite is either the first write to same location as pread by
1960 * pthread that is sequenced after pread or the value read by the
1961 * first read to the same lcoation as pread by pthread that is
1962 * sequenced after pread..
1964 * 1. If tid=pthread, then we check what other threads are reachable
1965 * through the mode order starting with pwrite. Those threads cannot
1966 * perform a write that will resolve the promise due to modification
1967 * order constraints.
1969 * 2. If the tid is not pthread, we check whether pwrite can reach the
1970 * action write through the modification order. If so, that thread
1971 * cannot perform a future write that will resolve the promise due to
1972 * modificatin order constraints.
1974 * @parem tid The thread that either read from the model action
1975 * write, or actually did the model action write.
1977 * @parem write The ModelAction representing the relevant write.
1980 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
1981 void * location = write->get_location();
1982 for (unsigned int i = 0; i < promises->size(); i++) {
1983 Promise *promise = (*promises)[i];
1984 const ModelAction *act = promise->get_action();
1986 //Is this promise on the same location?
1987 if ( act->get_location() != location )
1990 //same thread as the promise
1991 if ( act->get_tid()==tid ) {
1993 //do we have a pwrite for the promise, if not, set it
1994 if (promise->get_write() == NULL ) {
1995 promise->set_write(write);
1996 //The pwrite cannot happen before the promise
1997 if (write->happens_before(act) && (write != act)) {
1998 failed_promise = true;
2002 if (mo_graph->checkPromise(write, promise)) {
2003 failed_promise = true;
2008 //Don't do any lookups twice for the same thread
2009 if (promise->has_sync_thread(tid))
2012 if (promise->get_write()&&mo_graph->checkReachable(promise->get_write(), write)) {
2013 if (promise->increment_threads(tid)) {
2014 failed_promise = true;
2022 * Compute the set of writes that may break the current pending release
2023 * sequence. This information is extracted from previou release sequence
2026 * @param curr The current ModelAction. Must be a release sequence fixup
2029 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2031 if (pending_rel_seqs->empty())
2034 struct release_seq *pending = pending_rel_seqs->back();
2035 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2036 const ModelAction *write = pending->writes[i];
2037 curr->get_node()->add_relseq_break(write);
2040 /* NULL means don't break the sequence; just synchronize */
2041 curr->get_node()->add_relseq_break(NULL);
2045 * Build up an initial set of all past writes that this 'read' action may read
2046 * from. This set is determined by the clock vector's "happens before"
2048 * @param curr is the current ModelAction that we are exploring; it must be a
2051 void ModelChecker::build_reads_from_past(ModelAction *curr)
2053 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2055 ASSERT(curr->is_read());
2057 ModelAction *last_seq_cst = NULL;
2059 /* Track whether this object has been initialized */
2060 bool initialized = false;
2062 if (curr->is_seqcst()) {
2063 last_seq_cst = get_last_seq_cst(curr);
2064 /* We have to at least see the last sequentially consistent write,
2065 so we are initialized. */
2066 if (last_seq_cst != NULL)
2070 /* Iterate over all threads */
2071 for (i = 0; i < thrd_lists->size(); i++) {
2072 /* Iterate over actions in thread, starting from most recent */
2073 action_list_t *list = &(*thrd_lists)[i];
2074 action_list_t::reverse_iterator rit;
2075 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2076 ModelAction *act = *rit;
2078 /* Only consider 'write' actions */
2079 if (!act->is_write() || act == curr)
2082 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2083 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
2084 if (!curr->get_sleep_flag() || curr->is_seqcst() || sleep_can_read_from(curr, act)) {
2085 DEBUG("Adding action to may_read_from:\n");
2086 if (DBG_ENABLED()) {
2090 curr->get_node()->add_read_from(act);
2094 /* Include at most one act per-thread that "happens before" curr */
2095 if (act->happens_before(curr)) {
2103 /** @todo Need a more informative way of reporting errors. */
2104 printf("ERROR: may read from uninitialized atomic\n");
2108 if (DBG_ENABLED() || !initialized) {
2109 printf("Reached read action:\n");
2111 printf("Printing may_read_from\n");
2112 curr->get_node()->print_may_read_from();
2113 printf("End printing may_read_from\n");
2117 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2119 Node *prevnode=write->get_node()->get_parent();
2121 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2122 if (write->is_release()&&thread_sleep)
2124 if (!write->is_rmw()) {
2127 if (write->get_reads_from()==NULL)
2129 write=write->get_reads_from();
2133 static void print_list(action_list_t *list)
2135 action_list_t::iterator it;
2137 printf("---------------------------------------------------------------------\n");
2139 unsigned int hash=0;
2141 for (it = list->begin(); it != list->end(); it++) {
2143 hash=hash^(hash<<3)^((*it)->hash());
2145 printf("HASH %u\n", hash);
2146 printf("---------------------------------------------------------------------\n");
2149 #if SUPPORT_MOD_ORDER_DUMP
2150 void ModelChecker::dumpGraph(char *filename) {
2152 sprintf(buffer, "%s.dot",filename);
2153 FILE *file=fopen(buffer, "w");
2154 fprintf(file, "digraph %s {\n",filename);
2155 mo_graph->dumpNodes(file);
2156 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2158 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2159 ModelAction *action=*it;
2160 if (action->is_read()) {
2161 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2162 if (action->get_reads_from()!=NULL)
2163 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2165 if (thread_array[action->get_tid()] != NULL) {
2166 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2169 thread_array[action->get_tid()]=action;
2171 fprintf(file,"}\n");
2172 model_free(thread_array);
2177 void ModelChecker::print_summary()
2180 printf("Number of executions: %d\n", num_executions);
2181 printf("Number of feasible executions: %d\n", num_feasible_executions);
2182 printf("Total nodes created: %d\n", node_stack->get_total_nodes());
2184 #if SUPPORT_MOD_ORDER_DUMP
2186 char buffername[100];
2187 sprintf(buffername, "exec%04u", num_executions);
2188 mo_graph->dumpGraphToFile(buffername);
2189 sprintf(buffername, "graph%04u", num_executions);
2190 dumpGraph(buffername);
2193 if (!isfinalfeasible())
2194 printf("INFEASIBLE EXECUTION!\n");
2195 print_list(action_trace);
2200 * Add a Thread to the system for the first time. Should only be called once
2202 * @param t The Thread to add
2204 void ModelChecker::add_thread(Thread *t)
2206 thread_map->put(id_to_int(t->get_id()), t);
2207 scheduler->add_thread(t);
2211 * Removes a thread from the scheduler.
2212 * @param the thread to remove.
2214 void ModelChecker::remove_thread(Thread *t)
2216 scheduler->remove_thread(t);
2220 * @brief Get a Thread reference by its ID
2221 * @param tid The Thread's ID
2222 * @return A Thread reference
2224 Thread * ModelChecker::get_thread(thread_id_t tid) const
2226 return thread_map->get(id_to_int(tid));
2230 * @brief Get a reference to the Thread in which a ModelAction was executed
2231 * @param act The ModelAction
2232 * @return A Thread reference
2234 Thread * ModelChecker::get_thread(ModelAction *act) const
2236 return get_thread(act->get_tid());
2240 * @brief Check if a Thread is currently enabled
2241 * @param t The Thread to check
2242 * @return True if the Thread is currently enabled
2244 bool ModelChecker::is_enabled(Thread *t) const
2246 return scheduler->is_enabled(t);
2250 * @brief Check if a Thread is currently enabled
2251 * @param tid The ID of the Thread to check
2252 * @return True if the Thread is currently enabled
2254 bool ModelChecker::is_enabled(thread_id_t tid) const
2256 return scheduler->is_enabled(tid);
2260 * Switch from a user-context to the "master thread" context (a.k.a. system
2261 * context). This switch is made with the intention of exploring a particular
2262 * model-checking action (described by a ModelAction object). Must be called
2263 * from a user-thread context.
2265 * @param act The current action that will be explored. May be NULL only if
2266 * trace is exiting via an assertion (see ModelChecker::set_assert and
2267 * ModelChecker::has_asserted).
2268 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2270 int ModelChecker::switch_to_master(ModelAction *act)
2273 Thread *old = thread_current();
2274 set_current_action(act);
2275 old->set_state(THREAD_READY);
2276 return Thread::swap(old, &system_context);
2280 * Takes the next step in the execution, if possible.
2281 * @return Returns true (success) if a step was taken and false otherwise.
2283 bool ModelChecker::take_step() {
2287 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2289 if (curr->get_state() == THREAD_READY) {
2290 ASSERT(priv->current_action);
2292 priv->nextThread = check_current_action(priv->current_action);
2293 priv->current_action = NULL;
2295 if (curr->is_blocked() || curr->is_complete())
2296 scheduler->remove_thread(curr);
2301 Thread *next = scheduler->next_thread(priv->nextThread);
2303 /* Infeasible -> don't take any more steps */
2307 if (params.bound != 0) {
2308 if (priv->used_sequence_numbers > params.bound) {
2313 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2314 next ? id_to_int(next->get_id()) : -1);
2317 * Launch end-of-execution release sequence fixups only when there are:
2319 * (1) no more user threads to run (or when execution replay chooses
2320 * the 'model_thread')
2321 * (2) pending release sequences
2322 * (3) pending assertions (i.e., data races)
2323 * (4) no pending promises
2325 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2326 isfinalfeasible() && !unrealizedraces.empty()) {
2327 printf("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2328 pending_rel_seqs->size());
2329 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2330 std::memory_order_seq_cst, NULL, VALUE_NONE,
2332 set_current_action(fixup);
2336 /* next == NULL -> don't take any more steps */
2340 next->set_state(THREAD_RUNNING);
2342 if (next->get_pending() != NULL) {
2343 /* restart a pending action */
2344 set_current_action(next->get_pending());
2345 next->set_pending(NULL);
2346 next->set_state(THREAD_READY);
2350 /* Return false only if swap fails with an error */
2351 return (Thread::swap(&system_context, next) == 0);
2354 /** Runs the current execution until threre are no more steps to take. */
2355 void ModelChecker::finish_execution() {
2358 while (take_step());