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 bug_message(const char *str) {
23 const char *fmt = " [BUG] %s\n";
24 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
25 sprintf(msg, fmt, str);
27 ~bug_message() { if (msg) snapshot_free(msg); }
30 void print() { model_print("%s", msg); }
36 * Structure for holding small ModelChecker members that should be snapshotted
38 struct model_snapshot_members {
39 ModelAction *current_action;
40 unsigned int next_thread_id;
41 modelclock_t used_sequence_numbers;
43 ModelAction *next_backtrack;
44 std::vector< bug_message *, SnapshotAlloc<bug_message *> > bugs;
45 struct execution_stats stats;
48 /** @brief Constructor */
49 ModelChecker::ModelChecker(struct model_params params) :
50 /* Initialize default scheduler */
52 scheduler(new Scheduler()),
54 earliest_diverge(NULL),
55 action_trace(new action_list_t()),
56 thread_map(new HashTable<int, Thread *, int>()),
57 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
58 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
59 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
60 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
61 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
62 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
63 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
64 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
65 node_stack(new NodeStack()),
66 mo_graph(new CycleGraph()),
67 failed_promise(false),
68 too_many_reads(false),
70 bad_synchronization(false)
72 /* Allocate this "size" on the snapshotting heap */
73 priv = (struct model_snapshot_members *)snapshot_calloc(1, sizeof(*priv));
74 /* First thread created will have id INITIAL_THREAD_ID */
75 priv->next_thread_id = INITIAL_THREAD_ID;
77 /* Initialize a model-checker thread, for special ModelActions */
78 model_thread = new Thread(get_next_id());
79 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
82 /** @brief Destructor */
83 ModelChecker::~ModelChecker()
85 for (unsigned int i = 0; i < get_num_threads(); i++)
86 delete thread_map->get(i);
91 delete lock_waiters_map;
92 delete condvar_waiters_map;
95 for (unsigned int i = 0; i < promises->size(); i++)
96 delete (*promises)[i];
99 delete pending_rel_seqs;
101 delete thrd_last_action;
106 for (unsigned int i = 0; i < priv->bugs.size(); i++)
107 delete priv->bugs[i];
112 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr) {
113 action_list_t * tmp=hash->get(ptr);
115 tmp=new action_list_t();
121 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr) {
122 std::vector<action_list_t> * tmp=hash->get(ptr);
124 tmp=new std::vector<action_list_t>();
131 * Restores user program to initial state and resets all model-checker data
134 void ModelChecker::reset_to_initial_state()
136 DEBUG("+++ Resetting to initial state +++\n");
137 node_stack->reset_execution();
138 failed_promise = false;
139 too_many_reads = false;
140 bad_synchronization = false;
142 snapshotObject->backTrackBeforeStep(0);
145 /** @return a thread ID for a new Thread */
146 thread_id_t ModelChecker::get_next_id()
148 return priv->next_thread_id++;
151 /** @return the number of user threads created during this execution */
152 unsigned int ModelChecker::get_num_threads() const
154 return priv->next_thread_id;
157 /** @return The currently executing Thread. */
158 Thread * ModelChecker::get_current_thread()
160 return scheduler->get_current_thread();
163 /** @return a sequence number for a new ModelAction */
164 modelclock_t ModelChecker::get_next_seq_num()
166 return ++priv->used_sequence_numbers;
169 Node * ModelChecker::get_curr_node() {
170 return node_stack->get_head();
174 * @brief Choose the next thread to execute.
176 * This function chooses the next thread that should execute. It can force the
177 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
178 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
179 * The model-checker may have no preference regarding the next thread (i.e.,
180 * when exploring a new execution ordering), in which case this will return
182 * @param curr The current ModelAction. This action might guide the choice of
184 * @return The next thread to run. If the model-checker has no preference, NULL.
186 Thread * ModelChecker::get_next_thread(ModelAction *curr)
191 /* Do not split atomic actions. */
193 return thread_current();
194 /* The THREAD_CREATE action points to the created Thread */
195 else if (curr->get_type() == THREAD_CREATE)
196 return (Thread *)curr->get_location();
199 /* Have we completed exploring the preselected path? */
203 /* Else, we are trying to replay an execution */
204 ModelAction *next = node_stack->get_next()->get_action();
206 if (next == diverge) {
207 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
208 earliest_diverge=diverge;
210 Node *nextnode = next->get_node();
211 Node *prevnode = nextnode->get_parent();
212 scheduler->update_sleep_set(prevnode);
214 /* Reached divergence point */
215 if (nextnode->increment_misc()) {
216 /* The next node will try to satisfy a different misc_index values. */
217 tid = next->get_tid();
218 node_stack->pop_restofstack(2);
219 } else if (nextnode->increment_promise()) {
220 /* The next node will try to satisfy a different set of promises. */
221 tid = next->get_tid();
222 node_stack->pop_restofstack(2);
223 } else if (nextnode->increment_read_from()) {
224 /* The next node will read from a different value. */
225 tid = next->get_tid();
226 node_stack->pop_restofstack(2);
227 } else if (nextnode->increment_future_value()) {
228 /* The next node will try to read from a different future value. */
229 tid = next->get_tid();
230 node_stack->pop_restofstack(2);
231 } else if (nextnode->increment_relseq_break()) {
232 /* The next node will try to resolve a release sequence differently */
233 tid = next->get_tid();
234 node_stack->pop_restofstack(2);
236 /* Make a different thread execute for next step */
237 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
238 tid = prevnode->get_next_backtrack();
239 /* Make sure the backtracked thread isn't sleeping. */
240 node_stack->pop_restofstack(1);
241 if (diverge==earliest_diverge) {
242 earliest_diverge=prevnode->get_action();
245 /* The correct sleep set is in the parent node. */
248 DEBUG("*** Divergence point ***\n");
252 tid = next->get_tid();
254 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
255 ASSERT(tid != THREAD_ID_T_NONE);
256 return thread_map->get(id_to_int(tid));
260 * We need to know what the next actions of all threads in the sleep
261 * set will be. This method computes them and stores the actions at
262 * the corresponding thread object's pending action.
265 void ModelChecker::execute_sleep_set() {
266 for(unsigned int i=0;i<get_num_threads();i++) {
267 thread_id_t tid=int_to_id(i);
268 Thread *thr=get_thread(tid);
269 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET &&
270 thr->get_pending() == NULL ) {
271 thr->set_state(THREAD_RUNNING);
272 scheduler->next_thread(thr);
273 Thread::swap(&system_context, thr);
274 priv->current_action->set_sleep_flag();
275 thr->set_pending(priv->current_action);
278 priv->current_action = NULL;
281 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
282 for(unsigned int i=0;i<get_num_threads();i++) {
283 thread_id_t tid=int_to_id(i);
284 Thread *thr=get_thread(tid);
285 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
286 ModelAction *pending_act=thr->get_pending();
287 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
288 //Remove this thread from sleep set
289 scheduler->remove_sleep(thr);
296 * Check if we are in a deadlock. Should only be called at the end of an
297 * execution, although it should not give false positives in the middle of an
298 * execution (there should be some ENABLED thread).
300 * @return True if program is in a deadlock; false otherwise
302 bool ModelChecker::is_deadlocked() const
304 bool blocking_threads = false;
305 for (unsigned int i = 0; i < get_num_threads(); i++) {
306 thread_id_t tid = int_to_id(i);
309 Thread *t = get_thread(tid);
310 if (!t->is_model_thread() && t->get_pending())
311 blocking_threads = true;
313 return blocking_threads;
317 * Check if this is a complete execution. That is, have all thread completed
318 * execution (rather than exiting because sleep sets have forced a redundant
321 * @return True if the execution is complete.
323 bool ModelChecker::is_complete_execution() const
325 for (unsigned int i = 0; i < get_num_threads(); i++)
326 if (is_enabled(int_to_id(i)))
332 * @brief Assert a bug in the executing program.
334 * Use this function to assert any sort of bug in the user program. If the
335 * current trace is feasible (actually, a prefix of some feasible execution),
336 * then this execution will be aborted, printing the appropriate message. If
337 * the current trace is not yet feasible, the error message will be stashed and
338 * printed if the execution ever becomes feasible.
340 * @param msg Descriptive message for the bug (do not include newline char)
341 * @return True if bug is immediately-feasible
343 bool ModelChecker::assert_bug(const char *msg)
345 priv->bugs.push_back(new bug_message(msg));
347 if (isfeasibleprefix()) {
355 * @brief Assert a bug in the executing program, asserted by a user thread
356 * @see ModelChecker::assert_bug
357 * @param msg Descriptive message for the bug (do not include newline char)
359 void ModelChecker::assert_user_bug(const char *msg)
361 /* If feasible bug, bail out now */
363 switch_to_master(NULL);
366 /** @return True, if any bugs have been reported for this execution */
367 bool ModelChecker::have_bug_reports() const
369 return priv->bugs.size() != 0;
372 /** @brief Print bug report listing for this execution (if any bugs exist) */
373 void ModelChecker::print_bugs() const
375 if (have_bug_reports()) {
376 model_print("Bug report: %zu bug%s detected\n",
378 priv->bugs.size() > 1 ? "s" : "");
379 for (unsigned int i = 0; i < priv->bugs.size(); i++)
380 priv->bugs[i]->print();
385 * @brief Record end-of-execution stats
387 * Must be run when exiting an execution. Records various stats.
388 * @see struct execution_stats
390 void ModelChecker::record_stats()
393 if (!isfinalfeasible())
394 stats.num_infeasible++;
395 else if (have_bug_reports())
396 stats.num_buggy_executions++;
397 else if (is_complete_execution())
398 stats.num_complete++;
401 /** @brief Print execution stats */
402 void ModelChecker::print_stats() const
404 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
405 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
406 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
407 model_print("Total executions: %d\n", stats.num_total);
408 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
412 * Queries the model-checker for more executions to explore and, if one
413 * exists, resets the model-checker state to execute a new execution.
415 * @return If there are more executions to explore, return true. Otherwise,
418 bool ModelChecker::next_execution()
422 if (isfinalfeasible() && (is_complete_execution() || have_bug_reports())) {
423 model_print("Earliest divergence point since last feasible execution:\n");
424 if (earliest_diverge)
425 earliest_diverge->print();
427 model_print("(Not set)\n");
429 earliest_diverge = NULL;
432 assert_bug("Deadlock detected");
438 } else if (DBG_ENABLED()) {
445 if ((diverge = get_next_backtrack()) == NULL)
449 model_print("Next execution will diverge at:\n");
453 reset_to_initial_state();
457 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
459 switch (act->get_type()) {
463 /* linear search: from most recent to oldest */
464 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
465 action_list_t::reverse_iterator rit;
466 for (rit = list->rbegin(); rit != list->rend(); rit++) {
467 ModelAction *prev = *rit;
468 if (prev->could_synchronize_with(act))
474 case ATOMIC_TRYLOCK: {
475 /* linear search: from most recent to oldest */
476 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
477 action_list_t::reverse_iterator rit;
478 for (rit = list->rbegin(); rit != list->rend(); rit++) {
479 ModelAction *prev = *rit;
480 if (act->is_conflicting_lock(prev))
485 case ATOMIC_UNLOCK: {
486 /* linear search: from most recent to oldest */
487 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
488 action_list_t::reverse_iterator rit;
489 for (rit = list->rbegin(); rit != list->rend(); rit++) {
490 ModelAction *prev = *rit;
491 if (!act->same_thread(prev)&&prev->is_failed_trylock())
497 /* linear search: from most recent to oldest */
498 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
499 action_list_t::reverse_iterator rit;
500 for (rit = list->rbegin(); rit != list->rend(); rit++) {
501 ModelAction *prev = *rit;
502 if (!act->same_thread(prev)&&prev->is_failed_trylock())
504 if (!act->same_thread(prev)&&prev->is_notify())
510 case ATOMIC_NOTIFY_ALL:
511 case ATOMIC_NOTIFY_ONE: {
512 /* linear search: from most recent to oldest */
513 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
514 action_list_t::reverse_iterator rit;
515 for (rit = list->rbegin(); rit != list->rend(); rit++) {
516 ModelAction *prev = *rit;
517 if (!act->same_thread(prev)&&prev->is_wait())
528 /** This method finds backtracking points where we should try to
529 * reorder the parameter ModelAction against.
531 * @param the ModelAction to find backtracking points for.
533 void ModelChecker::set_backtracking(ModelAction *act)
535 Thread *t = get_thread(act);
536 ModelAction * prev = get_last_conflict(act);
540 Node * node = prev->get_node()->get_parent();
542 int low_tid, high_tid;
543 if (node->is_enabled(t)) {
544 low_tid = id_to_int(act->get_tid());
545 high_tid = low_tid+1;
548 high_tid = get_num_threads();
551 for(int i = low_tid; i < high_tid; i++) {
552 thread_id_t tid = int_to_id(i);
554 /* Make sure this thread can be enabled here. */
555 if (i >= node->get_num_threads())
558 /* Don't backtrack into a point where the thread is disabled or sleeping. */
559 if (node->enabled_status(tid)!=THREAD_ENABLED)
562 /* Check if this has been explored already */
563 if (node->has_been_explored(tid))
566 /* See if fairness allows */
567 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
569 for(int t=0;t<node->get_num_threads();t++) {
570 thread_id_t tother=int_to_id(t);
571 if (node->is_enabled(tother) && node->has_priority(tother)) {
579 /* Cache the latest backtracking point */
580 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
581 priv->next_backtrack = prev;
583 /* If this is a new backtracking point, mark the tree */
584 if (!node->set_backtrack(tid))
586 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
587 id_to_int(prev->get_tid()),
588 id_to_int(t->get_id()));
597 * Returns last backtracking point. The model checker will explore a different
598 * path for this point in the next execution.
599 * @return The ModelAction at which the next execution should diverge.
601 ModelAction * ModelChecker::get_next_backtrack()
603 ModelAction *next = priv->next_backtrack;
604 priv->next_backtrack = NULL;
609 * Processes a read or rmw model action.
610 * @param curr is the read model action to process.
611 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
612 * @return True if processing this read updates the mo_graph.
614 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
616 uint64_t value = VALUE_NONE;
617 bool updated = false;
619 const ModelAction *reads_from = curr->get_node()->get_read_from();
620 if (reads_from != NULL) {
621 mo_graph->startChanges();
623 value = reads_from->get_value();
624 bool r_status = false;
626 if (!second_part_of_rmw) {
627 check_recency(curr, reads_from);
628 r_status = r_modification_order(curr, reads_from);
632 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
633 mo_graph->rollbackChanges();
634 too_many_reads = false;
638 curr->read_from(reads_from);
639 mo_graph->commitChanges();
640 mo_check_promises(curr->get_tid(), reads_from);
643 } else if (!second_part_of_rmw) {
644 /* Read from future value */
645 value = curr->get_node()->get_future_value();
646 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
647 curr->read_from(NULL);
648 Promise *valuepromise = new Promise(curr, value, expiration);
649 promises->push_back(valuepromise);
651 get_thread(curr)->set_return_value(value);
657 * Processes a lock, trylock, or unlock model action. @param curr is
658 * the read model action to process.
660 * The try lock operation checks whether the lock is taken. If not,
661 * it falls to the normal lock operation case. If so, it returns
664 * The lock operation has already been checked that it is enabled, so
665 * it just grabs the lock and synchronizes with the previous unlock.
667 * The unlock operation has to re-enable all of the threads that are
668 * waiting on the lock.
670 * @return True if synchronization was updated; false otherwise
672 bool ModelChecker::process_mutex(ModelAction *curr) {
673 std::mutex *mutex=NULL;
674 struct std::mutex_state *state=NULL;
676 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
677 mutex = (std::mutex *)curr->get_location();
678 state = mutex->get_state();
679 } else if(curr->is_wait()) {
680 mutex = (std::mutex *)curr->get_value();
681 state = mutex->get_state();
684 switch (curr->get_type()) {
685 case ATOMIC_TRYLOCK: {
686 bool success = !state->islocked;
687 curr->set_try_lock(success);
689 get_thread(curr)->set_return_value(0);
692 get_thread(curr)->set_return_value(1);
694 //otherwise fall into the lock case
696 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
697 assert_bug("Lock access before initialization");
698 state->islocked = true;
699 ModelAction *unlock = get_last_unlock(curr);
700 //synchronize with the previous unlock statement
701 if (unlock != NULL) {
702 curr->synchronize_with(unlock);
707 case ATOMIC_UNLOCK: {
709 state->islocked = false;
710 //wake up the other threads
711 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
712 //activate all the waiting threads
713 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
714 scheduler->wake(get_thread(*rit));
721 state->islocked = false;
722 //wake up the other threads
723 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
724 //activate all the waiting threads
725 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
726 scheduler->wake(get_thread(*rit));
729 //check whether we should go to sleep or not...simulate spurious failures
730 if (curr->get_node()->get_misc()==0) {
731 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
733 scheduler->sleep(get_current_thread());
737 case ATOMIC_NOTIFY_ALL: {
738 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
739 //activate all the waiting threads
740 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
741 scheduler->wake(get_thread(*rit));
746 case ATOMIC_NOTIFY_ONE: {
747 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
748 int wakeupthread=curr->get_node()->get_misc();
749 action_list_t::iterator it = waiters->begin();
750 advance(it, wakeupthread);
751 scheduler->wake(get_thread(*it));
763 * Process a write ModelAction
764 * @param curr The ModelAction to process
765 * @return True if the mo_graph was updated or promises were resolved
767 bool ModelChecker::process_write(ModelAction *curr)
769 bool updated_mod_order = w_modification_order(curr);
770 bool updated_promises = resolve_promises(curr);
772 if (promises->size() == 0) {
773 for (unsigned int i = 0; i < futurevalues->size(); i++) {
774 struct PendingFutureValue pfv = (*futurevalues)[i];
775 //Do more ambitious checks now that mo is more complete
776 if (mo_may_allow(pfv.writer, pfv.act)&&
777 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
778 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
779 priv->next_backtrack = pfv.act;
781 futurevalues->resize(0);
784 mo_graph->commitChanges();
785 mo_check_promises(curr->get_tid(), curr);
787 get_thread(curr)->set_return_value(VALUE_NONE);
788 return updated_mod_order || updated_promises;
792 * @brief Process the current action for thread-related activity
794 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
795 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
796 * synchronization, etc. This function is a no-op for non-THREAD actions
797 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
799 * @param curr The current action
800 * @return True if synchronization was updated or a thread completed
802 bool ModelChecker::process_thread_action(ModelAction *curr)
804 bool updated = false;
806 switch (curr->get_type()) {
807 case THREAD_CREATE: {
808 Thread *th = (Thread *)curr->get_location();
809 th->set_creation(curr);
813 Thread *blocking = (Thread *)curr->get_location();
814 ModelAction *act = get_last_action(blocking->get_id());
815 curr->synchronize_with(act);
816 updated = true; /* trigger rel-seq checks */
819 case THREAD_FINISH: {
820 Thread *th = get_thread(curr);
821 while (!th->wait_list_empty()) {
822 ModelAction *act = th->pop_wait_list();
823 scheduler->wake(get_thread(act));
826 updated = true; /* trigger rel-seq checks */
830 check_promises(curr->get_tid(), NULL, curr->get_cv());
841 * @brief Process the current action for release sequence fixup activity
843 * Performs model-checker release sequence fixups for the current action,
844 * forcing a single pending release sequence to break (with a given, potential
845 * "loose" write) or to complete (i.e., synchronize). If a pending release
846 * sequence forms a complete release sequence, then we must perform the fixup
847 * synchronization, mo_graph additions, etc.
849 * @param curr The current action; must be a release sequence fixup action
850 * @param work_queue The work queue to which to add work items as they are
853 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
855 const ModelAction *write = curr->get_node()->get_relseq_break();
856 struct release_seq *sequence = pending_rel_seqs->back();
857 pending_rel_seqs->pop_back();
859 ModelAction *acquire = sequence->acquire;
860 const ModelAction *rf = sequence->rf;
861 const ModelAction *release = sequence->release;
865 ASSERT(release->same_thread(rf));
869 * @todo Forcing a synchronization requires that we set
870 * modification order constraints. For instance, we can't allow
871 * a fixup sequence in which two separate read-acquire
872 * operations read from the same sequence, where the first one
873 * synchronizes and the other doesn't. Essentially, we can't
874 * allow any writes to insert themselves between 'release' and
878 /* Must synchronize */
879 if (!acquire->synchronize_with(release)) {
880 set_bad_synchronization();
883 /* Re-check all pending release sequences */
884 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
885 /* Re-check act for mo_graph edges */
886 work_queue->push_back(MOEdgeWorkEntry(acquire));
888 /* propagate synchronization to later actions */
889 action_list_t::reverse_iterator rit = action_trace->rbegin();
890 for (; (*rit) != acquire; rit++) {
891 ModelAction *propagate = *rit;
892 if (acquire->happens_before(propagate)) {
893 propagate->synchronize_with(acquire);
894 /* Re-check 'propagate' for mo_graph edges */
895 work_queue->push_back(MOEdgeWorkEntry(propagate));
899 /* Break release sequence with new edges:
900 * release --mo--> write --mo--> rf */
901 mo_graph->addEdge(release, write);
902 mo_graph->addEdge(write, rf);
905 /* See if we have realized a data race */
910 * Initialize the current action by performing one or more of the following
911 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
912 * in the NodeStack, manipulating backtracking sets, allocating and
913 * initializing clock vectors, and computing the promises to fulfill.
915 * @param curr The current action, as passed from the user context; may be
916 * freed/invalidated after the execution of this function, with a different
917 * action "returned" its place (pass-by-reference)
918 * @return True if curr is a newly-explored action; false otherwise
920 bool ModelChecker::initialize_curr_action(ModelAction **curr)
922 ModelAction *newcurr;
924 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
925 newcurr = process_rmw(*curr);
928 if (newcurr->is_rmw())
929 compute_promises(newcurr);
935 (*curr)->set_seq_number(get_next_seq_num());
937 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled());
939 /* First restore type and order in case of RMW operation */
940 if ((*curr)->is_rmwr())
941 newcurr->copy_typeandorder(*curr);
943 ASSERT((*curr)->get_location() == newcurr->get_location());
944 newcurr->copy_from_new(*curr);
946 /* Discard duplicate ModelAction; use action from NodeStack */
949 /* Always compute new clock vector */
950 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
953 return false; /* Action was explored previously */
957 /* Always compute new clock vector */
958 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
960 * Perform one-time actions when pushing new ModelAction onto
963 if (newcurr->is_write())
964 compute_promises(newcurr);
965 else if (newcurr->is_relseq_fixup())
966 compute_relseq_breakwrites(newcurr);
967 else if (newcurr->is_wait())
968 newcurr->get_node()->set_misc_max(2);
969 else if (newcurr->is_notify_one()) {
970 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
972 return true; /* This was a new ModelAction */
977 * @brief Check whether a model action is enabled.
979 * Checks whether a lock or join operation would be successful (i.e., is the
980 * lock already locked, or is the joined thread already complete). If not, put
981 * the action in a waiter list.
983 * @param curr is the ModelAction to check whether it is enabled.
984 * @return a bool that indicates whether the action is enabled.
986 bool ModelChecker::check_action_enabled(ModelAction *curr) {
987 if (curr->is_lock()) {
988 std::mutex * lock = (std::mutex *)curr->get_location();
989 struct std::mutex_state * state = lock->get_state();
990 if (state->islocked) {
991 //Stick the action in the appropriate waiting queue
992 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
995 } else if (curr->get_type() == THREAD_JOIN) {
996 Thread *blocking = (Thread *)curr->get_location();
997 if (!blocking->is_complete()) {
998 blocking->push_wait_list(curr);
1007 * Stores the ModelAction for the current thread action. Call this
1008 * immediately before switching from user- to system-context to pass
1009 * data between them.
1010 * @param act The ModelAction created by the user-thread action
1012 void ModelChecker::set_current_action(ModelAction *act) {
1013 priv->current_action = act;
1017 * This is the heart of the model checker routine. It performs model-checking
1018 * actions corresponding to a given "current action." Among other processes, it
1019 * calculates reads-from relationships, updates synchronization clock vectors,
1020 * forms a memory_order constraints graph, and handles replay/backtrack
1021 * execution when running permutations of previously-observed executions.
1023 * @param curr The current action to process
1024 * @return The next Thread that must be executed. May be NULL if ModelChecker
1025 * makes no choice (e.g., according to replay execution, combining RMW actions,
1028 Thread * ModelChecker::check_current_action(ModelAction *curr)
1031 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1033 if (!check_action_enabled(curr)) {
1034 /* Make the execution look like we chose to run this action
1035 * much later, when a lock/join can succeed */
1036 get_current_thread()->set_pending(curr);
1037 scheduler->sleep(get_current_thread());
1038 return get_next_thread(NULL);
1041 bool newly_explored = initialize_curr_action(&curr);
1043 wake_up_sleeping_actions(curr);
1045 /* Add the action to lists before any other model-checking tasks */
1046 if (!second_part_of_rmw)
1047 add_action_to_lists(curr);
1049 /* Build may_read_from set for newly-created actions */
1050 if (newly_explored && curr->is_read())
1051 build_reads_from_past(curr);
1053 /* Initialize work_queue with the "current action" work */
1054 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1055 while (!work_queue.empty() && !has_asserted()) {
1056 WorkQueueEntry work = work_queue.front();
1057 work_queue.pop_front();
1059 switch (work.type) {
1060 case WORK_CHECK_CURR_ACTION: {
1061 ModelAction *act = work.action;
1062 bool update = false; /* update this location's release seq's */
1063 bool update_all = false; /* update all release seq's */
1065 if (process_thread_action(curr))
1068 if (act->is_read() && process_read(act, second_part_of_rmw))
1071 if (act->is_write() && process_write(act))
1074 if (act->is_mutex_op() && process_mutex(act))
1077 if (act->is_relseq_fixup())
1078 process_relseq_fixup(curr, &work_queue);
1081 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1083 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1086 case WORK_CHECK_RELEASE_SEQ:
1087 resolve_release_sequences(work.location, &work_queue);
1089 case WORK_CHECK_MO_EDGES: {
1090 /** @todo Complete verification of work_queue */
1091 ModelAction *act = work.action;
1092 bool updated = false;
1094 if (act->is_read()) {
1095 const ModelAction *rf = act->get_reads_from();
1096 if (rf != NULL && r_modification_order(act, rf))
1099 if (act->is_write()) {
1100 if (w_modification_order(act))
1103 mo_graph->commitChanges();
1106 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1115 check_curr_backtracking(curr);
1116 set_backtracking(curr);
1117 return get_next_thread(curr);
1120 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
1121 Node *currnode = curr->get_node();
1122 Node *parnode = currnode->get_parent();
1124 if ((!parnode->backtrack_empty() ||
1125 !currnode->misc_empty() ||
1126 !currnode->read_from_empty() ||
1127 !currnode->future_value_empty() ||
1128 !currnode->promise_empty() ||
1129 !currnode->relseq_break_empty())
1130 && (!priv->next_backtrack ||
1131 *curr > *priv->next_backtrack)) {
1132 priv->next_backtrack = curr;
1136 bool ModelChecker::promises_expired() const
1138 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
1139 Promise *promise = (*promises)[promise_index];
1140 if (promise->get_expiration()<priv->used_sequence_numbers) {
1147 /** @return whether the current partial trace must be a prefix of a
1148 * feasible trace. */
1149 bool ModelChecker::isfeasibleprefix() const
1151 return promises->size() == 0 && pending_rel_seqs->size() == 0 && isfeasible();
1154 /** @return whether the current partial trace is feasible. */
1155 bool ModelChecker::isfeasible() const
1157 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
1158 DEBUG("Infeasible: RMW violation\n");
1160 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
1163 /** @return whether the current partial trace is feasible other than
1164 * multiple RMW reading from the same store. */
1165 bool ModelChecker::isfeasibleotherthanRMW() const
1167 if (DBG_ENABLED()) {
1168 if (mo_graph->checkForCycles())
1169 DEBUG("Infeasible: modification order cycles\n");
1171 DEBUG("Infeasible: failed promise\n");
1173 DEBUG("Infeasible: too many reads\n");
1174 if (bad_synchronization)
1175 DEBUG("Infeasible: bad synchronization ordering\n");
1176 if (promises_expired())
1177 DEBUG("Infeasible: promises expired\n");
1179 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
1182 /** Returns whether the current completed trace is feasible. */
1183 bool ModelChecker::isfinalfeasible() const
1185 if (DBG_ENABLED() && promises->size() != 0)
1186 DEBUG("Infeasible: unrevolved promises\n");
1188 return isfeasible() && promises->size() == 0;
1191 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1192 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1193 ModelAction *lastread = get_last_action(act->get_tid());
1194 lastread->process_rmw(act);
1195 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1196 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1197 mo_graph->commitChanges();
1203 * Checks whether a thread has read from the same write for too many times
1204 * without seeing the effects of a later write.
1207 * 1) there must a different write that we could read from that would satisfy the modification order,
1208 * 2) we must have read from the same value in excess of maxreads times, and
1209 * 3) that other write must have been in the reads_from set for maxreads times.
1211 * If so, we decide that the execution is no longer feasible.
1213 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1214 if (params.maxreads != 0) {
1216 if (curr->get_node()->get_read_from_size() <= 1)
1218 //Must make sure that execution is currently feasible... We could
1219 //accidentally clear by rolling back
1222 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1223 int tid = id_to_int(curr->get_tid());
1226 if ((int)thrd_lists->size() <= tid)
1228 action_list_t *list = &(*thrd_lists)[tid];
1230 action_list_t::reverse_iterator rit = list->rbegin();
1231 /* Skip past curr */
1232 for (; (*rit) != curr; rit++)
1234 /* go past curr now */
1237 action_list_t::reverse_iterator ritcopy = rit;
1238 //See if we have enough reads from the same value
1240 for (; count < params.maxreads; rit++,count++) {
1241 if (rit==list->rend())
1243 ModelAction *act = *rit;
1244 if (!act->is_read())
1247 if (act->get_reads_from() != rf)
1249 if (act->get_node()->get_read_from_size() <= 1)
1252 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1254 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1256 //Need a different write
1260 /* Test to see whether this is a feasible write to read from*/
1261 mo_graph->startChanges();
1262 r_modification_order(curr, write);
1263 bool feasiblereadfrom = isfeasible();
1264 mo_graph->rollbackChanges();
1266 if (!feasiblereadfrom)
1270 bool feasiblewrite = true;
1271 //new we need to see if this write works for everyone
1273 for (int loop = count; loop>0; loop--,rit++) {
1274 ModelAction *act=*rit;
1275 bool foundvalue = false;
1276 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1277 if (act->get_node()->get_read_from_at(j)==write) {
1283 feasiblewrite = false;
1287 if (feasiblewrite) {
1288 too_many_reads = true;
1296 * Updates the mo_graph with the constraints imposed from the current
1299 * Basic idea is the following: Go through each other thread and find
1300 * the lastest action that happened before our read. Two cases:
1302 * (1) The action is a write => that write must either occur before
1303 * the write we read from or be the write we read from.
1305 * (2) The action is a read => the write that that action read from
1306 * must occur before the write we read from or be the same write.
1308 * @param curr The current action. Must be a read.
1309 * @param rf The action that curr reads from. Must be a write.
1310 * @return True if modification order edges were added; false otherwise
1312 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1314 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1317 ASSERT(curr->is_read());
1319 /* Iterate over all threads */
1320 for (i = 0; i < thrd_lists->size(); i++) {
1321 /* Iterate over actions in thread, starting from most recent */
1322 action_list_t *list = &(*thrd_lists)[i];
1323 action_list_t::reverse_iterator rit;
1324 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1325 ModelAction *act = *rit;
1328 * Include at most one act per-thread that "happens
1329 * before" curr. Don't consider reflexively.
1331 if (act->happens_before(curr) && act != curr) {
1332 if (act->is_write()) {
1334 mo_graph->addEdge(act, rf);
1338 const ModelAction *prevreadfrom = act->get_reads_from();
1339 //if the previous read is unresolved, keep going...
1340 if (prevreadfrom == NULL)
1343 if (rf != prevreadfrom) {
1344 mo_graph->addEdge(prevreadfrom, rf);
1356 /** This method fixes up the modification order when we resolve a
1357 * promises. The basic problem is that actions that occur after the
1358 * read curr could not property add items to the modification order
1361 * So for each thread, we find the earliest item that happens after
1362 * the read curr. This is the item we have to fix up with additional
1363 * constraints. If that action is write, we add a MO edge between
1364 * the Action rf and that action. If the action is a read, we add a
1365 * MO edge between the Action rf, and whatever the read accessed.
1367 * @param curr is the read ModelAction that we are fixing up MO edges for.
1368 * @param rf is the write ModelAction that curr reads from.
1371 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1373 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1375 ASSERT(curr->is_read());
1377 /* Iterate over all threads */
1378 for (i = 0; i < thrd_lists->size(); i++) {
1379 /* Iterate over actions in thread, starting from most recent */
1380 action_list_t *list = &(*thrd_lists)[i];
1381 action_list_t::reverse_iterator rit;
1382 ModelAction *lastact = NULL;
1384 /* Find last action that happens after curr that is either not curr or a rmw */
1385 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1386 ModelAction *act = *rit;
1387 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1393 /* Include at most one act per-thread that "happens before" curr */
1394 if (lastact != NULL) {
1395 if (lastact==curr) {
1396 //Case 1: The resolved read is a RMW, and we need to make sure
1397 //that the write portion of the RMW mod order after rf
1399 mo_graph->addEdge(rf, lastact);
1400 } else if (lastact->is_read()) {
1401 //Case 2: The resolved read is a normal read and the next
1402 //operation is a read, and we need to make sure the value read
1403 //is mod ordered after rf
1405 const ModelAction *postreadfrom = lastact->get_reads_from();
1406 if (postreadfrom != NULL&&rf != postreadfrom)
1407 mo_graph->addEdge(rf, postreadfrom);
1409 //Case 3: The resolved read is a normal read and the next
1410 //operation is a write, and we need to make sure that the
1411 //write is mod ordered after rf
1413 mo_graph->addEdge(rf, lastact);
1421 * Updates the mo_graph with the constraints imposed from the current write.
1423 * Basic idea is the following: Go through each other thread and find
1424 * the lastest action that happened before our write. Two cases:
1426 * (1) The action is a write => that write must occur before
1429 * (2) The action is a read => the write that that action read from
1430 * must occur before the current write.
1432 * This method also handles two other issues:
1434 * (I) Sequential Consistency: Making sure that if the current write is
1435 * seq_cst, that it occurs after the previous seq_cst write.
1437 * (II) Sending the write back to non-synchronizing reads.
1439 * @param curr The current action. Must be a write.
1440 * @return True if modification order edges were added; false otherwise
1442 bool ModelChecker::w_modification_order(ModelAction *curr)
1444 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1447 ASSERT(curr->is_write());
1449 if (curr->is_seqcst()) {
1450 /* We have to at least see the last sequentially consistent write,
1451 so we are initialized. */
1452 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1453 if (last_seq_cst != NULL) {
1454 mo_graph->addEdge(last_seq_cst, curr);
1459 /* Iterate over all threads */
1460 for (i = 0; i < thrd_lists->size(); i++) {
1461 /* Iterate over actions in thread, starting from most recent */
1462 action_list_t *list = &(*thrd_lists)[i];
1463 action_list_t::reverse_iterator rit;
1464 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1465 ModelAction *act = *rit;
1468 * 1) If RMW and it actually read from something, then we
1469 * already have all relevant edges, so just skip to next
1472 * 2) If RMW and it didn't read from anything, we should
1473 * whatever edge we can get to speed up convergence.
1475 * 3) If normal write, we need to look at earlier actions, so
1476 * continue processing list.
1478 if (curr->is_rmw()) {
1479 if (curr->get_reads_from()!=NULL)
1488 * Include at most one act per-thread that "happens
1491 if (act->happens_before(curr)) {
1493 * Note: if act is RMW, just add edge:
1495 * The following edge should be handled elsewhere:
1496 * readfrom(act) --mo--> act
1498 if (act->is_write())
1499 mo_graph->addEdge(act, curr);
1500 else if (act->is_read()) {
1501 //if previous read accessed a null, just keep going
1502 if (act->get_reads_from() == NULL)
1504 mo_graph->addEdge(act->get_reads_from(), curr);
1508 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1509 !act->same_thread(curr)) {
1510 /* We have an action that:
1511 (1) did not happen before us
1512 (2) is a read and we are a write
1513 (3) cannot synchronize with us
1514 (4) is in a different thread
1516 that read could potentially read from our write. Note that
1517 these checks are overly conservative at this point, we'll
1518 do more checks before actually removing the
1522 if (thin_air_constraint_may_allow(curr, act)) {
1524 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1525 struct PendingFutureValue pfv = {curr,act};
1526 futurevalues->push_back(pfv);
1536 /** Arbitrary reads from the future are not allowed. Section 29.3
1537 * part 9 places some constraints. This method checks one result of constraint
1538 * constraint. Others require compiler support. */
1539 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1540 if (!writer->is_rmw())
1543 if (!reader->is_rmw())
1546 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1547 if (search == reader)
1549 if (search->get_tid() == reader->get_tid() &&
1550 search->happens_before(reader))
1558 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1559 * some constraints. This method checks one the following constraint (others
1560 * require compiler support):
1562 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1564 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1566 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1568 /* Iterate over all threads */
1569 for (i = 0; i < thrd_lists->size(); i++) {
1570 const ModelAction *write_after_read = NULL;
1572 /* Iterate over actions in thread, starting from most recent */
1573 action_list_t *list = &(*thrd_lists)[i];
1574 action_list_t::reverse_iterator rit;
1575 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1576 ModelAction *act = *rit;
1578 if (!reader->happens_before(act))
1580 else if (act->is_write())
1581 write_after_read = act;
1582 else if (act->is_read() && act->get_reads_from() != NULL && act != reader) {
1583 write_after_read = act->get_reads_from();
1587 if (write_after_read && write_after_read!=writer && mo_graph->checkReachable(write_after_read, writer))
1594 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1595 * The ModelAction under consideration is expected to be taking part in
1596 * release/acquire synchronization as an object of the "reads from" relation.
1597 * Note that this can only provide release sequence support for RMW chains
1598 * which do not read from the future, as those actions cannot be traced until
1599 * their "promise" is fulfilled. Similarly, we may not even establish the
1600 * presence of a release sequence with certainty, as some modification order
1601 * constraints may be decided further in the future. Thus, this function
1602 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1603 * and a boolean representing certainty.
1605 * @param rf The action that might be part of a release sequence. Must be a
1607 * @param release_heads A pass-by-reference style return parameter. After
1608 * execution of this function, release_heads will contain the heads of all the
1609 * relevant release sequences, if any exists with certainty
1610 * @param pending A pass-by-reference style return parameter which is only used
1611 * when returning false (i.e., uncertain). Returns most information regarding
1612 * an uncertain release sequence, including any write operations that might
1613 * break the sequence.
1614 * @return true, if the ModelChecker is certain that release_heads is complete;
1617 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1618 rel_heads_list_t *release_heads,
1619 struct release_seq *pending) const
1621 /* Only check for release sequences if there are no cycles */
1622 if (mo_graph->checkForCycles())
1626 ASSERT(rf->is_write());
1628 if (rf->is_release())
1629 release_heads->push_back(rf);
1631 break; /* End of RMW chain */
1633 /** @todo Need to be smarter here... In the linux lock
1634 * example, this will run to the beginning of the program for
1636 /** @todo The way to be smarter here is to keep going until 1
1637 * thread has a release preceded by an acquire and you've seen
1640 /* acq_rel RMW is a sufficient stopping condition */
1641 if (rf->is_acquire() && rf->is_release())
1642 return true; /* complete */
1644 rf = rf->get_reads_from();
1647 /* read from future: need to settle this later */
1649 return false; /* incomplete */
1652 if (rf->is_release())
1653 return true; /* complete */
1655 /* else relaxed write; check modification order for contiguous subsequence
1656 * -> rf must be same thread as release */
1657 int tid = id_to_int(rf->get_tid());
1658 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1659 action_list_t *list = &(*thrd_lists)[tid];
1660 action_list_t::const_reverse_iterator rit;
1662 /* Find rf in the thread list */
1663 rit = std::find(list->rbegin(), list->rend(), rf);
1664 ASSERT(rit != list->rend());
1666 /* Find the last write/release */
1667 for (; rit != list->rend(); rit++)
1668 if ((*rit)->is_release())
1670 if (rit == list->rend()) {
1671 /* No write-release in this thread */
1672 return true; /* complete */
1674 ModelAction *release = *rit;
1676 ASSERT(rf->same_thread(release));
1678 pending->writes.clear();
1680 bool certain = true;
1681 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1682 if (id_to_int(rf->get_tid()) == (int)i)
1684 list = &(*thrd_lists)[i];
1686 /* Can we ensure no future writes from this thread may break
1687 * the release seq? */
1688 bool future_ordered = false;
1690 ModelAction *last = get_last_action(int_to_id(i));
1691 Thread *th = get_thread(int_to_id(i));
1692 if ((last && rf->happens_before(last)) ||
1695 future_ordered = true;
1697 ASSERT(!th->is_model_thread() || future_ordered);
1699 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1700 const ModelAction *act = *rit;
1701 /* Reach synchronization -> this thread is complete */
1702 if (act->happens_before(release))
1704 if (rf->happens_before(act)) {
1705 future_ordered = true;
1709 /* Only non-RMW writes can break release sequences */
1710 if (!act->is_write() || act->is_rmw())
1713 /* Check modification order */
1714 if (mo_graph->checkReachable(rf, act)) {
1715 /* rf --mo--> act */
1716 future_ordered = true;
1719 if (mo_graph->checkReachable(act, release))
1720 /* act --mo--> release */
1722 if (mo_graph->checkReachable(release, act) &&
1723 mo_graph->checkReachable(act, rf)) {
1724 /* release --mo-> act --mo--> rf */
1725 return true; /* complete */
1727 /* act may break release sequence */
1728 pending->writes.push_back(act);
1731 if (!future_ordered)
1732 certain = false; /* This thread is uncertain */
1736 release_heads->push_back(release);
1737 pending->writes.clear();
1739 pending->release = release;
1746 * A public interface for getting the release sequence head(s) with which a
1747 * given ModelAction must synchronize. This function only returns a non-empty
1748 * result when it can locate a release sequence head with certainty. Otherwise,
1749 * it may mark the internal state of the ModelChecker so that it will handle
1750 * the release sequence at a later time, causing @a act to update its
1751 * synchronization at some later point in execution.
1752 * @param act The 'acquire' action that may read from a release sequence
1753 * @param release_heads A pass-by-reference return parameter. Will be filled
1754 * with the head(s) of the release sequence(s), if they exists with certainty.
1755 * @see ModelChecker::release_seq_heads
1757 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1759 const ModelAction *rf = act->get_reads_from();
1760 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1761 sequence->acquire = act;
1763 if (!release_seq_heads(rf, release_heads, sequence)) {
1764 /* add act to 'lazy checking' list */
1765 pending_rel_seqs->push_back(sequence);
1767 snapshot_free(sequence);
1772 * Attempt to resolve all stashed operations that might synchronize with a
1773 * release sequence for a given location. This implements the "lazy" portion of
1774 * determining whether or not a release sequence was contiguous, since not all
1775 * modification order information is present at the time an action occurs.
1777 * @param location The location/object that should be checked for release
1778 * sequence resolutions. A NULL value means to check all locations.
1779 * @param work_queue The work queue to which to add work items as they are
1781 * @return True if any updates occurred (new synchronization, new mo_graph
1784 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1786 bool updated = false;
1787 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1788 while (it != pending_rel_seqs->end()) {
1789 struct release_seq *pending = *it;
1790 ModelAction *act = pending->acquire;
1792 /* Only resolve sequences on the given location, if provided */
1793 if (location && act->get_location() != location) {
1798 const ModelAction *rf = act->get_reads_from();
1799 rel_heads_list_t release_heads;
1801 complete = release_seq_heads(rf, &release_heads, pending);
1802 for (unsigned int i = 0; i < release_heads.size(); i++) {
1803 if (!act->has_synchronized_with(release_heads[i])) {
1804 if (act->synchronize_with(release_heads[i]))
1807 set_bad_synchronization();
1812 /* Re-check all pending release sequences */
1813 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1814 /* Re-check act for mo_graph edges */
1815 work_queue->push_back(MOEdgeWorkEntry(act));
1817 /* propagate synchronization to later actions */
1818 action_list_t::reverse_iterator rit = action_trace->rbegin();
1819 for (; (*rit) != act; rit++) {
1820 ModelAction *propagate = *rit;
1821 if (act->happens_before(propagate)) {
1822 propagate->synchronize_with(act);
1823 /* Re-check 'propagate' for mo_graph edges */
1824 work_queue->push_back(MOEdgeWorkEntry(propagate));
1829 it = pending_rel_seqs->erase(it);
1830 snapshot_free(pending);
1836 // If we resolved promises or data races, see if we have realized a data race.
1843 * Performs various bookkeeping operations for the current ModelAction. For
1844 * instance, adds action to the per-object, per-thread action vector and to the
1845 * action trace list of all thread actions.
1847 * @param act is the ModelAction to add.
1849 void ModelChecker::add_action_to_lists(ModelAction *act)
1851 int tid = id_to_int(act->get_tid());
1852 action_trace->push_back(act);
1854 get_safe_ptr_action(obj_map, act->get_location())->push_back(act);
1856 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
1857 if (tid >= (int)vec->size())
1858 vec->resize(priv->next_thread_id);
1859 (*vec)[tid].push_back(act);
1861 if ((int)thrd_last_action->size() <= tid)
1862 thrd_last_action->resize(get_num_threads());
1863 (*thrd_last_action)[tid] = act;
1865 if (act->is_wait()) {
1866 void *mutex_loc=(void *) act->get_value();
1867 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
1869 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
1870 if (tid >= (int)vec->size())
1871 vec->resize(priv->next_thread_id);
1872 (*vec)[tid].push_back(act);
1874 if ((int)thrd_last_action->size() <= tid)
1875 thrd_last_action->resize(get_num_threads());
1876 (*thrd_last_action)[tid] = act;
1881 * @brief Get the last action performed by a particular Thread
1882 * @param tid The thread ID of the Thread in question
1883 * @return The last action in the thread
1885 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1887 int threadid = id_to_int(tid);
1888 if (threadid < (int)thrd_last_action->size())
1889 return (*thrd_last_action)[id_to_int(tid)];
1895 * Gets the last memory_order_seq_cst write (in the total global sequence)
1896 * performed on a particular object (i.e., memory location), not including the
1898 * @param curr The current ModelAction; also denotes the object location to
1900 * @return The last seq_cst write
1902 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1904 void *location = curr->get_location();
1905 action_list_t *list = get_safe_ptr_action(obj_map, location);
1906 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1907 action_list_t::reverse_iterator rit;
1908 for (rit = list->rbegin(); rit != list->rend(); rit++)
1909 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1915 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1916 * location). This function identifies the mutex according to the current
1917 * action, which is presumed to perform on the same mutex.
1918 * @param curr The current ModelAction; also denotes the object location to
1920 * @return The last unlock operation
1922 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1924 void *location = curr->get_location();
1925 action_list_t *list = get_safe_ptr_action(obj_map, location);
1926 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1927 action_list_t::reverse_iterator rit;
1928 for (rit = list->rbegin(); rit != list->rend(); rit++)
1929 if ((*rit)->is_unlock() || (*rit)->is_wait())
1934 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1936 ModelAction *parent = get_last_action(tid);
1938 parent = get_thread(tid)->get_creation();
1943 * Returns the clock vector for a given thread.
1944 * @param tid The thread whose clock vector we want
1945 * @return Desired clock vector
1947 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1949 return get_parent_action(tid)->get_cv();
1953 * Resolve a set of Promises with a current write. The set is provided in the
1954 * Node corresponding to @a write.
1955 * @param write The ModelAction that is fulfilling Promises
1956 * @return True if promises were resolved; false otherwise
1958 bool ModelChecker::resolve_promises(ModelAction *write)
1960 bool resolved = false;
1961 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1963 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1964 Promise *promise = (*promises)[promise_index];
1965 if (write->get_node()->get_promise(i)) {
1966 ModelAction *read = promise->get_action();
1967 if (read->is_rmw()) {
1968 mo_graph->addRMWEdge(write, read);
1970 read->read_from(write);
1971 //First fix up the modification order for actions that happened
1973 r_modification_order(read, write);
1974 //Next fix up the modification order for actions that happened
1976 post_r_modification_order(read, write);
1977 //Make sure the promise's value matches the write's value
1978 ASSERT(promise->get_value() == write->get_value());
1981 promises->erase(promises->begin() + promise_index);
1982 threads_to_check.push_back(read->get_tid());
1989 //Check whether reading these writes has made threads unable to
1992 for(unsigned int i=0;i<threads_to_check.size();i++)
1993 mo_check_promises(threads_to_check[i], write);
1999 * Compute the set of promises that could potentially be satisfied by this
2000 * action. Note that the set computation actually appears in the Node, not in
2002 * @param curr The ModelAction that may satisfy promises
2004 void ModelChecker::compute_promises(ModelAction *curr)
2006 for (unsigned int i = 0; i < promises->size(); i++) {
2007 Promise *promise = (*promises)[i];
2008 const ModelAction *act = promise->get_action();
2009 if (!act->happens_before(curr) &&
2011 !act->could_synchronize_with(curr) &&
2012 !act->same_thread(curr) &&
2013 act->get_location() == curr->get_location() &&
2014 promise->get_value() == curr->get_value()) {
2015 curr->get_node()->set_promise(i, act->is_rmw());
2020 /** Checks promises in response to change in ClockVector Threads. */
2021 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2023 for (unsigned int i = 0; i < promises->size(); i++) {
2024 Promise *promise = (*promises)[i];
2025 const ModelAction *act = promise->get_action();
2026 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
2027 merge_cv->synchronized_since(act)) {
2028 if (promise->increment_threads(tid)) {
2029 //Promise has failed
2030 failed_promise = true;
2037 void ModelChecker::check_promises_thread_disabled() {
2038 for (unsigned int i = 0; i < promises->size(); i++) {
2039 Promise *promise = (*promises)[i];
2040 if (promise->check_promise()) {
2041 failed_promise = true;
2047 /** Checks promises in response to addition to modification order for threads.
2049 * pthread is the thread that performed the read that created the promise
2051 * pread is the read that created the promise
2053 * pwrite is either the first write to same location as pread by
2054 * pthread that is sequenced after pread or the value read by the
2055 * first read to the same lcoation as pread by pthread that is
2056 * sequenced after pread..
2058 * 1. If tid=pthread, then we check what other threads are reachable
2059 * through the mode order starting with pwrite. Those threads cannot
2060 * perform a write that will resolve the promise due to modification
2061 * order constraints.
2063 * 2. If the tid is not pthread, we check whether pwrite can reach the
2064 * action write through the modification order. If so, that thread
2065 * cannot perform a future write that will resolve the promise due to
2066 * modificatin order constraints.
2068 * @parem tid The thread that either read from the model action
2069 * write, or actually did the model action write.
2071 * @parem write The ModelAction representing the relevant write.
2074 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
2075 void * location = write->get_location();
2076 for (unsigned int i = 0; i < promises->size(); i++) {
2077 Promise *promise = (*promises)[i];
2078 const ModelAction *act = promise->get_action();
2080 //Is this promise on the same location?
2081 if ( act->get_location() != location )
2084 //same thread as the promise
2085 if ( act->get_tid()==tid ) {
2087 //do we have a pwrite for the promise, if not, set it
2088 if (promise->get_write() == NULL ) {
2089 promise->set_write(write);
2090 //The pwrite cannot happen before the promise
2091 if (write->happens_before(act) && (write != act)) {
2092 failed_promise = true;
2096 if (mo_graph->checkPromise(write, promise)) {
2097 failed_promise = true;
2102 //Don't do any lookups twice for the same thread
2103 if (promise->has_sync_thread(tid))
2106 if (promise->get_write()&&mo_graph->checkReachable(promise->get_write(), write)) {
2107 if (promise->increment_threads(tid)) {
2108 failed_promise = true;
2116 * Compute the set of writes that may break the current pending release
2117 * sequence. This information is extracted from previou release sequence
2120 * @param curr The current ModelAction. Must be a release sequence fixup
2123 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2125 if (pending_rel_seqs->empty())
2128 struct release_seq *pending = pending_rel_seqs->back();
2129 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2130 const ModelAction *write = pending->writes[i];
2131 curr->get_node()->add_relseq_break(write);
2134 /* NULL means don't break the sequence; just synchronize */
2135 curr->get_node()->add_relseq_break(NULL);
2139 * Build up an initial set of all past writes that this 'read' action may read
2140 * from. This set is determined by the clock vector's "happens before"
2142 * @param curr is the current ModelAction that we are exploring; it must be a
2145 void ModelChecker::build_reads_from_past(ModelAction *curr)
2147 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2149 ASSERT(curr->is_read());
2151 ModelAction *last_seq_cst = NULL;
2153 /* Track whether this object has been initialized */
2154 bool initialized = false;
2156 if (curr->is_seqcst()) {
2157 last_seq_cst = get_last_seq_cst(curr);
2158 /* We have to at least see the last sequentially consistent write,
2159 so we are initialized. */
2160 if (last_seq_cst != NULL)
2164 /* Iterate over all threads */
2165 for (i = 0; i < thrd_lists->size(); i++) {
2166 /* Iterate over actions in thread, starting from most recent */
2167 action_list_t *list = &(*thrd_lists)[i];
2168 action_list_t::reverse_iterator rit;
2169 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2170 ModelAction *act = *rit;
2172 /* Only consider 'write' actions */
2173 if (!act->is_write() || act == curr)
2176 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2177 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
2178 if (!curr->get_sleep_flag() || curr->is_seqcst() || sleep_can_read_from(curr, act)) {
2179 DEBUG("Adding action to may_read_from:\n");
2180 if (DBG_ENABLED()) {
2184 curr->get_node()->add_read_from(act);
2188 /* Include at most one act per-thread that "happens before" curr */
2189 if (act->happens_before(curr)) {
2197 assert_bug("May read from uninitialized atomic");
2199 if (DBG_ENABLED() || !initialized) {
2200 model_print("Reached read action:\n");
2202 model_print("Printing may_read_from\n");
2203 curr->get_node()->print_may_read_from();
2204 model_print("End printing may_read_from\n");
2208 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2210 Node *prevnode=write->get_node()->get_parent();
2212 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2213 if (write->is_release()&&thread_sleep)
2215 if (!write->is_rmw()) {
2218 if (write->get_reads_from()==NULL)
2220 write=write->get_reads_from();
2224 static void print_list(action_list_t *list, int exec_num = -1)
2226 action_list_t::iterator it;
2228 model_print("---------------------------------------------------------------------\n");
2230 model_print("Execution %d:\n", exec_num);
2232 unsigned int hash=0;
2234 for (it = list->begin(); it != list->end(); it++) {
2236 hash=hash^(hash<<3)^((*it)->hash());
2238 model_print("HASH %u\n", hash);
2239 model_print("---------------------------------------------------------------------\n");
2242 #if SUPPORT_MOD_ORDER_DUMP
2243 void ModelChecker::dumpGraph(char *filename) {
2245 sprintf(buffer, "%s.dot",filename);
2246 FILE *file=fopen(buffer, "w");
2247 fprintf(file, "digraph %s {\n",filename);
2248 mo_graph->dumpNodes(file);
2249 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2251 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2252 ModelAction *action=*it;
2253 if (action->is_read()) {
2254 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2255 if (action->get_reads_from()!=NULL)
2256 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2258 if (thread_array[action->get_tid()] != NULL) {
2259 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2262 thread_array[action->get_tid()]=action;
2264 fprintf(file,"}\n");
2265 model_free(thread_array);
2270 void ModelChecker::print_summary()
2272 #if SUPPORT_MOD_ORDER_DUMP
2274 char buffername[100];
2275 sprintf(buffername, "exec%04u", stats.num_total);
2276 mo_graph->dumpGraphToFile(buffername);
2277 sprintf(buffername, "graph%04u", stats.num_total);
2278 dumpGraph(buffername);
2281 if (!isfinalfeasible())
2282 model_print("INFEASIBLE EXECUTION!\n");
2283 print_list(action_trace, stats.num_total);
2288 * Add a Thread to the system for the first time. Should only be called once
2290 * @param t The Thread to add
2292 void ModelChecker::add_thread(Thread *t)
2294 thread_map->put(id_to_int(t->get_id()), t);
2295 scheduler->add_thread(t);
2299 * Removes a thread from the scheduler.
2300 * @param the thread to remove.
2302 void ModelChecker::remove_thread(Thread *t)
2304 scheduler->remove_thread(t);
2308 * @brief Get a Thread reference by its ID
2309 * @param tid The Thread's ID
2310 * @return A Thread reference
2312 Thread * ModelChecker::get_thread(thread_id_t tid) const
2314 return thread_map->get(id_to_int(tid));
2318 * @brief Get a reference to the Thread in which a ModelAction was executed
2319 * @param act The ModelAction
2320 * @return A Thread reference
2322 Thread * ModelChecker::get_thread(ModelAction *act) const
2324 return get_thread(act->get_tid());
2328 * @brief Check if a Thread is currently enabled
2329 * @param t The Thread to check
2330 * @return True if the Thread is currently enabled
2332 bool ModelChecker::is_enabled(Thread *t) const
2334 return scheduler->is_enabled(t);
2338 * @brief Check if a Thread is currently enabled
2339 * @param tid The ID of the Thread to check
2340 * @return True if the Thread is currently enabled
2342 bool ModelChecker::is_enabled(thread_id_t tid) const
2344 return scheduler->is_enabled(tid);
2348 * Switch from a user-context to the "master thread" context (a.k.a. system
2349 * context). This switch is made with the intention of exploring a particular
2350 * model-checking action (described by a ModelAction object). Must be called
2351 * from a user-thread context.
2353 * @param act The current action that will be explored. May be NULL only if
2354 * trace is exiting via an assertion (see ModelChecker::set_assert and
2355 * ModelChecker::has_asserted).
2356 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2358 int ModelChecker::switch_to_master(ModelAction *act)
2361 Thread *old = thread_current();
2362 set_current_action(act);
2363 old->set_state(THREAD_READY);
2364 return Thread::swap(old, &system_context);
2368 * Takes the next step in the execution, if possible.
2369 * @return Returns true (success) if a step was taken and false otherwise.
2371 bool ModelChecker::take_step() {
2375 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2377 if (curr->get_state() == THREAD_READY) {
2378 ASSERT(priv->current_action);
2380 priv->nextThread = check_current_action(priv->current_action);
2381 priv->current_action = NULL;
2383 if (curr->is_blocked() || curr->is_complete())
2384 scheduler->remove_thread(curr);
2389 Thread *next = scheduler->next_thread(priv->nextThread);
2391 /* Infeasible -> don't take any more steps */
2394 else if (isfeasibleprefix() && have_bug_reports()) {
2399 if (params.bound != 0) {
2400 if (priv->used_sequence_numbers > params.bound) {
2405 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2406 next ? id_to_int(next->get_id()) : -1);
2409 * Launch end-of-execution release sequence fixups only when there are:
2411 * (1) no more user threads to run (or when execution replay chooses
2412 * the 'model_thread')
2413 * (2) pending release sequences
2414 * (3) pending assertions (i.e., data races)
2415 * (4) no pending promises
2417 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2418 isfinalfeasible() && !unrealizedraces.empty()) {
2419 model_print("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2420 pending_rel_seqs->size());
2421 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2422 std::memory_order_seq_cst, NULL, VALUE_NONE,
2424 set_current_action(fixup);
2428 /* next == NULL -> don't take any more steps */
2432 next->set_state(THREAD_RUNNING);
2434 if (next->get_pending() != NULL) {
2435 /* restart a pending action */
2436 set_current_action(next->get_pending());
2437 next->set_pending(NULL);
2438 next->set_state(THREAD_READY);
2442 /* Return false only if swap fails with an error */
2443 return (Thread::swap(&system_context, next) == 0);
2446 /** Runs the current execution until threre are no more steps to take. */
2447 void ModelChecker::finish_execution() {
2450 while (take_step());