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() { printf("%s", msg); }
34 * Structure for holding small ModelChecker members that should be snapshotted
36 struct model_snapshot_members {
37 ModelAction *current_action;
38 unsigned int next_thread_id;
39 modelclock_t used_sequence_numbers;
41 ModelAction *next_backtrack;
42 std::vector< bug_message *, SnapshotAlloc<bug_message *> > bugs;
45 /** @brief Constructor */
46 ModelChecker::ModelChecker(struct model_params params) :
47 /* Initialize default scheduler */
49 scheduler(new Scheduler()),
51 num_feasible_executions(0),
53 earliest_diverge(NULL),
54 action_trace(new action_list_t()),
55 thread_map(new HashTable<int, Thread *, int>()),
56 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
57 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
58 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
59 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
60 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
61 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
62 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
63 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
64 node_stack(new NodeStack()),
65 mo_graph(new CycleGraph()),
66 failed_promise(false),
67 too_many_reads(false),
69 bad_synchronization(false)
71 /* Allocate this "size" on the snapshotting heap */
72 priv = (struct model_snapshot_members *)snapshot_calloc(1, sizeof(*priv));
73 /* First thread created will have id INITIAL_THREAD_ID */
74 priv->next_thread_id = INITIAL_THREAD_ID;
76 /* Initialize a model-checker thread, for special ModelActions */
77 model_thread = new Thread(get_next_id());
78 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
81 /** @brief Destructor */
82 ModelChecker::~ModelChecker()
84 for (unsigned int i = 0; i < get_num_threads(); i++)
85 delete thread_map->get(i);
90 delete lock_waiters_map;
91 delete condvar_waiters_map;
94 for (unsigned int i = 0; i < promises->size(); i++)
95 delete (*promises)[i];
98 delete pending_rel_seqs;
100 delete thrd_last_action;
105 for (unsigned int i = 0; i < priv->bugs.size(); i++)
106 delete priv->bugs[i];
111 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr) {
112 action_list_t * tmp=hash->get(ptr);
114 tmp=new action_list_t();
120 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr) {
121 std::vector<action_list_t> * tmp=hash->get(ptr);
123 tmp=new std::vector<action_list_t>();
130 * Restores user program to initial state and resets all model-checker data
133 void ModelChecker::reset_to_initial_state()
135 DEBUG("+++ Resetting to initial state +++\n");
136 node_stack->reset_execution();
137 failed_promise = false;
138 too_many_reads = false;
139 bad_synchronization = false;
141 snapshotObject->backTrackBeforeStep(0);
144 /** @return a thread ID for a new Thread */
145 thread_id_t ModelChecker::get_next_id()
147 return priv->next_thread_id++;
150 /** @return the number of user threads created during this execution */
151 unsigned int ModelChecker::get_num_threads() const
153 return priv->next_thread_id;
156 /** @return The currently executing Thread. */
157 Thread * ModelChecker::get_current_thread()
159 return scheduler->get_current_thread();
162 /** @return a sequence number for a new ModelAction */
163 modelclock_t ModelChecker::get_next_seq_num()
165 return ++priv->used_sequence_numbers;
168 Node * ModelChecker::get_curr_node() {
169 return node_stack->get_head();
173 * @brief Choose the next thread to execute.
175 * This function chooses the next thread that should execute. It can force the
176 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
177 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
178 * The model-checker may have no preference regarding the next thread (i.e.,
179 * when exploring a new execution ordering), in which case this will return
181 * @param curr The current ModelAction. This action might guide the choice of
183 * @return The next thread to run. If the model-checker has no preference, NULL.
185 Thread * ModelChecker::get_next_thread(ModelAction *curr)
190 /* Do not split atomic actions. */
192 return thread_current();
193 /* The THREAD_CREATE action points to the created Thread */
194 else if (curr->get_type() == THREAD_CREATE)
195 return (Thread *)curr->get_location();
198 /* Have we completed exploring the preselected path? */
202 /* Else, we are trying to replay an execution */
203 ModelAction *next = node_stack->get_next()->get_action();
205 if (next == diverge) {
206 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
207 earliest_diverge=diverge;
209 Node *nextnode = next->get_node();
210 Node *prevnode = nextnode->get_parent();
211 scheduler->update_sleep_set(prevnode);
213 /* Reached divergence point */
214 if (nextnode->increment_misc()) {
215 /* The next node will try to satisfy a different misc_index values. */
216 tid = next->get_tid();
217 node_stack->pop_restofstack(2);
218 } else if (nextnode->increment_promise()) {
219 /* The next node will try to satisfy a different set of promises. */
220 tid = next->get_tid();
221 node_stack->pop_restofstack(2);
222 } else if (nextnode->increment_read_from()) {
223 /* The next node will read from a different value. */
224 tid = next->get_tid();
225 node_stack->pop_restofstack(2);
226 } else if (nextnode->increment_future_value()) {
227 /* The next node will try to read from a different future value. */
228 tid = next->get_tid();
229 node_stack->pop_restofstack(2);
230 } else if (nextnode->increment_relseq_break()) {
231 /* The next node will try to resolve a release sequence differently */
232 tid = next->get_tid();
233 node_stack->pop_restofstack(2);
235 /* Make a different thread execute for next step */
236 scheduler->add_sleep(thread_map->get(id_to_int(next->get_tid())));
237 tid = prevnode->get_next_backtrack();
238 /* Make sure the backtracked thread isn't sleeping. */
239 node_stack->pop_restofstack(1);
240 if (diverge==earliest_diverge) {
241 earliest_diverge=prevnode->get_action();
244 /* The correct sleep set is in the parent node. */
247 DEBUG("*** Divergence point ***\n");
251 tid = next->get_tid();
253 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
254 ASSERT(tid != THREAD_ID_T_NONE);
255 return thread_map->get(id_to_int(tid));
259 * We need to know what the next actions of all threads in the sleep
260 * set will be. This method computes them and stores the actions at
261 * the corresponding thread object's pending action.
264 void ModelChecker::execute_sleep_set() {
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 thr->get_pending() == NULL ) {
270 thr->set_state(THREAD_RUNNING);
271 scheduler->next_thread(thr);
272 Thread::swap(&system_context, thr);
273 priv->current_action->set_sleep_flag();
274 thr->set_pending(priv->current_action);
277 priv->current_action = NULL;
280 void ModelChecker::wake_up_sleeping_actions(ModelAction * curr) {
281 for(unsigned int i=0;i<get_num_threads();i++) {
282 thread_id_t tid=int_to_id(i);
283 Thread *thr=get_thread(tid);
284 if ( scheduler->get_enabled(thr) == THREAD_SLEEP_SET ) {
285 ModelAction *pending_act=thr->get_pending();
286 if ((!curr->is_rmwr())&&pending_act->could_synchronize_with(curr)) {
287 //Remove this thread from sleep set
288 scheduler->remove_sleep(thr);
295 * Check if we are in a deadlock. Should only be called at the end of an
296 * execution, although it should not give false positives in the middle of an
297 * execution (there should be some ENABLED thread).
299 * @return True if program is in a deadlock; false otherwise
301 bool ModelChecker::is_deadlocked() const
303 bool blocking_threads = false;
304 for (unsigned int i = 0; i < get_num_threads(); i++) {
305 thread_id_t tid = int_to_id(i);
308 Thread *t = get_thread(tid);
309 if (!t->is_model_thread() && t->get_pending())
310 blocking_threads = true;
312 return blocking_threads;
316 * Check if this is a complete execution. That is, have all thread completed
317 * execution (rather than exiting because sleep sets have forced a redundant
320 * @return True if the execution is complete.
322 bool ModelChecker::is_complete_execution() const
324 for (unsigned int i = 0; i < get_num_threads(); i++)
325 if (is_enabled(int_to_id(i)))
331 * @brief Assert a bug in the executing program.
333 * Use this function to assert any sort of bug in the user program. If the
334 * current trace is feasible (actually, a prefix of some feasible execution),
335 * then this execution will be aborted, printing the appropriate message. If
336 * the current trace is not yet feasible, the error message will be stashed and
337 * printed if the execution ever becomes feasible.
339 * This function can also be used to immediately trigger the bug; that is, we
340 * don't wait for a feasible execution before aborting. Only use the
341 * "immediate" option when you know that the infeasibility is justified (e.g.,
342 * pending release sequences are not a problem)
344 * @param msg Descriptive message for the bug (do not include newline char)
345 * @param user_thread Was this assertion triggered from a user thread?
346 * @param immediate Should this bug be triggered immediately?
348 void ModelChecker::assert_bug(const char *msg, bool user_thread, bool immediate)
350 priv->bugs.push_back(new bug_message(msg));
352 if (immediate || isfeasibleprefix()) {
355 switch_to_master(NULL);
360 * @brief Assert a bug in the executing program, with a default message
361 * @see ModelChecker::assert_bug
362 * @param user_thread Was this assertion triggered from a user thread?
364 void ModelChecker::assert_bug(bool user_thread)
366 assert_bug("bug detected", user_thread);
370 * @brief Assert a bug in the executing program immediately
371 * @see ModelChecker::assert_bug
372 * @param msg Descriptive message for the bug (do not include newline char)
374 void ModelChecker::assert_bug_immediate(const char *msg)
376 printf("Feasible: %s\n", isfeasibleprefix() ? "yes" : "no");
377 assert_bug(msg, false, true);
380 /** @return True, if any bugs have been reported for this execution */
381 bool ModelChecker::have_bug_reports() const
383 return priv->bugs.size() != 0;
386 /** @brief Print bug report listing for this execution (if any bugs exist) */
387 void ModelChecker::print_bugs() const
389 if (have_bug_reports()) {
390 printf("Bug report: %zu bugs detected\n", priv->bugs.size());
391 for (unsigned int i = 0; i < priv->bugs.size(); i++)
392 priv->bugs[i]->print();
397 * Queries the model-checker for more executions to explore and, if one
398 * exists, resets the model-checker state to execute a new execution.
400 * @return If there are more executions to explore, return true. Otherwise,
403 bool ModelChecker::next_execution()
409 if (isfinalfeasible() && (is_complete_execution() || have_bug_reports())) {
410 printf("Earliest divergence point since last feasible execution:\n");
411 if (earliest_diverge)
412 earliest_diverge->print();
414 printf("(Not set)\n");
416 earliest_diverge = NULL;
417 num_feasible_executions++;
420 assert_bug("Deadlock detected", false);
425 } else if (DBG_ENABLED()) {
429 if ((diverge = get_next_backtrack()) == NULL)
433 printf("Next execution will diverge at:\n");
437 reset_to_initial_state();
441 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
443 switch (act->get_type()) {
447 /* linear search: from most recent to oldest */
448 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
449 action_list_t::reverse_iterator rit;
450 for (rit = list->rbegin(); rit != list->rend(); rit++) {
451 ModelAction *prev = *rit;
452 if (prev->could_synchronize_with(act))
458 case ATOMIC_TRYLOCK: {
459 /* linear search: from most recent to oldest */
460 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
461 action_list_t::reverse_iterator rit;
462 for (rit = list->rbegin(); rit != list->rend(); rit++) {
463 ModelAction *prev = *rit;
464 if (act->is_conflicting_lock(prev))
469 case ATOMIC_UNLOCK: {
470 /* linear search: from most recent to oldest */
471 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
472 action_list_t::reverse_iterator rit;
473 for (rit = list->rbegin(); rit != list->rend(); rit++) {
474 ModelAction *prev = *rit;
475 if (!act->same_thread(prev)&&prev->is_failed_trylock())
481 /* linear search: from most recent to oldest */
482 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
483 action_list_t::reverse_iterator rit;
484 for (rit = list->rbegin(); rit != list->rend(); rit++) {
485 ModelAction *prev = *rit;
486 if (!act->same_thread(prev)&&prev->is_failed_trylock())
488 if (!act->same_thread(prev)&&prev->is_notify())
494 case ATOMIC_NOTIFY_ALL:
495 case ATOMIC_NOTIFY_ONE: {
496 /* linear search: from most recent to oldest */
497 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
498 action_list_t::reverse_iterator rit;
499 for (rit = list->rbegin(); rit != list->rend(); rit++) {
500 ModelAction *prev = *rit;
501 if (!act->same_thread(prev)&&prev->is_wait())
512 /** This method finds backtracking points where we should try to
513 * reorder the parameter ModelAction against.
515 * @param the ModelAction to find backtracking points for.
517 void ModelChecker::set_backtracking(ModelAction *act)
519 Thread *t = get_thread(act);
520 ModelAction * prev = get_last_conflict(act);
524 Node * node = prev->get_node()->get_parent();
526 int low_tid, high_tid;
527 if (node->is_enabled(t)) {
528 low_tid = id_to_int(act->get_tid());
529 high_tid = low_tid+1;
532 high_tid = get_num_threads();
535 for(int i = low_tid; i < high_tid; i++) {
536 thread_id_t tid = int_to_id(i);
538 /* Make sure this thread can be enabled here. */
539 if (i >= node->get_num_threads())
542 /* Don't backtrack into a point where the thread is disabled or sleeping. */
543 if (node->enabled_status(tid)!=THREAD_ENABLED)
546 /* Check if this has been explored already */
547 if (node->has_been_explored(tid))
550 /* See if fairness allows */
551 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
553 for(int t=0;t<node->get_num_threads();t++) {
554 thread_id_t tother=int_to_id(t);
555 if (node->is_enabled(tother) && node->has_priority(tother)) {
563 /* Cache the latest backtracking point */
564 if (!priv->next_backtrack || *prev > *priv->next_backtrack)
565 priv->next_backtrack = prev;
567 /* If this is a new backtracking point, mark the tree */
568 if (!node->set_backtrack(tid))
570 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
571 id_to_int(prev->get_tid()),
572 id_to_int(t->get_id()));
581 * Returns last backtracking point. The model checker will explore a different
582 * path for this point in the next execution.
583 * @return The ModelAction at which the next execution should diverge.
585 ModelAction * ModelChecker::get_next_backtrack()
587 ModelAction *next = priv->next_backtrack;
588 priv->next_backtrack = NULL;
593 * Processes a read or rmw model action.
594 * @param curr is the read model action to process.
595 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
596 * @return True if processing this read updates the mo_graph.
598 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
600 uint64_t value = VALUE_NONE;
601 bool updated = false;
603 const ModelAction *reads_from = curr->get_node()->get_read_from();
604 if (reads_from != NULL) {
605 mo_graph->startChanges();
607 value = reads_from->get_value();
608 bool r_status = false;
610 if (!second_part_of_rmw) {
611 check_recency(curr, reads_from);
612 r_status = r_modification_order(curr, reads_from);
616 if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
617 mo_graph->rollbackChanges();
618 too_many_reads = false;
622 curr->read_from(reads_from);
623 mo_graph->commitChanges();
624 mo_check_promises(curr->get_tid(), reads_from);
627 } else if (!second_part_of_rmw) {
628 /* Read from future value */
629 value = curr->get_node()->get_future_value();
630 modelclock_t expiration = curr->get_node()->get_future_value_expiration();
631 curr->read_from(NULL);
632 Promise *valuepromise = new Promise(curr, value, expiration);
633 promises->push_back(valuepromise);
635 get_thread(curr)->set_return_value(value);
641 * Processes a lock, trylock, or unlock model action. @param curr is
642 * the read model action to process.
644 * The try lock operation checks whether the lock is taken. If not,
645 * it falls to the normal lock operation case. If so, it returns
648 * The lock operation has already been checked that it is enabled, so
649 * it just grabs the lock and synchronizes with the previous unlock.
651 * The unlock operation has to re-enable all of the threads that are
652 * waiting on the lock.
654 * @return True if synchronization was updated; false otherwise
656 bool ModelChecker::process_mutex(ModelAction *curr) {
657 std::mutex *mutex=NULL;
658 struct std::mutex_state *state=NULL;
660 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
661 mutex = (std::mutex *)curr->get_location();
662 state = mutex->get_state();
663 } else if(curr->is_wait()) {
664 mutex = (std::mutex *)curr->get_value();
665 state = mutex->get_state();
668 switch (curr->get_type()) {
669 case ATOMIC_TRYLOCK: {
670 bool success = !state->islocked;
671 curr->set_try_lock(success);
673 get_thread(curr)->set_return_value(0);
676 get_thread(curr)->set_return_value(1);
678 //otherwise fall into the lock case
680 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock) {
681 printf("Lock access before initialization\n");
684 state->islocked = true;
685 ModelAction *unlock = get_last_unlock(curr);
686 //synchronize with the previous unlock statement
687 if (unlock != NULL) {
688 curr->synchronize_with(unlock);
693 case ATOMIC_UNLOCK: {
695 state->islocked = false;
696 //wake up the other threads
697 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
698 //activate all the waiting threads
699 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
700 scheduler->wake(get_thread(*rit));
707 state->islocked = false;
708 //wake up the other threads
709 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
710 //activate all the waiting threads
711 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
712 scheduler->wake(get_thread(*rit));
715 //check whether we should go to sleep or not...simulate spurious failures
716 if (curr->get_node()->get_misc()==0) {
717 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
719 scheduler->sleep(get_current_thread());
723 case ATOMIC_NOTIFY_ALL: {
724 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
725 //activate all the waiting threads
726 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
727 scheduler->wake(get_thread(*rit));
732 case ATOMIC_NOTIFY_ONE: {
733 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
734 int wakeupthread=curr->get_node()->get_misc();
735 action_list_t::iterator it = waiters->begin();
736 advance(it, wakeupthread);
737 scheduler->wake(get_thread(*it));
749 * Process a write ModelAction
750 * @param curr The ModelAction to process
751 * @return True if the mo_graph was updated or promises were resolved
753 bool ModelChecker::process_write(ModelAction *curr)
755 bool updated_mod_order = w_modification_order(curr);
756 bool updated_promises = resolve_promises(curr);
758 if (promises->size() == 0) {
759 for (unsigned int i = 0; i < futurevalues->size(); i++) {
760 struct PendingFutureValue pfv = (*futurevalues)[i];
761 //Do more ambitious checks now that mo is more complete
762 if (mo_may_allow(pfv.writer, pfv.act)&&
763 pfv.act->get_node()->add_future_value(pfv.writer->get_value(), pfv.writer->get_seq_number()+params.maxfuturedelay) &&
764 (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
765 priv->next_backtrack = pfv.act;
767 futurevalues->resize(0);
770 mo_graph->commitChanges();
771 mo_check_promises(curr->get_tid(), curr);
773 get_thread(curr)->set_return_value(VALUE_NONE);
774 return updated_mod_order || updated_promises;
778 * @brief Process the current action for thread-related activity
780 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
781 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
782 * synchronization, etc. This function is a no-op for non-THREAD actions
783 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
785 * @param curr The current action
786 * @return True if synchronization was updated or a thread completed
788 bool ModelChecker::process_thread_action(ModelAction *curr)
790 bool updated = false;
792 switch (curr->get_type()) {
793 case THREAD_CREATE: {
794 Thread *th = (Thread *)curr->get_location();
795 th->set_creation(curr);
799 Thread *blocking = (Thread *)curr->get_location();
800 ModelAction *act = get_last_action(blocking->get_id());
801 curr->synchronize_with(act);
802 updated = true; /* trigger rel-seq checks */
805 case THREAD_FINISH: {
806 Thread *th = get_thread(curr);
807 while (!th->wait_list_empty()) {
808 ModelAction *act = th->pop_wait_list();
809 scheduler->wake(get_thread(act));
812 updated = true; /* trigger rel-seq checks */
816 check_promises(curr->get_tid(), NULL, curr->get_cv());
827 * @brief Process the current action for release sequence fixup activity
829 * Performs model-checker release sequence fixups for the current action,
830 * forcing a single pending release sequence to break (with a given, potential
831 * "loose" write) or to complete (i.e., synchronize). If a pending release
832 * sequence forms a complete release sequence, then we must perform the fixup
833 * synchronization, mo_graph additions, etc.
835 * @param curr The current action; must be a release sequence fixup action
836 * @param work_queue The work queue to which to add work items as they are
839 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
841 const ModelAction *write = curr->get_node()->get_relseq_break();
842 struct release_seq *sequence = pending_rel_seqs->back();
843 pending_rel_seqs->pop_back();
845 ModelAction *acquire = sequence->acquire;
846 const ModelAction *rf = sequence->rf;
847 const ModelAction *release = sequence->release;
851 ASSERT(release->same_thread(rf));
855 * @todo Forcing a synchronization requires that we set
856 * modification order constraints. For instance, we can't allow
857 * a fixup sequence in which two separate read-acquire
858 * operations read from the same sequence, where the first one
859 * synchronizes and the other doesn't. Essentially, we can't
860 * allow any writes to insert themselves between 'release' and
864 /* Must synchronize */
865 if (!acquire->synchronize_with(release)) {
866 set_bad_synchronization();
869 /* Re-check all pending release sequences */
870 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
871 /* Re-check act for mo_graph edges */
872 work_queue->push_back(MOEdgeWorkEntry(acquire));
874 /* propagate synchronization to later actions */
875 action_list_t::reverse_iterator rit = action_trace->rbegin();
876 for (; (*rit) != acquire; rit++) {
877 ModelAction *propagate = *rit;
878 if (acquire->happens_before(propagate)) {
879 propagate->synchronize_with(acquire);
880 /* Re-check 'propagate' for mo_graph edges */
881 work_queue->push_back(MOEdgeWorkEntry(propagate));
885 /* Break release sequence with new edges:
886 * release --mo--> write --mo--> rf */
887 mo_graph->addEdge(release, write);
888 mo_graph->addEdge(write, rf);
891 /* See if we have realized a data race */
892 if (checkDataRaces())
897 * Initialize the current action by performing one or more of the following
898 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
899 * in the NodeStack, manipulating backtracking sets, allocating and
900 * initializing clock vectors, and computing the promises to fulfill.
902 * @param curr The current action, as passed from the user context; may be
903 * freed/invalidated after the execution of this function, with a different
904 * action "returned" its place (pass-by-reference)
905 * @return True if curr is a newly-explored action; false otherwise
907 bool ModelChecker::initialize_curr_action(ModelAction **curr)
909 ModelAction *newcurr;
911 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
912 newcurr = process_rmw(*curr);
915 if (newcurr->is_rmw())
916 compute_promises(newcurr);
922 (*curr)->set_seq_number(get_next_seq_num());
924 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled());
926 /* First restore type and order in case of RMW operation */
927 if ((*curr)->is_rmwr())
928 newcurr->copy_typeandorder(*curr);
930 ASSERT((*curr)->get_location() == newcurr->get_location());
931 newcurr->copy_from_new(*curr);
933 /* Discard duplicate ModelAction; use action from NodeStack */
936 /* Always compute new clock vector */
937 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
940 return false; /* Action was explored previously */
944 /* Always compute new clock vector */
945 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
947 * Perform one-time actions when pushing new ModelAction onto
950 if (newcurr->is_write())
951 compute_promises(newcurr);
952 else if (newcurr->is_relseq_fixup())
953 compute_relseq_breakwrites(newcurr);
954 else if (newcurr->is_wait())
955 newcurr->get_node()->set_misc_max(2);
956 else if (newcurr->is_notify_one()) {
957 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
959 return true; /* This was a new ModelAction */
964 * @brief Check whether a model action is enabled.
966 * Checks whether a lock or join operation would be successful (i.e., is the
967 * lock already locked, or is the joined thread already complete). If not, put
968 * the action in a waiter list.
970 * @param curr is the ModelAction to check whether it is enabled.
971 * @return a bool that indicates whether the action is enabled.
973 bool ModelChecker::check_action_enabled(ModelAction *curr) {
974 if (curr->is_lock()) {
975 std::mutex * lock = (std::mutex *)curr->get_location();
976 struct std::mutex_state * state = lock->get_state();
977 if (state->islocked) {
978 //Stick the action in the appropriate waiting queue
979 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
982 } else if (curr->get_type() == THREAD_JOIN) {
983 Thread *blocking = (Thread *)curr->get_location();
984 if (!blocking->is_complete()) {
985 blocking->push_wait_list(curr);
994 * Stores the ModelAction for the current thread action. Call this
995 * immediately before switching from user- to system-context to pass
997 * @param act The ModelAction created by the user-thread action
999 void ModelChecker::set_current_action(ModelAction *act) {
1000 priv->current_action = act;
1004 * This is the heart of the model checker routine. It performs model-checking
1005 * actions corresponding to a given "current action." Among other processes, it
1006 * calculates reads-from relationships, updates synchronization clock vectors,
1007 * forms a memory_order constraints graph, and handles replay/backtrack
1008 * execution when running permutations of previously-observed executions.
1010 * @param curr The current action to process
1011 * @return The next Thread that must be executed. May be NULL if ModelChecker
1012 * makes no choice (e.g., according to replay execution, combining RMW actions,
1015 Thread * ModelChecker::check_current_action(ModelAction *curr)
1018 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1020 if (!check_action_enabled(curr)) {
1021 /* Make the execution look like we chose to run this action
1022 * much later, when a lock/join can succeed */
1023 get_current_thread()->set_pending(curr);
1024 scheduler->sleep(get_current_thread());
1025 return get_next_thread(NULL);
1028 bool newly_explored = initialize_curr_action(&curr);
1030 wake_up_sleeping_actions(curr);
1032 /* Add the action to lists before any other model-checking tasks */
1033 if (!second_part_of_rmw)
1034 add_action_to_lists(curr);
1036 /* Build may_read_from set for newly-created actions */
1037 if (newly_explored && curr->is_read())
1038 build_reads_from_past(curr);
1040 /* Initialize work_queue with the "current action" work */
1041 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1042 while (!work_queue.empty()) {
1043 WorkQueueEntry work = work_queue.front();
1044 work_queue.pop_front();
1046 switch (work.type) {
1047 case WORK_CHECK_CURR_ACTION: {
1048 ModelAction *act = work.action;
1049 bool update = false; /* update this location's release seq's */
1050 bool update_all = false; /* update all release seq's */
1052 if (process_thread_action(curr))
1055 if (act->is_read() && process_read(act, second_part_of_rmw))
1058 if (act->is_write() && process_write(act))
1061 if (act->is_mutex_op() && process_mutex(act))
1064 if (act->is_relseq_fixup())
1065 process_relseq_fixup(curr, &work_queue);
1068 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1070 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1073 case WORK_CHECK_RELEASE_SEQ:
1074 resolve_release_sequences(work.location, &work_queue);
1076 case WORK_CHECK_MO_EDGES: {
1077 /** @todo Complete verification of work_queue */
1078 ModelAction *act = work.action;
1079 bool updated = false;
1081 if (act->is_read()) {
1082 const ModelAction *rf = act->get_reads_from();
1083 if (rf != NULL && r_modification_order(act, rf))
1086 if (act->is_write()) {
1087 if (w_modification_order(act))
1090 mo_graph->commitChanges();
1093 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1102 check_curr_backtracking(curr);
1103 set_backtracking(curr);
1104 return get_next_thread(curr);
1107 void ModelChecker::check_curr_backtracking(ModelAction * curr) {
1108 Node *currnode = curr->get_node();
1109 Node *parnode = currnode->get_parent();
1111 if ((!parnode->backtrack_empty() ||
1112 !currnode->misc_empty() ||
1113 !currnode->read_from_empty() ||
1114 !currnode->future_value_empty() ||
1115 !currnode->promise_empty() ||
1116 !currnode->relseq_break_empty())
1117 && (!priv->next_backtrack ||
1118 *curr > *priv->next_backtrack)) {
1119 priv->next_backtrack = curr;
1123 bool ModelChecker::promises_expired() const
1125 for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
1126 Promise *promise = (*promises)[promise_index];
1127 if (promise->get_expiration()<priv->used_sequence_numbers) {
1134 /** @return whether the current partial trace must be a prefix of a
1135 * feasible trace. */
1136 bool ModelChecker::isfeasibleprefix() const
1138 return promises->size() == 0 && pending_rel_seqs->size() == 0 && isfeasible();
1141 /** @return whether the current partial trace is feasible. */
1142 bool ModelChecker::isfeasible() const
1144 if (DBG_ENABLED() && mo_graph->checkForRMWViolation())
1145 DEBUG("Infeasible: RMW violation\n");
1147 return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
1150 /** @return whether the current partial trace is feasible other than
1151 * multiple RMW reading from the same store. */
1152 bool ModelChecker::isfeasibleotherthanRMW() const
1154 if (DBG_ENABLED()) {
1155 if (mo_graph->checkForCycles())
1156 DEBUG("Infeasible: modification order cycles\n");
1158 DEBUG("Infeasible: failed promise\n");
1160 DEBUG("Infeasible: too many reads\n");
1161 if (bad_synchronization)
1162 DEBUG("Infeasible: bad synchronization ordering\n");
1163 if (promises_expired())
1164 DEBUG("Infeasible: promises expired\n");
1166 return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !bad_synchronization && !promises_expired();
1169 /** Returns whether the current completed trace is feasible. */
1170 bool ModelChecker::isfinalfeasible() const
1172 if (DBG_ENABLED() && promises->size() != 0)
1173 DEBUG("Infeasible: unrevolved promises\n");
1175 return isfeasible() && promises->size() == 0;
1178 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1179 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1180 ModelAction *lastread = get_last_action(act->get_tid());
1181 lastread->process_rmw(act);
1182 if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
1183 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1184 mo_graph->commitChanges();
1190 * Checks whether a thread has read from the same write for too many times
1191 * without seeing the effects of a later write.
1194 * 1) there must a different write that we could read from that would satisfy the modification order,
1195 * 2) we must have read from the same value in excess of maxreads times, and
1196 * 3) that other write must have been in the reads_from set for maxreads times.
1198 * If so, we decide that the execution is no longer feasible.
1200 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf) {
1201 if (params.maxreads != 0) {
1203 if (curr->get_node()->get_read_from_size() <= 1)
1205 //Must make sure that execution is currently feasible... We could
1206 //accidentally clear by rolling back
1209 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1210 int tid = id_to_int(curr->get_tid());
1213 if ((int)thrd_lists->size() <= tid)
1215 action_list_t *list = &(*thrd_lists)[tid];
1217 action_list_t::reverse_iterator rit = list->rbegin();
1218 /* Skip past curr */
1219 for (; (*rit) != curr; rit++)
1221 /* go past curr now */
1224 action_list_t::reverse_iterator ritcopy = rit;
1225 //See if we have enough reads from the same value
1227 for (; count < params.maxreads; rit++,count++) {
1228 if (rit==list->rend())
1230 ModelAction *act = *rit;
1231 if (!act->is_read())
1234 if (act->get_reads_from() != rf)
1236 if (act->get_node()->get_read_from_size() <= 1)
1239 for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
1241 const ModelAction * write = curr->get_node()->get_read_from_at(i);
1243 //Need a different write
1247 /* Test to see whether this is a feasible write to read from*/
1248 mo_graph->startChanges();
1249 r_modification_order(curr, write);
1250 bool feasiblereadfrom = isfeasible();
1251 mo_graph->rollbackChanges();
1253 if (!feasiblereadfrom)
1257 bool feasiblewrite = true;
1258 //new we need to see if this write works for everyone
1260 for (int loop = count; loop>0; loop--,rit++) {
1261 ModelAction *act=*rit;
1262 bool foundvalue = false;
1263 for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
1264 if (act->get_node()->get_read_from_at(j)==write) {
1270 feasiblewrite = false;
1274 if (feasiblewrite) {
1275 too_many_reads = true;
1283 * Updates the mo_graph with the constraints imposed from the current
1286 * Basic idea is the following: Go through each other thread and find
1287 * the lastest action that happened before our read. Two cases:
1289 * (1) The action is a write => that write must either occur before
1290 * the write we read from or be the write we read from.
1292 * (2) The action is a read => the write that that action read from
1293 * must occur before the write we read from or be the same write.
1295 * @param curr The current action. Must be a read.
1296 * @param rf The action that curr reads from. Must be a write.
1297 * @return True if modification order edges were added; false otherwise
1299 bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
1301 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1304 ASSERT(curr->is_read());
1306 /* Iterate over all threads */
1307 for (i = 0; i < thrd_lists->size(); i++) {
1308 /* Iterate over actions in thread, starting from most recent */
1309 action_list_t *list = &(*thrd_lists)[i];
1310 action_list_t::reverse_iterator rit;
1311 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1312 ModelAction *act = *rit;
1315 * Include at most one act per-thread that "happens
1316 * before" curr. Don't consider reflexively.
1318 if (act->happens_before(curr) && act != curr) {
1319 if (act->is_write()) {
1321 mo_graph->addEdge(act, rf);
1325 const ModelAction *prevreadfrom = act->get_reads_from();
1326 //if the previous read is unresolved, keep going...
1327 if (prevreadfrom == NULL)
1330 if (rf != prevreadfrom) {
1331 mo_graph->addEdge(prevreadfrom, rf);
1343 /** This method fixes up the modification order when we resolve a
1344 * promises. The basic problem is that actions that occur after the
1345 * read curr could not property add items to the modification order
1348 * So for each thread, we find the earliest item that happens after
1349 * the read curr. This is the item we have to fix up with additional
1350 * constraints. If that action is write, we add a MO edge between
1351 * the Action rf and that action. If the action is a read, we add a
1352 * MO edge between the Action rf, and whatever the read accessed.
1354 * @param curr is the read ModelAction that we are fixing up MO edges for.
1355 * @param rf is the write ModelAction that curr reads from.
1358 void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
1360 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1362 ASSERT(curr->is_read());
1364 /* Iterate over all threads */
1365 for (i = 0; i < thrd_lists->size(); i++) {
1366 /* Iterate over actions in thread, starting from most recent */
1367 action_list_t *list = &(*thrd_lists)[i];
1368 action_list_t::reverse_iterator rit;
1369 ModelAction *lastact = NULL;
1371 /* Find last action that happens after curr that is either not curr or a rmw */
1372 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1373 ModelAction *act = *rit;
1374 if (curr->happens_before(act) && (curr != act || curr->is_rmw())) {
1380 /* Include at most one act per-thread that "happens before" curr */
1381 if (lastact != NULL) {
1382 if (lastact==curr) {
1383 //Case 1: The resolved read is a RMW, and we need to make sure
1384 //that the write portion of the RMW mod order after rf
1386 mo_graph->addEdge(rf, lastact);
1387 } else if (lastact->is_read()) {
1388 //Case 2: The resolved read is a normal read and the next
1389 //operation is a read, and we need to make sure the value read
1390 //is mod ordered after rf
1392 const ModelAction *postreadfrom = lastact->get_reads_from();
1393 if (postreadfrom != NULL&&rf != postreadfrom)
1394 mo_graph->addEdge(rf, postreadfrom);
1396 //Case 3: The resolved read is a normal read and the next
1397 //operation is a write, and we need to make sure that the
1398 //write is mod ordered after rf
1400 mo_graph->addEdge(rf, lastact);
1408 * Updates the mo_graph with the constraints imposed from the current write.
1410 * Basic idea is the following: Go through each other thread and find
1411 * the lastest action that happened before our write. Two cases:
1413 * (1) The action is a write => that write must occur before
1416 * (2) The action is a read => the write that that action read from
1417 * must occur before the current write.
1419 * This method also handles two other issues:
1421 * (I) Sequential Consistency: Making sure that if the current write is
1422 * seq_cst, that it occurs after the previous seq_cst write.
1424 * (II) Sending the write back to non-synchronizing reads.
1426 * @param curr The current action. Must be a write.
1427 * @return True if modification order edges were added; false otherwise
1429 bool ModelChecker::w_modification_order(ModelAction *curr)
1431 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1434 ASSERT(curr->is_write());
1436 if (curr->is_seqcst()) {
1437 /* We have to at least see the last sequentially consistent write,
1438 so we are initialized. */
1439 ModelAction *last_seq_cst = get_last_seq_cst(curr);
1440 if (last_seq_cst != NULL) {
1441 mo_graph->addEdge(last_seq_cst, curr);
1446 /* Iterate over all threads */
1447 for (i = 0; i < thrd_lists->size(); i++) {
1448 /* Iterate over actions in thread, starting from most recent */
1449 action_list_t *list = &(*thrd_lists)[i];
1450 action_list_t::reverse_iterator rit;
1451 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1452 ModelAction *act = *rit;
1455 * 1) If RMW and it actually read from something, then we
1456 * already have all relevant edges, so just skip to next
1459 * 2) If RMW and it didn't read from anything, we should
1460 * whatever edge we can get to speed up convergence.
1462 * 3) If normal write, we need to look at earlier actions, so
1463 * continue processing list.
1465 if (curr->is_rmw()) {
1466 if (curr->get_reads_from()!=NULL)
1475 * Include at most one act per-thread that "happens
1478 if (act->happens_before(curr)) {
1480 * Note: if act is RMW, just add edge:
1482 * The following edge should be handled elsewhere:
1483 * readfrom(act) --mo--> act
1485 if (act->is_write())
1486 mo_graph->addEdge(act, curr);
1487 else if (act->is_read()) {
1488 //if previous read accessed a null, just keep going
1489 if (act->get_reads_from() == NULL)
1491 mo_graph->addEdge(act->get_reads_from(), curr);
1495 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1496 !act->same_thread(curr)) {
1497 /* We have an action that:
1498 (1) did not happen before us
1499 (2) is a read and we are a write
1500 (3) cannot synchronize with us
1501 (4) is in a different thread
1503 that read could potentially read from our write. Note that
1504 these checks are overly conservative at this point, we'll
1505 do more checks before actually removing the
1509 if (thin_air_constraint_may_allow(curr, act)) {
1511 (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() == act->get_reads_from() && isfeasibleotherthanRMW())) {
1512 struct PendingFutureValue pfv = {curr,act};
1513 futurevalues->push_back(pfv);
1523 /** Arbitrary reads from the future are not allowed. Section 29.3
1524 * part 9 places some constraints. This method checks one result of constraint
1525 * constraint. Others require compiler support. */
1526 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
1527 if (!writer->is_rmw())
1530 if (!reader->is_rmw())
1533 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1534 if (search == reader)
1536 if (search->get_tid() == reader->get_tid() &&
1537 search->happens_before(reader))
1545 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1546 * some constraints. This method checks one the following constraint (others
1547 * require compiler support):
1549 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1551 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1553 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1555 /* Iterate over all threads */
1556 for (i = 0; i < thrd_lists->size(); i++) {
1557 const ModelAction *write_after_read = NULL;
1559 /* Iterate over actions in thread, starting from most recent */
1560 action_list_t *list = &(*thrd_lists)[i];
1561 action_list_t::reverse_iterator rit;
1562 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1563 ModelAction *act = *rit;
1565 if (!reader->happens_before(act))
1567 else if (act->is_write())
1568 write_after_read = act;
1569 else if (act->is_read() && act->get_reads_from() != NULL && act != reader) {
1570 write_after_read = act->get_reads_from();
1574 if (write_after_read && write_after_read!=writer && mo_graph->checkReachable(write_after_read, writer))
1581 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1582 * The ModelAction under consideration is expected to be taking part in
1583 * release/acquire synchronization as an object of the "reads from" relation.
1584 * Note that this can only provide release sequence support for RMW chains
1585 * which do not read from the future, as those actions cannot be traced until
1586 * their "promise" is fulfilled. Similarly, we may not even establish the
1587 * presence of a release sequence with certainty, as some modification order
1588 * constraints may be decided further in the future. Thus, this function
1589 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1590 * and a boolean representing certainty.
1592 * @param rf The action that might be part of a release sequence. Must be a
1594 * @param release_heads A pass-by-reference style return parameter. After
1595 * execution of this function, release_heads will contain the heads of all the
1596 * relevant release sequences, if any exists with certainty
1597 * @param pending A pass-by-reference style return parameter which is only used
1598 * when returning false (i.e., uncertain). Returns most information regarding
1599 * an uncertain release sequence, including any write operations that might
1600 * break the sequence.
1601 * @return true, if the ModelChecker is certain that release_heads is complete;
1604 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1605 rel_heads_list_t *release_heads,
1606 struct release_seq *pending) const
1608 /* Only check for release sequences if there are no cycles */
1609 if (mo_graph->checkForCycles())
1613 ASSERT(rf->is_write());
1615 if (rf->is_release())
1616 release_heads->push_back(rf);
1618 break; /* End of RMW chain */
1620 /** @todo Need to be smarter here... In the linux lock
1621 * example, this will run to the beginning of the program for
1623 /** @todo The way to be smarter here is to keep going until 1
1624 * thread has a release preceded by an acquire and you've seen
1627 /* acq_rel RMW is a sufficient stopping condition */
1628 if (rf->is_acquire() && rf->is_release())
1629 return true; /* complete */
1631 rf = rf->get_reads_from();
1634 /* read from future: need to settle this later */
1636 return false; /* incomplete */
1639 if (rf->is_release())
1640 return true; /* complete */
1642 /* else relaxed write; check modification order for contiguous subsequence
1643 * -> rf must be same thread as release */
1644 int tid = id_to_int(rf->get_tid());
1645 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1646 action_list_t *list = &(*thrd_lists)[tid];
1647 action_list_t::const_reverse_iterator rit;
1649 /* Find rf in the thread list */
1650 rit = std::find(list->rbegin(), list->rend(), rf);
1651 ASSERT(rit != list->rend());
1653 /* Find the last write/release */
1654 for (; rit != list->rend(); rit++)
1655 if ((*rit)->is_release())
1657 if (rit == list->rend()) {
1658 /* No write-release in this thread */
1659 return true; /* complete */
1661 ModelAction *release = *rit;
1663 ASSERT(rf->same_thread(release));
1665 pending->writes.clear();
1667 bool certain = true;
1668 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1669 if (id_to_int(rf->get_tid()) == (int)i)
1671 list = &(*thrd_lists)[i];
1673 /* Can we ensure no future writes from this thread may break
1674 * the release seq? */
1675 bool future_ordered = false;
1677 ModelAction *last = get_last_action(int_to_id(i));
1678 Thread *th = get_thread(int_to_id(i));
1679 if ((last && rf->happens_before(last)) ||
1682 future_ordered = true;
1684 ASSERT(!th->is_model_thread() || future_ordered);
1686 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1687 const ModelAction *act = *rit;
1688 /* Reach synchronization -> this thread is complete */
1689 if (act->happens_before(release))
1691 if (rf->happens_before(act)) {
1692 future_ordered = true;
1696 /* Only non-RMW writes can break release sequences */
1697 if (!act->is_write() || act->is_rmw())
1700 /* Check modification order */
1701 if (mo_graph->checkReachable(rf, act)) {
1702 /* rf --mo--> act */
1703 future_ordered = true;
1706 if (mo_graph->checkReachable(act, release))
1707 /* act --mo--> release */
1709 if (mo_graph->checkReachable(release, act) &&
1710 mo_graph->checkReachable(act, rf)) {
1711 /* release --mo-> act --mo--> rf */
1712 return true; /* complete */
1714 /* act may break release sequence */
1715 pending->writes.push_back(act);
1718 if (!future_ordered)
1719 certain = false; /* This thread is uncertain */
1723 release_heads->push_back(release);
1724 pending->writes.clear();
1726 pending->release = release;
1733 * A public interface for getting the release sequence head(s) with which a
1734 * given ModelAction must synchronize. This function only returns a non-empty
1735 * result when it can locate a release sequence head with certainty. Otherwise,
1736 * it may mark the internal state of the ModelChecker so that it will handle
1737 * the release sequence at a later time, causing @a act to update its
1738 * synchronization at some later point in execution.
1739 * @param act The 'acquire' action that may read from a release sequence
1740 * @param release_heads A pass-by-reference return parameter. Will be filled
1741 * with the head(s) of the release sequence(s), if they exists with certainty.
1742 * @see ModelChecker::release_seq_heads
1744 void ModelChecker::get_release_seq_heads(ModelAction *act, rel_heads_list_t *release_heads)
1746 const ModelAction *rf = act->get_reads_from();
1747 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
1748 sequence->acquire = act;
1750 if (!release_seq_heads(rf, release_heads, sequence)) {
1751 /* add act to 'lazy checking' list */
1752 pending_rel_seqs->push_back(sequence);
1754 snapshot_free(sequence);
1759 * Attempt to resolve all stashed operations that might synchronize with a
1760 * release sequence for a given location. This implements the "lazy" portion of
1761 * determining whether or not a release sequence was contiguous, since not all
1762 * modification order information is present at the time an action occurs.
1764 * @param location The location/object that should be checked for release
1765 * sequence resolutions. A NULL value means to check all locations.
1766 * @param work_queue The work queue to which to add work items as they are
1768 * @return True if any updates occurred (new synchronization, new mo_graph
1771 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
1773 bool updated = false;
1774 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
1775 while (it != pending_rel_seqs->end()) {
1776 struct release_seq *pending = *it;
1777 ModelAction *act = pending->acquire;
1779 /* Only resolve sequences on the given location, if provided */
1780 if (location && act->get_location() != location) {
1785 const ModelAction *rf = act->get_reads_from();
1786 rel_heads_list_t release_heads;
1788 complete = release_seq_heads(rf, &release_heads, pending);
1789 for (unsigned int i = 0; i < release_heads.size(); i++) {
1790 if (!act->has_synchronized_with(release_heads[i])) {
1791 if (act->synchronize_with(release_heads[i]))
1794 set_bad_synchronization();
1799 /* Re-check all pending release sequences */
1800 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1801 /* Re-check act for mo_graph edges */
1802 work_queue->push_back(MOEdgeWorkEntry(act));
1804 /* propagate synchronization to later actions */
1805 action_list_t::reverse_iterator rit = action_trace->rbegin();
1806 for (; (*rit) != act; rit++) {
1807 ModelAction *propagate = *rit;
1808 if (act->happens_before(propagate)) {
1809 propagate->synchronize_with(act);
1810 /* Re-check 'propagate' for mo_graph edges */
1811 work_queue->push_back(MOEdgeWorkEntry(propagate));
1816 it = pending_rel_seqs->erase(it);
1817 snapshot_free(pending);
1823 // If we resolved promises or data races, see if we have realized a data race.
1824 if (checkDataRaces()) {
1832 * Performs various bookkeeping operations for the current ModelAction. For
1833 * instance, adds action to the per-object, per-thread action vector and to the
1834 * action trace list of all thread actions.
1836 * @param act is the ModelAction to add.
1838 void ModelChecker::add_action_to_lists(ModelAction *act)
1840 int tid = id_to_int(act->get_tid());
1841 action_trace->push_back(act);
1843 get_safe_ptr_action(obj_map, act->get_location())->push_back(act);
1845 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
1846 if (tid >= (int)vec->size())
1847 vec->resize(priv->next_thread_id);
1848 (*vec)[tid].push_back(act);
1850 if ((int)thrd_last_action->size() <= tid)
1851 thrd_last_action->resize(get_num_threads());
1852 (*thrd_last_action)[tid] = act;
1854 if (act->is_wait()) {
1855 void *mutex_loc=(void *) act->get_value();
1856 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
1858 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
1859 if (tid >= (int)vec->size())
1860 vec->resize(priv->next_thread_id);
1861 (*vec)[tid].push_back(act);
1863 if ((int)thrd_last_action->size() <= tid)
1864 thrd_last_action->resize(get_num_threads());
1865 (*thrd_last_action)[tid] = act;
1870 * @brief Get the last action performed by a particular Thread
1871 * @param tid The thread ID of the Thread in question
1872 * @return The last action in the thread
1874 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
1876 int threadid = id_to_int(tid);
1877 if (threadid < (int)thrd_last_action->size())
1878 return (*thrd_last_action)[id_to_int(tid)];
1884 * Gets the last memory_order_seq_cst write (in the total global sequence)
1885 * performed on a particular object (i.e., memory location), not including the
1887 * @param curr The current ModelAction; also denotes the object location to
1889 * @return The last seq_cst write
1891 ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr) const
1893 void *location = curr->get_location();
1894 action_list_t *list = get_safe_ptr_action(obj_map, location);
1895 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1896 action_list_t::reverse_iterator rit;
1897 for (rit = list->rbegin(); rit != list->rend(); rit++)
1898 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
1904 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1905 * location). This function identifies the mutex according to the current
1906 * action, which is presumed to perform on the same mutex.
1907 * @param curr The current ModelAction; also denotes the object location to
1909 * @return The last unlock operation
1911 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
1913 void *location = curr->get_location();
1914 action_list_t *list = get_safe_ptr_action(obj_map, location);
1915 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1916 action_list_t::reverse_iterator rit;
1917 for (rit = list->rbegin(); rit != list->rend(); rit++)
1918 if ((*rit)->is_unlock() || (*rit)->is_wait())
1923 ModelAction * ModelChecker::get_parent_action(thread_id_t tid)
1925 ModelAction *parent = get_last_action(tid);
1927 parent = get_thread(tid)->get_creation();
1932 * Returns the clock vector for a given thread.
1933 * @param tid The thread whose clock vector we want
1934 * @return Desired clock vector
1936 ClockVector * ModelChecker::get_cv(thread_id_t tid)
1938 return get_parent_action(tid)->get_cv();
1942 * Resolve a set of Promises with a current write. The set is provided in the
1943 * Node corresponding to @a write.
1944 * @param write The ModelAction that is fulfilling Promises
1945 * @return True if promises were resolved; false otherwise
1947 bool ModelChecker::resolve_promises(ModelAction *write)
1949 bool resolved = false;
1950 std::vector< thread_id_t, ModelAlloc<thread_id_t> > threads_to_check;
1952 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
1953 Promise *promise = (*promises)[promise_index];
1954 if (write->get_node()->get_promise(i)) {
1955 ModelAction *read = promise->get_action();
1956 if (read->is_rmw()) {
1957 mo_graph->addRMWEdge(write, read);
1959 read->read_from(write);
1960 //First fix up the modification order for actions that happened
1962 r_modification_order(read, write);
1963 //Next fix up the modification order for actions that happened
1965 post_r_modification_order(read, write);
1966 //Make sure the promise's value matches the write's value
1967 ASSERT(promise->get_value() == write->get_value());
1970 promises->erase(promises->begin() + promise_index);
1971 threads_to_check.push_back(read->get_tid());
1978 //Check whether reading these writes has made threads unable to
1981 for(unsigned int i=0;i<threads_to_check.size();i++)
1982 mo_check_promises(threads_to_check[i], write);
1988 * Compute the set of promises that could potentially be satisfied by this
1989 * action. Note that the set computation actually appears in the Node, not in
1991 * @param curr The ModelAction that may satisfy promises
1993 void ModelChecker::compute_promises(ModelAction *curr)
1995 for (unsigned int i = 0; i < promises->size(); i++) {
1996 Promise *promise = (*promises)[i];
1997 const ModelAction *act = promise->get_action();
1998 if (!act->happens_before(curr) &&
2000 !act->could_synchronize_with(curr) &&
2001 !act->same_thread(curr) &&
2002 act->get_location() == curr->get_location() &&
2003 promise->get_value() == curr->get_value()) {
2004 curr->get_node()->set_promise(i, act->is_rmw());
2009 /** Checks promises in response to change in ClockVector Threads. */
2010 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2012 for (unsigned int i = 0; i < promises->size(); i++) {
2013 Promise *promise = (*promises)[i];
2014 const ModelAction *act = promise->get_action();
2015 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
2016 merge_cv->synchronized_since(act)) {
2017 if (promise->increment_threads(tid)) {
2018 //Promise has failed
2019 failed_promise = true;
2026 void ModelChecker::check_promises_thread_disabled() {
2027 for (unsigned int i = 0; i < promises->size(); i++) {
2028 Promise *promise = (*promises)[i];
2029 if (promise->check_promise()) {
2030 failed_promise = true;
2036 /** Checks promises in response to addition to modification order for threads.
2038 * pthread is the thread that performed the read that created the promise
2040 * pread is the read that created the promise
2042 * pwrite is either the first write to same location as pread by
2043 * pthread that is sequenced after pread or the value read by the
2044 * first read to the same lcoation as pread by pthread that is
2045 * sequenced after pread..
2047 * 1. If tid=pthread, then we check what other threads are reachable
2048 * through the mode order starting with pwrite. Those threads cannot
2049 * perform a write that will resolve the promise due to modification
2050 * order constraints.
2052 * 2. If the tid is not pthread, we check whether pwrite can reach the
2053 * action write through the modification order. If so, that thread
2054 * cannot perform a future write that will resolve the promise due to
2055 * modificatin order constraints.
2057 * @parem tid The thread that either read from the model action
2058 * write, or actually did the model action write.
2060 * @parem write The ModelAction representing the relevant write.
2063 void ModelChecker::mo_check_promises(thread_id_t tid, const ModelAction *write) {
2064 void * location = write->get_location();
2065 for (unsigned int i = 0; i < promises->size(); i++) {
2066 Promise *promise = (*promises)[i];
2067 const ModelAction *act = promise->get_action();
2069 //Is this promise on the same location?
2070 if ( act->get_location() != location )
2073 //same thread as the promise
2074 if ( act->get_tid()==tid ) {
2076 //do we have a pwrite for the promise, if not, set it
2077 if (promise->get_write() == NULL ) {
2078 promise->set_write(write);
2079 //The pwrite cannot happen before the promise
2080 if (write->happens_before(act) && (write != act)) {
2081 failed_promise = true;
2085 if (mo_graph->checkPromise(write, promise)) {
2086 failed_promise = true;
2091 //Don't do any lookups twice for the same thread
2092 if (promise->has_sync_thread(tid))
2095 if (promise->get_write()&&mo_graph->checkReachable(promise->get_write(), write)) {
2096 if (promise->increment_threads(tid)) {
2097 failed_promise = true;
2105 * Compute the set of writes that may break the current pending release
2106 * sequence. This information is extracted from previou release sequence
2109 * @param curr The current ModelAction. Must be a release sequence fixup
2112 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2114 if (pending_rel_seqs->empty())
2117 struct release_seq *pending = pending_rel_seqs->back();
2118 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2119 const ModelAction *write = pending->writes[i];
2120 curr->get_node()->add_relseq_break(write);
2123 /* NULL means don't break the sequence; just synchronize */
2124 curr->get_node()->add_relseq_break(NULL);
2128 * Build up an initial set of all past writes that this 'read' action may read
2129 * from. This set is determined by the clock vector's "happens before"
2131 * @param curr is the current ModelAction that we are exploring; it must be a
2134 void ModelChecker::build_reads_from_past(ModelAction *curr)
2136 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2138 ASSERT(curr->is_read());
2140 ModelAction *last_seq_cst = NULL;
2142 /* Track whether this object has been initialized */
2143 bool initialized = false;
2145 if (curr->is_seqcst()) {
2146 last_seq_cst = get_last_seq_cst(curr);
2147 /* We have to at least see the last sequentially consistent write,
2148 so we are initialized. */
2149 if (last_seq_cst != NULL)
2153 /* Iterate over all threads */
2154 for (i = 0; i < thrd_lists->size(); i++) {
2155 /* Iterate over actions in thread, starting from most recent */
2156 action_list_t *list = &(*thrd_lists)[i];
2157 action_list_t::reverse_iterator rit;
2158 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2159 ModelAction *act = *rit;
2161 /* Only consider 'write' actions */
2162 if (!act->is_write() || act == curr)
2165 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2166 if (!curr->is_seqcst() || (!act->is_seqcst() && (last_seq_cst == NULL || !act->happens_before(last_seq_cst))) || act == last_seq_cst) {
2167 if (!curr->get_sleep_flag() || curr->is_seqcst() || sleep_can_read_from(curr, act)) {
2168 DEBUG("Adding action to may_read_from:\n");
2169 if (DBG_ENABLED()) {
2173 curr->get_node()->add_read_from(act);
2177 /* Include at most one act per-thread that "happens before" curr */
2178 if (act->happens_before(curr)) {
2186 /** @todo Need a more informative way of reporting errors. */
2187 printf("ERROR: may read from uninitialized atomic\n");
2191 if (DBG_ENABLED() || !initialized) {
2192 printf("Reached read action:\n");
2194 printf("Printing may_read_from\n");
2195 curr->get_node()->print_may_read_from();
2196 printf("End printing may_read_from\n");
2200 bool ModelChecker::sleep_can_read_from(ModelAction * curr, const ModelAction *write) {
2202 Node *prevnode=write->get_node()->get_parent();
2204 bool thread_sleep=prevnode->enabled_status(curr->get_tid())==THREAD_SLEEP_SET;
2205 if (write->is_release()&&thread_sleep)
2207 if (!write->is_rmw()) {
2210 if (write->get_reads_from()==NULL)
2212 write=write->get_reads_from();
2216 static void print_list(action_list_t *list)
2218 action_list_t::iterator it;
2220 printf("---------------------------------------------------------------------\n");
2222 unsigned int hash=0;
2224 for (it = list->begin(); it != list->end(); it++) {
2226 hash=hash^(hash<<3)^((*it)->hash());
2228 printf("HASH %u\n", hash);
2229 printf("---------------------------------------------------------------------\n");
2232 #if SUPPORT_MOD_ORDER_DUMP
2233 void ModelChecker::dumpGraph(char *filename) {
2235 sprintf(buffer, "%s.dot",filename);
2236 FILE *file=fopen(buffer, "w");
2237 fprintf(file, "digraph %s {\n",filename);
2238 mo_graph->dumpNodes(file);
2239 ModelAction ** thread_array=(ModelAction **)model_calloc(1, sizeof(ModelAction *)*get_num_threads());
2241 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2242 ModelAction *action=*it;
2243 if (action->is_read()) {
2244 fprintf(file, "N%u [label=\"%u, T%u\"];\n", action->get_seq_number(),action->get_seq_number(), action->get_tid());
2245 if (action->get_reads_from()!=NULL)
2246 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2248 if (thread_array[action->get_tid()] != NULL) {
2249 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2252 thread_array[action->get_tid()]=action;
2254 fprintf(file,"}\n");
2255 model_free(thread_array);
2260 void ModelChecker::print_summary()
2263 printf("Number of executions: %d\n", num_executions);
2264 printf("Number of feasible executions: %d\n", num_feasible_executions);
2265 printf("Total nodes created: %d\n", node_stack->get_total_nodes());
2267 #if SUPPORT_MOD_ORDER_DUMP
2269 char buffername[100];
2270 sprintf(buffername, "exec%04u", num_executions);
2271 mo_graph->dumpGraphToFile(buffername);
2272 sprintf(buffername, "graph%04u", num_executions);
2273 dumpGraph(buffername);
2276 if (!isfinalfeasible())
2277 printf("INFEASIBLE EXECUTION!\n");
2278 print_list(action_trace);
2283 * Add a Thread to the system for the first time. Should only be called once
2285 * @param t The Thread to add
2287 void ModelChecker::add_thread(Thread *t)
2289 thread_map->put(id_to_int(t->get_id()), t);
2290 scheduler->add_thread(t);
2294 * Removes a thread from the scheduler.
2295 * @param the thread to remove.
2297 void ModelChecker::remove_thread(Thread *t)
2299 scheduler->remove_thread(t);
2303 * @brief Get a Thread reference by its ID
2304 * @param tid The Thread's ID
2305 * @return A Thread reference
2307 Thread * ModelChecker::get_thread(thread_id_t tid) const
2309 return thread_map->get(id_to_int(tid));
2313 * @brief Get a reference to the Thread in which a ModelAction was executed
2314 * @param act The ModelAction
2315 * @return A Thread reference
2317 Thread * ModelChecker::get_thread(ModelAction *act) const
2319 return get_thread(act->get_tid());
2323 * @brief Check if a Thread is currently enabled
2324 * @param t The Thread to check
2325 * @return True if the Thread is currently enabled
2327 bool ModelChecker::is_enabled(Thread *t) const
2329 return scheduler->is_enabled(t);
2333 * @brief Check if a Thread is currently enabled
2334 * @param tid The ID of the Thread to check
2335 * @return True if the Thread is currently enabled
2337 bool ModelChecker::is_enabled(thread_id_t tid) const
2339 return scheduler->is_enabled(tid);
2343 * Switch from a user-context to the "master thread" context (a.k.a. system
2344 * context). This switch is made with the intention of exploring a particular
2345 * model-checking action (described by a ModelAction object). Must be called
2346 * from a user-thread context.
2348 * @param act The current action that will be explored. May be NULL only if
2349 * trace is exiting via an assertion (see ModelChecker::set_assert and
2350 * ModelChecker::has_asserted).
2351 * @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
2353 int ModelChecker::switch_to_master(ModelAction *act)
2356 Thread *old = thread_current();
2357 set_current_action(act);
2358 old->set_state(THREAD_READY);
2359 return Thread::swap(old, &system_context);
2363 * Takes the next step in the execution, if possible.
2364 * @return Returns true (success) if a step was taken and false otherwise.
2366 bool ModelChecker::take_step() {
2370 Thread *curr = priv->current_action ? get_thread(priv->current_action) : NULL;
2372 if (curr->get_state() == THREAD_READY) {
2373 ASSERT(priv->current_action);
2375 priv->nextThread = check_current_action(priv->current_action);
2376 priv->current_action = NULL;
2378 if (curr->is_blocked() || curr->is_complete())
2379 scheduler->remove_thread(curr);
2384 Thread *next = scheduler->next_thread(priv->nextThread);
2386 /* Infeasible -> don't take any more steps */
2389 else if (isfeasibleprefix() && have_bug_reports()) {
2394 if (params.bound != 0) {
2395 if (priv->used_sequence_numbers > params.bound) {
2400 DEBUG("(%d, %d)\n", curr ? id_to_int(curr->get_id()) : -1,
2401 next ? id_to_int(next->get_id()) : -1);
2404 * Launch end-of-execution release sequence fixups only when there are:
2406 * (1) no more user threads to run (or when execution replay chooses
2407 * the 'model_thread')
2408 * (2) pending release sequences
2409 * (3) pending assertions (i.e., data races)
2410 * (4) no pending promises
2412 if (!pending_rel_seqs->empty() && (!next || next->is_model_thread()) &&
2413 isfinalfeasible() && !unrealizedraces.empty()) {
2414 printf("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2415 pending_rel_seqs->size());
2416 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2417 std::memory_order_seq_cst, NULL, VALUE_NONE,
2419 set_current_action(fixup);
2423 /* next == NULL -> don't take any more steps */
2427 next->set_state(THREAD_RUNNING);
2429 if (next->get_pending() != NULL) {
2430 /* restart a pending action */
2431 set_current_action(next->get_pending());
2432 next->set_pending(NULL);
2433 next->set_state(THREAD_READY);
2437 /* Return false only if swap fails with an error */
2438 return (Thread::swap(&system_context, next) == 0);
2441 /** Runs the current execution until threre are no more steps to take. */
2442 void ModelChecker::finish_execution() {
2445 while (take_step());