10 #include "snapshot-interface.h"
12 #include "clockvector.h"
13 #include "cyclegraph.h"
16 #include "threads-model.h"
19 #define INITIAL_THREAD_ID 0
24 bug_message(const char *str) {
25 const char *fmt = " [BUG] %s\n";
26 msg = (char *)snapshot_malloc(strlen(fmt) + strlen(str));
27 sprintf(msg, fmt, str);
29 ~bug_message() { if (msg) snapshot_free(msg); }
32 void print() { model_print("%s", msg); }
38 * Structure for holding small ModelChecker members that should be snapshotted
40 struct model_snapshot_members {
41 model_snapshot_members() :
42 /* First thread created will have id INITIAL_THREAD_ID */
43 next_thread_id(INITIAL_THREAD_ID),
44 used_sequence_numbers(0),
48 failed_promise(false),
49 too_many_reads(false),
50 no_valid_reads(false),
51 bad_synchronization(false),
55 ~model_snapshot_members() {
56 for (unsigned int i = 0; i < bugs.size(); i++)
61 unsigned int next_thread_id;
62 modelclock_t used_sequence_numbers;
63 ModelAction *next_backtrack;
64 std::vector< bug_message *, SnapshotAlloc<bug_message *> > bugs;
65 struct execution_stats stats;
69 /** @brief Incorrectly-ordered synchronization was made */
70 bool bad_synchronization;
76 /** @brief Constructor */
77 ModelChecker::ModelChecker(struct model_params params) :
78 /* Initialize default scheduler */
80 scheduler(new Scheduler()),
82 earliest_diverge(NULL),
83 action_trace(new action_list_t()),
84 thread_map(new HashTable<int, Thread *, int>()),
85 obj_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
86 lock_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
87 condvar_waiters_map(new HashTable<const void *, action_list_t *, uintptr_t, 4>()),
88 obj_thrd_map(new HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4 >()),
89 promises(new std::vector< Promise *, SnapshotAlloc<Promise *> >()),
90 futurevalues(new std::vector< struct PendingFutureValue, SnapshotAlloc<struct PendingFutureValue> >()),
91 pending_rel_seqs(new std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >()),
92 thrd_last_action(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >(1)),
93 thrd_last_fence_release(new std::vector< ModelAction *, SnapshotAlloc<ModelAction *> >()),
94 node_stack(new NodeStack()),
95 priv(new struct model_snapshot_members()),
96 mo_graph(new CycleGraph())
98 /* Initialize a model-checker thread, for special ModelActions */
99 model_thread = new Thread(get_next_id());
100 thread_map->put(id_to_int(model_thread->get_id()), model_thread);
103 /** @brief Destructor */
104 ModelChecker::~ModelChecker()
106 for (unsigned int i = 0; i < get_num_threads(); i++)
107 delete thread_map->get(i);
112 delete lock_waiters_map;
113 delete condvar_waiters_map;
116 for (unsigned int i = 0; i < promises->size(); i++)
117 delete (*promises)[i];
120 delete pending_rel_seqs;
122 delete thrd_last_action;
123 delete thrd_last_fence_release;
130 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
132 action_list_t *tmp = hash->get(ptr);
134 tmp = new action_list_t();
140 static std::vector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, std::vector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
142 std::vector<action_list_t> *tmp = hash->get(ptr);
144 tmp = new std::vector<action_list_t>();
151 * Restores user program to initial state and resets all model-checker data
154 void ModelChecker::reset_to_initial_state()
156 DEBUG("+++ Resetting to initial state +++\n");
157 node_stack->reset_execution();
159 /* Print all model-checker output before rollback */
163 * FIXME: if we utilize partial rollback, we will need to free only
164 * those pending actions which were NOT pending before the rollback
167 for (unsigned int i = 0; i < get_num_threads(); i++)
168 delete get_thread(int_to_id(i))->get_pending();
170 snapshot_backtrack_before(0);
173 /** @return a thread ID for a new Thread */
174 thread_id_t ModelChecker::get_next_id()
176 return priv->next_thread_id++;
179 /** @return the number of user threads created during this execution */
180 unsigned int ModelChecker::get_num_threads() const
182 return priv->next_thread_id;
186 * Must be called from user-thread context (e.g., through the global
187 * thread_current() interface)
189 * @return The currently executing Thread.
191 Thread * ModelChecker::get_current_thread() const
193 return scheduler->get_current_thread();
196 /** @return a sequence number for a new ModelAction */
197 modelclock_t ModelChecker::get_next_seq_num()
199 return ++priv->used_sequence_numbers;
202 Node * ModelChecker::get_curr_node() const
204 return node_stack->get_head();
208 * @brief Choose the next thread to execute.
210 * This function chooses the next thread that should execute. It can force the
211 * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
212 * followed by a THREAD_START, or it can enforce execution replay/backtracking.
213 * The model-checker may have no preference regarding the next thread (i.e.,
214 * when exploring a new execution ordering), in which case this will return
216 * @param curr The current ModelAction. This action might guide the choice of
218 * @return The next thread to run. If the model-checker has no preference, NULL.
220 Thread * ModelChecker::get_next_thread(ModelAction *curr)
225 /* Do not split atomic actions. */
227 return get_thread(curr);
228 else if (curr->get_type() == THREAD_CREATE)
229 return curr->get_thread_operand();
232 /* Have we completed exploring the preselected path? */
236 /* Else, we are trying to replay an execution */
237 ModelAction *next = node_stack->get_next()->get_action();
239 if (next == diverge) {
240 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
241 earliest_diverge = diverge;
243 Node *nextnode = next->get_node();
244 Node *prevnode = nextnode->get_parent();
245 scheduler->update_sleep_set(prevnode);
247 /* Reached divergence point */
248 if (nextnode->increment_misc()) {
249 /* The next node will try to satisfy a different misc_index values. */
250 tid = next->get_tid();
251 node_stack->pop_restofstack(2);
252 } else if (nextnode->increment_promise()) {
253 /* The next node will try to satisfy a different set of promises. */
254 tid = next->get_tid();
255 node_stack->pop_restofstack(2);
256 } else if (nextnode->increment_read_from()) {
257 /* The next node will read from a different value. */
258 tid = next->get_tid();
259 node_stack->pop_restofstack(2);
260 } else if (nextnode->increment_future_value()) {
261 /* The next node will try to read from a different future value. */
262 tid = next->get_tid();
263 node_stack->pop_restofstack(2);
264 } else if (nextnode->increment_relseq_break()) {
265 /* The next node will try to resolve a release sequence differently */
266 tid = next->get_tid();
267 node_stack->pop_restofstack(2);
270 /* Make a different thread execute for next step */
271 scheduler->add_sleep(get_thread(next->get_tid()));
272 tid = prevnode->get_next_backtrack();
273 /* Make sure the backtracked thread isn't sleeping. */
274 node_stack->pop_restofstack(1);
275 if (diverge == earliest_diverge) {
276 earliest_diverge = prevnode->get_action();
279 /* The correct sleep set is in the parent node. */
282 DEBUG("*** Divergence point ***\n");
286 tid = next->get_tid();
288 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
289 ASSERT(tid != THREAD_ID_T_NONE);
290 return thread_map->get(id_to_int(tid));
294 * We need to know what the next actions of all threads in the sleep
295 * set will be. This method computes them and stores the actions at
296 * the corresponding thread object's pending action.
299 void ModelChecker::execute_sleep_set()
301 for (unsigned int i = 0; i < get_num_threads(); i++) {
302 thread_id_t tid = int_to_id(i);
303 Thread *thr = get_thread(tid);
304 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
305 thr->get_pending()->set_sleep_flag();
310 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
312 for (unsigned int i = 0; i < get_num_threads(); i++) {
313 Thread *thr = get_thread(int_to_id(i));
314 if (scheduler->is_sleep_set(thr)) {
315 ModelAction *pending_act = thr->get_pending();
316 if ((!curr->is_rmwr()) && pending_act->could_synchronize_with(curr))
317 //Remove this thread from sleep set
318 scheduler->remove_sleep(thr);
323 /** @brief Alert the model-checker that an incorrectly-ordered
324 * synchronization was made */
325 void ModelChecker::set_bad_synchronization()
327 priv->bad_synchronization = true;
330 bool ModelChecker::has_asserted() const
332 return priv->asserted;
335 void ModelChecker::set_assert()
337 priv->asserted = true;
341 * Check if we are in a deadlock. Should only be called at the end of an
342 * execution, although it should not give false positives in the middle of an
343 * execution (there should be some ENABLED thread).
345 * @return True if program is in a deadlock; false otherwise
347 bool ModelChecker::is_deadlocked() const
349 bool blocking_threads = false;
350 for (unsigned int i = 0; i < get_num_threads(); i++) {
351 thread_id_t tid = int_to_id(i);
354 Thread *t = get_thread(tid);
355 if (!t->is_model_thread() && t->get_pending())
356 blocking_threads = true;
358 return blocking_threads;
362 * Check if this is a complete execution. That is, have all thread completed
363 * execution (rather than exiting because sleep sets have forced a redundant
366 * @return True if the execution is complete.
368 bool ModelChecker::is_complete_execution() const
370 for (unsigned int i = 0; i < get_num_threads(); i++)
371 if (is_enabled(int_to_id(i)))
377 * @brief Assert a bug in the executing program.
379 * Use this function to assert any sort of bug in the user program. If the
380 * current trace is feasible (actually, a prefix of some feasible execution),
381 * then this execution will be aborted, printing the appropriate message. If
382 * the current trace is not yet feasible, the error message will be stashed and
383 * printed if the execution ever becomes feasible.
385 * @param msg Descriptive message for the bug (do not include newline char)
386 * @return True if bug is immediately-feasible
388 bool ModelChecker::assert_bug(const char *msg)
390 priv->bugs.push_back(new bug_message(msg));
392 if (isfeasibleprefix()) {
400 * @brief Assert a bug in the executing program, asserted by a user thread
401 * @see ModelChecker::assert_bug
402 * @param msg Descriptive message for the bug (do not include newline char)
404 void ModelChecker::assert_user_bug(const char *msg)
406 /* If feasible bug, bail out now */
408 switch_to_master(NULL);
411 /** @return True, if any bugs have been reported for this execution */
412 bool ModelChecker::have_bug_reports() const
414 return priv->bugs.size() != 0;
417 /** @brief Print bug report listing for this execution (if any bugs exist) */
418 void ModelChecker::print_bugs() const
420 if (have_bug_reports()) {
421 model_print("Bug report: %zu bug%s detected\n",
423 priv->bugs.size() > 1 ? "s" : "");
424 for (unsigned int i = 0; i < priv->bugs.size(); i++)
425 priv->bugs[i]->print();
430 * @brief Record end-of-execution stats
432 * Must be run when exiting an execution. Records various stats.
433 * @see struct execution_stats
435 void ModelChecker::record_stats()
438 if (!isfeasibleprefix())
439 stats.num_infeasible++;
440 else if (have_bug_reports())
441 stats.num_buggy_executions++;
442 else if (is_complete_execution())
443 stats.num_complete++;
445 stats.num_redundant++;
448 /** @brief Print execution stats */
449 void ModelChecker::print_stats() const
451 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
452 model_print("Number of redundant executions: %d\n", stats.num_redundant);
453 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
454 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
455 model_print("Total executions: %d\n", stats.num_total);
456 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
460 * @brief End-of-exeuction print
461 * @param printbugs Should any existing bugs be printed?
463 void ModelChecker::print_execution(bool printbugs) const
465 print_program_output();
467 if (DBG_ENABLED() || params.verbose) {
468 model_print("Earliest divergence point since last feasible execution:\n");
469 if (earliest_diverge)
470 earliest_diverge->print();
472 model_print("(Not set)\n");
478 /* Don't print invalid bugs */
487 * Queries the model-checker for more executions to explore and, if one
488 * exists, resets the model-checker state to execute a new execution.
490 * @return If there are more executions to explore, return true. Otherwise,
493 bool ModelChecker::next_execution()
496 /* Is this execution a feasible execution that's worth bug-checking? */
497 bool complete = isfeasibleprefix() && (is_complete_execution() ||
500 /* End-of-execution bug checks */
503 assert_bug("Deadlock detected");
511 if (DBG_ENABLED() || params.verbose || (complete && have_bug_reports()))
512 print_execution(complete);
514 clear_program_output();
517 earliest_diverge = NULL;
519 if ((diverge = get_next_backtrack()) == NULL)
523 model_print("Next execution will diverge at:\n");
527 reset_to_initial_state();
531 ModelAction * ModelChecker::get_last_conflict(ModelAction *act)
533 switch (act->get_type()) {
538 /* Optimization: relaxed operations don't need backtracking */
539 if (act->is_relaxed())
541 /* linear search: from most recent to oldest */
542 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
543 action_list_t::reverse_iterator rit;
544 for (rit = list->rbegin(); rit != list->rend(); rit++) {
545 ModelAction *prev = *rit;
546 if (prev->could_synchronize_with(act))
552 case ATOMIC_TRYLOCK: {
553 /* linear search: from most recent to oldest */
554 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
555 action_list_t::reverse_iterator rit;
556 for (rit = list->rbegin(); rit != list->rend(); rit++) {
557 ModelAction *prev = *rit;
558 if (act->is_conflicting_lock(prev))
563 case ATOMIC_UNLOCK: {
564 /* linear search: from most recent to oldest */
565 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
566 action_list_t::reverse_iterator rit;
567 for (rit = list->rbegin(); rit != list->rend(); rit++) {
568 ModelAction *prev = *rit;
569 if (!act->same_thread(prev) && prev->is_failed_trylock())
575 /* linear search: from most recent to oldest */
576 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
577 action_list_t::reverse_iterator rit;
578 for (rit = list->rbegin(); rit != list->rend(); rit++) {
579 ModelAction *prev = *rit;
580 if (!act->same_thread(prev) && prev->is_failed_trylock())
582 if (!act->same_thread(prev) && prev->is_notify())
588 case ATOMIC_NOTIFY_ALL:
589 case ATOMIC_NOTIFY_ONE: {
590 /* linear search: from most recent to oldest */
591 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
592 action_list_t::reverse_iterator rit;
593 for (rit = list->rbegin(); rit != list->rend(); rit++) {
594 ModelAction *prev = *rit;
595 if (!act->same_thread(prev) && prev->is_wait())
606 /** This method finds backtracking points where we should try to
607 * reorder the parameter ModelAction against.
609 * @param the ModelAction to find backtracking points for.
611 void ModelChecker::set_backtracking(ModelAction *act)
613 Thread *t = get_thread(act);
614 ModelAction *prev = get_last_conflict(act);
618 Node *node = prev->get_node()->get_parent();
620 int low_tid, high_tid;
621 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
622 low_tid = id_to_int(act->get_tid());
623 high_tid = low_tid + 1;
626 high_tid = get_num_threads();
629 for (int i = low_tid; i < high_tid; i++) {
630 thread_id_t tid = int_to_id(i);
632 /* Make sure this thread can be enabled here. */
633 if (i >= node->get_num_threads())
636 /* Don't backtrack into a point where the thread is disabled or sleeping. */
637 if (node->enabled_status(tid) != THREAD_ENABLED)
640 /* Check if this has been explored already */
641 if (node->has_been_explored(tid))
644 /* See if fairness allows */
645 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
647 for (int t = 0; t < node->get_num_threads(); t++) {
648 thread_id_t tother = int_to_id(t);
649 if (node->is_enabled(tother) && node->has_priority(tother)) {
657 /* Cache the latest backtracking point */
658 set_latest_backtrack(prev);
660 /* If this is a new backtracking point, mark the tree */
661 if (!node->set_backtrack(tid))
663 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
664 id_to_int(prev->get_tid()),
665 id_to_int(t->get_id()));
674 * @brief Cache the a backtracking point as the "most recent", if eligible
676 * Note that this does not prepare the NodeStack for this backtracking
677 * operation, it only caches the action on a per-execution basis
679 * @param act The operation at which we should explore a different next action
680 * (i.e., backtracking point)
681 * @return True, if this action is now the most recent backtracking point;
684 bool ModelChecker::set_latest_backtrack(ModelAction *act)
686 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
687 priv->next_backtrack = act;
694 * Returns last backtracking point. The model checker will explore a different
695 * path for this point in the next execution.
696 * @return The ModelAction at which the next execution should diverge.
698 ModelAction * ModelChecker::get_next_backtrack()
700 ModelAction *next = priv->next_backtrack;
701 priv->next_backtrack = NULL;
706 * Processes a read or rmw model action.
707 * @param curr is the read model action to process.
708 * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
709 * @return True if processing this read updates the mo_graph.
711 bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
713 uint64_t value = VALUE_NONE;
714 bool updated = false;
716 const ModelAction *reads_from = curr->get_node()->get_read_from();
717 if (reads_from != NULL) {
718 mo_graph->startChanges();
720 value = reads_from->get_value();
721 bool r_status = false;
723 if (!second_part_of_rmw) {
724 check_recency(curr, reads_from);
725 r_status = r_modification_order(curr, reads_from);
728 if (!second_part_of_rmw && is_infeasible() && (curr->get_node()->increment_read_from() || curr->get_node()->increment_future_value())) {
729 mo_graph->rollbackChanges();
730 priv->too_many_reads = false;
734 read_from(curr, reads_from);
735 mo_graph->commitChanges();
736 mo_check_promises(curr, true);
739 } else if (!second_part_of_rmw) {
740 /* Read from future value */
741 struct future_value fv = curr->get_node()->get_future_value();
742 Promise *promise = new Promise(curr, fv);
744 curr->set_read_from_promise(promise);
745 promises->push_back(promise);
746 mo_graph->startChanges();
747 updated = r_modification_order(curr, promise);
748 mo_graph->commitChanges();
750 get_thread(curr)->set_return_value(value);
756 * Processes a lock, trylock, or unlock model action. @param curr is
757 * the read model action to process.
759 * The try lock operation checks whether the lock is taken. If not,
760 * it falls to the normal lock operation case. If so, it returns
763 * The lock operation has already been checked that it is enabled, so
764 * it just grabs the lock and synchronizes with the previous unlock.
766 * The unlock operation has to re-enable all of the threads that are
767 * waiting on the lock.
769 * @return True if synchronization was updated; false otherwise
771 bool ModelChecker::process_mutex(ModelAction *curr)
773 std::mutex *mutex = NULL;
774 struct std::mutex_state *state = NULL;
776 if (curr->is_trylock() || curr->is_lock() || curr->is_unlock()) {
777 mutex = (std::mutex *)curr->get_location();
778 state = mutex->get_state();
779 } else if (curr->is_wait()) {
780 mutex = (std::mutex *)curr->get_value();
781 state = mutex->get_state();
784 switch (curr->get_type()) {
785 case ATOMIC_TRYLOCK: {
786 bool success = !state->islocked;
787 curr->set_try_lock(success);
789 get_thread(curr)->set_return_value(0);
792 get_thread(curr)->set_return_value(1);
794 //otherwise fall into the lock case
796 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
797 assert_bug("Lock access before initialization");
798 state->islocked = true;
799 ModelAction *unlock = get_last_unlock(curr);
800 //synchronize with the previous unlock statement
801 if (unlock != NULL) {
802 curr->synchronize_with(unlock);
807 case ATOMIC_UNLOCK: {
809 state->islocked = false;
810 //wake up the other threads
811 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
812 //activate all the waiting threads
813 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
814 scheduler->wake(get_thread(*rit));
821 state->islocked = false;
822 //wake up the other threads
823 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
824 //activate all the waiting threads
825 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
826 scheduler->wake(get_thread(*rit));
829 //check whether we should go to sleep or not...simulate spurious failures
830 if (curr->get_node()->get_misc() == 0) {
831 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
833 scheduler->sleep(get_thread(curr));
837 case ATOMIC_NOTIFY_ALL: {
838 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
839 //activate all the waiting threads
840 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
841 scheduler->wake(get_thread(*rit));
846 case ATOMIC_NOTIFY_ONE: {
847 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
848 int wakeupthread = curr->get_node()->get_misc();
849 action_list_t::iterator it = waiters->begin();
850 advance(it, wakeupthread);
851 scheduler->wake(get_thread(*it));
862 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
864 /* Do more ambitious checks now that mo is more complete */
865 if (mo_may_allow(writer, reader)) {
866 Node *node = reader->get_node();
868 /* Find an ancestor thread which exists at the time of the reader */
869 Thread *write_thread = get_thread(writer);
870 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
871 write_thread = write_thread->get_parent();
873 struct future_value fv = {
875 writer->get_seq_number() + params.maxfuturedelay,
876 write_thread->get_id(),
878 if (node->add_future_value(fv))
879 set_latest_backtrack(reader);
884 * Process a write ModelAction
885 * @param curr The ModelAction to process
886 * @return True if the mo_graph was updated or promises were resolved
888 bool ModelChecker::process_write(ModelAction *curr)
890 bool updated_mod_order = w_modification_order(curr);
891 bool updated_promises = resolve_promises(curr);
893 if (promises->size() == 0) {
894 for (unsigned int i = 0; i < futurevalues->size(); i++) {
895 struct PendingFutureValue pfv = (*futurevalues)[i];
896 add_future_value(pfv.writer, pfv.act);
898 futurevalues->clear();
901 mo_graph->commitChanges();
902 mo_check_promises(curr, false);
904 get_thread(curr)->set_return_value(VALUE_NONE);
905 return updated_mod_order || updated_promises;
909 * Process a fence ModelAction
910 * @param curr The ModelAction to process
911 * @return True if synchronization was updated
913 bool ModelChecker::process_fence(ModelAction *curr)
916 * fence-relaxed: no-op
917 * fence-release: only log the occurence (not in this function), for
918 * use in later synchronization
919 * fence-acquire (this function): search for hypothetical release
922 bool updated = false;
923 if (curr->is_acquire()) {
924 action_list_t *list = action_trace;
925 action_list_t::reverse_iterator rit;
926 /* Find X : is_read(X) && X --sb-> curr */
927 for (rit = list->rbegin(); rit != list->rend(); rit++) {
928 ModelAction *act = *rit;
931 if (act->get_tid() != curr->get_tid())
933 /* Stop at the beginning of the thread */
934 if (act->is_thread_start())
936 /* Stop once we reach a prior fence-acquire */
937 if (act->is_fence() && act->is_acquire())
941 /* read-acquire will find its own release sequences */
942 if (act->is_acquire())
945 /* Establish hypothetical release sequences */
946 rel_heads_list_t release_heads;
947 get_release_seq_heads(curr, act, &release_heads);
948 for (unsigned int i = 0; i < release_heads.size(); i++)
949 if (!curr->synchronize_with(release_heads[i]))
950 set_bad_synchronization();
951 if (release_heads.size() != 0)
959 * @brief Process the current action for thread-related activity
961 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
962 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
963 * synchronization, etc. This function is a no-op for non-THREAD actions
964 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
966 * @param curr The current action
967 * @return True if synchronization was updated or a thread completed
969 bool ModelChecker::process_thread_action(ModelAction *curr)
971 bool updated = false;
973 switch (curr->get_type()) {
974 case THREAD_CREATE: {
975 thrd_t *thrd = (thrd_t *)curr->get_location();
976 struct thread_params *params = (struct thread_params *)curr->get_value();
977 Thread *th = new Thread(thrd, params->func, params->arg);
979 th->set_creation(curr);
980 /* Promises can be satisfied by children */
981 for (unsigned int i = 0; i < promises->size(); i++) {
982 Promise *promise = (*promises)[i];
983 if (promise->thread_is_available(curr->get_tid()))
984 promise->add_thread(th->get_id());
989 Thread *blocking = curr->get_thread_operand();
990 ModelAction *act = get_last_action(blocking->get_id());
991 curr->synchronize_with(act);
992 updated = true; /* trigger rel-seq checks */
995 case THREAD_FINISH: {
996 Thread *th = get_thread(curr);
997 while (!th->wait_list_empty()) {
998 ModelAction *act = th->pop_wait_list();
999 scheduler->wake(get_thread(act));
1002 /* Completed thread can't satisfy promises */
1003 for (unsigned int i = 0; i < promises->size(); i++) {
1004 Promise *promise = (*promises)[i];
1005 if (promise->thread_is_available(th->get_id()))
1006 if (promise->eliminate_thread(th->get_id()))
1007 priv->failed_promise = true;
1009 updated = true; /* trigger rel-seq checks */
1012 case THREAD_START: {
1013 check_promises(curr->get_tid(), NULL, curr->get_cv());
1024 * @brief Process the current action for release sequence fixup activity
1026 * Performs model-checker release sequence fixups for the current action,
1027 * forcing a single pending release sequence to break (with a given, potential
1028 * "loose" write) or to complete (i.e., synchronize). If a pending release
1029 * sequence forms a complete release sequence, then we must perform the fixup
1030 * synchronization, mo_graph additions, etc.
1032 * @param curr The current action; must be a release sequence fixup action
1033 * @param work_queue The work queue to which to add work items as they are
1036 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1038 const ModelAction *write = curr->get_node()->get_relseq_break();
1039 struct release_seq *sequence = pending_rel_seqs->back();
1040 pending_rel_seqs->pop_back();
1042 ModelAction *acquire = sequence->acquire;
1043 const ModelAction *rf = sequence->rf;
1044 const ModelAction *release = sequence->release;
1048 ASSERT(release->same_thread(rf));
1050 if (write == NULL) {
1052 * @todo Forcing a synchronization requires that we set
1053 * modification order constraints. For instance, we can't allow
1054 * a fixup sequence in which two separate read-acquire
1055 * operations read from the same sequence, where the first one
1056 * synchronizes and the other doesn't. Essentially, we can't
1057 * allow any writes to insert themselves between 'release' and
1061 /* Must synchronize */
1062 if (!acquire->synchronize_with(release)) {
1063 set_bad_synchronization();
1066 /* Re-check all pending release sequences */
1067 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1068 /* Re-check act for mo_graph edges */
1069 work_queue->push_back(MOEdgeWorkEntry(acquire));
1071 /* propagate synchronization to later actions */
1072 action_list_t::reverse_iterator rit = action_trace->rbegin();
1073 for (; (*rit) != acquire; rit++) {
1074 ModelAction *propagate = *rit;
1075 if (acquire->happens_before(propagate)) {
1076 propagate->synchronize_with(acquire);
1077 /* Re-check 'propagate' for mo_graph edges */
1078 work_queue->push_back(MOEdgeWorkEntry(propagate));
1082 /* Break release sequence with new edges:
1083 * release --mo--> write --mo--> rf */
1084 mo_graph->addEdge(release, write);
1085 mo_graph->addEdge(write, rf);
1088 /* See if we have realized a data race */
1093 * Initialize the current action by performing one or more of the following
1094 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1095 * in the NodeStack, manipulating backtracking sets, allocating and
1096 * initializing clock vectors, and computing the promises to fulfill.
1098 * @param curr The current action, as passed from the user context; may be
1099 * freed/invalidated after the execution of this function, with a different
1100 * action "returned" its place (pass-by-reference)
1101 * @return True if curr is a newly-explored action; false otherwise
1103 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1105 ModelAction *newcurr;
1107 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1108 newcurr = process_rmw(*curr);
1111 if (newcurr->is_rmw())
1112 compute_promises(newcurr);
1118 (*curr)->set_seq_number(get_next_seq_num());
1120 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1122 /* First restore type and order in case of RMW operation */
1123 if ((*curr)->is_rmwr())
1124 newcurr->copy_typeandorder(*curr);
1126 ASSERT((*curr)->get_location() == newcurr->get_location());
1127 newcurr->copy_from_new(*curr);
1129 /* Discard duplicate ModelAction; use action from NodeStack */
1132 /* Always compute new clock vector */
1133 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1136 return false; /* Action was explored previously */
1140 /* Always compute new clock vector */
1141 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1143 /* Assign most recent release fence */
1144 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1147 * Perform one-time actions when pushing new ModelAction onto
1150 if (newcurr->is_write())
1151 compute_promises(newcurr);
1152 else if (newcurr->is_relseq_fixup())
1153 compute_relseq_breakwrites(newcurr);
1154 else if (newcurr->is_wait())
1155 newcurr->get_node()->set_misc_max(2);
1156 else if (newcurr->is_notify_one()) {
1157 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1159 return true; /* This was a new ModelAction */
1164 * @brief Establish reads-from relation between two actions
1166 * Perform basic operations involved with establishing a concrete rf relation,
1167 * including setting the ModelAction data and checking for release sequences.
1169 * @param act The action that is reading (must be a read)
1170 * @param rf The action from which we are reading (must be a write)
1172 * @return True if this read established synchronization
1174 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1176 act->set_read_from(rf);
1177 if (rf != NULL && act->is_acquire()) {
1178 rel_heads_list_t release_heads;
1179 get_release_seq_heads(act, act, &release_heads);
1180 int num_heads = release_heads.size();
1181 for (unsigned int i = 0; i < release_heads.size(); i++)
1182 if (!act->synchronize_with(release_heads[i])) {
1183 set_bad_synchronization();
1186 return num_heads > 0;
1192 * @brief Check whether a model action is enabled.
1194 * Checks whether a lock or join operation would be successful (i.e., is the
1195 * lock already locked, or is the joined thread already complete). If not, put
1196 * the action in a waiter list.
1198 * @param curr is the ModelAction to check whether it is enabled.
1199 * @return a bool that indicates whether the action is enabled.
1201 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1202 if (curr->is_lock()) {
1203 std::mutex *lock = (std::mutex *)curr->get_location();
1204 struct std::mutex_state *state = lock->get_state();
1205 if (state->islocked) {
1206 //Stick the action in the appropriate waiting queue
1207 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
1210 } else if (curr->get_type() == THREAD_JOIN) {
1211 Thread *blocking = (Thread *)curr->get_location();
1212 if (!blocking->is_complete()) {
1213 blocking->push_wait_list(curr);
1222 * This is the heart of the model checker routine. It performs model-checking
1223 * actions corresponding to a given "current action." Among other processes, it
1224 * calculates reads-from relationships, updates synchronization clock vectors,
1225 * forms a memory_order constraints graph, and handles replay/backtrack
1226 * execution when running permutations of previously-observed executions.
1228 * @param curr The current action to process
1229 * @return The ModelAction that is actually executed; may be different than
1230 * curr; may be NULL, if the current action is not enabled to run
1232 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1235 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1237 if (!check_action_enabled(curr)) {
1238 /* Make the execution look like we chose to run this action
1239 * much later, when a lock/join can succeed */
1240 get_thread(curr)->set_pending(curr);
1241 scheduler->sleep(get_thread(curr));
1245 bool newly_explored = initialize_curr_action(&curr);
1251 wake_up_sleeping_actions(curr);
1253 /* Add the action to lists before any other model-checking tasks */
1254 if (!second_part_of_rmw)
1255 add_action_to_lists(curr);
1257 /* Build may_read_from set for newly-created actions */
1258 if (newly_explored && curr->is_read())
1259 build_reads_from_past(curr);
1261 /* Initialize work_queue with the "current action" work */
1262 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1263 while (!work_queue.empty() && !has_asserted()) {
1264 WorkQueueEntry work = work_queue.front();
1265 work_queue.pop_front();
1267 switch (work.type) {
1268 case WORK_CHECK_CURR_ACTION: {
1269 ModelAction *act = work.action;
1270 bool update = false; /* update this location's release seq's */
1271 bool update_all = false; /* update all release seq's */
1273 if (process_thread_action(curr))
1276 if (act->is_read() && process_read(act, second_part_of_rmw))
1279 if (act->is_write() && process_write(act))
1282 if (act->is_fence() && process_fence(act))
1285 if (act->is_mutex_op() && process_mutex(act))
1288 if (act->is_relseq_fixup())
1289 process_relseq_fixup(curr, &work_queue);
1292 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1294 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1297 case WORK_CHECK_RELEASE_SEQ:
1298 resolve_release_sequences(work.location, &work_queue);
1300 case WORK_CHECK_MO_EDGES: {
1301 /** @todo Complete verification of work_queue */
1302 ModelAction *act = work.action;
1303 bool updated = false;
1305 if (act->is_read()) {
1306 const ModelAction *rf = act->get_reads_from();
1307 const Promise *promise = act->get_reads_from_promise();
1309 if (r_modification_order(act, rf))
1311 } else if (promise) {
1312 if (r_modification_order(act, promise))
1316 if (act->is_write()) {
1317 if (w_modification_order(act))
1320 mo_graph->commitChanges();
1323 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1332 check_curr_backtracking(curr);
1333 set_backtracking(curr);
1337 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1339 Node *currnode = curr->get_node();
1340 Node *parnode = currnode->get_parent();
1342 if ((parnode && !parnode->backtrack_empty()) ||
1343 !currnode->misc_empty() ||
1344 !currnode->read_from_empty() ||
1345 !currnode->future_value_empty() ||
1346 !currnode->promise_empty() ||
1347 !currnode->relseq_break_empty()) {
1348 set_latest_backtrack(curr);
1352 bool ModelChecker::promises_expired() const
1354 for (unsigned int i = 0; i < promises->size(); i++) {
1355 Promise *promise = (*promises)[i];
1356 if (promise->get_expiration() < priv->used_sequence_numbers)
1363 * This is the strongest feasibility check available.
1364 * @return whether the current trace (partial or complete) must be a prefix of
1367 bool ModelChecker::isfeasibleprefix() const
1369 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1373 * Print disagnostic information about an infeasible execution
1374 * @param prefix A string to prefix the output with; if NULL, then a default
1375 * message prefix will be provided
1377 void ModelChecker::print_infeasibility(const char *prefix) const
1381 if (mo_graph->checkForCycles())
1382 ptr += sprintf(ptr, "[mo cycle]");
1383 if (priv->failed_promise)
1384 ptr += sprintf(ptr, "[failed promise]");
1385 if (priv->too_many_reads)
1386 ptr += sprintf(ptr, "[too many reads]");
1387 if (priv->no_valid_reads)
1388 ptr += sprintf(ptr, "[no valid reads-from]");
1389 if (priv->bad_synchronization)
1390 ptr += sprintf(ptr, "[bad sw ordering]");
1391 if (promises_expired())
1392 ptr += sprintf(ptr, "[promise expired]");
1393 if (promises->size() != 0)
1394 ptr += sprintf(ptr, "[unresolved promise]");
1396 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1400 * Returns whether the current completed trace is feasible, except for pending
1401 * release sequences.
1403 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1405 return !is_infeasible() && promises->size() == 0;
1409 * Check if the current partial trace is infeasible. Does not check any
1410 * end-of-execution flags, which might rule out the execution. Thus, this is
1411 * useful only for ruling an execution as infeasible.
1412 * @return whether the current partial trace is infeasible.
1414 bool ModelChecker::is_infeasible() const
1416 return mo_graph->checkForCycles() ||
1417 priv->no_valid_reads ||
1418 priv->failed_promise ||
1419 priv->too_many_reads ||
1420 priv->bad_synchronization ||
1424 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1425 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1426 ModelAction *lastread = get_last_action(act->get_tid());
1427 lastread->process_rmw(act);
1428 if (act->is_rmw()) {
1429 if (lastread->get_reads_from())
1430 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1432 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1433 mo_graph->commitChanges();
1439 * Checks whether a thread has read from the same write for too many times
1440 * without seeing the effects of a later write.
1443 * 1) there must a different write that we could read from that would satisfy the modification order,
1444 * 2) we must have read from the same value in excess of maxreads times, and
1445 * 3) that other write must have been in the reads_from set for maxreads times.
1447 * If so, we decide that the execution is no longer feasible.
1449 void ModelChecker::check_recency(ModelAction *curr, const ModelAction *rf)
1451 if (params.maxreads != 0) {
1452 if (curr->get_node()->get_read_from_size() <= 1)
1454 //Must make sure that execution is currently feasible... We could
1455 //accidentally clear by rolling back
1456 if (is_infeasible())
1458 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1459 int tid = id_to_int(curr->get_tid());
1462 if ((int)thrd_lists->size() <= tid)
1464 action_list_t *list = &(*thrd_lists)[tid];
1466 action_list_t::reverse_iterator rit = list->rbegin();
1467 /* Skip past curr */
1468 for (; (*rit) != curr; rit++)
1470 /* go past curr now */
1473 action_list_t::reverse_iterator ritcopy = rit;
1474 //See if we have enough reads from the same value
1476 for (; count < params.maxreads; rit++, count++) {
1477 if (rit == list->rend())
1479 ModelAction *act = *rit;
1480 if (!act->is_read())
1483 if (act->get_reads_from() != rf)
1485 if (act->get_node()->get_read_from_size() <= 1)
1488 for (int i = 0; i < curr->get_node()->get_read_from_size(); i++) {
1490 const ModelAction *write = curr->get_node()->get_read_from_at(i);
1492 /* Need a different write */
1496 /* Test to see whether this is a feasible write to read from */
1497 /** NOTE: all members of read-from set should be
1498 * feasible, so we no longer check it here **/
1502 bool feasiblewrite = true;
1503 //new we need to see if this write works for everyone
1505 for (int loop = count; loop > 0; loop--, rit++) {
1506 ModelAction *act = *rit;
1507 bool foundvalue = false;
1508 for (int j = 0; j < act->get_node()->get_read_from_size(); j++) {
1509 if (act->get_node()->get_read_from_at(j) == write) {
1515 feasiblewrite = false;
1519 if (feasiblewrite) {
1520 priv->too_many_reads = true;
1528 * Updates the mo_graph with the constraints imposed from the current
1531 * Basic idea is the following: Go through each other thread and find
1532 * the last action that happened before our read. Two cases:
1534 * (1) The action is a write => that write must either occur before
1535 * the write we read from or be the write we read from.
1537 * (2) The action is a read => the write that that action read from
1538 * must occur before the write we read from or be the same write.
1540 * @param curr The current action. Must be a read.
1541 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1542 * @return True if modification order edges were added; false otherwise
1544 template <typename rf_type>
1545 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1547 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1550 ASSERT(curr->is_read());
1552 /* Last SC fence in the current thread */
1553 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1555 /* Iterate over all threads */
1556 for (i = 0; i < thrd_lists->size(); i++) {
1557 /* Last SC fence in thread i */
1558 ModelAction *last_sc_fence_thread_local = NULL;
1559 if (int_to_id((int)i) != curr->get_tid())
1560 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1562 /* Last SC fence in thread i, before last SC fence in current thread */
1563 ModelAction *last_sc_fence_thread_before = NULL;
1564 if (last_sc_fence_local)
1565 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1567 /* Iterate over actions in thread, starting from most recent */
1568 action_list_t *list = &(*thrd_lists)[i];
1569 action_list_t::reverse_iterator rit;
1570 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1571 ModelAction *act = *rit;
1573 if (act->is_write() && !act->equals(rf) && act != curr) {
1574 /* C++, Section 29.3 statement 5 */
1575 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1576 *act < *last_sc_fence_thread_local) {
1577 added = mo_graph->addEdge(act, rf) || added;
1580 /* C++, Section 29.3 statement 4 */
1581 else if (act->is_seqcst() && last_sc_fence_local &&
1582 *act < *last_sc_fence_local) {
1583 added = mo_graph->addEdge(act, rf) || added;
1586 /* C++, Section 29.3 statement 6 */
1587 else if (last_sc_fence_thread_before &&
1588 *act < *last_sc_fence_thread_before) {
1589 added = mo_graph->addEdge(act, rf) || added;
1595 * Include at most one act per-thread that "happens
1596 * before" curr. Don't consider reflexively.
1598 if (act->happens_before(curr) && act != curr) {
1599 if (act->is_write()) {
1600 if (!act->equals(rf)) {
1601 added = mo_graph->addEdge(act, rf) || added;
1604 const ModelAction *prevreadfrom = act->get_reads_from();
1605 //if the previous read is unresolved, keep going...
1606 if (prevreadfrom == NULL)
1609 if (!prevreadfrom->equals(rf)) {
1610 added = mo_graph->addEdge(prevreadfrom, rf) || added;
1619 * All compatible, thread-exclusive promises must be ordered after any
1620 * concrete loads from the same thread
1622 for (unsigned int i = 0; i < promises->size(); i++)
1623 if ((*promises)[i]->is_compatible_exclusive(curr))
1624 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1630 * Updates the mo_graph with the constraints imposed from the current write.
1632 * Basic idea is the following: Go through each other thread and find
1633 * the lastest action that happened before our write. Two cases:
1635 * (1) The action is a write => that write must occur before
1638 * (2) The action is a read => the write that that action read from
1639 * must occur before the current write.
1641 * This method also handles two other issues:
1643 * (I) Sequential Consistency: Making sure that if the current write is
1644 * seq_cst, that it occurs after the previous seq_cst write.
1646 * (II) Sending the write back to non-synchronizing reads.
1648 * @param curr The current action. Must be a write.
1649 * @return True if modification order edges were added; false otherwise
1651 bool ModelChecker::w_modification_order(ModelAction *curr)
1653 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1656 ASSERT(curr->is_write());
1658 if (curr->is_seqcst()) {
1659 /* We have to at least see the last sequentially consistent write,
1660 so we are initialized. */
1661 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1662 if (last_seq_cst != NULL) {
1663 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1667 /* Last SC fence in the current thread */
1668 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1670 /* Iterate over all threads */
1671 for (i = 0; i < thrd_lists->size(); i++) {
1672 /* Last SC fence in thread i, before last SC fence in current thread */
1673 ModelAction *last_sc_fence_thread_before = NULL;
1674 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1675 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1677 /* Iterate over actions in thread, starting from most recent */
1678 action_list_t *list = &(*thrd_lists)[i];
1679 action_list_t::reverse_iterator rit;
1680 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1681 ModelAction *act = *rit;
1684 * 1) If RMW and it actually read from something, then we
1685 * already have all relevant edges, so just skip to next
1688 * 2) If RMW and it didn't read from anything, we should
1689 * whatever edge we can get to speed up convergence.
1691 * 3) If normal write, we need to look at earlier actions, so
1692 * continue processing list.
1694 if (curr->is_rmw()) {
1695 if (curr->get_reads_from() != NULL)
1703 /* C++, Section 29.3 statement 7 */
1704 if (last_sc_fence_thread_before && act->is_write() &&
1705 *act < *last_sc_fence_thread_before) {
1706 added = mo_graph->addEdge(act, curr) || added;
1711 * Include at most one act per-thread that "happens
1714 if (act->happens_before(curr)) {
1716 * Note: if act is RMW, just add edge:
1718 * The following edge should be handled elsewhere:
1719 * readfrom(act) --mo--> act
1721 if (act->is_write())
1722 added = mo_graph->addEdge(act, curr) || added;
1723 else if (act->is_read()) {
1724 //if previous read accessed a null, just keep going
1725 if (act->get_reads_from() == NULL)
1727 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
1730 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1731 !act->same_thread(curr)) {
1732 /* We have an action that:
1733 (1) did not happen before us
1734 (2) is a read and we are a write
1735 (3) cannot synchronize with us
1736 (4) is in a different thread
1738 that read could potentially read from our write. Note that
1739 these checks are overly conservative at this point, we'll
1740 do more checks before actually removing the
1744 if (thin_air_constraint_may_allow(curr, act)) {
1745 if (!is_infeasible())
1746 futurevalues->push_back(PendingFutureValue(curr, act));
1747 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
1748 add_future_value(curr, act);
1755 * All compatible, thread-exclusive promises must be ordered after any
1756 * concrete stores to the same thread, or else they can be merged with
1759 for (unsigned int i = 0; i < promises->size(); i++)
1760 if ((*promises)[i]->is_compatible_exclusive(curr))
1761 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
1766 /** Arbitrary reads from the future are not allowed. Section 29.3
1767 * part 9 places some constraints. This method checks one result of constraint
1768 * constraint. Others require compiler support. */
1769 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader)
1771 if (!writer->is_rmw())
1774 if (!reader->is_rmw())
1777 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
1778 if (search == reader)
1780 if (search->get_tid() == reader->get_tid() &&
1781 search->happens_before(reader))
1789 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
1790 * some constraints. This method checks one the following constraint (others
1791 * require compiler support):
1793 * If X --hb-> Y --mo-> Z, then X should not read from Z.
1795 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
1797 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
1799 /* Iterate over all threads */
1800 for (i = 0; i < thrd_lists->size(); i++) {
1801 const ModelAction *write_after_read = NULL;
1803 /* Iterate over actions in thread, starting from most recent */
1804 action_list_t *list = &(*thrd_lists)[i];
1805 action_list_t::reverse_iterator rit;
1806 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1807 ModelAction *act = *rit;
1809 /* Don't disallow due to act == reader */
1810 if (!reader->happens_before(act) || reader == act)
1812 else if (act->is_write())
1813 write_after_read = act;
1814 else if (act->is_read() && act->get_reads_from() != NULL)
1815 write_after_read = act->get_reads_from();
1818 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
1825 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
1826 * The ModelAction under consideration is expected to be taking part in
1827 * release/acquire synchronization as an object of the "reads from" relation.
1828 * Note that this can only provide release sequence support for RMW chains
1829 * which do not read from the future, as those actions cannot be traced until
1830 * their "promise" is fulfilled. Similarly, we may not even establish the
1831 * presence of a release sequence with certainty, as some modification order
1832 * constraints may be decided further in the future. Thus, this function
1833 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
1834 * and a boolean representing certainty.
1836 * @param rf The action that might be part of a release sequence. Must be a
1838 * @param release_heads A pass-by-reference style return parameter. After
1839 * execution of this function, release_heads will contain the heads of all the
1840 * relevant release sequences, if any exists with certainty
1841 * @param pending A pass-by-reference style return parameter which is only used
1842 * when returning false (i.e., uncertain). Returns most information regarding
1843 * an uncertain release sequence, including any write operations that might
1844 * break the sequence.
1845 * @return true, if the ModelChecker is certain that release_heads is complete;
1848 bool ModelChecker::release_seq_heads(const ModelAction *rf,
1849 rel_heads_list_t *release_heads,
1850 struct release_seq *pending) const
1852 /* Only check for release sequences if there are no cycles */
1853 if (mo_graph->checkForCycles())
1857 ASSERT(rf->is_write());
1859 if (rf->is_release())
1860 release_heads->push_back(rf);
1861 else if (rf->get_last_fence_release())
1862 release_heads->push_back(rf->get_last_fence_release());
1864 break; /* End of RMW chain */
1866 /** @todo Need to be smarter here... In the linux lock
1867 * example, this will run to the beginning of the program for
1869 /** @todo The way to be smarter here is to keep going until 1
1870 * thread has a release preceded by an acquire and you've seen
1873 /* acq_rel RMW is a sufficient stopping condition */
1874 if (rf->is_acquire() && rf->is_release())
1875 return true; /* complete */
1877 rf = rf->get_reads_from();
1880 /* read from future: need to settle this later */
1882 return false; /* incomplete */
1885 if (rf->is_release())
1886 return true; /* complete */
1888 /* else relaxed write
1889 * - check for fence-release in the same thread (29.8, stmt. 3)
1890 * - check modification order for contiguous subsequence
1891 * -> rf must be same thread as release */
1893 const ModelAction *fence_release = rf->get_last_fence_release();
1894 /* Synchronize with a fence-release unconditionally; we don't need to
1895 * find any more "contiguous subsequence..." for it */
1897 release_heads->push_back(fence_release);
1899 int tid = id_to_int(rf->get_tid());
1900 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
1901 action_list_t *list = &(*thrd_lists)[tid];
1902 action_list_t::const_reverse_iterator rit;
1904 /* Find rf in the thread list */
1905 rit = std::find(list->rbegin(), list->rend(), rf);
1906 ASSERT(rit != list->rend());
1908 /* Find the last {write,fence}-release */
1909 for (; rit != list->rend(); rit++) {
1910 if (fence_release && *(*rit) < *fence_release)
1912 if ((*rit)->is_release())
1915 if (rit == list->rend()) {
1916 /* No write-release in this thread */
1917 return true; /* complete */
1918 } else if (fence_release && *(*rit) < *fence_release) {
1919 /* The fence-release is more recent (and so, "stronger") than
1920 * the most recent write-release */
1921 return true; /* complete */
1922 } /* else, need to establish contiguous release sequence */
1923 ModelAction *release = *rit;
1925 ASSERT(rf->same_thread(release));
1927 pending->writes.clear();
1929 bool certain = true;
1930 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
1931 if (id_to_int(rf->get_tid()) == (int)i)
1933 list = &(*thrd_lists)[i];
1935 /* Can we ensure no future writes from this thread may break
1936 * the release seq? */
1937 bool future_ordered = false;
1939 ModelAction *last = get_last_action(int_to_id(i));
1940 Thread *th = get_thread(int_to_id(i));
1941 if ((last && rf->happens_before(last)) ||
1944 future_ordered = true;
1946 ASSERT(!th->is_model_thread() || future_ordered);
1948 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1949 const ModelAction *act = *rit;
1950 /* Reach synchronization -> this thread is complete */
1951 if (act->happens_before(release))
1953 if (rf->happens_before(act)) {
1954 future_ordered = true;
1958 /* Only non-RMW writes can break release sequences */
1959 if (!act->is_write() || act->is_rmw())
1962 /* Check modification order */
1963 if (mo_graph->checkReachable(rf, act)) {
1964 /* rf --mo--> act */
1965 future_ordered = true;
1968 if (mo_graph->checkReachable(act, release))
1969 /* act --mo--> release */
1971 if (mo_graph->checkReachable(release, act) &&
1972 mo_graph->checkReachable(act, rf)) {
1973 /* release --mo-> act --mo--> rf */
1974 return true; /* complete */
1976 /* act may break release sequence */
1977 pending->writes.push_back(act);
1980 if (!future_ordered)
1981 certain = false; /* This thread is uncertain */
1985 release_heads->push_back(release);
1986 pending->writes.clear();
1988 pending->release = release;
1995 * An interface for getting the release sequence head(s) with which a
1996 * given ModelAction must synchronize. This function only returns a non-empty
1997 * result when it can locate a release sequence head with certainty. Otherwise,
1998 * it may mark the internal state of the ModelChecker so that it will handle
1999 * the release sequence at a later time, causing @a acquire to update its
2000 * synchronization at some later point in execution.
2002 * @param acquire The 'acquire' action that may synchronize with a release
2004 * @param read The read action that may read from a release sequence; this may
2005 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2006 * when 'acquire' is a fence-acquire)
2007 * @param release_heads A pass-by-reference return parameter. Will be filled
2008 * with the head(s) of the release sequence(s), if they exists with certainty.
2009 * @see ModelChecker::release_seq_heads
2011 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2012 ModelAction *read, rel_heads_list_t *release_heads)
2014 const ModelAction *rf = read->get_reads_from();
2015 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2016 sequence->acquire = acquire;
2017 sequence->read = read;
2019 if (!release_seq_heads(rf, release_heads, sequence)) {
2020 /* add act to 'lazy checking' list */
2021 pending_rel_seqs->push_back(sequence);
2023 snapshot_free(sequence);
2028 * Attempt to resolve all stashed operations that might synchronize with a
2029 * release sequence for a given location. This implements the "lazy" portion of
2030 * determining whether or not a release sequence was contiguous, since not all
2031 * modification order information is present at the time an action occurs.
2033 * @param location The location/object that should be checked for release
2034 * sequence resolutions. A NULL value means to check all locations.
2035 * @param work_queue The work queue to which to add work items as they are
2037 * @return True if any updates occurred (new synchronization, new mo_graph
2040 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2042 bool updated = false;
2043 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
2044 while (it != pending_rel_seqs->end()) {
2045 struct release_seq *pending = *it;
2046 ModelAction *acquire = pending->acquire;
2047 const ModelAction *read = pending->read;
2049 /* Only resolve sequences on the given location, if provided */
2050 if (location && read->get_location() != location) {
2055 const ModelAction *rf = read->get_reads_from();
2056 rel_heads_list_t release_heads;
2058 complete = release_seq_heads(rf, &release_heads, pending);
2059 for (unsigned int i = 0; i < release_heads.size(); i++) {
2060 if (!acquire->has_synchronized_with(release_heads[i])) {
2061 if (acquire->synchronize_with(release_heads[i]))
2064 set_bad_synchronization();
2069 /* Re-check all pending release sequences */
2070 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2071 /* Re-check read-acquire for mo_graph edges */
2072 if (acquire->is_read())
2073 work_queue->push_back(MOEdgeWorkEntry(acquire));
2075 /* propagate synchronization to later actions */
2076 action_list_t::reverse_iterator rit = action_trace->rbegin();
2077 for (; (*rit) != acquire; rit++) {
2078 ModelAction *propagate = *rit;
2079 if (acquire->happens_before(propagate)) {
2080 propagate->synchronize_with(acquire);
2081 /* Re-check 'propagate' for mo_graph edges */
2082 work_queue->push_back(MOEdgeWorkEntry(propagate));
2087 it = pending_rel_seqs->erase(it);
2088 snapshot_free(pending);
2094 // If we resolved promises or data races, see if we have realized a data race.
2101 * Performs various bookkeeping operations for the current ModelAction. For
2102 * instance, adds action to the per-object, per-thread action vector and to the
2103 * action trace list of all thread actions.
2105 * @param act is the ModelAction to add.
2107 void ModelChecker::add_action_to_lists(ModelAction *act)
2109 int tid = id_to_int(act->get_tid());
2110 ModelAction *uninit = NULL;
2112 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2113 if (list->empty() && act->is_atomic_var()) {
2114 uninit = new_uninitialized_action(act->get_location());
2115 uninit_id = id_to_int(uninit->get_tid());
2116 list->push_back(uninit);
2118 list->push_back(act);
2120 action_trace->push_back(act);
2122 action_trace->push_front(uninit);
2124 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2125 if (tid >= (int)vec->size())
2126 vec->resize(priv->next_thread_id);
2127 (*vec)[tid].push_back(act);
2129 (*vec)[uninit_id].push_front(uninit);
2131 if ((int)thrd_last_action->size() <= tid)
2132 thrd_last_action->resize(get_num_threads());
2133 (*thrd_last_action)[tid] = act;
2135 (*thrd_last_action)[uninit_id] = uninit;
2137 if (act->is_fence() && act->is_release()) {
2138 if ((int)thrd_last_fence_release->size() <= tid)
2139 thrd_last_fence_release->resize(get_num_threads());
2140 (*thrd_last_fence_release)[tid] = act;
2143 if (act->is_wait()) {
2144 void *mutex_loc = (void *) act->get_value();
2145 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2147 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2148 if (tid >= (int)vec->size())
2149 vec->resize(priv->next_thread_id);
2150 (*vec)[tid].push_back(act);
2155 * @brief Get the last action performed by a particular Thread
2156 * @param tid The thread ID of the Thread in question
2157 * @return The last action in the thread
2159 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2161 int threadid = id_to_int(tid);
2162 if (threadid < (int)thrd_last_action->size())
2163 return (*thrd_last_action)[id_to_int(tid)];
2169 * @brief Get the last fence release performed by a particular Thread
2170 * @param tid The thread ID of the Thread in question
2171 * @return The last fence release in the thread, if one exists; NULL otherwise
2173 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2175 int threadid = id_to_int(tid);
2176 if (threadid < (int)thrd_last_fence_release->size())
2177 return (*thrd_last_fence_release)[id_to_int(tid)];
2183 * Gets the last memory_order_seq_cst write (in the total global sequence)
2184 * performed on a particular object (i.e., memory location), not including the
2186 * @param curr The current ModelAction; also denotes the object location to
2188 * @return The last seq_cst write
2190 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2192 void *location = curr->get_location();
2193 action_list_t *list = get_safe_ptr_action(obj_map, location);
2194 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2195 action_list_t::reverse_iterator rit;
2196 for (rit = list->rbegin(); rit != list->rend(); rit++)
2197 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
2203 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2204 * performed in a particular thread, prior to a particular fence.
2205 * @param tid The ID of the thread to check
2206 * @param before_fence The fence from which to begin the search; if NULL, then
2207 * search for the most recent fence in the thread.
2208 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2210 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2212 /* All fences should have NULL location */
2213 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2214 action_list_t::reverse_iterator rit = list->rbegin();
2217 for (; rit != list->rend(); rit++)
2218 if (*rit == before_fence)
2221 ASSERT(*rit == before_fence);
2225 for (; rit != list->rend(); rit++)
2226 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2232 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2233 * location). This function identifies the mutex according to the current
2234 * action, which is presumed to perform on the same mutex.
2235 * @param curr The current ModelAction; also denotes the object location to
2237 * @return The last unlock operation
2239 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2241 void *location = curr->get_location();
2242 action_list_t *list = get_safe_ptr_action(obj_map, location);
2243 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2244 action_list_t::reverse_iterator rit;
2245 for (rit = list->rbegin(); rit != list->rend(); rit++)
2246 if ((*rit)->is_unlock() || (*rit)->is_wait())
2251 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2253 ModelAction *parent = get_last_action(tid);
2255 parent = get_thread(tid)->get_creation();
2260 * Returns the clock vector for a given thread.
2261 * @param tid The thread whose clock vector we want
2262 * @return Desired clock vector
2264 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2266 return get_parent_action(tid)->get_cv();
2270 * Resolve a set of Promises with a current write. The set is provided in the
2271 * Node corresponding to @a write.
2272 * @param write The ModelAction that is fulfilling Promises
2273 * @return True if promises were resolved; false otherwise
2275 bool ModelChecker::resolve_promises(ModelAction *write)
2277 bool haveResolved = false;
2278 std::vector< ModelAction *, ModelAlloc<ModelAction *> > actions_to_check;
2279 promise_list_t mustResolve, resolved;
2281 for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
2282 Promise *promise = (*promises)[promise_index];
2283 if (write->get_node()->get_promise(i)) {
2284 ModelAction *read = promise->get_action();
2285 read_from(read, write);
2286 //Make sure the promise's value matches the write's value
2287 ASSERT(promise->is_compatible(write));
2288 mo_graph->resolvePromise(read, write, &mustResolve);
2290 resolved.push_back(promise);
2291 promises->erase(promises->begin() + promise_index);
2292 actions_to_check.push_back(read);
2294 haveResolved = true;
2299 for (unsigned int i = 0; i < mustResolve.size(); i++) {
2300 if (std::find(resolved.begin(), resolved.end(), mustResolve[i])
2302 priv->failed_promise = true;
2304 for (unsigned int i = 0; i < resolved.size(); i++)
2306 //Check whether reading these writes has made threads unable to
2309 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2310 ModelAction *read = actions_to_check[i];
2311 mo_check_promises(read, true);
2314 return haveResolved;
2318 * Compute the set of promises that could potentially be satisfied by this
2319 * action. Note that the set computation actually appears in the Node, not in
2321 * @param curr The ModelAction that may satisfy promises
2323 void ModelChecker::compute_promises(ModelAction *curr)
2325 for (unsigned int i = 0; i < promises->size(); i++) {
2326 Promise *promise = (*promises)[i];
2327 const ModelAction *act = promise->get_action();
2328 if (!act->happens_before(curr) &&
2330 !act->could_synchronize_with(curr) &&
2331 !act->same_thread(curr) &&
2332 act->get_location() == curr->get_location() &&
2333 promise->get_value() == curr->get_value()) {
2334 curr->get_node()->set_promise(i, act->is_rmw());
2339 /** Checks promises in response to change in ClockVector Threads. */
2340 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2342 for (unsigned int i = 0; i < promises->size(); i++) {
2343 Promise *promise = (*promises)[i];
2344 const ModelAction *act = promise->get_action();
2345 if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
2346 merge_cv->synchronized_since(act)) {
2347 if (promise->eliminate_thread(tid)) {
2348 //Promise has failed
2349 priv->failed_promise = true;
2356 void ModelChecker::check_promises_thread_disabled()
2358 for (unsigned int i = 0; i < promises->size(); i++) {
2359 Promise *promise = (*promises)[i];
2360 if (promise->has_failed()) {
2361 priv->failed_promise = true;
2368 * @brief Checks promises in response to addition to modification order for
2371 * We test whether threads are still available for satisfying promises after an
2372 * addition to our modification order constraints. Those that are unavailable
2373 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2374 * that promise has failed.
2376 * @param act The ModelAction which updated the modification order
2377 * @param is_read_check Should be true if act is a read and we must check for
2378 * updates to the store from which it read (there is a distinction here for
2379 * RMW's, which are both a load and a store)
2381 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2383 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2385 for (unsigned int i = 0; i < promises->size(); i++) {
2386 Promise *promise = (*promises)[i];
2387 const ModelAction *pread = promise->get_action();
2389 // Is this promise on the same location?
2390 if (!pread->same_var(write))
2393 if (pread->happens_before(act) && mo_graph->checkPromise(write, promise)) {
2394 priv->failed_promise = true;
2398 // Don't do any lookups twice for the same thread
2399 if (!promise->thread_is_available(act->get_tid()))
2402 if (mo_graph->checkReachable(promise, write)) {
2403 if (mo_graph->checkPromise(write, promise)) {
2404 priv->failed_promise = true;
2412 * Compute the set of writes that may break the current pending release
2413 * sequence. This information is extracted from previou release sequence
2416 * @param curr The current ModelAction. Must be a release sequence fixup
2419 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2421 if (pending_rel_seqs->empty())
2424 struct release_seq *pending = pending_rel_seqs->back();
2425 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2426 const ModelAction *write = pending->writes[i];
2427 curr->get_node()->add_relseq_break(write);
2430 /* NULL means don't break the sequence; just synchronize */
2431 curr->get_node()->add_relseq_break(NULL);
2435 * Build up an initial set of all past writes that this 'read' action may read
2436 * from. This set is determined by the clock vector's "happens before"
2438 * @param curr is the current ModelAction that we are exploring; it must be a
2441 void ModelChecker::build_reads_from_past(ModelAction *curr)
2443 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2445 ASSERT(curr->is_read());
2447 ModelAction *last_sc_write = NULL;
2449 if (curr->is_seqcst())
2450 last_sc_write = get_last_seq_cst_write(curr);
2452 /* Iterate over all threads */
2453 for (i = 0; i < thrd_lists->size(); i++) {
2454 /* Iterate over actions in thread, starting from most recent */
2455 action_list_t *list = &(*thrd_lists)[i];
2456 action_list_t::reverse_iterator rit;
2457 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2458 ModelAction *act = *rit;
2460 /* Only consider 'write' actions */
2461 if (!act->is_write() || act == curr)
2464 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2465 bool allow_read = true;
2467 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2469 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2473 /* Only add feasible reads */
2474 mo_graph->startChanges();
2475 r_modification_order(curr, act);
2476 if (!is_infeasible())
2477 curr->get_node()->add_read_from(act);
2478 mo_graph->rollbackChanges();
2481 /* Include at most one act per-thread that "happens before" curr */
2482 if (act->happens_before(curr))
2486 /* We may find no valid may-read-from only if the execution is doomed */
2487 if (!curr->get_node()->get_read_from_size()) {
2488 priv->no_valid_reads = true;
2492 if (DBG_ENABLED()) {
2493 model_print("Reached read action:\n");
2495 model_print("Printing may_read_from\n");
2496 curr->get_node()->print_may_read_from();
2497 model_print("End printing may_read_from\n");
2501 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2504 /* UNINIT actions don't have a Node, and they never sleep */
2505 if (write->is_uninitialized())
2507 Node *prevnode = write->get_node()->get_parent();
2509 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2510 if (write->is_release() && thread_sleep)
2512 if (!write->is_rmw()) {
2515 if (write->get_reads_from() == NULL)
2517 write = write->get_reads_from();
2522 * @brief Create a new action representing an uninitialized atomic
2523 * @param location The memory location of the atomic object
2524 * @return A pointer to a new ModelAction
2526 ModelAction * ModelChecker::new_uninitialized_action(void *location) const
2528 ModelAction *act = (ModelAction *)snapshot_malloc(sizeof(class ModelAction));
2529 act = new (act) ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, location, 0, model_thread);
2530 act->create_cv(NULL);
2534 static void print_list(action_list_t *list)
2536 action_list_t::iterator it;
2538 model_print("---------------------------------------------------------------------\n");
2540 unsigned int hash = 0;
2542 for (it = list->begin(); it != list->end(); it++) {
2544 hash = hash^(hash<<3)^((*it)->hash());
2546 model_print("HASH %u\n", hash);
2547 model_print("---------------------------------------------------------------------\n");
2550 #if SUPPORT_MOD_ORDER_DUMP
2551 void ModelChecker::dumpGraph(char *filename) const
2554 sprintf(buffer, "%s.dot", filename);
2555 FILE *file = fopen(buffer, "w");
2556 fprintf(file, "digraph %s {\n", filename);
2557 mo_graph->dumpNodes(file);
2558 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2560 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2561 ModelAction *action = *it;
2562 if (action->is_read()) {
2563 fprintf(file, "N%u [label=\"N%u, T%u\"];\n", action->get_seq_number(), action->get_seq_number(), action->get_tid());
2564 if (action->get_reads_from() != NULL)
2565 fprintf(file, "N%u -> N%u[label=\"rf\", color=red];\n", action->get_seq_number(), action->get_reads_from()->get_seq_number());
2567 if (thread_array[action->get_tid()] != NULL) {
2568 fprintf(file, "N%u -> N%u[label=\"sb\", color=blue];\n", thread_array[action->get_tid()]->get_seq_number(), action->get_seq_number());
2571 thread_array[action->get_tid()] = action;
2573 fprintf(file, "}\n");
2574 model_free(thread_array);
2579 /** @brief Prints an execution trace summary. */
2580 void ModelChecker::print_summary() const
2582 #if SUPPORT_MOD_ORDER_DUMP
2583 char buffername[100];
2584 sprintf(buffername, "exec%04u", stats.num_total);
2585 mo_graph->dumpGraphToFile(buffername);
2586 sprintf(buffername, "graph%04u", stats.num_total);
2587 dumpGraph(buffername);
2590 model_print("Execution %d:", stats.num_total);
2591 if (isfeasibleprefix())
2594 print_infeasibility(" INFEASIBLE");
2595 print_list(action_trace);
2600 * Add a Thread to the system for the first time. Should only be called once
2602 * @param t The Thread to add
2604 void ModelChecker::add_thread(Thread *t)
2606 thread_map->put(id_to_int(t->get_id()), t);
2607 scheduler->add_thread(t);
2611 * Removes a thread from the scheduler.
2612 * @param the thread to remove.
2614 void ModelChecker::remove_thread(Thread *t)
2616 scheduler->remove_thread(t);
2620 * @brief Get a Thread reference by its ID
2621 * @param tid The Thread's ID
2622 * @return A Thread reference
2624 Thread * ModelChecker::get_thread(thread_id_t tid) const
2626 return thread_map->get(id_to_int(tid));
2630 * @brief Get a reference to the Thread in which a ModelAction was executed
2631 * @param act The ModelAction
2632 * @return A Thread reference
2634 Thread * ModelChecker::get_thread(const ModelAction *act) const
2636 return get_thread(act->get_tid());
2640 * @brief Check if a Thread is currently enabled
2641 * @param t The Thread to check
2642 * @return True if the Thread is currently enabled
2644 bool ModelChecker::is_enabled(Thread *t) const
2646 return scheduler->is_enabled(t);
2650 * @brief Check if a Thread is currently enabled
2651 * @param tid The ID of the Thread to check
2652 * @return True if the Thread is currently enabled
2654 bool ModelChecker::is_enabled(thread_id_t tid) const
2656 return scheduler->is_enabled(tid);
2660 * Switch from a user-context to the "master thread" context (a.k.a. system
2661 * context). This switch is made with the intention of exploring a particular
2662 * model-checking action (described by a ModelAction object). Must be called
2663 * from a user-thread context.
2665 * @param act The current action that will be explored. May be NULL only if
2666 * trace is exiting via an assertion (see ModelChecker::set_assert and
2667 * ModelChecker::has_asserted).
2668 * @return Return the value returned by the current action
2670 uint64_t ModelChecker::switch_to_master(ModelAction *act)
2673 Thread *old = thread_current();
2674 ASSERT(!old->get_pending());
2675 old->set_pending(act);
2676 if (Thread::swap(old, &system_context) < 0) {
2677 perror("swap threads");
2680 return old->get_return_value();
2684 * Takes the next step in the execution, if possible.
2685 * @param curr The current step to take
2686 * @return Returns the next Thread to run, if any; NULL if this execution
2689 Thread * ModelChecker::take_step(ModelAction *curr)
2691 Thread *curr_thrd = get_thread(curr);
2692 ASSERT(curr_thrd->get_state() == THREAD_READY);
2694 curr = check_current_action(curr);
2696 /* Infeasible -> don't take any more steps */
2697 if (is_infeasible())
2699 else if (isfeasibleprefix() && have_bug_reports()) {
2704 if (params.bound != 0 && priv->used_sequence_numbers > params.bound)
2707 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
2708 scheduler->remove_thread(curr_thrd);
2710 Thread *next_thrd = get_next_thread(curr);
2711 next_thrd = scheduler->next_thread(next_thrd);
2713 DEBUG("(%d, %d)\n", curr_thrd ? id_to_int(curr_thrd->get_id()) : -1,
2714 next_thrd ? id_to_int(next_thrd->get_id()) : -1);
2717 * Launch end-of-execution release sequence fixups only when there are:
2719 * (1) no more user threads to run (or when execution replay chooses
2720 * the 'model_thread')
2721 * (2) pending release sequences
2722 * (3) pending assertions (i.e., data races)
2723 * (4) no pending promises
2725 if (!pending_rel_seqs->empty() && (!next_thrd || next_thrd->is_model_thread()) &&
2726 is_feasible_prefix_ignore_relseq() && !unrealizedraces.empty()) {
2727 model_print("*** WARNING: release sequence fixup action (%zu pending release seuqences) ***\n",
2728 pending_rel_seqs->size());
2729 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
2730 std::memory_order_seq_cst, NULL, VALUE_NONE,
2732 model_thread->set_pending(fixup);
2733 return model_thread;
2736 /* next_thrd == NULL -> don't take any more steps */
2743 /** Wrapper to run the user's main function, with appropriate arguments */
2744 void user_main_wrapper(void *)
2746 user_main(model->params.argc, model->params.argv);
2749 /** @brief Run ModelChecker for the user program */
2750 void ModelChecker::run()
2754 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL);
2758 for (unsigned int i = 0; i < get_num_threads(); i++) {
2759 thread_id_t tid = int_to_id(i);
2760 Thread *thr = get_thread(tid);
2761 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
2762 scheduler->next_thread(thr);
2763 Thread::swap(&system_context, thr);
2767 /* Catch assertions from prior take_step or from
2768 * between-ModelAction bugs (e.g., data races) */
2772 /* Consume the next action for a Thread */
2773 ModelAction *curr = t->get_pending();
2774 t->set_pending(NULL);
2775 t = take_step(curr);
2776 } while (t && !t->is_model_thread());
2777 /** @TODO Re-write release sequence fixups here */
2778 } while (next_execution());