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 we defer to the
217 * @param curr Optional: The current ModelAction. Only used if non-NULL and it
218 * might guide the choice of next thread (i.e., THREAD_CREATE should be
219 * followed by THREAD_START, or ATOMIC_RMWR followed by ATOMIC_{RMW,RMWC})
220 * @return The next chosen thread to run, if any exist. Or else if no threads
221 * remain to be executed, return NULL.
223 Thread * ModelChecker::get_next_thread(ModelAction *curr)
228 /* Do not split atomic actions. */
230 return get_thread(curr);
231 else if (curr->get_type() == THREAD_CREATE)
232 return curr->get_thread_operand();
236 * Have we completed exploring the preselected path? Then let the
240 return scheduler->select_next_thread();
242 /* Else, we are trying to replay an execution */
243 ModelAction *next = node_stack->get_next()->get_action();
245 if (next == diverge) {
246 if (earliest_diverge == NULL || *diverge < *earliest_diverge)
247 earliest_diverge = diverge;
249 Node *nextnode = next->get_node();
250 Node *prevnode = nextnode->get_parent();
251 scheduler->update_sleep_set(prevnode);
253 /* Reached divergence point */
254 if (nextnode->increment_misc()) {
255 /* The next node will try to satisfy a different misc_index values. */
256 tid = next->get_tid();
257 node_stack->pop_restofstack(2);
258 } else if (nextnode->increment_promise()) {
259 /* The next node will try to satisfy a different set of promises. */
260 tid = next->get_tid();
261 node_stack->pop_restofstack(2);
262 } else if (nextnode->increment_read_from()) {
263 /* The next node will read from a different value. */
264 tid = next->get_tid();
265 node_stack->pop_restofstack(2);
266 } else if (nextnode->increment_relseq_break()) {
267 /* The next node will try to resolve a release sequence differently */
268 tid = next->get_tid();
269 node_stack->pop_restofstack(2);
272 /* Make a different thread execute for next step */
273 scheduler->add_sleep(get_thread(next->get_tid()));
274 tid = prevnode->get_next_backtrack();
275 /* Make sure the backtracked thread isn't sleeping. */
276 node_stack->pop_restofstack(1);
277 if (diverge == earliest_diverge) {
278 earliest_diverge = prevnode->get_action();
281 /* Start the round robin scheduler from this thread id */
282 scheduler->set_scheduler_thread(tid);
283 /* The correct sleep set is in the parent node. */
286 DEBUG("*** Divergence point ***\n");
290 tid = next->get_tid();
292 DEBUG("*** ModelChecker chose next thread = %d ***\n", id_to_int(tid));
293 ASSERT(tid != THREAD_ID_T_NONE);
294 return thread_map->get(id_to_int(tid));
298 * We need to know what the next actions of all threads in the sleep
299 * set will be. This method computes them and stores the actions at
300 * the corresponding thread object's pending action.
303 void ModelChecker::execute_sleep_set()
305 for (unsigned int i = 0; i < get_num_threads(); i++) {
306 thread_id_t tid = int_to_id(i);
307 Thread *thr = get_thread(tid);
308 if (scheduler->is_sleep_set(thr) && thr->get_pending()) {
309 thr->get_pending()->set_sleep_flag();
315 * @brief Should the current action wake up a given thread?
317 * @param curr The current action
318 * @param thread The thread that we might wake up
319 * @return True, if we should wake up the sleeping thread; false otherwise
321 bool ModelChecker::should_wake_up(const ModelAction *curr, const Thread *thread) const
323 const ModelAction *asleep = thread->get_pending();
324 /* Don't allow partial RMW to wake anyone up */
327 /* Synchronizing actions may have been backtracked */
328 if (asleep->could_synchronize_with(curr))
330 /* All acquire/release fences and fence-acquire/store-release */
331 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
333 /* Fence-release + store can awake load-acquire on the same location */
334 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
335 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
336 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
342 void ModelChecker::wake_up_sleeping_actions(ModelAction *curr)
344 for (unsigned int i = 0; i < get_num_threads(); i++) {
345 Thread *thr = get_thread(int_to_id(i));
346 if (scheduler->is_sleep_set(thr)) {
347 if (should_wake_up(curr, thr))
348 /* Remove this thread from sleep set */
349 scheduler->remove_sleep(thr);
354 /** @brief Alert the model-checker that an incorrectly-ordered
355 * synchronization was made */
356 void ModelChecker::set_bad_synchronization()
358 priv->bad_synchronization = true;
362 * Check whether the current trace has triggered an assertion which should halt
365 * @return True, if the execution should be aborted; false otherwise
367 bool ModelChecker::has_asserted() const
369 return priv->asserted;
373 * Trigger a trace assertion which should cause this execution to be halted.
374 * This can be due to a detected bug or due to an infeasibility that should
377 void ModelChecker::set_assert()
379 priv->asserted = true;
383 * Check if we are in a deadlock. Should only be called at the end of an
384 * execution, although it should not give false positives in the middle of an
385 * execution (there should be some ENABLED thread).
387 * @return True if program is in a deadlock; false otherwise
389 bool ModelChecker::is_deadlocked() const
391 bool blocking_threads = false;
392 for (unsigned int i = 0; i < get_num_threads(); i++) {
393 thread_id_t tid = int_to_id(i);
396 Thread *t = get_thread(tid);
397 if (!t->is_model_thread() && t->get_pending())
398 blocking_threads = true;
400 return blocking_threads;
404 * Check if a Thread has entered a deadlock situation. This will not check
405 * other threads for potential deadlock situations, and may miss deadlocks
408 * @param t The thread which may have entered a deadlock
409 * @return True if this Thread entered a deadlock; false otherwise
411 bool ModelChecker::check_deadlock(const Thread *t) const
413 for (Thread *waiting = t->waiting_on() ; waiting != NULL; waiting = waiting->waiting_on())
420 * Check if this is a complete execution. That is, have all thread completed
421 * execution (rather than exiting because sleep sets have forced a redundant
424 * @return True if the execution is complete.
426 bool ModelChecker::is_complete_execution() const
428 for (unsigned int i = 0; i < get_num_threads(); i++)
429 if (is_enabled(int_to_id(i)))
435 * @brief Assert a bug in the executing program.
437 * Use this function to assert any sort of bug in the user program. If the
438 * current trace is feasible (actually, a prefix of some feasible execution),
439 * then this execution will be aborted, printing the appropriate message. If
440 * the current trace is not yet feasible, the error message will be stashed and
441 * printed if the execution ever becomes feasible.
443 * @param msg Descriptive message for the bug (do not include newline char)
444 * @return True if bug is immediately-feasible
446 bool ModelChecker::assert_bug(const char *msg)
448 priv->bugs.push_back(new bug_message(msg));
450 if (isfeasibleprefix()) {
458 * @brief Assert a bug in the executing program, asserted by a user thread
459 * @see ModelChecker::assert_bug
460 * @param msg Descriptive message for the bug (do not include newline char)
462 void ModelChecker::assert_user_bug(const char *msg)
464 /* If feasible bug, bail out now */
466 switch_to_master(NULL);
469 /** @return True, if any bugs have been reported for this execution */
470 bool ModelChecker::have_bug_reports() const
472 return priv->bugs.size() != 0;
475 /** @brief Print bug report listing for this execution (if any bugs exist) */
476 void ModelChecker::print_bugs() const
478 if (have_bug_reports()) {
479 model_print("Bug report: %zu bug%s detected\n",
481 priv->bugs.size() > 1 ? "s" : "");
482 for (unsigned int i = 0; i < priv->bugs.size(); i++)
483 priv->bugs[i]->print();
488 * @brief Record end-of-execution stats
490 * Must be run when exiting an execution. Records various stats.
491 * @see struct execution_stats
493 void ModelChecker::record_stats()
496 if (!isfeasibleprefix())
497 stats.num_infeasible++;
498 else if (have_bug_reports())
499 stats.num_buggy_executions++;
500 else if (is_complete_execution())
501 stats.num_complete++;
503 stats.num_redundant++;
506 * @todo We can violate this ASSERT() when fairness/sleep sets
507 * conflict to cause an execution to terminate, e.g. with:
508 * Scheduler: [0: disabled][1: disabled][2: sleep][3: current, enabled]
510 //ASSERT(scheduler->all_threads_sleeping());
514 /** @brief Print execution stats */
515 void ModelChecker::print_stats() const
517 model_print("Number of complete, bug-free executions: %d\n", stats.num_complete);
518 model_print("Number of redundant executions: %d\n", stats.num_redundant);
519 model_print("Number of buggy executions: %d\n", stats.num_buggy_executions);
520 model_print("Number of infeasible executions: %d\n", stats.num_infeasible);
521 model_print("Total executions: %d\n", stats.num_total);
522 model_print("Total nodes created: %d\n", node_stack->get_total_nodes());
526 * @brief End-of-exeuction print
527 * @param printbugs Should any existing bugs be printed?
529 void ModelChecker::print_execution(bool printbugs) const
531 print_program_output();
533 if (DBG_ENABLED() || params.verbose) {
534 model_print("Earliest divergence point since last feasible execution:\n");
535 if (earliest_diverge)
536 earliest_diverge->print();
538 model_print("(Not set)\n");
544 /* Don't print invalid bugs */
553 * Queries the model-checker for more executions to explore and, if one
554 * exists, resets the model-checker state to execute a new execution.
556 * @return If there are more executions to explore, return true. Otherwise,
559 bool ModelChecker::next_execution()
562 /* Is this execution a feasible execution that's worth bug-checking? */
563 bool complete = isfeasibleprefix() && (is_complete_execution() ||
566 /* End-of-execution bug checks */
569 assert_bug("Deadlock detected");
577 if (DBG_ENABLED() || params.verbose || (complete && have_bug_reports()))
578 print_execution(complete);
580 clear_program_output();
583 earliest_diverge = NULL;
585 if ((diverge = get_next_backtrack()) == NULL)
589 model_print("Next execution will diverge at:\n");
593 reset_to_initial_state();
598 * @brief Find the last fence-related backtracking conflict for a ModelAction
600 * This function performs the search for the most recent conflicting action
601 * against which we should perform backtracking, as affected by fence
602 * operations. This includes pairs of potentially-synchronizing actions which
603 * occur due to fence-acquire or fence-release, and hence should be explored in
604 * the opposite execution order.
606 * @param act The current action
607 * @return The most recent action which conflicts with act due to fences
609 ModelAction * ModelChecker::get_last_fence_conflict(ModelAction *act) const
611 /* Only perform release/acquire fence backtracking for stores */
612 if (!act->is_write())
615 /* Find a fence-release (or, act is a release) */
616 ModelAction *last_release;
617 if (act->is_release())
620 last_release = get_last_fence_release(act->get_tid());
624 /* Skip past the release */
625 action_list_t *list = action_trace;
626 action_list_t::reverse_iterator rit;
627 for (rit = list->rbegin(); rit != list->rend(); rit++)
628 if (*rit == last_release)
630 ASSERT(rit != list->rend());
635 * load --sb-> fence-acquire */
636 std::vector< ModelAction *, ModelAlloc<ModelAction *> > acquire_fences(get_num_threads(), NULL);
637 std::vector< ModelAction *, ModelAlloc<ModelAction *> > prior_loads(get_num_threads(), NULL);
638 bool found_acquire_fences = false;
639 for ( ; rit != list->rend(); rit++) {
640 ModelAction *prev = *rit;
641 if (act->same_thread(prev))
644 int tid = id_to_int(prev->get_tid());
646 if (prev->is_read() && act->same_var(prev)) {
647 if (prev->is_acquire()) {
648 /* Found most recent load-acquire, don't need
649 * to search for more fences */
650 if (!found_acquire_fences)
653 prior_loads[tid] = prev;
656 if (prev->is_acquire() && prev->is_fence() && !acquire_fences[tid]) {
657 found_acquire_fences = true;
658 acquire_fences[tid] = prev;
662 ModelAction *latest_backtrack = NULL;
663 for (unsigned int i = 0; i < acquire_fences.size(); i++)
664 if (acquire_fences[i] && prior_loads[i])
665 if (!latest_backtrack || *latest_backtrack < *acquire_fences[i])
666 latest_backtrack = acquire_fences[i];
667 return latest_backtrack;
671 * @brief Find the last backtracking conflict for a ModelAction
673 * This function performs the search for the most recent conflicting action
674 * against which we should perform backtracking. This primary includes pairs of
675 * synchronizing actions which should be explored in the opposite execution
678 * @param act The current action
679 * @return The most recent action which conflicts with act
681 ModelAction * ModelChecker::get_last_conflict(ModelAction *act) const
683 switch (act->get_type()) {
684 /* case ATOMIC_FENCE: fences don't directly cause backtracking */
688 ModelAction *ret = NULL;
690 /* linear search: from most recent to oldest */
691 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
692 action_list_t::reverse_iterator rit;
693 for (rit = list->rbegin(); rit != list->rend(); rit++) {
694 ModelAction *prev = *rit;
695 if (prev->could_synchronize_with(act)) {
701 ModelAction *ret2 = get_last_fence_conflict(act);
711 case ATOMIC_TRYLOCK: {
712 /* linear search: from most recent to oldest */
713 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
714 action_list_t::reverse_iterator rit;
715 for (rit = list->rbegin(); rit != list->rend(); rit++) {
716 ModelAction *prev = *rit;
717 if (act->is_conflicting_lock(prev))
722 case ATOMIC_UNLOCK: {
723 /* linear search: from most recent to oldest */
724 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
725 action_list_t::reverse_iterator rit;
726 for (rit = list->rbegin(); rit != list->rend(); rit++) {
727 ModelAction *prev = *rit;
728 if (!act->same_thread(prev) && prev->is_failed_trylock())
734 /* linear search: from most recent to oldest */
735 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
736 action_list_t::reverse_iterator rit;
737 for (rit = list->rbegin(); rit != list->rend(); rit++) {
738 ModelAction *prev = *rit;
739 if (!act->same_thread(prev) && prev->is_failed_trylock())
741 if (!act->same_thread(prev) && prev->is_notify())
747 case ATOMIC_NOTIFY_ALL:
748 case ATOMIC_NOTIFY_ONE: {
749 /* linear search: from most recent to oldest */
750 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
751 action_list_t::reverse_iterator rit;
752 for (rit = list->rbegin(); rit != list->rend(); rit++) {
753 ModelAction *prev = *rit;
754 if (!act->same_thread(prev) && prev->is_wait())
765 /** This method finds backtracking points where we should try to
766 * reorder the parameter ModelAction against.
768 * @param the ModelAction to find backtracking points for.
770 void ModelChecker::set_backtracking(ModelAction *act)
772 Thread *t = get_thread(act);
773 ModelAction *prev = get_last_conflict(act);
777 Node *node = prev->get_node()->get_parent();
779 int low_tid, high_tid;
780 if (node->enabled_status(t->get_id()) == THREAD_ENABLED) {
781 low_tid = id_to_int(act->get_tid());
782 high_tid = low_tid + 1;
785 high_tid = get_num_threads();
788 for (int i = low_tid; i < high_tid; i++) {
789 thread_id_t tid = int_to_id(i);
791 /* Make sure this thread can be enabled here. */
792 if (i >= node->get_num_threads())
795 /* Don't backtrack into a point where the thread is disabled or sleeping. */
796 if (node->enabled_status(tid) != THREAD_ENABLED)
799 /* Check if this has been explored already */
800 if (node->has_been_explored(tid))
803 /* See if fairness allows */
804 if (model->params.fairwindow != 0 && !node->has_priority(tid)) {
806 for (int t = 0; t < node->get_num_threads(); t++) {
807 thread_id_t tother = int_to_id(t);
808 if (node->is_enabled(tother) && node->has_priority(tother)) {
816 /* Cache the latest backtracking point */
817 set_latest_backtrack(prev);
819 /* If this is a new backtracking point, mark the tree */
820 if (!node->set_backtrack(tid))
822 DEBUG("Setting backtrack: conflict = %d, instead tid = %d\n",
823 id_to_int(prev->get_tid()),
824 id_to_int(t->get_id()));
833 * @brief Cache the a backtracking point as the "most recent", if eligible
835 * Note that this does not prepare the NodeStack for this backtracking
836 * operation, it only caches the action on a per-execution basis
838 * @param act The operation at which we should explore a different next action
839 * (i.e., backtracking point)
840 * @return True, if this action is now the most recent backtracking point;
843 bool ModelChecker::set_latest_backtrack(ModelAction *act)
845 if (!priv->next_backtrack || *act > *priv->next_backtrack) {
846 priv->next_backtrack = act;
853 * Returns last backtracking point. The model checker will explore a different
854 * path for this point in the next execution.
855 * @return The ModelAction at which the next execution should diverge.
857 ModelAction * ModelChecker::get_next_backtrack()
859 ModelAction *next = priv->next_backtrack;
860 priv->next_backtrack = NULL;
865 * Processes a read model action.
866 * @param curr is the read model action to process.
867 * @return True if processing this read updates the mo_graph.
869 bool ModelChecker::process_read(ModelAction *curr)
871 Node *node = curr->get_node();
873 bool updated = false;
874 switch (node->get_read_from_status()) {
875 case READ_FROM_PAST: {
876 const ModelAction *rf = node->get_read_from_past();
879 mo_graph->startChanges();
881 ASSERT(!is_infeasible());
882 if (!check_recency(curr, rf)) {
883 if (node->increment_read_from()) {
884 mo_graph->rollbackChanges();
887 priv->too_many_reads = true;
891 updated = r_modification_order(curr, rf);
893 mo_graph->commitChanges();
894 mo_check_promises(curr, true);
897 case READ_FROM_PROMISE: {
898 Promise *promise = curr->get_node()->get_read_from_promise();
899 if (promise->add_reader(curr))
900 priv->failed_promise = true;
901 curr->set_read_from_promise(promise);
902 mo_graph->startChanges();
903 if (!check_recency(curr, promise))
904 priv->too_many_reads = true;
905 updated = r_modification_order(curr, promise);
906 mo_graph->commitChanges();
909 case READ_FROM_FUTURE: {
910 /* Read from future value */
911 struct future_value fv = node->get_future_value();
912 Promise *promise = new Promise(curr, fv);
913 curr->set_read_from_promise(promise);
914 promises->push_back(promise);
915 mo_graph->startChanges();
916 updated = r_modification_order(curr, promise);
917 mo_graph->commitChanges();
923 get_thread(curr)->set_return_value(curr->get_return_value());
929 * Processes a lock, trylock, or unlock model action. @param curr is
930 * the read model action to process.
932 * The try lock operation checks whether the lock is taken. If not,
933 * it falls to the normal lock operation case. If so, it returns
936 * The lock operation has already been checked that it is enabled, so
937 * it just grabs the lock and synchronizes with the previous unlock.
939 * The unlock operation has to re-enable all of the threads that are
940 * waiting on the lock.
942 * @return True if synchronization was updated; false otherwise
944 bool ModelChecker::process_mutex(ModelAction *curr)
946 std::mutex *mutex = curr->get_mutex();
947 struct std::mutex_state *state = NULL;
950 state = mutex->get_state();
952 switch (curr->get_type()) {
953 case ATOMIC_TRYLOCK: {
954 bool success = !state->locked;
955 curr->set_try_lock(success);
957 get_thread(curr)->set_return_value(0);
960 get_thread(curr)->set_return_value(1);
962 //otherwise fall into the lock case
964 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
965 assert_bug("Lock access before initialization");
966 state->locked = get_thread(curr);
967 ModelAction *unlock = get_last_unlock(curr);
968 //synchronize with the previous unlock statement
969 if (unlock != NULL) {
970 curr->synchronize_with(unlock);
975 case ATOMIC_UNLOCK: {
977 state->locked = NULL;
978 //wake up the other threads
979 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, curr->get_location());
980 //activate all the waiting threads
981 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
982 scheduler->wake(get_thread(*rit));
989 state->locked = NULL;
990 //wake up the other threads
991 action_list_t *waiters = get_safe_ptr_action(lock_waiters_map, (void *) curr->get_value());
992 //activate all the waiting threads
993 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
994 scheduler->wake(get_thread(*rit));
997 //check whether we should go to sleep or not...simulate spurious failures
998 if (curr->get_node()->get_misc() == 0) {
999 get_safe_ptr_action(condvar_waiters_map, curr->get_location())->push_back(curr);
1001 scheduler->sleep(get_thread(curr));
1005 case ATOMIC_NOTIFY_ALL: {
1006 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1007 //activate all the waiting threads
1008 for (action_list_t::iterator rit = waiters->begin(); rit != waiters->end(); rit++) {
1009 scheduler->wake(get_thread(*rit));
1014 case ATOMIC_NOTIFY_ONE: {
1015 action_list_t *waiters = get_safe_ptr_action(condvar_waiters_map, curr->get_location());
1016 int wakeupthread = curr->get_node()->get_misc();
1017 action_list_t::iterator it = waiters->begin();
1018 advance(it, wakeupthread);
1019 scheduler->wake(get_thread(*it));
1030 void ModelChecker::add_future_value(const ModelAction *writer, ModelAction *reader)
1032 /* Do more ambitious checks now that mo is more complete */
1033 if (mo_may_allow(writer, reader)) {
1034 Node *node = reader->get_node();
1036 /* Find an ancestor thread which exists at the time of the reader */
1037 Thread *write_thread = get_thread(writer);
1038 while (id_to_int(write_thread->get_id()) >= node->get_num_threads())
1039 write_thread = write_thread->get_parent();
1041 struct future_value fv = {
1042 writer->get_write_value(),
1043 writer->get_seq_number() + params.maxfuturedelay,
1044 write_thread->get_id(),
1046 if (node->add_future_value(fv))
1047 set_latest_backtrack(reader);
1052 * Process a write ModelAction
1053 * @param curr The ModelAction to process
1054 * @return True if the mo_graph was updated or promises were resolved
1056 bool ModelChecker::process_write(ModelAction *curr)
1058 /* Readers to which we may send our future value */
1059 std::vector< ModelAction *, ModelAlloc<ModelAction *> > send_fv;
1061 bool updated_mod_order = w_modification_order(curr, &send_fv);
1062 int promise_idx = get_promise_to_resolve(curr);
1063 const ModelAction *earliest_promise_reader;
1064 bool updated_promises = false;
1066 if (promise_idx >= 0) {
1067 earliest_promise_reader = (*promises)[promise_idx]->get_reader(0);
1068 updated_promises = resolve_promise(curr, promise_idx);
1070 earliest_promise_reader = NULL;
1072 /* Don't send future values to reads after the Promise we resolve */
1073 for (unsigned int i = 0; i < send_fv.size(); i++) {
1074 ModelAction *read = send_fv[i];
1075 if (!earliest_promise_reader || *read < *earliest_promise_reader)
1076 futurevalues->push_back(PendingFutureValue(curr, read));
1079 if (promises->size() == 0) {
1080 for (unsigned int i = 0; i < futurevalues->size(); i++) {
1081 struct PendingFutureValue pfv = (*futurevalues)[i];
1082 add_future_value(pfv.writer, pfv.act);
1084 futurevalues->clear();
1087 mo_graph->commitChanges();
1088 mo_check_promises(curr, false);
1090 get_thread(curr)->set_return_value(VALUE_NONE);
1091 return updated_mod_order || updated_promises;
1095 * Process a fence ModelAction
1096 * @param curr The ModelAction to process
1097 * @return True if synchronization was updated
1099 bool ModelChecker::process_fence(ModelAction *curr)
1102 * fence-relaxed: no-op
1103 * fence-release: only log the occurence (not in this function), for
1104 * use in later synchronization
1105 * fence-acquire (this function): search for hypothetical release
1108 bool updated = false;
1109 if (curr->is_acquire()) {
1110 action_list_t *list = action_trace;
1111 action_list_t::reverse_iterator rit;
1112 /* Find X : is_read(X) && X --sb-> curr */
1113 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1114 ModelAction *act = *rit;
1117 if (act->get_tid() != curr->get_tid())
1119 /* Stop at the beginning of the thread */
1120 if (act->is_thread_start())
1122 /* Stop once we reach a prior fence-acquire */
1123 if (act->is_fence() && act->is_acquire())
1125 if (!act->is_read())
1127 /* read-acquire will find its own release sequences */
1128 if (act->is_acquire())
1131 /* Establish hypothetical release sequences */
1132 rel_heads_list_t release_heads;
1133 get_release_seq_heads(curr, act, &release_heads);
1134 for (unsigned int i = 0; i < release_heads.size(); i++)
1135 if (!curr->synchronize_with(release_heads[i]))
1136 set_bad_synchronization();
1137 if (release_heads.size() != 0)
1145 * @brief Process the current action for thread-related activity
1147 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
1148 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
1149 * synchronization, etc. This function is a no-op for non-THREAD actions
1150 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
1152 * @param curr The current action
1153 * @return True if synchronization was updated or a thread completed
1155 bool ModelChecker::process_thread_action(ModelAction *curr)
1157 bool updated = false;
1159 switch (curr->get_type()) {
1160 case THREAD_CREATE: {
1161 thrd_t *thrd = (thrd_t *)curr->get_location();
1162 struct thread_params *params = (struct thread_params *)curr->get_value();
1163 Thread *th = new Thread(thrd, params->func, params->arg, get_thread(curr));
1165 th->set_creation(curr);
1166 /* Promises can be satisfied by children */
1167 for (unsigned int i = 0; i < promises->size(); i++) {
1168 Promise *promise = (*promises)[i];
1169 if (promise->thread_is_available(curr->get_tid()))
1170 promise->add_thread(th->get_id());
1175 Thread *blocking = curr->get_thread_operand();
1176 ModelAction *act = get_last_action(blocking->get_id());
1177 curr->synchronize_with(act);
1178 updated = true; /* trigger rel-seq checks */
1181 case THREAD_FINISH: {
1182 Thread *th = get_thread(curr);
1183 while (!th->wait_list_empty()) {
1184 ModelAction *act = th->pop_wait_list();
1185 scheduler->wake(get_thread(act));
1188 /* Completed thread can't satisfy promises */
1189 for (unsigned int i = 0; i < promises->size(); i++) {
1190 Promise *promise = (*promises)[i];
1191 if (promise->thread_is_available(th->get_id()))
1192 if (promise->eliminate_thread(th->get_id()))
1193 priv->failed_promise = true;
1195 updated = true; /* trigger rel-seq checks */
1198 case THREAD_START: {
1199 check_promises(curr->get_tid(), NULL, curr->get_cv());
1210 * @brief Process the current action for release sequence fixup activity
1212 * Performs model-checker release sequence fixups for the current action,
1213 * forcing a single pending release sequence to break (with a given, potential
1214 * "loose" write) or to complete (i.e., synchronize). If a pending release
1215 * sequence forms a complete release sequence, then we must perform the fixup
1216 * synchronization, mo_graph additions, etc.
1218 * @param curr The current action; must be a release sequence fixup action
1219 * @param work_queue The work queue to which to add work items as they are
1222 void ModelChecker::process_relseq_fixup(ModelAction *curr, work_queue_t *work_queue)
1224 const ModelAction *write = curr->get_node()->get_relseq_break();
1225 struct release_seq *sequence = pending_rel_seqs->back();
1226 pending_rel_seqs->pop_back();
1228 ModelAction *acquire = sequence->acquire;
1229 const ModelAction *rf = sequence->rf;
1230 const ModelAction *release = sequence->release;
1234 ASSERT(release->same_thread(rf));
1236 if (write == NULL) {
1238 * @todo Forcing a synchronization requires that we set
1239 * modification order constraints. For instance, we can't allow
1240 * a fixup sequence in which two separate read-acquire
1241 * operations read from the same sequence, where the first one
1242 * synchronizes and the other doesn't. Essentially, we can't
1243 * allow any writes to insert themselves between 'release' and
1247 /* Must synchronize */
1248 if (!acquire->synchronize_with(release)) {
1249 set_bad_synchronization();
1252 /* Re-check all pending release sequences */
1253 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
1254 /* Re-check act for mo_graph edges */
1255 work_queue->push_back(MOEdgeWorkEntry(acquire));
1257 /* propagate synchronization to later actions */
1258 action_list_t::reverse_iterator rit = action_trace->rbegin();
1259 for (; (*rit) != acquire; rit++) {
1260 ModelAction *propagate = *rit;
1261 if (acquire->happens_before(propagate)) {
1262 propagate->synchronize_with(acquire);
1263 /* Re-check 'propagate' for mo_graph edges */
1264 work_queue->push_back(MOEdgeWorkEntry(propagate));
1268 /* Break release sequence with new edges:
1269 * release --mo--> write --mo--> rf */
1270 mo_graph->addEdge(release, write);
1271 mo_graph->addEdge(write, rf);
1274 /* See if we have realized a data race */
1279 * Initialize the current action by performing one or more of the following
1280 * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
1281 * in the NodeStack, manipulating backtracking sets, allocating and
1282 * initializing clock vectors, and computing the promises to fulfill.
1284 * @param curr The current action, as passed from the user context; may be
1285 * freed/invalidated after the execution of this function, with a different
1286 * action "returned" its place (pass-by-reference)
1287 * @return True if curr is a newly-explored action; false otherwise
1289 bool ModelChecker::initialize_curr_action(ModelAction **curr)
1291 ModelAction *newcurr;
1293 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
1294 newcurr = process_rmw(*curr);
1297 if (newcurr->is_rmw())
1298 compute_promises(newcurr);
1304 (*curr)->set_seq_number(get_next_seq_num());
1306 newcurr = node_stack->explore_action(*curr, scheduler->get_enabled_array());
1308 /* First restore type and order in case of RMW operation */
1309 if ((*curr)->is_rmwr())
1310 newcurr->copy_typeandorder(*curr);
1312 ASSERT((*curr)->get_location() == newcurr->get_location());
1313 newcurr->copy_from_new(*curr);
1315 /* Discard duplicate ModelAction; use action from NodeStack */
1318 /* Always compute new clock vector */
1319 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1322 return false; /* Action was explored previously */
1326 /* Always compute new clock vector */
1327 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
1329 /* Assign most recent release fence */
1330 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
1333 * Perform one-time actions when pushing new ModelAction onto
1336 if (newcurr->is_write())
1337 compute_promises(newcurr);
1338 else if (newcurr->is_relseq_fixup())
1339 compute_relseq_breakwrites(newcurr);
1340 else if (newcurr->is_wait())
1341 newcurr->get_node()->set_misc_max(2);
1342 else if (newcurr->is_notify_one()) {
1343 newcurr->get_node()->set_misc_max(get_safe_ptr_action(condvar_waiters_map, newcurr->get_location())->size());
1345 return true; /* This was a new ModelAction */
1350 * @brief Establish reads-from relation between two actions
1352 * Perform basic operations involved with establishing a concrete rf relation,
1353 * including setting the ModelAction data and checking for release sequences.
1355 * @param act The action that is reading (must be a read)
1356 * @param rf The action from which we are reading (must be a write)
1358 * @return True if this read established synchronization
1360 bool ModelChecker::read_from(ModelAction *act, const ModelAction *rf)
1363 act->set_read_from(rf);
1364 if (act->is_acquire()) {
1365 rel_heads_list_t release_heads;
1366 get_release_seq_heads(act, act, &release_heads);
1367 int num_heads = release_heads.size();
1368 for (unsigned int i = 0; i < release_heads.size(); i++)
1369 if (!act->synchronize_with(release_heads[i])) {
1370 set_bad_synchronization();
1373 return num_heads > 0;
1379 * Check promises and eliminate potentially-satisfying threads when a thread is
1380 * blocked (e.g., join, lock). A thread which is waiting on another thread can
1381 * no longer satisfy a promise generated from that thread.
1383 * @param blocker The thread on which a thread is waiting
1384 * @param waiting The waiting thread
1386 void ModelChecker::thread_blocking_check_promises(Thread *blocker, Thread *waiting)
1388 for (unsigned int i = 0; i < promises->size(); i++) {
1389 Promise *promise = (*promises)[i];
1390 if (!promise->thread_is_available(waiting->get_id()))
1392 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
1393 ModelAction *reader = promise->get_reader(j);
1394 if (reader->get_tid() != blocker->get_id())
1396 if (promise->eliminate_thread(waiting->get_id())) {
1397 /* Promise has failed */
1398 priv->failed_promise = true;
1400 /* Only eliminate the 'waiting' thread once */
1408 * @brief Check whether a model action is enabled.
1410 * Checks whether a lock or join operation would be successful (i.e., is the
1411 * lock already locked, or is the joined thread already complete). If not, put
1412 * the action in a waiter list.
1414 * @param curr is the ModelAction to check whether it is enabled.
1415 * @return a bool that indicates whether the action is enabled.
1417 bool ModelChecker::check_action_enabled(ModelAction *curr) {
1418 if (curr->is_lock()) {
1419 std::mutex *lock = (std::mutex *)curr->get_location();
1420 struct std::mutex_state *state = lock->get_state();
1421 if (state->locked) {
1422 //Stick the action in the appropriate waiting queue
1423 get_safe_ptr_action(lock_waiters_map, curr->get_location())->push_back(curr);
1426 } else if (curr->get_type() == THREAD_JOIN) {
1427 Thread *blocking = (Thread *)curr->get_location();
1428 if (!blocking->is_complete()) {
1429 blocking->push_wait_list(curr);
1430 thread_blocking_check_promises(blocking, get_thread(curr));
1439 * This is the heart of the model checker routine. It performs model-checking
1440 * actions corresponding to a given "current action." Among other processes, it
1441 * calculates reads-from relationships, updates synchronization clock vectors,
1442 * forms a memory_order constraints graph, and handles replay/backtrack
1443 * execution when running permutations of previously-observed executions.
1445 * @param curr The current action to process
1446 * @return The ModelAction that is actually executed; may be different than
1447 * curr; may be NULL, if the current action is not enabled to run
1449 ModelAction * ModelChecker::check_current_action(ModelAction *curr)
1452 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
1454 if (!check_action_enabled(curr)) {
1455 /* Make the execution look like we chose to run this action
1456 * much later, when a lock/join can succeed */
1457 get_thread(curr)->set_pending(curr);
1458 scheduler->sleep(get_thread(curr));
1462 bool newly_explored = initialize_curr_action(&curr);
1468 wake_up_sleeping_actions(curr);
1470 /* Add the action to lists before any other model-checking tasks */
1471 if (!second_part_of_rmw)
1472 add_action_to_lists(curr);
1474 /* Build may_read_from set for newly-created actions */
1475 if (newly_explored && curr->is_read())
1476 build_may_read_from(curr);
1478 /* Initialize work_queue with the "current action" work */
1479 work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
1480 while (!work_queue.empty() && !has_asserted()) {
1481 WorkQueueEntry work = work_queue.front();
1482 work_queue.pop_front();
1484 switch (work.type) {
1485 case WORK_CHECK_CURR_ACTION: {
1486 ModelAction *act = work.action;
1487 bool update = false; /* update this location's release seq's */
1488 bool update_all = false; /* update all release seq's */
1490 if (process_thread_action(curr))
1493 if (act->is_read() && !second_part_of_rmw && process_read(act))
1496 if (act->is_write() && process_write(act))
1499 if (act->is_fence() && process_fence(act))
1502 if (act->is_mutex_op() && process_mutex(act))
1505 if (act->is_relseq_fixup())
1506 process_relseq_fixup(curr, &work_queue);
1509 work_queue.push_back(CheckRelSeqWorkEntry(NULL));
1511 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1514 case WORK_CHECK_RELEASE_SEQ:
1515 resolve_release_sequences(work.location, &work_queue);
1517 case WORK_CHECK_MO_EDGES: {
1518 /** @todo Complete verification of work_queue */
1519 ModelAction *act = work.action;
1520 bool updated = false;
1522 if (act->is_read()) {
1523 const ModelAction *rf = act->get_reads_from();
1524 const Promise *promise = act->get_reads_from_promise();
1526 if (r_modification_order(act, rf))
1528 } else if (promise) {
1529 if (r_modification_order(act, promise))
1533 if (act->is_write()) {
1534 if (w_modification_order(act, NULL))
1537 mo_graph->commitChanges();
1540 work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
1549 check_curr_backtracking(curr);
1550 set_backtracking(curr);
1554 void ModelChecker::check_curr_backtracking(ModelAction *curr)
1556 Node *currnode = curr->get_node();
1557 Node *parnode = currnode->get_parent();
1559 if ((parnode && !parnode->backtrack_empty()) ||
1560 !currnode->misc_empty() ||
1561 !currnode->read_from_empty() ||
1562 !currnode->promise_empty() ||
1563 !currnode->relseq_break_empty()) {
1564 set_latest_backtrack(curr);
1568 bool ModelChecker::promises_expired() const
1570 for (unsigned int i = 0; i < promises->size(); i++) {
1571 Promise *promise = (*promises)[i];
1572 if (promise->get_expiration() < priv->used_sequence_numbers)
1579 * This is the strongest feasibility check available.
1580 * @return whether the current trace (partial or complete) must be a prefix of
1583 bool ModelChecker::isfeasibleprefix() const
1585 return pending_rel_seqs->size() == 0 && is_feasible_prefix_ignore_relseq();
1589 * Print disagnostic information about an infeasible execution
1590 * @param prefix A string to prefix the output with; if NULL, then a default
1591 * message prefix will be provided
1593 void ModelChecker::print_infeasibility(const char *prefix) const
1597 if (mo_graph->checkForCycles())
1598 ptr += sprintf(ptr, "[mo cycle]");
1599 if (priv->failed_promise)
1600 ptr += sprintf(ptr, "[failed promise]");
1601 if (priv->too_many_reads)
1602 ptr += sprintf(ptr, "[too many reads]");
1603 if (priv->no_valid_reads)
1604 ptr += sprintf(ptr, "[no valid reads-from]");
1605 if (priv->bad_synchronization)
1606 ptr += sprintf(ptr, "[bad sw ordering]");
1607 if (promises_expired())
1608 ptr += sprintf(ptr, "[promise expired]");
1609 if (promises->size() != 0)
1610 ptr += sprintf(ptr, "[unresolved promise]");
1612 model_print("%s: %s\n", prefix ? prefix : "Infeasible", buf);
1616 * Returns whether the current completed trace is feasible, except for pending
1617 * release sequences.
1619 bool ModelChecker::is_feasible_prefix_ignore_relseq() const
1621 return !is_infeasible() && promises->size() == 0;
1625 * Check if the current partial trace is infeasible. Does not check any
1626 * end-of-execution flags, which might rule out the execution. Thus, this is
1627 * useful only for ruling an execution as infeasible.
1628 * @return whether the current partial trace is infeasible.
1630 bool ModelChecker::is_infeasible() const
1632 return mo_graph->checkForCycles() ||
1633 priv->no_valid_reads ||
1634 priv->failed_promise ||
1635 priv->too_many_reads ||
1636 priv->bad_synchronization ||
1640 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
1641 ModelAction * ModelChecker::process_rmw(ModelAction *act) {
1642 ModelAction *lastread = get_last_action(act->get_tid());
1643 lastread->process_rmw(act);
1644 if (act->is_rmw()) {
1645 if (lastread->get_reads_from())
1646 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
1648 mo_graph->addRMWEdge(lastread->get_reads_from_promise(), lastread);
1649 mo_graph->commitChanges();
1655 * A helper function for ModelChecker::check_recency, to check if the current
1656 * thread is able to read from a different write/promise for 'params.maxreads'
1657 * number of steps and if that write/promise should become visible (i.e., is
1658 * ordered later in the modification order). This helps model memory liveness.
1660 * @param curr The current action. Must be a read.
1661 * @param rf The write/promise from which we plan to read
1662 * @param other_rf The write/promise from which we may read
1663 * @return True if we were able to read from other_rf for params.maxreads steps
1665 template <typename T, typename U>
1666 bool ModelChecker::should_read_instead(const ModelAction *curr, const T *rf, const U *other_rf) const
1668 /* Need a different write/promise */
1669 if (other_rf->equals(rf))
1672 /* Only look for "newer" writes/promises */
1673 if (!mo_graph->checkReachable(rf, other_rf))
1676 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1677 action_list_t *list = &(*thrd_lists)[id_to_int(curr->get_tid())];
1678 action_list_t::reverse_iterator rit = list->rbegin();
1679 ASSERT((*rit) == curr);
1680 /* Skip past curr */
1683 /* Does this write/promise work for everyone? */
1684 for (int i = 0; i < params.maxreads; i++, rit++) {
1685 ModelAction *act = *rit;
1686 if (!act->may_read_from(other_rf))
1693 * Checks whether a thread has read from the same write or Promise for too many
1694 * times without seeing the effects of a later write/Promise.
1697 * 1) there must a different write/promise that we could read from,
1698 * 2) we must have read from the same write/promise in excess of maxreads times,
1699 * 3) that other write/promise must have been in the reads_from set for maxreads times, and
1700 * 4) that other write/promise must be mod-ordered after the write/promise we are reading.
1702 * If so, we decide that the execution is no longer feasible.
1704 * @param curr The current action. Must be a read.
1705 * @param rf The ModelAction/Promise from which we might read.
1706 * @return True if the read should succeed; false otherwise
1708 template <typename T>
1709 bool ModelChecker::check_recency(ModelAction *curr, const T *rf) const
1711 if (!params.maxreads)
1714 //NOTE: Next check is just optimization, not really necessary....
1715 if (curr->get_node()->get_read_from_past_size() +
1716 curr->get_node()->get_read_from_promise_size() <= 1)
1719 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1720 int tid = id_to_int(curr->get_tid());
1721 ASSERT(tid < (int)thrd_lists->size());
1722 action_list_t *list = &(*thrd_lists)[tid];
1723 action_list_t::reverse_iterator rit = list->rbegin();
1724 ASSERT((*rit) == curr);
1725 /* Skip past curr */
1728 action_list_t::reverse_iterator ritcopy = rit;
1729 /* See if we have enough reads from the same value */
1730 for (int count = 0; count < params.maxreads; ritcopy++, count++) {
1731 if (ritcopy == list->rend())
1733 ModelAction *act = *ritcopy;
1734 if (!act->is_read())
1736 if (act->get_reads_from_promise() && !act->get_reads_from_promise()->equals(rf))
1738 if (act->get_reads_from() && !act->get_reads_from()->equals(rf))
1740 if (act->get_node()->get_read_from_past_size() +
1741 act->get_node()->get_read_from_promise_size() <= 1)
1744 for (int i = 0; i < curr->get_node()->get_read_from_past_size(); i++) {
1745 const ModelAction *write = curr->get_node()->get_read_from_past(i);
1746 if (should_read_instead(curr, rf, write))
1747 return false; /* liveness failure */
1749 for (int i = 0; i < curr->get_node()->get_read_from_promise_size(); i++) {
1750 const Promise *promise = curr->get_node()->get_read_from_promise(i);
1751 if (should_read_instead(curr, rf, promise))
1752 return false; /* liveness failure */
1758 * Updates the mo_graph with the constraints imposed from the current
1761 * Basic idea is the following: Go through each other thread and find
1762 * the last action that happened before our read. Two cases:
1764 * (1) The action is a write => that write must either occur before
1765 * the write we read from or be the write we read from.
1767 * (2) The action is a read => the write that that action read from
1768 * must occur before the write we read from or be the same write.
1770 * @param curr The current action. Must be a read.
1771 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
1772 * @return True if modification order edges were added; false otherwise
1774 template <typename rf_type>
1775 bool ModelChecker::r_modification_order(ModelAction *curr, const rf_type *rf)
1777 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1780 ASSERT(curr->is_read());
1782 /* Last SC fence in the current thread */
1783 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1785 /* Iterate over all threads */
1786 for (i = 0; i < thrd_lists->size(); i++) {
1787 /* Last SC fence in thread i */
1788 ModelAction *last_sc_fence_thread_local = NULL;
1789 if (int_to_id((int)i) != curr->get_tid())
1790 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
1792 /* Last SC fence in thread i, before last SC fence in current thread */
1793 ModelAction *last_sc_fence_thread_before = NULL;
1794 if (last_sc_fence_local)
1795 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1797 /* Iterate over actions in thread, starting from most recent */
1798 action_list_t *list = &(*thrd_lists)[i];
1799 action_list_t::reverse_iterator rit;
1800 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1801 ModelAction *act = *rit;
1803 if (act->is_write() && !act->equals(rf) && act != curr) {
1804 /* C++, Section 29.3 statement 5 */
1805 if (curr->is_seqcst() && last_sc_fence_thread_local &&
1806 *act < *last_sc_fence_thread_local) {
1807 added = mo_graph->addEdge(act, rf) || added;
1810 /* C++, Section 29.3 statement 4 */
1811 else if (act->is_seqcst() && last_sc_fence_local &&
1812 *act < *last_sc_fence_local) {
1813 added = mo_graph->addEdge(act, rf) || added;
1816 /* C++, Section 29.3 statement 6 */
1817 else if (last_sc_fence_thread_before &&
1818 *act < *last_sc_fence_thread_before) {
1819 added = mo_graph->addEdge(act, rf) || added;
1825 * Include at most one act per-thread that "happens
1826 * before" curr. Don't consider reflexively.
1828 if (act->happens_before(curr) && act != curr) {
1829 if (act->is_write()) {
1830 if (!act->equals(rf)) {
1831 added = mo_graph->addEdge(act, rf) || added;
1834 const ModelAction *prevrf = act->get_reads_from();
1835 const Promise *prevrf_promise = act->get_reads_from_promise();
1837 if (!prevrf->equals(rf))
1838 added = mo_graph->addEdge(prevrf, rf) || added;
1839 } else if (!prevrf_promise->equals(rf)) {
1840 added = mo_graph->addEdge(prevrf_promise, rf) || added;
1849 * All compatible, thread-exclusive promises must be ordered after any
1850 * concrete loads from the same thread
1852 for (unsigned int i = 0; i < promises->size(); i++)
1853 if ((*promises)[i]->is_compatible_exclusive(curr))
1854 added = mo_graph->addEdge(rf, (*promises)[i]) || added;
1860 * Updates the mo_graph with the constraints imposed from the current write.
1862 * Basic idea is the following: Go through each other thread and find
1863 * the lastest action that happened before our write. Two cases:
1865 * (1) The action is a write => that write must occur before
1868 * (2) The action is a read => the write that that action read from
1869 * must occur before the current write.
1871 * This method also handles two other issues:
1873 * (I) Sequential Consistency: Making sure that if the current write is
1874 * seq_cst, that it occurs after the previous seq_cst write.
1876 * (II) Sending the write back to non-synchronizing reads.
1878 * @param curr The current action. Must be a write.
1879 * @param send_fv A vector for stashing reads to which we may pass our future
1880 * value. If NULL, then don't record any future values.
1881 * @return True if modification order edges were added; false otherwise
1883 bool ModelChecker::w_modification_order(ModelAction *curr, std::vector< ModelAction *, ModelAlloc<ModelAction *> > *send_fv)
1885 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
1888 ASSERT(curr->is_write());
1890 if (curr->is_seqcst()) {
1891 /* We have to at least see the last sequentially consistent write,
1892 so we are initialized. */
1893 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
1894 if (last_seq_cst != NULL) {
1895 added = mo_graph->addEdge(last_seq_cst, curr) || added;
1899 /* Last SC fence in the current thread */
1900 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
1902 /* Iterate over all threads */
1903 for (i = 0; i < thrd_lists->size(); i++) {
1904 /* Last SC fence in thread i, before last SC fence in current thread */
1905 ModelAction *last_sc_fence_thread_before = NULL;
1906 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
1907 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
1909 /* Iterate over actions in thread, starting from most recent */
1910 action_list_t *list = &(*thrd_lists)[i];
1911 action_list_t::reverse_iterator rit;
1912 for (rit = list->rbegin(); rit != list->rend(); rit++) {
1913 ModelAction *act = *rit;
1916 * 1) If RMW and it actually read from something, then we
1917 * already have all relevant edges, so just skip to next
1920 * 2) If RMW and it didn't read from anything, we should
1921 * whatever edge we can get to speed up convergence.
1923 * 3) If normal write, we need to look at earlier actions, so
1924 * continue processing list.
1926 if (curr->is_rmw()) {
1927 if (curr->get_reads_from() != NULL)
1935 /* C++, Section 29.3 statement 7 */
1936 if (last_sc_fence_thread_before && act->is_write() &&
1937 *act < *last_sc_fence_thread_before) {
1938 added = mo_graph->addEdge(act, curr) || added;
1943 * Include at most one act per-thread that "happens
1946 if (act->happens_before(curr)) {
1948 * Note: if act is RMW, just add edge:
1950 * The following edge should be handled elsewhere:
1951 * readfrom(act) --mo--> act
1953 if (act->is_write())
1954 added = mo_graph->addEdge(act, curr) || added;
1955 else if (act->is_read()) {
1956 //if previous read accessed a null, just keep going
1957 if (act->get_reads_from() == NULL)
1959 added = mo_graph->addEdge(act->get_reads_from(), curr) || added;
1962 } else if (act->is_read() && !act->could_synchronize_with(curr) &&
1963 !act->same_thread(curr)) {
1964 /* We have an action that:
1965 (1) did not happen before us
1966 (2) is a read and we are a write
1967 (3) cannot synchronize with us
1968 (4) is in a different thread
1970 that read could potentially read from our write. Note that
1971 these checks are overly conservative at this point, we'll
1972 do more checks before actually removing the
1976 if (send_fv && thin_air_constraint_may_allow(curr, act)) {
1977 if (!is_infeasible())
1978 send_fv->push_back(act);
1979 else if (curr->is_rmw() && act->is_rmw() && curr->get_reads_from() && curr->get_reads_from() == act->get_reads_from())
1980 add_future_value(curr, act);
1987 * All compatible, thread-exclusive promises must be ordered after any
1988 * concrete stores to the same thread, or else they can be merged with
1991 for (unsigned int i = 0; i < promises->size(); i++)
1992 if ((*promises)[i]->is_compatible_exclusive(curr))
1993 added = mo_graph->addEdge(curr, (*promises)[i]) || added;
1998 /** Arbitrary reads from the future are not allowed. Section 29.3
1999 * part 9 places some constraints. This method checks one result of constraint
2000 * constraint. Others require compiler support. */
2001 bool ModelChecker::thin_air_constraint_may_allow(const ModelAction *writer, const ModelAction *reader)
2003 if (!writer->is_rmw())
2006 if (!reader->is_rmw())
2009 for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
2010 if (search == reader)
2012 if (search->get_tid() == reader->get_tid() &&
2013 search->happens_before(reader))
2021 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
2022 * some constraints. This method checks one the following constraint (others
2023 * require compiler support):
2025 * If X --hb-> Y --mo-> Z, then X should not read from Z.
2027 bool ModelChecker::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
2029 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, reader->get_location());
2031 /* Iterate over all threads */
2032 for (i = 0; i < thrd_lists->size(); i++) {
2033 const ModelAction *write_after_read = NULL;
2035 /* Iterate over actions in thread, starting from most recent */
2036 action_list_t *list = &(*thrd_lists)[i];
2037 action_list_t::reverse_iterator rit;
2038 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2039 ModelAction *act = *rit;
2041 /* Don't disallow due to act == reader */
2042 if (!reader->happens_before(act) || reader == act)
2044 else if (act->is_write())
2045 write_after_read = act;
2046 else if (act->is_read() && act->get_reads_from() != NULL)
2047 write_after_read = act->get_reads_from();
2050 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
2057 * Finds the head(s) of the release sequence(s) containing a given ModelAction.
2058 * The ModelAction under consideration is expected to be taking part in
2059 * release/acquire synchronization as an object of the "reads from" relation.
2060 * Note that this can only provide release sequence support for RMW chains
2061 * which do not read from the future, as those actions cannot be traced until
2062 * their "promise" is fulfilled. Similarly, we may not even establish the
2063 * presence of a release sequence with certainty, as some modification order
2064 * constraints may be decided further in the future. Thus, this function
2065 * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
2066 * and a boolean representing certainty.
2068 * @param rf The action that might be part of a release sequence. Must be a
2070 * @param release_heads A pass-by-reference style return parameter. After
2071 * execution of this function, release_heads will contain the heads of all the
2072 * relevant release sequences, if any exists with certainty
2073 * @param pending A pass-by-reference style return parameter which is only used
2074 * when returning false (i.e., uncertain). Returns most information regarding
2075 * an uncertain release sequence, including any write operations that might
2076 * break the sequence.
2077 * @return true, if the ModelChecker is certain that release_heads is complete;
2080 bool ModelChecker::release_seq_heads(const ModelAction *rf,
2081 rel_heads_list_t *release_heads,
2082 struct release_seq *pending) const
2084 /* Only check for release sequences if there are no cycles */
2085 if (mo_graph->checkForCycles())
2088 for ( ; rf != NULL; rf = rf->get_reads_from()) {
2089 ASSERT(rf->is_write());
2091 if (rf->is_release())
2092 release_heads->push_back(rf);
2093 else if (rf->get_last_fence_release())
2094 release_heads->push_back(rf->get_last_fence_release());
2096 break; /* End of RMW chain */
2098 /** @todo Need to be smarter here... In the linux lock
2099 * example, this will run to the beginning of the program for
2101 /** @todo The way to be smarter here is to keep going until 1
2102 * thread has a release preceded by an acquire and you've seen
2105 /* acq_rel RMW is a sufficient stopping condition */
2106 if (rf->is_acquire() && rf->is_release())
2107 return true; /* complete */
2110 /* read from future: need to settle this later */
2112 return false; /* incomplete */
2115 if (rf->is_release())
2116 return true; /* complete */
2118 /* else relaxed write
2119 * - check for fence-release in the same thread (29.8, stmt. 3)
2120 * - check modification order for contiguous subsequence
2121 * -> rf must be same thread as release */
2123 const ModelAction *fence_release = rf->get_last_fence_release();
2124 /* Synchronize with a fence-release unconditionally; we don't need to
2125 * find any more "contiguous subsequence..." for it */
2127 release_heads->push_back(fence_release);
2129 int tid = id_to_int(rf->get_tid());
2130 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, rf->get_location());
2131 action_list_t *list = &(*thrd_lists)[tid];
2132 action_list_t::const_reverse_iterator rit;
2134 /* Find rf in the thread list */
2135 rit = std::find(list->rbegin(), list->rend(), rf);
2136 ASSERT(rit != list->rend());
2138 /* Find the last {write,fence}-release */
2139 for (; rit != list->rend(); rit++) {
2140 if (fence_release && *(*rit) < *fence_release)
2142 if ((*rit)->is_release())
2145 if (rit == list->rend()) {
2146 /* No write-release in this thread */
2147 return true; /* complete */
2148 } else if (fence_release && *(*rit) < *fence_release) {
2149 /* The fence-release is more recent (and so, "stronger") than
2150 * the most recent write-release */
2151 return true; /* complete */
2152 } /* else, need to establish contiguous release sequence */
2153 ModelAction *release = *rit;
2155 ASSERT(rf->same_thread(release));
2157 pending->writes.clear();
2159 bool certain = true;
2160 for (unsigned int i = 0; i < thrd_lists->size(); i++) {
2161 if (id_to_int(rf->get_tid()) == (int)i)
2163 list = &(*thrd_lists)[i];
2165 /* Can we ensure no future writes from this thread may break
2166 * the release seq? */
2167 bool future_ordered = false;
2169 ModelAction *last = get_last_action(int_to_id(i));
2170 Thread *th = get_thread(int_to_id(i));
2171 if ((last && rf->happens_before(last)) ||
2174 future_ordered = true;
2176 ASSERT(!th->is_model_thread() || future_ordered);
2178 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2179 const ModelAction *act = *rit;
2180 /* Reach synchronization -> this thread is complete */
2181 if (act->happens_before(release))
2183 if (rf->happens_before(act)) {
2184 future_ordered = true;
2188 /* Only non-RMW writes can break release sequences */
2189 if (!act->is_write() || act->is_rmw())
2192 /* Check modification order */
2193 if (mo_graph->checkReachable(rf, act)) {
2194 /* rf --mo--> act */
2195 future_ordered = true;
2198 if (mo_graph->checkReachable(act, release))
2199 /* act --mo--> release */
2201 if (mo_graph->checkReachable(release, act) &&
2202 mo_graph->checkReachable(act, rf)) {
2203 /* release --mo-> act --mo--> rf */
2204 return true; /* complete */
2206 /* act may break release sequence */
2207 pending->writes.push_back(act);
2210 if (!future_ordered)
2211 certain = false; /* This thread is uncertain */
2215 release_heads->push_back(release);
2216 pending->writes.clear();
2218 pending->release = release;
2225 * An interface for getting the release sequence head(s) with which a
2226 * given ModelAction must synchronize. This function only returns a non-empty
2227 * result when it can locate a release sequence head with certainty. Otherwise,
2228 * it may mark the internal state of the ModelChecker so that it will handle
2229 * the release sequence at a later time, causing @a acquire to update its
2230 * synchronization at some later point in execution.
2232 * @param acquire The 'acquire' action that may synchronize with a release
2234 * @param read The read action that may read from a release sequence; this may
2235 * be the same as acquire, or else an earlier action in the same thread (i.e.,
2236 * when 'acquire' is a fence-acquire)
2237 * @param release_heads A pass-by-reference return parameter. Will be filled
2238 * with the head(s) of the release sequence(s), if they exists with certainty.
2239 * @see ModelChecker::release_seq_heads
2241 void ModelChecker::get_release_seq_heads(ModelAction *acquire,
2242 ModelAction *read, rel_heads_list_t *release_heads)
2244 const ModelAction *rf = read->get_reads_from();
2245 struct release_seq *sequence = (struct release_seq *)snapshot_calloc(1, sizeof(struct release_seq));
2246 sequence->acquire = acquire;
2247 sequence->read = read;
2249 if (!release_seq_heads(rf, release_heads, sequence)) {
2250 /* add act to 'lazy checking' list */
2251 pending_rel_seqs->push_back(sequence);
2253 snapshot_free(sequence);
2258 * Attempt to resolve all stashed operations that might synchronize with a
2259 * release sequence for a given location. This implements the "lazy" portion of
2260 * determining whether or not a release sequence was contiguous, since not all
2261 * modification order information is present at the time an action occurs.
2263 * @param location The location/object that should be checked for release
2264 * sequence resolutions. A NULL value means to check all locations.
2265 * @param work_queue The work queue to which to add work items as they are
2267 * @return True if any updates occurred (new synchronization, new mo_graph
2270 bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
2272 bool updated = false;
2273 std::vector< struct release_seq *, SnapshotAlloc<struct release_seq *> >::iterator it = pending_rel_seqs->begin();
2274 while (it != pending_rel_seqs->end()) {
2275 struct release_seq *pending = *it;
2276 ModelAction *acquire = pending->acquire;
2277 const ModelAction *read = pending->read;
2279 /* Only resolve sequences on the given location, if provided */
2280 if (location && read->get_location() != location) {
2285 const ModelAction *rf = read->get_reads_from();
2286 rel_heads_list_t release_heads;
2288 complete = release_seq_heads(rf, &release_heads, pending);
2289 for (unsigned int i = 0; i < release_heads.size(); i++) {
2290 if (!acquire->has_synchronized_with(release_heads[i])) {
2291 if (acquire->synchronize_with(release_heads[i]))
2294 set_bad_synchronization();
2299 /* Re-check all pending release sequences */
2300 work_queue->push_back(CheckRelSeqWorkEntry(NULL));
2301 /* Re-check read-acquire for mo_graph edges */
2302 if (acquire->is_read())
2303 work_queue->push_back(MOEdgeWorkEntry(acquire));
2305 /* propagate synchronization to later actions */
2306 action_list_t::reverse_iterator rit = action_trace->rbegin();
2307 for (; (*rit) != acquire; rit++) {
2308 ModelAction *propagate = *rit;
2309 if (acquire->happens_before(propagate)) {
2310 propagate->synchronize_with(acquire);
2311 /* Re-check 'propagate' for mo_graph edges */
2312 work_queue->push_back(MOEdgeWorkEntry(propagate));
2317 it = pending_rel_seqs->erase(it);
2318 snapshot_free(pending);
2324 // If we resolved promises or data races, see if we have realized a data race.
2331 * Performs various bookkeeping operations for the current ModelAction. For
2332 * instance, adds action to the per-object, per-thread action vector and to the
2333 * action trace list of all thread actions.
2335 * @param act is the ModelAction to add.
2337 void ModelChecker::add_action_to_lists(ModelAction *act)
2339 int tid = id_to_int(act->get_tid());
2340 ModelAction *uninit = NULL;
2342 action_list_t *list = get_safe_ptr_action(obj_map, act->get_location());
2343 if (list->empty() && act->is_atomic_var()) {
2344 uninit = new_uninitialized_action(act->get_location());
2345 uninit_id = id_to_int(uninit->get_tid());
2346 list->push_back(uninit);
2348 list->push_back(act);
2350 action_trace->push_back(act);
2352 action_trace->push_front(uninit);
2354 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, act->get_location());
2355 if (tid >= (int)vec->size())
2356 vec->resize(priv->next_thread_id);
2357 (*vec)[tid].push_back(act);
2359 (*vec)[uninit_id].push_front(uninit);
2361 if ((int)thrd_last_action->size() <= tid)
2362 thrd_last_action->resize(get_num_threads());
2363 (*thrd_last_action)[tid] = act;
2365 (*thrd_last_action)[uninit_id] = uninit;
2367 if (act->is_fence() && act->is_release()) {
2368 if ((int)thrd_last_fence_release->size() <= tid)
2369 thrd_last_fence_release->resize(get_num_threads());
2370 (*thrd_last_fence_release)[tid] = act;
2373 if (act->is_wait()) {
2374 void *mutex_loc = (void *) act->get_value();
2375 get_safe_ptr_action(obj_map, mutex_loc)->push_back(act);
2377 std::vector<action_list_t> *vec = get_safe_ptr_vect_action(obj_thrd_map, mutex_loc);
2378 if (tid >= (int)vec->size())
2379 vec->resize(priv->next_thread_id);
2380 (*vec)[tid].push_back(act);
2385 * @brief Get the last action performed by a particular Thread
2386 * @param tid The thread ID of the Thread in question
2387 * @return The last action in the thread
2389 ModelAction * ModelChecker::get_last_action(thread_id_t tid) const
2391 int threadid = id_to_int(tid);
2392 if (threadid < (int)thrd_last_action->size())
2393 return (*thrd_last_action)[id_to_int(tid)];
2399 * @brief Get the last fence release performed by a particular Thread
2400 * @param tid The thread ID of the Thread in question
2401 * @return The last fence release in the thread, if one exists; NULL otherwise
2403 ModelAction * ModelChecker::get_last_fence_release(thread_id_t tid) const
2405 int threadid = id_to_int(tid);
2406 if (threadid < (int)thrd_last_fence_release->size())
2407 return (*thrd_last_fence_release)[id_to_int(tid)];
2413 * Gets the last memory_order_seq_cst write (in the total global sequence)
2414 * performed on a particular object (i.e., memory location), not including the
2416 * @param curr The current ModelAction; also denotes the object location to
2418 * @return The last seq_cst write
2420 ModelAction * ModelChecker::get_last_seq_cst_write(ModelAction *curr) const
2422 void *location = curr->get_location();
2423 action_list_t *list = get_safe_ptr_action(obj_map, location);
2424 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
2425 action_list_t::reverse_iterator rit;
2426 for (rit = list->rbegin(); rit != list->rend(); rit++)
2427 if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
2433 * Gets the last memory_order_seq_cst fence (in the total global sequence)
2434 * performed in a particular thread, prior to a particular fence.
2435 * @param tid The ID of the thread to check
2436 * @param before_fence The fence from which to begin the search; if NULL, then
2437 * search for the most recent fence in the thread.
2438 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
2440 ModelAction * ModelChecker::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
2442 /* All fences should have NULL location */
2443 action_list_t *list = get_safe_ptr_action(obj_map, NULL);
2444 action_list_t::reverse_iterator rit = list->rbegin();
2447 for (; rit != list->rend(); rit++)
2448 if (*rit == before_fence)
2451 ASSERT(*rit == before_fence);
2455 for (; rit != list->rend(); rit++)
2456 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
2462 * Gets the last unlock operation performed on a particular mutex (i.e., memory
2463 * location). This function identifies the mutex according to the current
2464 * action, which is presumed to perform on the same mutex.
2465 * @param curr The current ModelAction; also denotes the object location to
2467 * @return The last unlock operation
2469 ModelAction * ModelChecker::get_last_unlock(ModelAction *curr) const
2471 void *location = curr->get_location();
2472 action_list_t *list = get_safe_ptr_action(obj_map, location);
2473 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
2474 action_list_t::reverse_iterator rit;
2475 for (rit = list->rbegin(); rit != list->rend(); rit++)
2476 if ((*rit)->is_unlock() || (*rit)->is_wait())
2481 ModelAction * ModelChecker::get_parent_action(thread_id_t tid) const
2483 ModelAction *parent = get_last_action(tid);
2485 parent = get_thread(tid)->get_creation();
2490 * Returns the clock vector for a given thread.
2491 * @param tid The thread whose clock vector we want
2492 * @return Desired clock vector
2494 ClockVector * ModelChecker::get_cv(thread_id_t tid) const
2496 return get_parent_action(tid)->get_cv();
2500 * @brief Find the promise, if any to resolve for the current action
2501 * @param curr The current ModelAction. Should be a write.
2502 * @return The (non-negative) index for the Promise to resolve, if any;
2505 int ModelChecker::get_promise_to_resolve(const ModelAction *curr) const
2507 for (unsigned int i = 0; i < promises->size(); i++)
2508 if (curr->get_node()->get_promise(i))
2514 * Resolve a Promise with a current write.
2515 * @param write The ModelAction that is fulfilling Promises
2516 * @param promise_idx The index corresponding to the promise
2517 * @return True if the Promise was successfully resolved; false otherwise
2519 bool ModelChecker::resolve_promise(ModelAction *write, unsigned int promise_idx)
2521 std::vector< ModelAction *, ModelAlloc<ModelAction *> > actions_to_check;
2522 Promise *promise = (*promises)[promise_idx];
2524 for (unsigned int i = 0; i < promise->get_num_readers(); i++) {
2525 ModelAction *read = promise->get_reader(i);
2526 read_from(read, write);
2527 actions_to_check.push_back(read);
2529 /* Make sure the promise's value matches the write's value */
2530 ASSERT(promise->is_compatible(write) && promise->same_value(write));
2531 if (!mo_graph->resolvePromise(promise, write))
2532 priv->failed_promise = true;
2534 promises->erase(promises->begin() + promise_idx);
2536 * @todo It is possible to end up in an inconsistent state, where a
2537 * "resolved" promise may still be referenced if
2538 * CycleGraph::resolvePromise() failed, so don't delete 'promise'.
2540 * Note that the inconsistency only matters when dumping mo_graph to
2546 //Check whether reading these writes has made threads unable to
2548 for (unsigned int i = 0; i < actions_to_check.size(); i++) {
2549 ModelAction *read = actions_to_check[i];
2550 mo_check_promises(read, true);
2557 * Compute the set of promises that could potentially be satisfied by this
2558 * action. Note that the set computation actually appears in the Node, not in
2560 * @param curr The ModelAction that may satisfy promises
2562 void ModelChecker::compute_promises(ModelAction *curr)
2564 for (unsigned int i = 0; i < promises->size(); i++) {
2565 Promise *promise = (*promises)[i];
2566 if (!promise->is_compatible(curr) || !promise->same_value(curr))
2569 bool satisfy = true;
2570 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2571 const ModelAction *act = promise->get_reader(j);
2572 if (act->happens_before(curr) ||
2573 act->could_synchronize_with(curr)) {
2579 curr->get_node()->set_promise(i);
2583 /** Checks promises in response to change in ClockVector Threads. */
2584 void ModelChecker::check_promises(thread_id_t tid, ClockVector *old_cv, ClockVector *merge_cv)
2586 for (unsigned int i = 0; i < promises->size(); i++) {
2587 Promise *promise = (*promises)[i];
2588 if (!promise->thread_is_available(tid))
2590 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2591 const ModelAction *act = promise->get_reader(j);
2592 if ((!old_cv || !old_cv->synchronized_since(act)) &&
2593 merge_cv->synchronized_since(act)) {
2594 if (promise->eliminate_thread(tid)) {
2595 /* Promise has failed */
2596 priv->failed_promise = true;
2604 void ModelChecker::check_promises_thread_disabled()
2606 for (unsigned int i = 0; i < promises->size(); i++) {
2607 Promise *promise = (*promises)[i];
2608 if (promise->has_failed()) {
2609 priv->failed_promise = true;
2616 * @brief Checks promises in response to addition to modification order for
2619 * We test whether threads are still available for satisfying promises after an
2620 * addition to our modification order constraints. Those that are unavailable
2621 * are "eliminated". Once all threads are eliminated from satisfying a promise,
2622 * that promise has failed.
2624 * @param act The ModelAction which updated the modification order
2625 * @param is_read_check Should be true if act is a read and we must check for
2626 * updates to the store from which it read (there is a distinction here for
2627 * RMW's, which are both a load and a store)
2629 void ModelChecker::mo_check_promises(const ModelAction *act, bool is_read_check)
2631 const ModelAction *write = is_read_check ? act->get_reads_from() : act;
2633 for (unsigned int i = 0; i < promises->size(); i++) {
2634 Promise *promise = (*promises)[i];
2636 // Is this promise on the same location?
2637 if (!promise->same_location(write))
2640 for (unsigned int j = 0; j < promise->get_num_readers(); j++) {
2641 const ModelAction *pread = promise->get_reader(j);
2642 if (!pread->happens_before(act))
2644 if (mo_graph->checkPromise(write, promise)) {
2645 priv->failed_promise = true;
2651 // Don't do any lookups twice for the same thread
2652 if (!promise->thread_is_available(act->get_tid()))
2655 if (mo_graph->checkReachable(promise, write)) {
2656 if (mo_graph->checkPromise(write, promise)) {
2657 priv->failed_promise = true;
2665 * Compute the set of writes that may break the current pending release
2666 * sequence. This information is extracted from previou release sequence
2669 * @param curr The current ModelAction. Must be a release sequence fixup
2672 void ModelChecker::compute_relseq_breakwrites(ModelAction *curr)
2674 if (pending_rel_seqs->empty())
2677 struct release_seq *pending = pending_rel_seqs->back();
2678 for (unsigned int i = 0; i < pending->writes.size(); i++) {
2679 const ModelAction *write = pending->writes[i];
2680 curr->get_node()->add_relseq_break(write);
2683 /* NULL means don't break the sequence; just synchronize */
2684 curr->get_node()->add_relseq_break(NULL);
2688 * Build up an initial set of all past writes that this 'read' action may read
2689 * from, as well as any previously-observed future values that must still be valid.
2691 * @param curr is the current ModelAction that we are exploring; it must be a
2694 void ModelChecker::build_may_read_from(ModelAction *curr)
2696 std::vector<action_list_t> *thrd_lists = get_safe_ptr_vect_action(obj_thrd_map, curr->get_location());
2698 ASSERT(curr->is_read());
2700 ModelAction *last_sc_write = NULL;
2702 if (curr->is_seqcst())
2703 last_sc_write = get_last_seq_cst_write(curr);
2705 /* Iterate over all threads */
2706 for (i = 0; i < thrd_lists->size(); i++) {
2707 /* Iterate over actions in thread, starting from most recent */
2708 action_list_t *list = &(*thrd_lists)[i];
2709 action_list_t::reverse_iterator rit;
2710 for (rit = list->rbegin(); rit != list->rend(); rit++) {
2711 ModelAction *act = *rit;
2713 /* Only consider 'write' actions */
2714 if (!act->is_write() || act == curr)
2717 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
2718 bool allow_read = true;
2720 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
2722 else if (curr->get_sleep_flag() && !curr->is_seqcst() && !sleep_can_read_from(curr, act))
2726 /* Only add feasible reads */
2727 mo_graph->startChanges();
2728 r_modification_order(curr, act);
2729 if (!is_infeasible())
2730 curr->get_node()->add_read_from_past(act);
2731 mo_graph->rollbackChanges();
2734 /* Include at most one act per-thread that "happens before" curr */
2735 if (act->happens_before(curr))
2740 /* Inherit existing, promised future values */
2741 for (i = 0; i < promises->size(); i++) {
2742 const Promise *promise = (*promises)[i];
2743 const ModelAction *promise_read = promise->get_reader(0);
2744 if (promise_read->same_var(curr)) {
2745 /* Only add feasible future-values */
2746 mo_graph->startChanges();
2747 r_modification_order(curr, promise);
2748 if (!is_infeasible())
2749 curr->get_node()->add_read_from_promise(promise_read);
2750 mo_graph->rollbackChanges();
2754 /* We may find no valid may-read-from only if the execution is doomed */
2755 if (!curr->get_node()->read_from_size()) {
2756 priv->no_valid_reads = true;
2760 if (DBG_ENABLED()) {
2761 model_print("Reached read action:\n");
2763 model_print("Printing read_from_past\n");
2764 curr->get_node()->print_read_from_past();
2765 model_print("End printing read_from_past\n");
2769 bool ModelChecker::sleep_can_read_from(ModelAction *curr, const ModelAction *write)
2771 for ( ; write != NULL; write = write->get_reads_from()) {
2772 /* UNINIT actions don't have a Node, and they never sleep */
2773 if (write->is_uninitialized())
2775 Node *prevnode = write->get_node()->get_parent();
2777 bool thread_sleep = prevnode->enabled_status(curr->get_tid()) == THREAD_SLEEP_SET;
2778 if (write->is_release() && thread_sleep)
2780 if (!write->is_rmw())
2787 * @brief Create a new action representing an uninitialized atomic
2788 * @param location The memory location of the atomic object
2789 * @return A pointer to a new ModelAction
2791 ModelAction * ModelChecker::new_uninitialized_action(void *location) const
2793 ModelAction *act = (ModelAction *)snapshot_malloc(sizeof(class ModelAction));
2794 act = new (act) ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, location, 0, model_thread);
2795 act->create_cv(NULL);
2799 static void print_list(action_list_t *list)
2801 action_list_t::iterator it;
2803 model_print("---------------------------------------------------------------------\n");
2805 unsigned int hash = 0;
2807 for (it = list->begin(); it != list->end(); it++) {
2809 hash = hash^(hash<<3)^((*it)->hash());
2811 model_print("HASH %u\n", hash);
2812 model_print("---------------------------------------------------------------------\n");
2815 #if SUPPORT_MOD_ORDER_DUMP
2816 void ModelChecker::dumpGraph(char *filename) const
2819 sprintf(buffer, "%s.dot", filename);
2820 FILE *file = fopen(buffer, "w");
2821 fprintf(file, "digraph %s {\n", filename);
2822 mo_graph->dumpNodes(file);
2823 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
2825 for (action_list_t::iterator it = action_trace->begin(); it != action_trace->end(); it++) {
2826 ModelAction *act = *it;
2827 if (act->is_read()) {
2828 mo_graph->dot_print_node(file, act);
2829 if (act->get_reads_from())
2830 mo_graph->dot_print_edge(file,
2831 act->get_reads_from(),
2833 "label=\"rf\", color=red, weight=2");
2835 mo_graph->dot_print_edge(file,
2836 act->get_reads_from_promise(),
2838 "label=\"rf\", color=red");
2840 if (thread_array[act->get_tid()]) {
2841 mo_graph->dot_print_edge(file,
2842 thread_array[id_to_int(act->get_tid())],
2844 "label=\"sb\", color=blue, weight=400");
2847 thread_array[act->get_tid()] = act;
2849 fprintf(file, "}\n");
2850 model_free(thread_array);
2855 /** @brief Prints an execution trace summary. */
2856 void ModelChecker::print_summary() const
2858 #if SUPPORT_MOD_ORDER_DUMP
2859 char buffername[100];
2860 sprintf(buffername, "exec%04u", stats.num_total);
2861 mo_graph->dumpGraphToFile(buffername);
2862 sprintf(buffername, "graph%04u", stats.num_total);
2863 dumpGraph(buffername);
2866 model_print("Execution %d:", stats.num_total);
2867 if (isfeasibleprefix()) {
2868 if (scheduler->all_threads_sleeping())
2869 model_print(" SLEEP-SET REDUNDANT");
2872 print_infeasibility(" INFEASIBLE");
2873 print_list(action_trace);
2878 * Add a Thread to the system for the first time. Should only be called once
2880 * @param t The Thread to add
2882 void ModelChecker::add_thread(Thread *t)
2884 thread_map->put(id_to_int(t->get_id()), t);
2885 scheduler->add_thread(t);
2889 * Removes a thread from the scheduler.
2890 * @param the thread to remove.
2892 void ModelChecker::remove_thread(Thread *t)
2894 scheduler->remove_thread(t);
2898 * @brief Get a Thread reference by its ID
2899 * @param tid The Thread's ID
2900 * @return A Thread reference
2902 Thread * ModelChecker::get_thread(thread_id_t tid) const
2904 return thread_map->get(id_to_int(tid));
2908 * @brief Get a reference to the Thread in which a ModelAction was executed
2909 * @param act The ModelAction
2910 * @return A Thread reference
2912 Thread * ModelChecker::get_thread(const ModelAction *act) const
2914 return get_thread(act->get_tid());
2918 * @brief Get a Promise's "promise number"
2920 * A "promise number" is an index number that is unique to a promise, valid
2921 * only for a specific snapshot of an execution trace. Promises may come and go
2922 * as they are generated an resolved, so an index only retains meaning for the
2925 * @param promise The Promise to check
2926 * @return The promise index, if the promise still is valid; otherwise -1
2928 int ModelChecker::get_promise_number(const Promise *promise) const
2930 for (unsigned int i = 0; i < promises->size(); i++)
2931 if ((*promises)[i] == promise)
2938 * @brief Check if a Thread is currently enabled
2939 * @param t The Thread to check
2940 * @return True if the Thread is currently enabled
2942 bool ModelChecker::is_enabled(Thread *t) const
2944 return scheduler->is_enabled(t);
2948 * @brief Check if a Thread is currently enabled
2949 * @param tid The ID of the Thread to check
2950 * @return True if the Thread is currently enabled
2952 bool ModelChecker::is_enabled(thread_id_t tid) const
2954 return scheduler->is_enabled(tid);
2958 * Switch from a model-checker context to a user-thread context. This is the
2959 * complement of ModelChecker::switch_to_master and must be called from the
2960 * model-checker context
2962 * @param thread The user-thread to switch to
2964 void ModelChecker::switch_from_master(Thread *thread)
2966 scheduler->set_current_thread(thread);
2967 Thread::swap(&system_context, thread);
2971 * Switch from a user-context to the "master thread" context (a.k.a. system
2972 * context). This switch is made with the intention of exploring a particular
2973 * model-checking action (described by a ModelAction object). Must be called
2974 * from a user-thread context.
2976 * @param act The current action that will be explored. May be NULL only if
2977 * trace is exiting via an assertion (see ModelChecker::set_assert and
2978 * ModelChecker::has_asserted).
2979 * @return Return the value returned by the current action
2981 uint64_t ModelChecker::switch_to_master(ModelAction *act)
2984 Thread *old = thread_current();
2985 ASSERT(!old->get_pending());
2986 old->set_pending(act);
2987 if (Thread::swap(old, &system_context) < 0) {
2988 perror("swap threads");
2991 return old->get_return_value();
2995 * Takes the next step in the execution, if possible.
2996 * @param curr The current step to take
2997 * @return Returns the next Thread to run, if any; NULL if this execution
3000 Thread * ModelChecker::take_step(ModelAction *curr)
3002 Thread *curr_thrd = get_thread(curr);
3003 ASSERT(curr_thrd->get_state() == THREAD_READY);
3005 curr = check_current_action(curr);
3007 /* Infeasible -> don't take any more steps */
3008 if (is_infeasible())
3010 else if (isfeasibleprefix() && have_bug_reports()) {
3015 if (params.bound != 0 && priv->used_sequence_numbers > params.bound)
3018 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
3019 scheduler->remove_thread(curr_thrd);
3021 Thread *next_thrd = get_next_thread(curr);
3023 DEBUG("(%d, %d)\n", curr_thrd ? id_to_int(curr_thrd->get_id()) : -1,
3024 next_thrd ? id_to_int(next_thrd->get_id()) : -1);
3029 /** Wrapper to run the user's main function, with appropriate arguments */
3030 void user_main_wrapper(void *)
3032 user_main(model->params.argc, model->params.argv);
3035 /** @brief Run ModelChecker for the user program */
3036 void ModelChecker::run()
3040 Thread *t = new Thread(&user_thread, &user_main_wrapper, NULL, NULL);
3045 * Stash next pending action(s) for thread(s). There
3046 * should only need to stash one thread's action--the
3047 * thread which just took a step--plus the first step
3048 * for any newly-created thread
3050 for (unsigned int i = 0; i < get_num_threads(); i++) {
3051 thread_id_t tid = int_to_id(i);
3052 Thread *thr = get_thread(tid);
3053 if (!thr->is_model_thread() && !thr->is_complete() && !thr->get_pending()) {
3054 switch_from_master(thr);
3055 if (check_deadlock(thr))
3056 assert_bug("Deadlock detected");
3060 /* Catch assertions from prior take_step or from
3061 * between-ModelAction bugs (e.g., data races) */
3065 /* Consume the next action for a Thread */
3066 ModelAction *curr = t->get_pending();
3067 t->set_pending(NULL);
3068 t = take_step(curr);
3069 } while (t && !t->is_model_thread());
3072 * Launch end-of-execution release sequence fixups only when
3073 * the execution is otherwise feasible AND there are:
3075 * (1) pending release sequences
3076 * (2) pending assertions that could be invalidated by a change
3077 * in clock vectors (i.e., data races)
3078 * (3) no pending promises
3080 while (!pending_rel_seqs->empty() &&
3081 is_feasible_prefix_ignore_relseq() &&
3082 !unrealizedraces.empty()) {
3083 model_print("*** WARNING: release sequence fixup action "
3084 "(%zu pending release seuqence(s)) ***\n",
3085 pending_rel_seqs->size());
3086 ModelAction *fixup = new ModelAction(MODEL_FIXUP_RELSEQ,
3087 std::memory_order_seq_cst, NULL, VALUE_NONE,
3091 } while (next_execution());
3093 model_print("******* Model-checking complete: *******\n");