11 #include "clockvector.h"
12 #include "cyclegraph.h"
14 #include "threads-model.h"
15 #include "bugmessage.h"
18 #define INITIAL_THREAD_ID 0
21 * Structure for holding small ModelChecker members that should be snapshotted
23 struct model_snapshot_members {
24 model_snapshot_members() :
25 /* First thread created will have id INITIAL_THREAD_ID */
26 next_thread_id(INITIAL_THREAD_ID),
27 used_sequence_numbers(0),
29 bad_synchronization(false),
33 ~model_snapshot_members() {
34 for (unsigned int i = 0;i < bugs.size();i++)
39 unsigned int next_thread_id;
40 modelclock_t used_sequence_numbers;
41 SnapVector<bug_message *> bugs;
42 /** @brief Incorrectly-ordered synchronization was made */
43 bool bad_synchronization;
49 /** @brief Constructor */
50 ModelExecution::ModelExecution(ModelChecker *m, Scheduler *scheduler) :
55 thread_map(2), /* We'll always need at least 2 threads */
59 condvar_waiters_map(),
63 thrd_last_fence_release(),
64 priv(new struct model_snapshot_members ()),
65 mo_graph(new CycleGraph()),
68 /* Initialize a model-checker thread, for special ModelActions */
69 model_thread = new Thread(get_next_id());
70 add_thread(model_thread);
71 scheduler->register_engine(this);
74 /** @brief Destructor */
75 ModelExecution::~ModelExecution()
77 for (unsigned int i = 0;i < get_num_threads();i++)
78 delete get_thread(int_to_id(i));
84 int ModelExecution::get_execution_number() const
86 return model->get_execution_number();
89 static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t *, uintptr_t, 4> * hash, void * ptr)
91 action_list_t *tmp = hash->get(ptr);
93 tmp = new action_list_t();
99 static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<const void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
101 SnapVector<action_list_t> *tmp = hash->get(ptr);
103 tmp = new SnapVector<action_list_t>();
109 /** @return a thread ID for a new Thread */
110 thread_id_t ModelExecution::get_next_id()
112 return priv->next_thread_id++;
115 /** @return the number of user threads created during this execution */
116 unsigned int ModelExecution::get_num_threads() const
118 return priv->next_thread_id;
121 /** @return a sequence number for a new ModelAction */
122 modelclock_t ModelExecution::get_next_seq_num()
124 return ++priv->used_sequence_numbers;
128 * @brief Should the current action wake up a given thread?
130 * @param curr The current action
131 * @param thread The thread that we might wake up
132 * @return True, if we should wake up the sleeping thread; false otherwise
134 bool ModelExecution::should_wake_up(const ModelAction *curr, const Thread *thread) const
136 const ModelAction *asleep = thread->get_pending();
137 /* Don't allow partial RMW to wake anyone up */
140 /* Synchronizing actions may have been backtracked */
141 if (asleep->could_synchronize_with(curr))
143 /* All acquire/release fences and fence-acquire/store-release */
144 if (asleep->is_fence() && asleep->is_acquire() && curr->is_release())
146 /* Fence-release + store can awake load-acquire on the same location */
147 if (asleep->is_read() && asleep->is_acquire() && curr->same_var(asleep) && curr->is_write()) {
148 ModelAction *fence_release = get_last_fence_release(curr->get_tid());
149 if (fence_release && *(get_last_action(thread->get_id())) < *fence_release)
155 void ModelExecution::wake_up_sleeping_actions(ModelAction *curr)
157 for (unsigned int i = 0;i < get_num_threads();i++) {
158 Thread *thr = get_thread(int_to_id(i));
159 if (scheduler->is_sleep_set(thr)) {
160 if (should_wake_up(curr, thr))
161 /* Remove this thread from sleep set */
162 scheduler->remove_sleep(thr);
167 /** @brief Alert the model-checker that an incorrectly-ordered
168 * synchronization was made */
169 void ModelExecution::set_bad_synchronization()
171 priv->bad_synchronization = true;
174 bool ModelExecution::assert_bug(const char *msg)
176 priv->bugs.push_back(new bug_message(msg));
178 if (isfeasibleprefix()) {
185 /** @return True, if any bugs have been reported for this execution */
186 bool ModelExecution::have_bug_reports() const
188 return priv->bugs.size() != 0;
191 SnapVector<bug_message *> * ModelExecution::get_bugs() const
197 * Check whether the current trace has triggered an assertion which should halt
200 * @return True, if the execution should be aborted; false otherwise
202 bool ModelExecution::has_asserted() const
204 return priv->asserted;
208 * Trigger a trace assertion which should cause this execution to be halted.
209 * This can be due to a detected bug or due to an infeasibility that should
212 void ModelExecution::set_assert()
214 priv->asserted = true;
218 * Check if we are in a deadlock. Should only be called at the end of an
219 * execution, although it should not give false positives in the middle of an
220 * execution (there should be some ENABLED thread).
222 * @return True if program is in a deadlock; false otherwise
224 bool ModelExecution::is_deadlocked() const
226 bool blocking_threads = false;
227 for (unsigned int i = 0;i < get_num_threads();i++) {
228 thread_id_t tid = int_to_id(i);
231 Thread *t = get_thread(tid);
232 if (!t->is_model_thread() && t->get_pending())
233 blocking_threads = true;
235 return blocking_threads;
239 * Check if this is a complete execution. That is, have all thread completed
240 * execution (rather than exiting because sleep sets have forced a redundant
243 * @return True if the execution is complete.
245 bool ModelExecution::is_complete_execution() const
247 for (unsigned int i = 0;i < get_num_threads();i++)
248 if (is_enabled(int_to_id(i)))
255 * Processes a read model action.
256 * @param curr is the read model action to process.
257 * @param rf_set is the set of model actions we can possibly read from
258 * @return True if processing this read updates the mo_graph.
260 void ModelExecution::process_read(ModelAction *curr, SnapVector<ModelAction *> * rf_set)
262 SnapVector<const ModelAction *> * priorset = new SnapVector<const ModelAction *>();
265 int index = fuzzer->selectWrite(curr, rf_set);
266 ModelAction *rf = (*rf_set)[index];
270 bool canprune = false;
271 if (r_modification_order(curr, rf, priorset, &canprune)) {
272 for(unsigned int i=0;i<priorset->size();i++) {
273 mo_graph->addEdge((*priorset)[i], rf);
276 get_thread(curr)->set_return_value(curr->get_return_value());
278 if (canprune && curr->get_type() == ATOMIC_READ) {
279 int tid = id_to_int(curr->get_tid());
280 (*obj_thrd_map.get(curr->get_location()))[tid].pop_back();
285 (*rf_set)[index] = rf_set->back();
291 * Processes a lock, trylock, or unlock model action. @param curr is
292 * the read model action to process.
294 * The try lock operation checks whether the lock is taken. If not,
295 * it falls to the normal lock operation case. If so, it returns
298 * The lock operation has already been checked that it is enabled, so
299 * it just grabs the lock and synchronizes with the previous unlock.
301 * The unlock operation has to re-enable all of the threads that are
302 * waiting on the lock.
304 * @return True if synchronization was updated; false otherwise
306 bool ModelExecution::process_mutex(ModelAction *curr)
308 cdsc::mutex *mutex = curr->get_mutex();
309 struct cdsc::mutex_state *state = NULL;
312 state = mutex->get_state();
314 switch (curr->get_type()) {
315 case ATOMIC_TRYLOCK: {
316 bool success = !state->locked;
317 curr->set_try_lock(success);
319 get_thread(curr)->set_return_value(0);
322 get_thread(curr)->set_return_value(1);
324 //otherwise fall into the lock case
326 if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
327 assert_bug("Lock access before initialization");
328 state->locked = get_thread(curr);
329 ModelAction *unlock = get_last_unlock(curr);
330 //synchronize with the previous unlock statement
331 if (unlock != NULL) {
332 synchronize(unlock, curr);
338 case ATOMIC_UNLOCK: {
339 //TODO: FIX WAIT SITUATION...WAITS CAN SPURIOUSLY FAIL...TIMED WAITS SHOULD PROBABLY JUST BE THE SAME AS NORMAL WAITS...THINK ABOUT PROBABILITIES THOUGH....AS IN TIMED WAIT MUST FAIL TO GUARANTEE PROGRESS...NORMAL WAIT MAY FAIL...SO NEED NORMAL WAIT TO WORK CORRECTLY IN THE CASE IT SPURIOUSLY FAILS AND IN THE CASE IT DOESN'T... TIMED WAITS MUST EVENMTUALLY RELEASE...
341 /* wake up the other threads */
342 for (unsigned int i = 0;i < get_num_threads();i++) {
343 Thread *t = get_thread(int_to_id(i));
344 Thread *curr_thrd = get_thread(curr);
345 if (t->waiting_on() == curr_thrd && t->get_pending()->is_lock())
349 /* unlock the lock - after checking who was waiting on it */
350 state->locked = NULL;
352 if (!curr->is_wait())
353 break;/* The rest is only for ATOMIC_WAIT */
357 case ATOMIC_NOTIFY_ALL: {
358 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
359 //activate all the waiting threads
360 for (action_list_t::iterator rit = waiters->begin();rit != waiters->end();rit++) {
361 scheduler->wake(get_thread(*rit));
366 case ATOMIC_NOTIFY_ONE: {
367 action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
368 if (waiters->size() != 0) {
369 Thread * thread = fuzzer->selectNotify(waiters);
370 scheduler->wake(thread);
382 * Process a write ModelAction
383 * @param curr The ModelAction to process
384 * @return True if the mo_graph was updated or promises were resolved
386 void ModelExecution::process_write(ModelAction *curr)
389 w_modification_order(curr);
392 get_thread(curr)->set_return_value(VALUE_NONE);
396 * Process a fence ModelAction
397 * @param curr The ModelAction to process
398 * @return True if synchronization was updated
400 bool ModelExecution::process_fence(ModelAction *curr)
403 * fence-relaxed: no-op
404 * fence-release: only log the occurence (not in this function), for
405 * use in later synchronization
406 * fence-acquire (this function): search for hypothetical release
408 * fence-seq-cst: MO constraints formed in {r,w}_modification_order
410 bool updated = false;
411 if (curr->is_acquire()) {
412 action_list_t *list = &action_trace;
413 action_list_t::reverse_iterator rit;
414 /* Find X : is_read(X) && X --sb-> curr */
415 for (rit = list->rbegin();rit != list->rend();rit++) {
416 ModelAction *act = *rit;
419 if (act->get_tid() != curr->get_tid())
421 /* Stop at the beginning of the thread */
422 if (act->is_thread_start())
424 /* Stop once we reach a prior fence-acquire */
425 if (act->is_fence() && act->is_acquire())
429 /* read-acquire will find its own release sequences */
430 if (act->is_acquire())
433 /* Establish hypothetical release sequences */
434 ClockVector *cv = get_hb_from_write(act);
435 if (curr->get_cv()->merge(cv))
443 * @brief Process the current action for thread-related activity
445 * Performs current-action processing for a THREAD_* ModelAction. Proccesses
446 * may include setting Thread status, completing THREAD_FINISH/THREAD_JOIN
447 * synchronization, etc. This function is a no-op for non-THREAD actions
448 * (e.g., ATOMIC_{READ,WRITE,RMW,LOCK}, etc.)
450 * @param curr The current action
451 * @return True if synchronization was updated or a thread completed
453 bool ModelExecution::process_thread_action(ModelAction *curr)
455 bool updated = false;
457 switch (curr->get_type()) {
458 case THREAD_CREATE: {
459 thrd_t *thrd = (thrd_t *)curr->get_location();
460 struct thread_params *params = (struct thread_params *)curr->get_value();
461 Thread *th = new Thread(get_next_id(), thrd, params->func, params->arg, get_thread(curr));
462 curr->set_thread_operand(th);
464 th->set_creation(curr);
467 case PTHREAD_CREATE: {
468 (*(uint32_t *)curr->get_location()) = pthread_counter++;
470 struct pthread_params *params = (struct pthread_params *)curr->get_value();
471 Thread *th = new Thread(get_next_id(), NULL, params->func, params->arg, get_thread(curr));
472 curr->set_thread_operand(th);
474 th->set_creation(curr);
476 if ( pthread_map.size() < pthread_counter )
477 pthread_map.resize( pthread_counter );
478 pthread_map[ pthread_counter-1 ] = th;
483 Thread *blocking = curr->get_thread_operand();
484 ModelAction *act = get_last_action(blocking->get_id());
485 synchronize(act, curr);
486 updated = true; /* trigger rel-seq checks */
490 Thread *blocking = curr->get_thread_operand();
491 ModelAction *act = get_last_action(blocking->get_id());
492 synchronize(act, curr);
493 updated = true; /* trigger rel-seq checks */
494 break; // WL: to be add (modified)
497 case THREAD_FINISH: {
498 Thread *th = get_thread(curr);
499 /* Wake up any joining threads */
500 for (unsigned int i = 0;i < get_num_threads();i++) {
501 Thread *waiting = get_thread(int_to_id(i));
502 if (waiting->waiting_on() == th &&
503 waiting->get_pending()->is_thread_join())
504 scheduler->wake(waiting);
507 updated = true; /* trigger rel-seq checks */
521 * Initialize the current action by performing one or more of the following
522 * actions, as appropriate: merging RMWR and RMWC/RMW actions,
523 * manipulating backtracking sets, allocating and
524 * initializing clock vectors, and computing the promises to fulfill.
526 * @param curr The current action, as passed from the user context; may be
527 * freed/invalidated after the execution of this function, with a different
528 * action "returned" its place (pass-by-reference)
529 * @return True if curr is a newly-explored action; false otherwise
531 bool ModelExecution::initialize_curr_action(ModelAction **curr)
533 if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
534 ModelAction *newcurr = process_rmw(*curr);
540 ModelAction *newcurr = *curr;
542 newcurr->set_seq_number(get_next_seq_num());
543 /* Always compute new clock vector */
544 newcurr->create_cv(get_parent_action(newcurr->get_tid()));
546 /* Assign most recent release fence */
547 newcurr->set_last_fence_release(get_last_fence_release(newcurr->get_tid()));
549 return true; /* This was a new ModelAction */
554 * @brief Establish reads-from relation between two actions
556 * Perform basic operations involved with establishing a concrete rf relation,
557 * including setting the ModelAction data and checking for release sequences.
559 * @param act The action that is reading (must be a read)
560 * @param rf The action from which we are reading (must be a write)
562 * @return True if this read established synchronization
565 void ModelExecution::read_from(ModelAction *act, ModelAction *rf)
568 ASSERT(rf->is_write());
570 act->set_read_from(rf);
571 if (act->is_acquire()) {
572 ClockVector *cv = get_hb_from_write(rf);
575 act->get_cv()->merge(cv);
580 * @brief Synchronizes two actions
582 * When A synchronizes with B (or A --sw-> B), B inherits A's clock vector.
583 * This function performs the synchronization as well as providing other hooks
584 * for other checks along with synchronization.
586 * @param first The left-hand side of the synchronizes-with relation
587 * @param second The right-hand side of the synchronizes-with relation
588 * @return True if the synchronization was successful (i.e., was consistent
589 * with the execution order); false otherwise
591 bool ModelExecution::synchronize(const ModelAction *first, ModelAction *second)
593 if (*second < *first) {
594 set_bad_synchronization();
597 return second->synchronize_with(first);
601 * @brief Check whether a model action is enabled.
603 * Checks whether an operation would be successful (i.e., is a lock already
604 * locked, or is the joined thread already complete).
606 * For yield-blocking, yields are never enabled.
608 * @param curr is the ModelAction to check whether it is enabled.
609 * @return a bool that indicates whether the action is enabled.
611 bool ModelExecution::check_action_enabled(ModelAction *curr) {
612 if (curr->is_lock()) {
613 cdsc::mutex *lock = curr->get_mutex();
614 struct cdsc::mutex_state *state = lock->get_state();
617 } else if (curr->is_thread_join()) {
618 Thread *blocking = curr->get_thread_operand();
619 if (!blocking->is_complete()) {
628 * This is the heart of the model checker routine. It performs model-checking
629 * actions corresponding to a given "current action." Among other processes, it
630 * calculates reads-from relationships, updates synchronization clock vectors,
631 * forms a memory_order constraints graph, and handles replay/backtrack
632 * execution when running permutations of previously-observed executions.
634 * @param curr The current action to process
635 * @return The ModelAction that is actually executed; may be different than
638 ModelAction * ModelExecution::check_current_action(ModelAction *curr)
641 bool second_part_of_rmw = curr->is_rmwc() || curr->is_rmw();
642 bool newly_explored = initialize_curr_action(&curr);
646 wake_up_sleeping_actions(curr);
648 /* Add the action to lists before any other model-checking tasks */
649 if (!second_part_of_rmw && curr->get_type() != NOOP)
650 add_action_to_lists(curr);
652 SnapVector<ModelAction *> * rf_set = NULL;
653 /* Build may_read_from set for newly-created actions */
654 if (newly_explored && curr->is_read())
655 rf_set = build_may_read_from(curr);
657 process_thread_action(curr);
659 if (curr->is_read() && !second_part_of_rmw) {
660 process_read(curr, rf_set);
663 ASSERT(rf_set == NULL);
666 if (curr->is_write())
669 if (curr->is_fence())
672 if (curr->is_mutex_op())
679 * This is the strongest feasibility check available.
680 * @return whether the current trace (partial or complete) must be a prefix of
683 bool ModelExecution::isfeasibleprefix() const
685 return !is_infeasible();
689 * Print disagnostic information about an infeasible execution
690 * @param prefix A string to prefix the output with; if NULL, then a default
691 * message prefix will be provided
693 void ModelExecution::print_infeasibility(const char *prefix) const
697 if (priv->bad_synchronization)
698 ptr += sprintf(ptr, "[bad sw ordering]");
700 model_print("%s: %s", prefix ? prefix : "Infeasible", buf);
704 * Check if the current partial trace is infeasible. Does not check any
705 * end-of-execution flags, which might rule out the execution. Thus, this is
706 * useful only for ruling an execution as infeasible.
707 * @return whether the current partial trace is infeasible.
709 bool ModelExecution::is_infeasible() const
711 return priv->bad_synchronization;
714 /** Close out a RMWR by converting previous RMWR into a RMW or READ. */
715 ModelAction * ModelExecution::process_rmw(ModelAction *act) {
716 ModelAction *lastread = get_last_action(act->get_tid());
717 lastread->process_rmw(act);
719 mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
725 * @brief Updates the mo_graph with the constraints imposed from the current
728 * Basic idea is the following: Go through each other thread and find
729 * the last action that happened before our read. Two cases:
731 * -# The action is a write: that write must either occur before
732 * the write we read from or be the write we read from.
733 * -# The action is a read: the write that that action read from
734 * must occur before the write we read from or be the same write.
736 * @param curr The current action. Must be a read.
737 * @param rf The ModelAction or Promise that curr reads from. Must be a write.
738 * @return True if modification order edges were added; false otherwise
741 bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *rf, SnapVector<const ModelAction *> * priorset, bool * canprune)
743 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
745 ASSERT(curr->is_read());
747 /* Last SC fence in the current thread */
748 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
750 int tid = curr->get_tid();
751 ModelAction *prev_same_thread = NULL;
752 /* Iterate over all threads */
753 for (i = 0;i < thrd_lists->size();i++, tid = (((unsigned int)(tid+1)) == thrd_lists->size()) ? 0 : tid + 1) {
754 /* Last SC fence in thread tid */
755 ModelAction *last_sc_fence_thread_local = NULL;
757 last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(tid), NULL);
759 /* Last SC fence in thread tid, before last SC fence in current thread */
760 ModelAction *last_sc_fence_thread_before = NULL;
761 if (last_sc_fence_local)
762 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(tid), last_sc_fence_local);
764 //Only need to iterate if either hb has changed for thread in question or SC fence after last operation...
765 if (prev_same_thread != NULL &&
766 (prev_same_thread->get_cv()->getClock(tid) == curr->get_cv()->getClock(tid)) &&
767 (last_sc_fence_thread_local == NULL || *last_sc_fence_thread_local < *prev_same_thread)) {
771 /* Iterate over actions in thread, starting from most recent */
772 action_list_t *list = &(*thrd_lists)[tid];
773 action_list_t::reverse_iterator rit;
774 for (rit = list->rbegin();rit != list->rend();rit++) {
775 ModelAction *act = *rit;
780 /* Don't want to add reflexive edges on 'rf' */
781 if (act->equals(rf)) {
782 if (act->happens_before(curr))
788 if (act->is_write()) {
789 /* C++, Section 29.3 statement 5 */
790 if (curr->is_seqcst() && last_sc_fence_thread_local &&
791 *act < *last_sc_fence_thread_local) {
792 if (mo_graph->checkReachable(rf, act))
794 priorset->push_back(act);
797 /* C++, Section 29.3 statement 4 */
798 else if (act->is_seqcst() && last_sc_fence_local &&
799 *act < *last_sc_fence_local) {
800 if (mo_graph->checkReachable(rf, act))
802 priorset->push_back(act);
805 /* C++, Section 29.3 statement 6 */
806 else if (last_sc_fence_thread_before &&
807 *act < *last_sc_fence_thread_before) {
808 if (mo_graph->checkReachable(rf, act))
810 priorset->push_back(act);
816 * Include at most one act per-thread that "happens
819 if (act->happens_before(curr)) {
821 if (last_sc_fence_local == NULL ||
822 (*last_sc_fence_local < *prev_same_thread)) {
823 prev_same_thread = act;
826 if (act->is_write()) {
827 if (mo_graph->checkReachable(rf, act))
829 priorset->push_back(act);
831 const ModelAction *prevrf = act->get_reads_from();
832 if (!prevrf->equals(rf)) {
833 if (mo_graph->checkReachable(rf, prevrf))
835 priorset->push_back(prevrf);
837 if (act->get_tid() == curr->get_tid()) {
838 //Can prune curr from obj list
851 * Updates the mo_graph with the constraints imposed from the current write.
853 * Basic idea is the following: Go through each other thread and find
854 * the lastest action that happened before our write. Two cases:
856 * (1) The action is a write => that write must occur before
859 * (2) The action is a read => the write that that action read from
860 * must occur before the current write.
862 * This method also handles two other issues:
864 * (I) Sequential Consistency: Making sure that if the current write is
865 * seq_cst, that it occurs after the previous seq_cst write.
867 * (II) Sending the write back to non-synchronizing reads.
869 * @param curr The current action. Must be a write.
870 * @param send_fv A vector for stashing reads to which we may pass our future
871 * value. If NULL, then don't record any future values.
872 * @return True if modification order edges were added; false otherwise
874 void ModelExecution::w_modification_order(ModelAction *curr)
876 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
878 ASSERT(curr->is_write());
880 if (curr->is_seqcst()) {
881 /* We have to at least see the last sequentially consistent write,
882 so we are initialized. */
883 ModelAction *last_seq_cst = get_last_seq_cst_write(curr);
884 if (last_seq_cst != NULL) {
885 mo_graph->addEdge(last_seq_cst, curr);
889 /* Last SC fence in the current thread */
890 ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
892 /* Iterate over all threads */
893 for (i = 0;i < thrd_lists->size();i++) {
894 /* Last SC fence in thread i, before last SC fence in current thread */
895 ModelAction *last_sc_fence_thread_before = NULL;
896 if (last_sc_fence_local && int_to_id((int)i) != curr->get_tid())
897 last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
899 /* Iterate over actions in thread, starting from most recent */
900 action_list_t *list = &(*thrd_lists)[i];
901 action_list_t::reverse_iterator rit;
902 bool force_edge = false;
903 for (rit = list->rbegin();rit != list->rend();rit++) {
904 ModelAction *act = *rit;
907 * 1) If RMW and it actually read from something, then we
908 * already have all relevant edges, so just skip to next
911 * 2) If RMW and it didn't read from anything, we should
912 * whatever edge we can get to speed up convergence.
914 * 3) If normal write, we need to look at earlier actions, so
915 * continue processing list.
918 if (curr->is_rmw()) {
919 if (curr->get_reads_from() != NULL)
927 /* C++, Section 29.3 statement 7 */
928 if (last_sc_fence_thread_before && act->is_write() &&
929 *act < *last_sc_fence_thread_before) {
930 mo_graph->addEdge(act, curr, force_edge);
935 * Include at most one act per-thread that "happens
938 if (act->happens_before(curr)) {
940 * Note: if act is RMW, just add edge:
942 * The following edge should be handled elsewhere:
943 * readfrom(act) --mo--> act
946 mo_graph->addEdge(act, curr, force_edge);
947 else if (act->is_read()) {
948 //if previous read accessed a null, just keep going
949 mo_graph->addEdge(act->get_reads_from(), curr, force_edge);
958 * Arbitrary reads from the future are not allowed. Section 29.3 part 9 places
959 * some constraints. This method checks one the following constraint (others
960 * require compiler support):
962 * If X --hb-> Y --mo-> Z, then X should not read from Z.
963 * If X --hb-> Y, A --rf-> Y, and A --mo-> Z, then X should not read from Z.
965 bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *reader)
967 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(reader->get_location());
969 /* Iterate over all threads */
970 for (i = 0;i < thrd_lists->size();i++) {
971 const ModelAction *write_after_read = NULL;
973 /* Iterate over actions in thread, starting from most recent */
974 action_list_t *list = &(*thrd_lists)[i];
975 action_list_t::reverse_iterator rit;
976 for (rit = list->rbegin();rit != list->rend();rit++) {
977 ModelAction *act = *rit;
979 /* Don't disallow due to act == reader */
980 if (!reader->happens_before(act) || reader == act)
982 else if (act->is_write())
983 write_after_read = act;
984 else if (act->is_read() && act->get_reads_from() != NULL)
985 write_after_read = act->get_reads_from();
988 if (write_after_read && write_after_read != writer && mo_graph->checkReachable(write_after_read, writer))
995 * Computes the clock vector that happens before propagates from this write.
997 * @param rf The action that might be part of a release sequence. Must be a
999 * @return ClockVector of happens before relation.
1002 ClockVector * ModelExecution::get_hb_from_write(ModelAction *rf) const {
1003 SnapVector<ModelAction *> * processset = NULL;
1004 for ( ;rf != NULL;rf = rf->get_reads_from()) {
1005 ASSERT(rf->is_write());
1006 if (!rf->is_rmw() || (rf->is_acquire() && rf->is_release()) || rf->get_rfcv() != NULL)
1008 if (processset == NULL)
1009 processset = new SnapVector<ModelAction *>();
1010 processset->push_back(rf);
1013 int i = (processset == NULL) ? 0 : processset->size();
1015 ClockVector * vec = NULL;
1017 if (rf->get_rfcv() != NULL) {
1018 vec = rf->get_rfcv();
1019 } else if (rf->is_acquire() && rf->is_release()) {
1021 } else if (rf->is_release() && !rf->is_rmw()) {
1023 } else if (rf->is_release()) {
1024 //have rmw that is release and doesn't have a rfcv
1025 (vec = new ClockVector(vec, NULL))->merge(rf->get_cv());
1028 //operation that isn't release
1029 if (rf->get_last_fence_release()) {
1031 vec = rf->get_last_fence_release()->get_cv();
1033 (vec=new ClockVector(vec, NULL))->merge(rf->get_last_fence_release()->get_cv());
1039 rf = (*processset)[i];
1043 if (processset != NULL)
1049 * Performs various bookkeeping operations for the current ModelAction. For
1050 * instance, adds action to the per-object, per-thread action vector and to the
1051 * action trace list of all thread actions.
1053 * @param act is the ModelAction to add.
1055 void ModelExecution::add_action_to_lists(ModelAction *act)
1057 int tid = id_to_int(act->get_tid());
1058 ModelAction *uninit = NULL;
1060 action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
1061 if (list->empty() && act->is_atomic_var()) {
1062 uninit = get_uninitialized_action(act);
1063 uninit_id = id_to_int(uninit->get_tid());
1064 list->push_front(uninit);
1066 list->push_back(act);
1068 // Update action trace, a total order of all actions
1069 action_trace.push_back(act);
1071 action_trace.push_front(uninit);
1073 // Update obj_thrd_map, a per location, per thread, order of actions
1074 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
1075 if (tid >= (int)vec->size())
1076 vec->resize(priv->next_thread_id);
1077 (*vec)[tid].push_back(act);
1079 (*vec)[uninit_id].push_front(uninit);
1081 // Update thrd_last_action, the last action taken by each thrad
1082 if ((int)thrd_last_action.size() <= tid)
1083 thrd_last_action.resize(get_num_threads());
1084 thrd_last_action[tid] = act;
1086 thrd_last_action[uninit_id] = uninit;
1088 // Update thrd_last_fence_release, the last release fence taken by each thread
1089 if (act->is_fence() && act->is_release()) {
1090 if ((int)thrd_last_fence_release.size() <= tid)
1091 thrd_last_fence_release.resize(get_num_threads());
1092 thrd_last_fence_release[tid] = act;
1095 if (act->is_wait()) {
1096 void *mutex_loc = (void *) act->get_value();
1097 get_safe_ptr_action(&obj_map, mutex_loc)->push_back(act);
1099 SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, mutex_loc);
1100 if (tid >= (int)vec->size())
1101 vec->resize(priv->next_thread_id);
1102 (*vec)[tid].push_back(act);
1107 * @brief Get the last action performed by a particular Thread
1108 * @param tid The thread ID of the Thread in question
1109 * @return The last action in the thread
1111 ModelAction * ModelExecution::get_last_action(thread_id_t tid) const
1113 int threadid = id_to_int(tid);
1114 if (threadid < (int)thrd_last_action.size())
1115 return thrd_last_action[id_to_int(tid)];
1121 * @brief Get the last fence release performed by a particular Thread
1122 * @param tid The thread ID of the Thread in question
1123 * @return The last fence release in the thread, if one exists; NULL otherwise
1125 ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
1127 int threadid = id_to_int(tid);
1128 if (threadid < (int)thrd_last_fence_release.size())
1129 return thrd_last_fence_release[id_to_int(tid)];
1135 * Gets the last memory_order_seq_cst write (in the total global sequence)
1136 * performed on a particular object (i.e., memory location), not including the
1138 * @param curr The current ModelAction; also denotes the object location to
1140 * @return The last seq_cst write
1142 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
1144 void *location = curr->get_location();
1145 action_list_t *list = obj_map.get(location);
1146 /* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
1147 action_list_t::reverse_iterator rit;
1148 for (rit = list->rbegin();(*rit) != curr;rit++)
1150 rit++; /* Skip past curr */
1151 for ( ;rit != list->rend();rit++)
1152 if ((*rit)->is_write() && (*rit)->is_seqcst())
1158 * Gets the last memory_order_seq_cst fence (in the total global sequence)
1159 * performed in a particular thread, prior to a particular fence.
1160 * @param tid The ID of the thread to check
1161 * @param before_fence The fence from which to begin the search; if NULL, then
1162 * search for the most recent fence in the thread.
1163 * @return The last prior seq_cst fence in the thread, if exists; otherwise, NULL
1165 ModelAction * ModelExecution::get_last_seq_cst_fence(thread_id_t tid, const ModelAction *before_fence) const
1167 /* All fences should have location FENCE_LOCATION */
1168 action_list_t *list = obj_map.get(FENCE_LOCATION);
1173 action_list_t::reverse_iterator rit = list->rbegin();
1176 for (;rit != list->rend();rit++)
1177 if (*rit == before_fence)
1180 ASSERT(*rit == before_fence);
1184 for (;rit != list->rend();rit++)
1185 if ((*rit)->is_fence() && (tid == (*rit)->get_tid()) && (*rit)->is_seqcst())
1191 * Gets the last unlock operation performed on a particular mutex (i.e., memory
1192 * location). This function identifies the mutex according to the current
1193 * action, which is presumed to perform on the same mutex.
1194 * @param curr The current ModelAction; also denotes the object location to
1196 * @return The last unlock operation
1198 ModelAction * ModelExecution::get_last_unlock(ModelAction *curr) const
1200 void *location = curr->get_location();
1202 action_list_t *list = obj_map.get(location);
1203 /* Find: max({i in dom(S) | isUnlock(t_i) && samevar(t_i, t)}) */
1204 action_list_t::reverse_iterator rit;
1205 for (rit = list->rbegin();rit != list->rend();rit++)
1206 if ((*rit)->is_unlock() || (*rit)->is_wait())
1211 ModelAction * ModelExecution::get_parent_action(thread_id_t tid) const
1213 ModelAction *parent = get_last_action(tid);
1215 parent = get_thread(tid)->get_creation();
1220 * Returns the clock vector for a given thread.
1221 * @param tid The thread whose clock vector we want
1222 * @return Desired clock vector
1224 ClockVector * ModelExecution::get_cv(thread_id_t tid) const
1226 ModelAction *firstaction=get_parent_action(tid);
1227 return firstaction != NULL ? firstaction->get_cv() : NULL;
1230 bool valequals(uint64_t val1, uint64_t val2, int size) {
1233 return ((uint8_t)val1) == ((uint8_t)val2);
1235 return ((uint16_t)val1) == ((uint16_t)val2);
1237 return ((uint32_t)val1) == ((uint32_t)val2);
1247 * Build up an initial set of all past writes that this 'read' action may read
1248 * from, as well as any previously-observed future values that must still be valid.
1250 * @param curr is the current ModelAction that we are exploring; it must be a
1253 SnapVector<ModelAction *> * ModelExecution::build_may_read_from(ModelAction *curr)
1255 SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
1257 ASSERT(curr->is_read());
1259 ModelAction *last_sc_write = NULL;
1261 if (curr->is_seqcst())
1262 last_sc_write = get_last_seq_cst_write(curr);
1264 SnapVector<ModelAction *> * rf_set = new SnapVector<ModelAction *>();
1266 /* Iterate over all threads */
1267 for (i = 0;i < thrd_lists->size();i++) {
1268 /* Iterate over actions in thread, starting from most recent */
1269 action_list_t *list = &(*thrd_lists)[i];
1270 action_list_t::reverse_iterator rit;
1271 for (rit = list->rbegin();rit != list->rend();rit++) {
1272 ModelAction *act = *rit;
1274 /* Only consider 'write' actions */
1275 if (!act->is_write()) {
1276 if (act != curr && act->is_read() && act->happens_before(curr)) {
1277 ModelAction *tmp = act->get_reads_from();
1278 if (((unsigned int) id_to_int(tmp->get_tid()))==i)
1289 /* Don't consider more than one seq_cst write if we are a seq_cst read. */
1290 bool allow_read = true;
1292 if (curr->is_seqcst() && (act->is_seqcst() || (last_sc_write != NULL && act->happens_before(last_sc_write))) && act != last_sc_write)
1295 /* Need to check whether we will have two RMW reading from the same value */
1296 if (curr->is_rmwr()) {
1297 /* It is okay if we have a failing CAS */
1298 if (!curr->is_rmwrcas() ||
1299 valequals(curr->get_value(), act->get_value(), curr->getSize())) {
1300 //Need to make sure we aren't the second RMW
1301 CycleNode * node = mo_graph->getNode_noCreate(act);
1302 if (node != NULL && node->getRMW() != NULL) {
1303 //we are the second RMW
1310 /* Only add feasible reads */
1311 rf_set->push_back(act);
1314 /* Include at most one act per-thread that "happens before" curr */
1315 if (act->happens_before(curr))
1320 if (DBG_ENABLED()) {
1321 model_print("Reached read action:\n");
1323 model_print("End printing read_from_past\n");
1329 * @brief Get an action representing an uninitialized atomic
1331 * This function may create a new one.
1333 * @param curr The current action, which prompts the creation of an UNINIT action
1334 * @return A pointer to the UNINIT ModelAction
1336 ModelAction * ModelExecution::get_uninitialized_action(ModelAction *curr) const
1338 ModelAction *act = curr->get_uninit_action();
1340 act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), params->uninitvalue, model_thread);
1341 curr->set_uninit_action(act);
1343 act->create_cv(NULL);
1347 static void print_list(const action_list_t *list)
1349 action_list_t::const_iterator it;
1351 model_print("------------------------------------------------------------------------------------\n");
1352 model_print("# t Action type MO Location Value Rf CV\n");
1353 model_print("------------------------------------------------------------------------------------\n");
1355 unsigned int hash = 0;
1357 for (it = list->begin();it != list->end();it++) {
1358 const ModelAction *act = *it;
1359 if (act->get_seq_number() > 0)
1361 hash = hash^(hash<<3)^((*it)->hash());
1363 model_print("HASH %u\n", hash);
1364 model_print("------------------------------------------------------------------------------------\n");
1367 #if SUPPORT_MOD_ORDER_DUMP
1368 void ModelExecution::dumpGraph(char *filename) const
1371 sprintf(buffer, "%s.dot", filename);
1372 FILE *file = fopen(buffer, "w");
1373 fprintf(file, "digraph %s {\n", filename);
1374 mo_graph->dumpNodes(file);
1375 ModelAction **thread_array = (ModelAction **)model_calloc(1, sizeof(ModelAction *) * get_num_threads());
1377 for (action_list_t::const_iterator it = action_trace.begin();it != action_trace.end();it++) {
1378 ModelAction *act = *it;
1379 if (act->is_read()) {
1380 mo_graph->dot_print_node(file, act);
1381 mo_graph->dot_print_edge(file,
1382 act->get_reads_from(),
1384 "label=\"rf\", color=red, weight=2");
1386 if (thread_array[act->get_tid()]) {
1387 mo_graph->dot_print_edge(file,
1388 thread_array[id_to_int(act->get_tid())],
1390 "label=\"sb\", color=blue, weight=400");
1393 thread_array[act->get_tid()] = act;
1395 fprintf(file, "}\n");
1396 model_free(thread_array);
1401 /** @brief Prints an execution trace summary. */
1402 void ModelExecution::print_summary() const
1404 #if SUPPORT_MOD_ORDER_DUMP
1405 char buffername[100];
1406 sprintf(buffername, "exec%04u", get_execution_number());
1407 mo_graph->dumpGraphToFile(buffername);
1408 sprintf(buffername, "graph%04u", get_execution_number());
1409 dumpGraph(buffername);
1412 model_print("Execution trace %d:", get_execution_number());
1413 if (isfeasibleprefix()) {
1414 if (scheduler->all_threads_sleeping())
1415 model_print(" SLEEP-SET REDUNDANT");
1416 if (have_bug_reports())
1417 model_print(" DETECTED BUG(S)");
1419 print_infeasibility(" INFEASIBLE");
1422 print_list(&action_trace);
1428 * Add a Thread to the system for the first time. Should only be called once
1430 * @param t The Thread to add
1432 void ModelExecution::add_thread(Thread *t)
1434 unsigned int i = id_to_int(t->get_id());
1435 if (i >= thread_map.size())
1436 thread_map.resize(i + 1);
1438 if (!t->is_model_thread())
1439 scheduler->add_thread(t);
1443 * @brief Get a Thread reference by its ID
1444 * @param tid The Thread's ID
1445 * @return A Thread reference
1447 Thread * ModelExecution::get_thread(thread_id_t tid) const
1449 unsigned int i = id_to_int(tid);
1450 if (i < thread_map.size())
1451 return thread_map[i];
1456 * @brief Get a reference to the Thread in which a ModelAction was executed
1457 * @param act The ModelAction
1458 * @return A Thread reference
1460 Thread * ModelExecution::get_thread(const ModelAction *act) const
1462 return get_thread(act->get_tid());
1466 * @brief Get a Thread reference by its pthread ID
1467 * @param index The pthread's ID
1468 * @return A Thread reference
1470 Thread * ModelExecution::get_pthread(pthread_t pid) {
1476 uint32_t thread_id = x.v;
1477 if (thread_id < pthread_counter + 1) return pthread_map[thread_id];
1482 * @brief Check if a Thread is currently enabled
1483 * @param t The Thread to check
1484 * @return True if the Thread is currently enabled
1486 bool ModelExecution::is_enabled(Thread *t) const
1488 return scheduler->is_enabled(t);
1492 * @brief Check if a Thread is currently enabled
1493 * @param tid The ID of the Thread to check
1494 * @return True if the Thread is currently enabled
1496 bool ModelExecution::is_enabled(thread_id_t tid) const
1498 return scheduler->is_enabled(tid);
1502 * @brief Select the next thread to execute based on the curren action
1504 * RMW actions occur in two parts, and we cannot split them. And THREAD_CREATE
1505 * actions should be followed by the execution of their child thread. In either
1506 * case, the current action should determine the next thread schedule.
1508 * @param curr The current action
1509 * @return The next thread to run, if the current action will determine this
1510 * selection; otherwise NULL
1512 Thread * ModelExecution::action_select_next_thread(const ModelAction *curr) const
1514 /* Do not split atomic RMW */
1515 if (curr->is_rmwr())
1516 return get_thread(curr);
1517 /* Follow CREATE with the created thread */
1518 /* which is not needed, because model.cc takes care of this */
1519 if (curr->get_type() == THREAD_CREATE)
1520 return curr->get_thread_operand();
1521 if (curr->get_type() == PTHREAD_CREATE) {
1522 return curr->get_thread_operand();
1528 * Takes the next step in the execution, if possible.
1529 * @param curr The current step to take
1530 * @return Returns the next Thread to run, if any; NULL if this execution
1533 Thread * ModelExecution::take_step(ModelAction *curr)
1535 Thread *curr_thrd = get_thread(curr);
1536 ASSERT(curr_thrd->get_state() == THREAD_READY);
1538 ASSERT(check_action_enabled(curr)); /* May have side effects? */
1539 curr = check_current_action(curr);
1542 if (curr_thrd->is_blocked() || curr_thrd->is_complete())
1543 scheduler->remove_thread(curr_thrd);
1545 return action_select_next_thread(curr);
1548 Fuzzer * ModelExecution::getFuzzer() {