#include <stdio.h>
+#include <algorithm>
#include "model.h"
#include "action.h"
#include "clockvector.h"
#include "cyclegraph.h"
#include "promise.h"
+#include "datarace.h"
#define INITIAL_THREAD_ID 0
ModelChecker *model;
/** @brief Constructor */
-ModelChecker::ModelChecker() :
+ModelChecker::ModelChecker(struct model_params params) :
/* Initialize default scheduler */
scheduler(new Scheduler()),
- /* First thread created will have id INITIAL_THREAD_ID */
- next_thread_id(INITIAL_THREAD_ID),
- used_sequence_numbers(0),
num_executions(0),
- current_action(NULL),
+ num_feasible_executions(0),
+ params(params),
diverge(NULL),
- nextThread(THREAD_ID_T_NONE),
action_trace(new action_list_t()),
thread_map(new HashTable<int, Thread *, int>()),
obj_map(new HashTable<const void *, action_list_t, uintptr_t, 4>()),
obj_thrd_map(new HashTable<void *, std::vector<action_list_t>, uintptr_t, 4 >()),
promises(new std::vector<Promise *>()),
+ futurevalues(new std::vector<struct PendingFutureValue>()),
+ lazy_sync_with_release(new HashTable<void *, std::list<ModelAction *>, uintptr_t, 4>()),
thrd_last_action(new std::vector<ModelAction *>(1)),
node_stack(new NodeStack()),
- next_backtrack(NULL),
- cyclegraph(new CycleGraph()),
- failed_promise(false)
+ mo_graph(new CycleGraph()),
+ failed_promise(false),
+ too_many_reads(false),
+ asserted(false)
{
+ /* Allocate this "size" on the snapshotting heap */
+ priv = (struct model_snapshot_members *)calloc(1, sizeof(*priv));
+ /* First thread created will have id INITIAL_THREAD_ID */
+ priv->next_thread_id = INITIAL_THREAD_ID;
+
+ lazy_sync_size = &priv->lazy_sync_size;
}
/** @brief Destructor */
ModelChecker::~ModelChecker()
{
- /* std::map<int, Thread *>::iterator it;
- for (it = thread_map->begin(); it != thread_map->end(); it++)
- delete (*it).second;*/
+ for (int i = 0; i < get_num_threads(); i++)
+ delete thread_map->get(i);
delete thread_map;
delete obj_thrd_map;
delete obj_map;
delete action_trace;
+
+ for (unsigned int i = 0; i < promises->size(); i++)
+ delete (*promises)[i];
+ delete promises;
+
+ delete lazy_sync_with_release;
+
delete thrd_last_action;
delete node_stack;
delete scheduler;
- delete cyclegraph;
+ delete mo_graph;
}
/**
{
DEBUG("+++ Resetting to initial state +++\n");
node_stack->reset_execution();
- current_action = NULL;
- next_thread_id = INITIAL_THREAD_ID;
- used_sequence_numbers = 0;
- nextThread = 0;
- next_backtrack = NULL;
failed_promise = false;
+ too_many_reads = false;
+ reset_asserted();
snapshotObject->backTrackBeforeStep(0);
}
/** @returns a thread ID for a new Thread */
thread_id_t ModelChecker::get_next_id()
{
- return next_thread_id++;
+ return priv->next_thread_id++;
}
/** @returns the number of user threads created during this execution */
int ModelChecker::get_num_threads()
{
- return next_thread_id;
+ return priv->next_thread_id;
}
/** @returns a sequence number for a new ModelAction */
modelclock_t ModelChecker::get_next_seq_num()
{
- return ++used_sequence_numbers;
-}
-
-/**
- * Performs the "scheduling" for the model-checker. That is, it checks if the
- * model-checker has selected a "next thread to run" and returns it, if
- * available. This function should be called from the Scheduler routine, where
- * the Scheduler falls back to a default scheduling routine if needed.
- *
- * @return The next thread chosen by the model-checker. If the model-checker
- * makes no selection, retuns NULL.
- */
-Thread * ModelChecker::schedule_next_thread()
-{
- Thread *t;
- if (nextThread == THREAD_ID_T_NONE)
- return NULL;
- t = thread_map->get(id_to_int(nextThread));
-
- ASSERT(t != NULL);
-
- return t;
+ return ++priv->used_sequence_numbers;
}
/**
- * Choose the next thread in the replay sequence.
+ * @brief Choose the next thread to execute.
*
- * If the replay sequence has reached the 'diverge' point, returns a thread
- * from the backtracking set. Otherwise, simply returns the next thread in the
- * sequence that is being replayed.
+ * This function chooses the next thread that should execute. It can force the
+ * adjacency of read/write portions of a RMW action, force THREAD_CREATE to be
+ * followed by a THREAD_START, or it can enforce execution replay/backtracking.
+ * The model-checker may have no preference regarding the next thread (i.e.,
+ * when exploring a new execution ordering), in which case this will return
+ * NULL.
+ * @param curr The current ModelAction. This action might guide the choice of
+ * next thread.
+ * @return The next thread to run. If the model-checker has no preference, NULL.
*/
-thread_id_t ModelChecker::get_next_replay_thread()
+Thread * ModelChecker::get_next_thread(ModelAction *curr)
{
thread_id_t tid;
+ /* Do not split atomic actions. */
+ if (curr->is_rmwr())
+ return thread_current();
+ /* The THREAD_CREATE action points to the created Thread */
+ else if (curr->get_type() == THREAD_CREATE)
+ return (Thread *)curr->get_location();
+
/* Have we completed exploring the preselected path? */
if (diverge == NULL)
- return THREAD_ID_T_NONE;
+ return NULL;
/* Else, we are trying to replay an execution */
- ModelAction * next = node_stack->get_next()->get_action();
+ ModelAction *next = node_stack->get_next()->get_action();
if (next == diverge) {
Node *nextnode = next->get_node();
tid = next->get_tid();
}
DEBUG("*** ModelChecker chose next thread = %d ***\n", tid);
- return tid;
+ ASSERT(tid != THREAD_ID_T_NONE);
+ return thread_map->get(id_to_int(tid));
}
/**
DBG();
num_executions++;
+ if (isfinalfeasible())
+ num_feasible_executions++;
if (isfinalfeasible() || DBG_ENABLED())
print_summary();
- if ((diverge = model->get_next_backtrack()) == NULL)
+ if ((diverge = get_next_backtrack()) == NULL)
return false;
if (DBG_ENABLED()) {
diverge->print();
}
- model->reset_to_initial_state();
+ reset_to_initial_state();
return true;
}
return NULL;
}
/* linear search: from most recent to oldest */
- action_list_t *list = obj_map->ensureptr(act->get_location());
+ action_list_t *list = obj_map->get_safe_ptr(act->get_location());
action_list_t::reverse_iterator rit;
for (rit = list->rbegin(); rit != list->rend(); rit++) {
ModelAction *prev = *rit;
{
ModelAction *prev;
Node *node;
- Thread *t = get_thread(act->get_tid());
+ Thread *t = get_thread(act);
prev = get_last_conflict(act);
if (prev == NULL)
return;
/* Cache the latest backtracking point */
- if (!next_backtrack || *prev > *next_backtrack)
- next_backtrack = prev;
+ if (!priv->next_backtrack || *prev > *priv->next_backtrack)
+ priv->next_backtrack = prev;
/* If this is a new backtracking point, mark the tree */
if (!node->set_backtrack(t->get_id()))
*/
ModelAction * ModelChecker::get_next_backtrack()
{
- ModelAction *next = next_backtrack;
- next_backtrack = NULL;
+ ModelAction *next = priv->next_backtrack;
+ priv->next_backtrack = NULL;
return next;
}
-void ModelChecker::check_current_action(void)
+/**
+ * Processes a read or rmw model action.
+ * @param curr is the read model action to process.
+ * @param second_part_of_rmw is boolean that is true is this is the second action of a rmw.
+ * @return True if processing this read updates the mo_graph.
+ */
+bool ModelChecker::process_read(ModelAction *curr, bool second_part_of_rmw)
{
- ModelAction *curr = this->current_action;
- bool already_added = false;
- this->current_action = NULL;
- if (!curr) {
- DEBUG("trying to push NULL action...\n");
- return;
+ uint64_t value;
+ bool updated = false;
+ while (true) {
+ const ModelAction *reads_from = curr->get_node()->get_read_from();
+ if (reads_from != NULL) {
+ mo_graph->startChanges();
+
+ value = reads_from->get_value();
+ bool r_status = false;
+
+ if (!second_part_of_rmw) {
+ check_recency(curr);
+ r_status = r_modification_order(curr, reads_from);
+ }
+
+
+ if (!second_part_of_rmw&&!isfeasible()&&(curr->get_node()->increment_read_from()||curr->get_node()->increment_future_value())) {
+ mo_graph->rollbackChanges();
+ too_many_reads = false;
+ continue;
+ }
+
+ curr->read_from(reads_from);
+ mo_graph->commitChanges();
+ updated |= r_status;
+ } else if (!second_part_of_rmw) {
+ /* Read from future value */
+ value = curr->get_node()->get_future_value();
+ modelclock_t expiration = curr->get_node()->get_future_value_expiration();
+ curr->read_from(NULL);
+ Promise *valuepromise = new Promise(curr, value, expiration);
+ promises->push_back(valuepromise);
+ }
+ get_thread(curr)->set_return_value(value);
+ return updated;
+ }
+}
+
+/**
+ * Process a write ModelAction
+ * @param curr The ModelAction to process
+ * @return True if the mo_graph was updated or promises were resolved
+ */
+bool ModelChecker::process_write(ModelAction *curr)
+{
+ bool updated_mod_order = w_modification_order(curr);
+ bool updated_promises = resolve_promises(curr);
+
+ if (promises->size() == 0) {
+ for (unsigned int i = 0; i<futurevalues->size(); i++) {
+ struct PendingFutureValue pfv = (*futurevalues)[i];
+ if (pfv.act->get_node()->add_future_value(pfv.value, pfv.expiration) &&
+ (!priv->next_backtrack || *pfv.act > *priv->next_backtrack))
+ priv->next_backtrack = pfv.act;
+ }
+ futurevalues->resize(0);
}
- if (curr->is_rmwc()||curr->is_rmw()) {
- ModelAction *tmp=process_rmw(curr);
- already_added = true;
+ mo_graph->commitChanges();
+ get_thread(curr)->set_return_value(VALUE_NONE);
+ return updated_mod_order || updated_promises;
+}
+
+/**
+ * This is the heart of the model checker routine. It performs model-checking
+ * actions corresponding to a given "current action." Among other processes, it
+ * calculates reads-from relationships, updates synchronization clock vectors,
+ * forms a memory_order constraints graph, and handles replay/backtrack
+ * execution when running permutations of previously-observed executions.
+ *
+ * @param curr The current action to process
+ * @return The next Thread that must be executed. May be NULL if ModelChecker
+ * makes no choice (e.g., according to replay execution, combining RMW actions,
+ * etc.)
+ */
+Thread * ModelChecker::check_current_action(ModelAction *curr)
+{
+ bool second_part_of_rmw = false;
+
+ ASSERT(curr);
+
+ if (curr->is_rmwc() || curr->is_rmw()) {
+ ModelAction *tmp = process_rmw(curr);
+ second_part_of_rmw = true;
delete curr;
- curr=tmp;
+ curr = tmp;
+ compute_promises(curr);
} else {
- ModelAction * tmp = node_stack->explore_action(curr);
+ ModelAction *tmp = node_stack->explore_action(curr);
if (tmp) {
/* Discard duplicate ModelAction; use action from NodeStack */
/* First restore type and order in case of RMW operation */
tmp->copy_typeandorder(curr);
/* If we have diverged, we need to reset the clock vector. */
- if (diverge==NULL) {
+ if (diverge == NULL)
tmp->create_cv(get_parent_action(tmp->get_tid()));
- }
delete curr;
curr = tmp;
}
}
- /* Assign 'creation' parent */
- if (curr->get_type() == THREAD_CREATE) {
+ /* Thread specific actions */
+ switch (curr->get_type()) {
+ case THREAD_CREATE: {
Thread *th = (Thread *)curr->get_location();
th->set_creation(curr);
+ break;
}
-
- /* Deal with new thread */
- if (curr->get_type() == THREAD_START) {
- check_promises(NULL, curr->get_cv());
+ case THREAD_JOIN: {
+ Thread *waiting, *blocking;
+ waiting = get_thread(curr);
+ blocking = (Thread *)curr->get_location();
+ if (!blocking->is_complete()) {
+ blocking->push_wait_list(curr);
+ scheduler->sleep(waiting);
+ }
+ break;
}
-
- /* Assign reads_from values */
- Thread *th = get_thread(curr->get_tid());
- uint64_t value = VALUE_NONE;
- if (curr->is_read()) {
- const ModelAction *reads_from = curr->get_node()->get_read_from();
- if (reads_from!=NULL) {
- value = reads_from->get_value();
- /* Assign reads_from, perform release/acquire synchronization */
- curr->read_from(reads_from);
- r_modification_order(curr,reads_from);
- } else {
- /* Read from future value */
- value = curr->get_node()->get_future_value();
- curr->read_from(NULL);
- Promise * valuepromise=new Promise(curr, value);
- promises->push_back(valuepromise);
+ case THREAD_FINISH: {
+ Thread *th = get_thread(curr);
+ while (!th->wait_list_empty()) {
+ ModelAction *act = th->pop_wait_list();
+ Thread *wake = get_thread(act);
+ scheduler->wake(wake);
}
- } else if (curr->is_write()) {
- w_modification_order(curr);
- resolve_promises(curr);
+ th->complete();
+ break;
+ }
+ case THREAD_START: {
+ check_promises(NULL, curr->get_cv());
+ break;
+ }
+ default:
+ break;
}
- th->set_return_value(value);
-
- /* Add action to list. */
- if (!already_added)
+ /* Add current action to lists before work_queue loop */
+ if (!second_part_of_rmw)
add_action_to_lists(curr);
- /* Is there a better interface for setting the next thread rather
- than this field/convoluted approach? Perhaps like just returning
- it or something? */
+ work_queue_t work_queue(1, CheckCurrWorkEntry(curr));
- /* Do not split atomic actions. */
- if (curr->is_rmwr()) {
- nextThread = thread_current()->get_id();
- } else {
- nextThread = get_next_replay_thread();
+ while (!work_queue.empty()) {
+ WorkQueueEntry work = work_queue.front();
+ work_queue.pop_front();
+
+ switch (work.type) {
+ case WORK_CHECK_CURR_ACTION: {
+ ModelAction *act = work.action;
+ bool updated = false;
+ if (act->is_read() && process_read(act, second_part_of_rmw))
+ updated = true;
+
+ if (act->is_write() && process_write(act))
+ updated = true;
+
+ if (updated)
+ work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
+ break;
+ }
+ case WORK_CHECK_RELEASE_SEQ:
+ resolve_release_sequences(work.location, &work_queue);
+ break;
+ case WORK_CHECK_MO_EDGES: {
+ /** @todo Complete verification of work_queue */
+ ModelAction *act = work.action;
+ bool updated = false;
+
+ if (act->is_read()) {
+ if (r_modification_order(act, act->get_reads_from()))
+ updated = true;
+ }
+ if (act->is_write()) {
+ if (w_modification_order(act))
+ updated = true;
+ }
+
+ if (updated)
+ work_queue.push_back(CheckRelSeqWorkEntry(act->get_location()));
+ break;
+ }
+ default:
+ ASSERT(false);
+ break;
+ }
}
+ check_curr_backtracking(curr);
+
+ set_backtracking(curr);
+
+ return get_next_thread(curr);
+}
+
+void ModelChecker::check_curr_backtracking(ModelAction * curr) {
Node *currnode = curr->get_node();
Node *parnode = currnode->get_parent();
- if (!parnode->backtrack_empty()||!currnode->read_from_empty()||!currnode->future_value_empty()||!currnode->promise_empty())
- if (!next_backtrack || *curr > *next_backtrack)
- next_backtrack = curr;
+ if ((!parnode->backtrack_empty() ||
+ !currnode->read_from_empty() ||
+ !currnode->future_value_empty() ||
+ !currnode->promise_empty())
+ && (!priv->next_backtrack ||
+ *curr > *priv->next_backtrack)) {
+ priv->next_backtrack = curr;
+ }
+}
- set_backtracking(curr);
+bool ModelChecker::promises_expired() {
+ for (unsigned int promise_index = 0; promise_index < promises->size(); promise_index++) {
+ Promise *promise = (*promises)[promise_index];
+ if (promise->get_expiration()<priv->used_sequence_numbers) {
+ return true;
+ }
+ }
+ return false;
+}
+
+/** @returns whether the current partial trace must be a prefix of a
+ * feasible trace. */
+bool ModelChecker::isfeasibleprefix() {
+ return promises->size() == 0 && *lazy_sync_size == 0;
}
/** @returns whether the current partial trace is feasible. */
bool ModelChecker::isfeasible() {
- return !cyclegraph->checkForCycles() && !failed_promise;
+ return !mo_graph->checkForRMWViolation() && isfeasibleotherthanRMW();
+}
+
+/** @returns whether the current partial trace is feasible other than
+ * multiple RMW reading from the same store. */
+bool ModelChecker::isfeasibleotherthanRMW() {
+ if (DBG_ENABLED()) {
+ if (mo_graph->checkForCycles())
+ DEBUG("Infeasible: modification order cycles\n");
+ if (failed_promise)
+ DEBUG("Infeasible: failed promise\n");
+ if (too_many_reads)
+ DEBUG("Infeasible: too many reads\n");
+ if (promises_expired())
+ DEBUG("Infeasible: promises expired\n");
+ }
+ return !mo_graph->checkForCycles() && !failed_promise && !too_many_reads && !promises_expired();
}
/** Returns whether the current completed trace is feasible. */
bool ModelChecker::isfinalfeasible() {
- return isfeasible() && promises->size()==0;
+ if (DBG_ENABLED() && promises->size() != 0)
+ DEBUG("Infeasible: unrevolved promises\n");
+
+ return isfeasible() && promises->size() == 0;
}
/** Close out a RMWR by converting previous RMWR into a RMW or READ. */
-ModelAction * ModelChecker::process_rmw(ModelAction * act) {
+ModelAction * ModelChecker::process_rmw(ModelAction *act) {
int tid = id_to_int(act->get_tid());
- ModelAction *lastread=get_last_action(tid);
+ ModelAction *lastread = get_last_action(tid);
lastread->process_rmw(act);
- if (act->is_rmw())
- cyclegraph->addRMWEdge(lastread, lastread->get_reads_from());
+ if (act->is_rmw() && lastread->get_reads_from()!=NULL) {
+ mo_graph->addRMWEdge(lastread->get_reads_from(), lastread);
+ mo_graph->commitChanges();
+ }
return lastread;
}
/**
- * Updates the cyclegraph with the constraints imposed from the current read.
+ * Checks whether a thread has read from the same write for too many times
+ * without seeing the effects of a later write.
+ *
+ * Basic idea:
+ * 1) there must a different write that we could read from that would satisfy the modification order,
+ * 2) we must have read from the same value in excess of maxreads times, and
+ * 3) that other write must have been in the reads_from set for maxreads times.
+ *
+ * If so, we decide that the execution is no longer feasible.
+ */
+void ModelChecker::check_recency(ModelAction *curr) {
+ if (params.maxreads != 0) {
+ if (curr->get_node()->get_read_from_size() <= 1)
+ return;
+
+ //Must make sure that execution is currently feasible... We could
+ //accidentally clear by rolling back
+ if (!isfeasible())
+ return;
+
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
+ int tid = id_to_int(curr->get_tid());
+
+ /* Skip checks */
+ if ((int)thrd_lists->size() <= tid)
+ return;
+
+ action_list_t *list = &(*thrd_lists)[tid];
+
+ action_list_t::reverse_iterator rit = list->rbegin();
+ /* Skip past curr */
+ for (; (*rit) != curr; rit++)
+ ;
+ /* go past curr now */
+ rit++;
+
+ action_list_t::reverse_iterator ritcopy = rit;
+ //See if we have enough reads from the same value
+ int count = 0;
+ for (; count < params.maxreads; rit++,count++) {
+ if (rit==list->rend())
+ return;
+ ModelAction *act = *rit;
+ if (!act->is_read())
+ return;
+ if (act->get_reads_from() != curr->get_reads_from())
+ return;
+ if (act->get_node()->get_read_from_size() <= 1)
+ return;
+ }
+
+ for (int i = 0; i<curr->get_node()->get_read_from_size(); i++) {
+ //Get write
+ const ModelAction * write = curr->get_node()->get_read_from_at(i);
+ //Need a different write
+ if (write==curr->get_reads_from())
+ continue;
+
+ /* Test to see whether this is a feasible write to read from*/
+ mo_graph->startChanges();
+ r_modification_order(curr, write);
+ bool feasiblereadfrom = isfeasible();
+ mo_graph->rollbackChanges();
+
+ if (!feasiblereadfrom)
+ continue;
+ rit = ritcopy;
+
+ bool feasiblewrite = true;
+ //new we need to see if this write works for everyone
+
+ for (int loop = count; loop>0; loop--,rit++) {
+ ModelAction *act=*rit;
+ bool foundvalue = false;
+ for (int j = 0; j<act->get_node()->get_read_from_size(); j++) {
+ if (act->get_node()->get_read_from_at(i)==write) {
+ foundvalue = true;
+ break;
+ }
+ }
+ if (!foundvalue) {
+ feasiblewrite = false;
+ break;
+ }
+ }
+ if (feasiblewrite) {
+ too_many_reads = true;
+ return;
+ }
+ }
+ }
+}
+
+/**
+ * Updates the mo_graph with the constraints imposed from the current
+ * read.
+ *
+ * Basic idea is the following: Go through each other thread and find
+ * the lastest action that happened before our read. Two cases:
+ *
+ * (1) The action is a write => that write must either occur before
+ * the write we read from or be the write we read from.
+ *
+ * (2) The action is a read => the write that that action read from
+ * must occur before the write we read from or be the same write.
+ *
* @param curr The current action. Must be a read.
* @param rf The action that curr reads from. Must be a write.
+ * @return True if modification order edges were added; false otherwise
*/
-void ModelChecker::r_modification_order(ModelAction * curr, const ModelAction *rf) {
- std::vector<action_list_t> *thrd_lists = obj_thrd_map->ensureptr(curr->get_location());
+bool ModelChecker::r_modification_order(ModelAction *curr, const ModelAction *rf)
+{
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
unsigned int i;
+ bool added = false;
ASSERT(curr->is_read());
/* Iterate over all threads */
for (rit = list->rbegin(); rit != list->rend(); rit++) {
ModelAction *act = *rit;
- /* Include at most one act per-thread that "happens before" curr */
- if (act->happens_before(curr)) {
- if (act->is_read()) {
- const ModelAction * prevreadfrom=act->get_reads_from();
- if (prevreadfrom!=NULL&&rf!=prevreadfrom)
- cyclegraph->addEdge(rf, prevreadfrom);
- } else if (rf!=act) {
- cyclegraph->addEdge(rf, act);
+ /*
+ * Include at most one act per-thread that "happens
+ * before" curr. Don't consider reflexively.
+ */
+ if (act->happens_before(curr) && act != curr) {
+ if (act->is_write()) {
+ if (rf != act) {
+ mo_graph->addEdge(act, rf);
+ added = true;
+ }
+ } else {
+ const ModelAction *prevreadfrom = act->get_reads_from();
+ if (prevreadfrom != NULL && rf != prevreadfrom) {
+ mo_graph->addEdge(prevreadfrom, rf);
+ added = true;
+ }
}
break;
}
}
}
+
+ return added;
}
-/** Updates the cyclegraph with the constraints imposed from the
- * current read. */
-void ModelChecker::post_r_modification_order(ModelAction * curr, const ModelAction *rf) {
- std::vector<action_list_t> *thrd_lists = obj_thrd_map->ensureptr(curr->get_location());
+/** This method fixes up the modification order when we resolve a
+ * promises. The basic problem is that actions that occur after the
+ * read curr could not property add items to the modification order
+ * for our read.
+ *
+ * So for each thread, we find the earliest item that happens after
+ * the read curr. This is the item we have to fix up with additional
+ * constraints. If that action is write, we add a MO edge between
+ * the Action rf and that action. If the action is a read, we add a
+ * MO edge between the Action rf, and whatever the read accessed.
+ *
+ * @param curr is the read ModelAction that we are fixing up MO edges for.
+ * @param rf is the write ModelAction that curr reads from.
+ *
+ */
+
+void ModelChecker::post_r_modification_order(ModelAction *curr, const ModelAction *rf)
+{
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
unsigned int i;
ASSERT(curr->is_read());
/* Iterate over actions in thread, starting from most recent */
action_list_t *list = &(*thrd_lists)[i];
action_list_t::reverse_iterator rit;
- ModelAction *lastact=NULL;
+ ModelAction *lastact = NULL;
/* Find last action that happens after curr */
for (rit = list->rbegin(); rit != list->rend(); rit++) {
ModelAction *act = *rit;
if (curr->happens_before(act)) {
- lastact=act;
+ lastact = act;
} else
break;
}
/* Include at most one act per-thread that "happens before" curr */
- if (lastact!=NULL) {
+ if (lastact != NULL) {
if (lastact->is_read()) {
- const ModelAction * postreadfrom=lastact->get_reads_from();
- if (postreadfrom!=NULL&&rf!=postreadfrom)
- cyclegraph->addEdge(postreadfrom, rf);
- } else if (rf!=lastact) {
- cyclegraph->addEdge(lastact, rf);
+ const ModelAction *postreadfrom = lastact->get_reads_from();
+ if (postreadfrom != NULL&&rf != postreadfrom)
+ mo_graph->addEdge(rf, postreadfrom);
+ } else if (rf != lastact) {
+ mo_graph->addEdge(rf, lastact);
}
break;
}
}
/**
- * Updates the cyclegraph with the constraints imposed from the current write.
+ * Updates the mo_graph with the constraints imposed from the current write.
+ *
+ * Basic idea is the following: Go through each other thread and find
+ * the lastest action that happened before our write. Two cases:
+ *
+ * (1) The action is a write => that write must occur before
+ * the current write
+ *
+ * (2) The action is a read => the write that that action read from
+ * must occur before the current write.
+ *
+ * This method also handles two other issues:
+ *
+ * (I) Sequential Consistency: Making sure that if the current write is
+ * seq_cst, that it occurs after the previous seq_cst write.
+ *
+ * (II) Sending the write back to non-synchronizing reads.
+ *
* @param curr The current action. Must be a write.
+ * @return True if modification order edges were added; false otherwise
*/
-void ModelChecker::w_modification_order(ModelAction * curr) {
- std::vector<action_list_t> *thrd_lists = obj_thrd_map->ensureptr(curr->get_location());
+bool ModelChecker::w_modification_order(ModelAction *curr)
+{
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
unsigned int i;
+ bool added = false;
ASSERT(curr->is_write());
if (curr->is_seqcst()) {
/* We have to at least see the last sequentially consistent write,
so we are initialized. */
- ModelAction * last_seq_cst=get_last_seq_cst(curr->get_location());
- if (last_seq_cst!=NULL)
- cyclegraph->addEdge(curr, last_seq_cst);
+ ModelAction *last_seq_cst = get_last_seq_cst(curr);
+ if (last_seq_cst != NULL) {
+ mo_graph->addEdge(last_seq_cst, curr);
+ added = true;
+ }
}
/* Iterate over all threads */
action_list_t::reverse_iterator rit;
for (rit = list->rbegin(); rit != list->rend(); rit++) {
ModelAction *act = *rit;
+ if (act == curr) {
+ /*
+ * If RMW, we already have all relevant edges,
+ * so just skip to next thread.
+ * If normal write, we need to look at earlier
+ * actions, so continue processing list.
+ */
+ if (curr->is_rmw())
+ break;
+ else
+ continue;
+ }
- /* Include at most one act per-thread that "happens before" curr */
+ /*
+ * Include at most one act per-thread that "happens
+ * before" curr
+ */
if (act->happens_before(curr)) {
- if (act->is_read()) {
- cyclegraph->addEdge(curr, act->get_reads_from());
- } else
- cyclegraph->addEdge(curr, act);
+ /*
+ * Note: if act is RMW, just add edge:
+ * act --mo--> curr
+ * The following edge should be handled elsewhere:
+ * readfrom(act) --mo--> act
+ */
+ if (act->is_write())
+ mo_graph->addEdge(act, curr);
+ else if (act->is_read() && act->get_reads_from() != NULL)
+ mo_graph->addEdge(act->get_reads_from(), curr);
+ added = true;
+ break;
+ } else if (act->is_read() && !act->is_synchronizing(curr) &&
+ !act->same_thread(curr)) {
+ /* We have an action that:
+ (1) did not happen before us
+ (2) is a read and we are a write
+ (3) cannot synchronize with us
+ (4) is in a different thread
+ =>
+ that read could potentially read from our write.
+ */
+ if (thin_air_constraint_may_allow(curr, act)) {
+ if (isfeasible() ||
+ (curr->is_rmw() && act->is_rmw() && curr->get_reads_from()==act->get_reads_from() && isfeasibleotherthanRMW())) {
+ struct PendingFutureValue pfv = {curr->get_value(),curr->get_seq_number()+params.maxfuturedelay,act};
+ futurevalues->push_back(pfv);
+ }
+ }
+ }
+ }
+ }
+
+ return added;
+}
+
+/** Arbitrary reads from the future are not allowed. Section 29.3
+ * part 9 places some constraints. This method checks one result of constraint
+ * constraint. Others require compiler support. */
+
+bool ModelChecker::thin_air_constraint_may_allow(const ModelAction * writer, const ModelAction *reader) {
+ if (!writer->is_rmw())
+ return true;
+
+ if (!reader->is_rmw())
+ return true;
+
+ for (const ModelAction *search = writer->get_reads_from(); search != NULL; search = search->get_reads_from()) {
+ if (search==reader)
+ return false;
+ if (search->get_tid() == reader->get_tid() &&
+ search->happens_before(reader))
+ break;
+ }
+
+ return true;
+}
+
+/**
+ * Finds the head(s) of the release sequence(s) containing a given ModelAction.
+ * The ModelAction under consideration is expected to be taking part in
+ * release/acquire synchronization as an object of the "reads from" relation.
+ * Note that this can only provide release sequence support for RMW chains
+ * which do not read from the future, as those actions cannot be traced until
+ * their "promise" is fulfilled. Similarly, we may not even establish the
+ * presence of a release sequence with certainty, as some modification order
+ * constraints may be decided further in the future. Thus, this function
+ * "returns" two pieces of data: a pass-by-reference vector of @a release_heads
+ * and a boolean representing certainty.
+ *
+ * @todo Finish lazy updating, when promises are fulfilled in the future
+ * @param rf The action that might be part of a release sequence. Must be a
+ * write.
+ * @param release_heads A pass-by-reference style return parameter. After
+ * execution of this function, release_heads will contain the heads of all the
+ * relevant release sequences, if any exists
+ * @return true, if the ModelChecker is certain that release_heads is complete;
+ * false otherwise
+ */
+bool ModelChecker::release_seq_head(const ModelAction *rf,
+ std::vector< const ModelAction *, MyAlloc<const ModelAction *> > *release_heads) const
+{
+ if (!rf) {
+ /* read from future: need to settle this later */
+ return false; /* incomplete */
+ }
+
+ ASSERT(rf->is_write());
+
+ if (rf->is_release())
+ release_heads->push_back(rf);
+ if (rf->is_rmw()) {
+ /* We need a RMW action that is both an acquire and release to stop */
+ /** @todo Need to be smarter here... In the linux lock
+ * example, this will run to the beginning of the program for
+ * every acquire. */
+ if (rf->is_acquire() && rf->is_release())
+ return true; /* complete */
+ return release_seq_head(rf->get_reads_from(), release_heads);
+ }
+ if (rf->is_release())
+ return true; /* complete */
+
+ /* else relaxed write; check modification order for contiguous subsequence
+ * -> rf must be same thread as release */
+ int tid = id_to_int(rf->get_tid());
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(rf->get_location());
+ action_list_t *list = &(*thrd_lists)[tid];
+ action_list_t::const_reverse_iterator rit;
+
+ /* Find rf in the thread list */
+ rit = std::find(list->rbegin(), list->rend(), rf);
+ ASSERT(rit != list->rend());
+
+ /* Find the last write/release */
+ for (; rit != list->rend(); rit++)
+ if ((*rit)->is_release())
+ break;
+ if (rit == list->rend()) {
+ /* No write-release in this thread */
+ return true; /* complete */
+ }
+ ModelAction *release = *rit;
+
+ ASSERT(rf->same_thread(release));
+
+ bool certain = true;
+ for (unsigned int i = 0; i < thrd_lists->size(); i++) {
+ if (id_to_int(rf->get_tid()) == (int)i)
+ continue;
+ list = &(*thrd_lists)[i];
+
+ /* Can we ensure no future writes from this thread may break
+ * the release seq? */
+ bool future_ordered = false;
+
+ for (rit = list->rbegin(); rit != list->rend(); rit++) {
+ const ModelAction *act = *rit;
+ if (!act->is_write())
+ continue;
+ /* Reach synchronization -> this thread is complete */
+ if (act->happens_before(release))
+ break;
+ if (rf->happens_before(act)) {
+ future_ordered = true;
+ continue;
+ }
+
+ /* Check modification order */
+ if (mo_graph->checkReachable(rf, act)) {
+ /* rf --mo--> act */
+ future_ordered = true;
+ continue;
+ }
+ if (mo_graph->checkReachable(act, release))
+ /* act --mo--> release */
break;
- } else {
- if (act->is_read()&&!act->is_synchronizing(curr)&&!act->same_thread(curr)) {
- /* We have an action that:
- (1) did not happen before us
- (2) is a read and we are a write
- (3) cannot synchronize with us
- (4) is in a different thread
- =>
- that read could potentially read from our write.
- */
- if (act->get_node()->add_future_value(curr->get_value())&&
- (!next_backtrack || *act > * next_backtrack))
- next_backtrack = act;
+ if (mo_graph->checkReachable(release, act) &&
+ mo_graph->checkReachable(act, rf)) {
+ /* release --mo-> act --mo--> rf */
+ return true; /* complete */
+ }
+ certain = false;
+ }
+ if (!future_ordered)
+ return false; /* This thread is uncertain */
+ }
+
+ if (certain)
+ release_heads->push_back(release);
+ return certain;
+}
+
+/**
+ * A public interface for getting the release sequence head(s) with which a
+ * given ModelAction must synchronize. This function only returns a non-empty
+ * result when it can locate a release sequence head with certainty. Otherwise,
+ * it may mark the internal state of the ModelChecker so that it will handle
+ * the release sequence at a later time, causing @a act to update its
+ * synchronization at some later point in execution.
+ * @param act The 'acquire' action that may read from a release sequence
+ * @param release_heads A pass-by-reference return parameter. Will be filled
+ * with the head(s) of the release sequence(s), if they exists with certainty.
+ * @see ModelChecker::release_seq_head
+ */
+void ModelChecker::get_release_seq_heads(ModelAction *act,
+ std::vector< const ModelAction *, MyAlloc<const ModelAction *> > *release_heads)
+{
+ const ModelAction *rf = act->get_reads_from();
+ bool complete;
+ complete = release_seq_head(rf, release_heads);
+ if (!complete) {
+ /* add act to 'lazy checking' list */
+ std::list<ModelAction *> *list;
+ list = lazy_sync_with_release->get_safe_ptr(act->get_location());
+ list->push_back(act);
+ (*lazy_sync_size)++;
+ }
+}
+
+/**
+ * Attempt to resolve all stashed operations that might synchronize with a
+ * release sequence for a given location. This implements the "lazy" portion of
+ * determining whether or not a release sequence was contiguous, since not all
+ * modification order information is present at the time an action occurs.
+ *
+ * @param location The location/object that should be checked for release
+ * sequence resolutions
+ * @param work_queue The work queue to which to add work items as they are
+ * generated
+ * @return True if any updates occurred (new synchronization, new mo_graph
+ * edges)
+ */
+bool ModelChecker::resolve_release_sequences(void *location, work_queue_t *work_queue)
+{
+ std::list<ModelAction *> *list;
+ list = lazy_sync_with_release->getptr(location);
+ if (!list)
+ return false;
+
+ bool updated = false;
+ std::list<ModelAction *>::iterator it = list->begin();
+ while (it != list->end()) {
+ ModelAction *act = *it;
+ const ModelAction *rf = act->get_reads_from();
+ std::vector< const ModelAction *, MyAlloc<const ModelAction *> > release_heads;
+ bool complete;
+ complete = release_seq_head(rf, &release_heads);
+ for (unsigned int i = 0; i < release_heads.size(); i++) {
+ if (!act->has_synchronized_with(release_heads[i])) {
+ updated = true;
+ act->synchronize_with(release_heads[i]);
+ }
+ }
+
+ if (updated) {
+ /* Re-check act for mo_graph edges */
+ work_queue->push_back(MOEdgeWorkEntry(act));
+
+ /* propagate synchronization to later actions */
+ action_list_t::reverse_iterator it = action_trace->rbegin();
+ while ((*it) != act) {
+ ModelAction *propagate = *it;
+ if (act->happens_before(propagate)) {
+ propagate->synchronize_with(act);
+ /* Re-check 'propagate' for mo_graph edges */
+ work_queue->push_back(MOEdgeWorkEntry(propagate));
}
}
}
+ if (complete) {
+ it = list->erase(it);
+ (*lazy_sync_size)--;
+ } else
+ it++;
+ }
+
+ // If we resolved promises or data races, see if we have realized a data race.
+ if (checkDataRaces()) {
+ set_assert();
}
+
+ return updated;
}
/**
int tid = id_to_int(act->get_tid());
action_trace->push_back(act);
- obj_map->ensureptr(act->get_location())->push_back(act);
+ obj_map->get_safe_ptr(act->get_location())->push_back(act);
- std::vector<action_list_t> *vec = obj_thrd_map->ensureptr(act->get_location());
+ std::vector<action_list_t> *vec = obj_thrd_map->get_safe_ptr(act->get_location());
if (tid >= (int)vec->size())
- vec->resize(next_thread_id);
+ vec->resize(priv->next_thread_id);
(*vec)[tid].push_back(act);
if ((int)thrd_last_action->size() <= tid)
}
/**
- * Gets the last memory_order_seq_cst action (in the total global sequence)
- * performed on a particular object (i.e., memory location).
- * @param location The object location to check
- * @return The last seq_cst action performed
+ * Gets the last memory_order_seq_cst write (in the total global sequence)
+ * performed on a particular object (i.e., memory location), not including the
+ * current action.
+ * @param curr The current ModelAction; also denotes the object location to
+ * check
+ * @return The last seq_cst write
*/
-ModelAction * ModelChecker::get_last_seq_cst(const void *location)
+ModelAction * ModelChecker::get_last_seq_cst(ModelAction *curr)
{
- action_list_t *list = obj_map->ensureptr(location);
+ void *location = curr->get_location();
+ action_list_t *list = obj_map->get_safe_ptr(location);
/* Find: max({i in dom(S) | seq_cst(t_i) && isWrite(t_i) && samevar(t_i, t)}) */
action_list_t::reverse_iterator rit;
for (rit = list->rbegin(); rit != list->rend(); rit++)
- if ((*rit)->is_write() && (*rit)->is_seqcst())
+ if ((*rit)->is_write() && (*rit)->is_seqcst() && (*rit) != curr)
return *rit;
return NULL;
}
* @param tid The thread whose clock vector we want
* @return Desired clock vector
*/
-ClockVector * ModelChecker::get_cv(thread_id_t tid) {
+ClockVector * ModelChecker::get_cv(thread_id_t tid)
+{
return get_parent_action(tid)->get_cv();
}
+/**
+ * Resolve a set of Promises with a current write. The set is provided in the
+ * Node corresponding to @a write.
+ * @param write The ModelAction that is fulfilling Promises
+ * @return True if promises were resolved; false otherwise
+ */
+bool ModelChecker::resolve_promises(ModelAction *write)
+{
+ bool resolved = false;
-/** Resolve the given promises. */
-
-void ModelChecker::resolve_promises(ModelAction *write) {
- for(unsigned int i=0, promise_index=0;promise_index<promises->size(); i++) {
- Promise * promise=(*promises)[promise_index];
+ for (unsigned int i = 0, promise_index = 0; promise_index < promises->size(); i++) {
+ Promise *promise = (*promises)[promise_index];
if (write->get_node()->get_promise(i)) {
- ModelAction * read=promise->get_action();
+ ModelAction *read = promise->get_action();
read->read_from(write);
+ if (read->is_rmw()) {
+ mo_graph->addRMWEdge(write, read);
+ }
+ //First fix up the modification order for actions that happened
+ //before the read
r_modification_order(read, write);
+ //Next fix up the modification order for actions that happened
+ //after the read.
post_r_modification_order(read, write);
- promises->erase(promises->begin()+promise_index);
+ promises->erase(promises->begin() + promise_index);
+ resolved = true;
} else
promise_index++;
}
+ return resolved;
}
-/** Compute the set of promises that could potentially be satisfied by
- * this action. */
-
-void ModelChecker::compute_promises(ModelAction *curr) {
- for(unsigned int i=0;i<promises->size();i++) {
- Promise * promise=(*promises)[i];
- const ModelAction * act=promise->get_action();
- if (!act->happens_before(curr)&&
- act->is_read()&&
- !act->is_synchronizing(curr)&&
- !act->same_thread(curr)&&
- promise->get_value()==curr->get_value()) {
+/**
+ * Compute the set of promises that could potentially be satisfied by this
+ * action. Note that the set computation actually appears in the Node, not in
+ * ModelChecker.
+ * @param curr The ModelAction that may satisfy promises
+ */
+void ModelChecker::compute_promises(ModelAction *curr)
+{
+ for (unsigned int i = 0; i < promises->size(); i++) {
+ Promise *promise = (*promises)[i];
+ const ModelAction *act = promise->get_action();
+ if (!act->happens_before(curr) &&
+ act->is_read() &&
+ !act->is_synchronizing(curr) &&
+ !act->same_thread(curr) &&
+ promise->get_value() == curr->get_value()) {
curr->get_node()->set_promise(i);
}
}
}
/** Checks promises in response to change in ClockVector Threads. */
-
-void ModelChecker::check_promises(ClockVector *old_cv, ClockVector * merge_cv) {
- for(unsigned int i=0;i<promises->size();i++) {
- Promise * promise=(*promises)[i];
- const ModelAction * act=promise->get_action();
- if ((old_cv==NULL||!old_cv->synchronized_since(act))&&
+void ModelChecker::check_promises(ClockVector *old_cv, ClockVector *merge_cv)
+{
+ for (unsigned int i = 0; i < promises->size(); i++) {
+ Promise *promise = (*promises)[i];
+ const ModelAction *act = promise->get_action();
+ if ((old_cv == NULL || !old_cv->synchronized_since(act)) &&
merge_cv->synchronized_since(act)) {
//This thread is no longer able to send values back to satisfy the promise
- int num_synchronized_threads=promise->increment_threads();
- if (num_synchronized_threads==model->get_num_threads()) {
+ int num_synchronized_threads = promise->increment_threads();
+ if (num_synchronized_threads == get_num_threads()) {
//Promise has failed
failed_promise = true;
return;
*/
void ModelChecker::build_reads_from_past(ModelAction *curr)
{
- std::vector<action_list_t> *thrd_lists = obj_thrd_map->ensureptr(curr->get_location());
+ std::vector<action_list_t> *thrd_lists = obj_thrd_map->get_safe_ptr(curr->get_location());
unsigned int i;
ASSERT(curr->is_read());
bool initialized = false;
if (curr->is_seqcst()) {
- last_seq_cst = get_last_seq_cst(curr->get_location());
+ last_seq_cst = get_last_seq_cst(curr);
/* We have to at least see the last sequentially consistent write,
so we are initialized. */
if (last_seq_cst != NULL)
continue;
/* Don't consider more than one seq_cst write if we are a seq_cst read. */
- if (!act->is_seqcst() || !curr->is_seqcst() || act == last_seq_cst) {
+ if (!curr->is_seqcst()|| (!act->is_seqcst() && (last_seq_cst==NULL||!act->happens_before(last_seq_cst))) || act == last_seq_cst) {
DEBUG("Adding action to may_read_from:\n");
if (DBG_ENABLED()) {
act->print();
printf("---------------------------------------------------------------------\n");
}
-void ModelChecker::print_summary(void)
+void ModelChecker::print_summary()
{
printf("\n");
printf("Number of executions: %d\n", num_executions);
+ printf("Number of feasible executions: %d\n", num_feasible_executions);
printf("Total nodes created: %d\n", node_stack->get_total_nodes());
+#if SUPPORT_MOD_ORDER_DUMP
scheduler->print();
+ char buffername[100];
+ sprintf(buffername, "exec%u",num_executions);
+ mo_graph->dumpGraphToFile(buffername);
+#endif
if (!isfinalfeasible())
printf("INFEASIBLE EXECUTION!\n");
printf("\n");
}
-int ModelChecker::add_thread(Thread *t)
+/**
+ * Add a Thread to the system for the first time. Should only be called once
+ * per thread.
+ * @param t The Thread to add
+ */
+void ModelChecker::add_thread(Thread *t)
{
thread_map->put(id_to_int(t->get_id()), t);
scheduler->add_thread(t);
- return 0;
}
void ModelChecker::remove_thread(Thread *t)
* context). This switch is made with the intention of exploring a particular
* model-checking action (described by a ModelAction object). Must be called
* from a user-thread context.
- * @param act The current action that will be explored. May be NULL, although
- * there is little reason to switch to the model-checker without an action to
- * explore (note: act == NULL is sometimes used as a hack to allow a thread to
- * yield control without performing any progress; see thrd_join()).
+ * @param act The current action that will be explored. Must not be NULL.
* @return Return status from the 'swap' call (i.e., success/fail, 0/-1)
*/
int ModelChecker::switch_to_master(ModelAction *act)
{
DBG();
- Thread * old = thread_current();
+ Thread *old = thread_current();
set_current_action(act);
old->set_state(THREAD_READY);
- return Thread::swap(old, get_system_context());
+ return Thread::swap(old, &system_context);
+}
+
+/**
+ * Takes the next step in the execution, if possible.
+ * @return Returns true (success) if a step was taken and false otherwise.
+ */
+bool ModelChecker::take_step() {
+ Thread *curr, *next;
+
+ if (has_asserted())
+ return false;
+
+ curr = thread_current();
+ if (curr) {
+ if (curr->get_state() == THREAD_READY) {
+ ASSERT(priv->current_action);
+
+ priv->nextThread = check_current_action(priv->current_action);
+ priv->current_action = NULL;
+ if (!curr->is_blocked() && !curr->is_complete())
+ scheduler->add_thread(curr);
+ } else {
+ ASSERT(false);
+ }
+ }
+ next = scheduler->next_thread(priv->nextThread);
+
+ /* Infeasible -> don't take any more steps */
+ if (!isfeasible())
+ return false;
+
+ if (next)
+ next->set_state(THREAD_RUNNING);
+ DEBUG("(%d, %d)\n", curr ? curr->get_id() : -1, next ? next->get_id() : -1);
+
+ /* next == NULL -> don't take any more steps */
+ if (!next)
+ return false;
+ /* Return false only if swap fails with an error */
+ return (Thread::swap(&system_context, next) == 0);
+}
+
+/** Runs the current execution until threre are no more steps to take. */
+void ModelChecker::finish_execution() {
+ DBG();
+
+ while (take_step());
}