Bug fixes to atomic logic
[c11tester.git] / execution.cc
index 4a351a98f8397fb4634d1ae3dd2da6c2aa32e198..af54b54a54cc43e962c8606115eb10be467ddffa 100644 (file)
@@ -6,7 +6,6 @@
 #include "model.h"
 #include "execution.h"
 #include "action.h"
-#include "nodestack.h"
 #include "schedule.h"
 #include "common.h"
 #include "clockvector.h"
@@ -14,6 +13,7 @@
 #include "datarace.h"
 #include "threads-model.h"
 #include "bugmessage.h"
+#include "history.h"
 #include "fuzzer.h"
 
 #define INITIAL_THREAD_ID       0
@@ -47,33 +47,168 @@ struct model_snapshot_members {
        SNAPSHOTALLOC
 };
 
+
+#ifdef TLS
+#include <dlfcn.h>
+
+//Code taken from LLVM and licensed under the University of Illinois Open Source
+//License.
+static uintptr_t thread_descriptor_size;
+#if __LP64__ || defined(_WIN64)
+#  define SANITIZER_WORDSIZE 64
+#else
+#  define SANITIZER_WORDSIZE 32
+#endif
+
+#if SANITIZER_WORDSIZE == 64
+# define FIRST_32_SECOND_64(a, b) (b)
+#else
+# define FIRST_32_SECOND_64(a, b) (a)
+#endif
+
+#if defined(__x86_64__) && !defined(_LP64)
+# define SANITIZER_X32 1
+#else
+# define SANITIZER_X32 0
+#endif
+
+#if defined(__arm__)
+# define SANITIZER_ARM 1
+#else
+# define SANITIZER_ARM 0
+#endif
+
+uintptr_t ThreadDescriptorSize() {
+       uintptr_t val = thread_descriptor_size;
+       if (val)
+               return val;
+#if defined(__x86_64__) || defined(__i386__) || defined(__arm__)
+#ifdef _CS_GNU_LIBC_VERSION
+       char buf[64];
+       uintptr_t len = confstr(_CS_GNU_LIBC_VERSION, buf, sizeof(buf));
+       if (len < sizeof(buf) && strncmp(buf, "glibc 2.", 8) == 0) {
+               char *end;
+               int minor = strtoll(buf + 8, &end, 10);
+               if (end != buf + 8 && (*end == '\0' || *end == '.' || *end == '-')) {
+                       int patch = 0;
+                       if (*end == '.')
+                               // strtoll will return 0 if no valid conversion could be performed
+                               patch = strtoll(end + 1, nullptr, 10);
+
+                       /* sizeof(struct pthread) values from various glibc versions.  */
+                       if (SANITIZER_X32)
+                               val = 1728;// Assume only one particular version for x32.
+                       // For ARM sizeof(struct pthread) changed in Glibc 2.23.
+                       else if (SANITIZER_ARM)
+                               val = minor <= 22 ? 1120 : 1216;
+                       else if (minor <= 3)
+                               val = FIRST_32_SECOND_64(1104, 1696);
+                       else if (minor == 4)
+                               val = FIRST_32_SECOND_64(1120, 1728);
+                       else if (minor == 5)
+                               val = FIRST_32_SECOND_64(1136, 1728);
+                       else if (minor <= 9)
+                               val = FIRST_32_SECOND_64(1136, 1712);
+                       else if (minor == 10)
+                               val = FIRST_32_SECOND_64(1168, 1776);
+                       else if (minor == 11 || (minor == 12 && patch == 1))
+                               val = FIRST_32_SECOND_64(1168, 2288);
+                       else if (minor <= 13)
+                               val = FIRST_32_SECOND_64(1168, 2304);
+                       else
+                               val = FIRST_32_SECOND_64(1216, 2304);
+               }
+       }
+#endif
+#elif defined(__mips__)
+       // TODO(sagarthakur): add more values as per different glibc versions.
+       val = FIRST_32_SECOND_64(1152, 1776);
+#elif defined(__aarch64__)
+       // The sizeof (struct pthread) is the same from GLIBC 2.17 to 2.22.
+       val = 1776;
+#elif defined(__powerpc64__)
+       val = 1776;     // from glibc.ppc64le 2.20-8.fc21
+#elif defined(__s390__)
+       val = FIRST_32_SECOND_64(1152, 1776);   // valid for glibc 2.22
+#endif
+       if (val)
+               thread_descriptor_size = val;
+       return val;
+}
+
+#ifdef __i386__
+# define DL_INTERNAL_FUNCTION __attribute__((regparm(3), stdcall))
+#else
+# define DL_INTERNAL_FUNCTION
+#endif
+
+intptr_t RoundUpTo(uintptr_t size, uintptr_t boundary) {
+       return (size + boundary - 1) & ~(boundary - 1);
+}
+
+uintptr_t getTlsSize() {
+       // all current supported platforms have 16 bytes stack alignment
+       const size_t kStackAlign = 16;
+       typedef void (*get_tls_func)(size_t*, size_t*) DL_INTERNAL_FUNCTION;
+       get_tls_func get_tls;
+       void *get_tls_static_info_ptr = dlsym(RTLD_NEXT, "_dl_get_tls_static_info");
+       memcpy(&get_tls, &get_tls_static_info_ptr, sizeof(get_tls_static_info_ptr));
+       ASSERT(get_tls != 0);
+       size_t tls_size = 0;
+       size_t tls_align = 0;
+       get_tls(&tls_size, &tls_align);
+       if (tls_align < kStackAlign)
+               tls_align = kStackAlign;
+       return RoundUpTo(tls_size, tls_align);
+}
+
+#endif
+
 /** @brief Constructor */
-ModelExecution::ModelExecution(ModelChecker *m, Scheduler *scheduler, NodeStack *node_stack) :
+ModelExecution::ModelExecution(ModelChecker *m, Scheduler *scheduler) :
        model(m),
        params(NULL),
        scheduler(scheduler),
        action_trace(),
        thread_map(2),  /* We'll always need at least 2 threads */
        pthread_map(0),
-       pthread_counter(0),
+       pthread_counter(1),
        obj_map(),
        condvar_waiters_map(),
        obj_thrd_map(),
        mutex_map(),
        thrd_last_action(1),
        thrd_last_fence_release(),
-       node_stack(node_stack),
        priv(new struct model_snapshot_members ()),
-                        mo_graph(new CycleGraph()),
-       fuzzer(new Fuzzer())
+       mo_graph(new CycleGraph()),
+       fuzzer(new Fuzzer()),
+       thrd_func_list(),
+       thrd_func_inst_lists()
+#ifdef TLS
+       ,tls_base(NULL),
+       tls_addr(0),
+       tls_size(0),
+       thd_desc_size(0)
+#endif
 {
        /* Initialize a model-checker thread, for special ModelActions */
        model_thread = new Thread(get_next_id());
        add_thread(model_thread);
        scheduler->register_engine(this);
-       node_stack->register_engine(this);
 }
 
+#ifdef TLS
+void ModelExecution::initTLS() {
+       tls_addr = get_tls_addr();
+       tls_size = getTlsSize();
+       tls_addr -= tls_size;
+       thd_desc_size = ThreadDescriptorSize();
+       tls_addr += thd_desc_size;
+       tls_base = (char *) snapshot_calloc(tls_size,1);
+       memcpy(tls_base, reinterpret_cast<const char *>(tls_addr), tls_size);
+}
+#endif
+
 /** @brief Destructor */
 ModelExecution::~ModelExecution()
 {
@@ -99,7 +234,7 @@ static action_list_t * get_safe_ptr_action(HashTable<const void *, action_list_t
        return tmp;
 }
 
-static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
+static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<const void *, SnapVector<action_list_t> *, uintptr_t, 4> * hash, void * ptr)
 {
        SnapVector<action_list_t> *tmp = hash->get(ptr);
        if (tmp == NULL) {
@@ -109,18 +244,6 @@ static SnapVector<action_list_t> * get_safe_ptr_vect_action(HashTable<void *, Sn
        return tmp;
 }
 
-action_list_t * ModelExecution::get_actions_on_obj(void * obj, thread_id_t tid) const
-{
-       SnapVector<action_list_t> *wrv = obj_thrd_map.get(obj);
-       if (wrv==NULL)
-               return NULL;
-       unsigned int thread=id_to_int(tid);
-       if (thread < wrv->size())
-               return &(*wrv)[thread];
-       else
-               return NULL;
-}
-
 /** @return a thread ID for a new Thread */
 thread_id_t ModelExecution::get_next_id()
 {
@@ -203,6 +326,16 @@ bool ModelExecution::have_bug_reports() const
        return priv->bugs.size() != 0;
 }
 
+/** @return True, if any fatal bugs have been reported for this execution.
+ *  Any bug other than a data race is considered a fatal bug. Data races
+ *  are not considered fatal unless the number of races is exceeds
+ *  a threshold (temporarily set as 15).
+ */
+bool ModelExecution::have_fatal_bug_reports() const
+{
+       return priv->bugs.size() != 0;
+}
+
 SnapVector<bug_message *> * ModelExecution::get_bugs() const
 {
        return &priv->bugs;
@@ -265,6 +398,20 @@ bool ModelExecution::is_complete_execution() const
        return true;
 }
 
+ModelAction * ModelExecution::convertNonAtomicStore(void * location) {
+       uint64_t value = *((const uint64_t *) location);
+       modelclock_t storeclock;
+       thread_id_t storethread;
+       getStoreThreadAndClock(location, &storethread, &storeclock);
+       setAtomicStoreFlag(location);
+       ModelAction * act = new ModelAction(NONATOMIC_WRITE, memory_order_relaxed, location, value, get_thread(storethread));
+       act->set_seq_number(storeclock);
+       add_normal_write_to_lists(act);
+       add_write_to_lists(act);
+       w_modification_order(act);
+       return act;
+}
+
 
 /**
  * Processes a read model action.
@@ -275,21 +422,30 @@ bool ModelExecution::is_complete_execution() const
 void ModelExecution::process_read(ModelAction *curr, SnapVector<ModelAction *> * rf_set)
 {
        SnapVector<const ModelAction *> * priorset = new SnapVector<const ModelAction *>();
-       while(true) {
+       bool hasnonatomicstore = hasNonAtomicStore(curr->get_location());
+       if (hasnonatomicstore) {
+               ModelAction * nonatomicstore = convertNonAtomicStore(curr->get_location());
+               rf_set->push_back(nonatomicstore);
+       }
 
+       while(true) {
                int index = fuzzer->selectWrite(curr, rf_set);
-               const ModelAction *rf = (*rf_set)[index];
+               ModelAction *rf = (*rf_set)[index];
 
 
                ASSERT(rf);
-
-               if (r_modification_order(curr, rf, priorset)) {
+               bool canprune = false;
+               if (r_modification_order(curr, rf, priorset, &canprune)) {
                        for(unsigned int i=0;i<priorset->size();i++) {
                                mo_graph->addEdge((*priorset)[i], rf);
                        }
                        read_from(curr, rf);
                        get_thread(curr)->set_return_value(curr->get_return_value());
                        delete priorset;
+                       if (canprune && curr->get_type() == ATOMIC_READ) {
+                               int tid = id_to_int(curr->get_tid());
+                               (*obj_thrd_map.get(curr->get_location()))[tid].pop_back();
+                       }
                        return;
                }
                priorset->clear();
@@ -334,8 +490,9 @@ bool ModelExecution::process_mutex(ModelAction *curr)
        }
        //otherwise fall into the lock case
        case ATOMIC_LOCK: {
-               if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
-                       assert_bug("Lock access before initialization");
+               //TODO: FIND SOME BETTER WAY TO CHECK LOCK INITIALIZED OR NOT
+               //if (curr->get_cv()->getClock(state->alloc_tid) <= state->alloc_clock)
+               //      assert_bug("Lock access before initialization");
                state->locked = get_thread(curr);
                ModelAction *unlock = get_last_unlock(curr);
                //synchronize with the previous unlock statement
@@ -347,6 +504,8 @@ bool ModelExecution::process_mutex(ModelAction *curr)
        }
        case ATOMIC_WAIT:
        case ATOMIC_UNLOCK: {
+               //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...
+
                /* wake up the other threads */
                for (unsigned int i = 0;i < get_num_threads();i++) {
                        Thread *t = get_thread(int_to_id(i));
@@ -374,8 +533,10 @@ bool ModelExecution::process_mutex(ModelAction *curr)
        }
        case ATOMIC_NOTIFY_ONE: {
                action_list_t *waiters = get_safe_ptr_action(&condvar_waiters_map, curr->get_location());
-               Thread * thread = fuzzer->selectNotify(waiters);
-               scheduler->wake(thread);
+               if (waiters->size() != 0) {
+                       Thread * thread = fuzzer->selectNotify(waiters);
+                       scheduler->wake(thread);
+               }
                break;
        }
 
@@ -392,10 +553,7 @@ bool ModelExecution::process_mutex(ModelAction *curr)
  */
 void ModelExecution::process_write(ModelAction *curr)
 {
-
        w_modification_order(curr);
-
-
        get_thread(curr)->set_return_value(VALUE_NONE);
 }
 
@@ -438,11 +596,8 @@ bool ModelExecution::process_fence(ModelAction *curr)
                                continue;
 
                        /* Establish hypothetical release sequences */
-                       rel_heads_list_t release_heads;
-                       get_release_seq_heads(curr, act, &release_heads);
-                       for (unsigned int i = 0;i < release_heads.size();i++)
-                               synchronize(release_heads[i], curr);
-                       if (release_heads.size() != 0)
+                       ClockVector *cv = get_hb_from_write(act->get_reads_from());
+                       if (cv != NULL && curr->get_cv()->merge(cv))
                                updated = true;
                }
        }
@@ -529,8 +684,8 @@ bool ModelExecution::process_thread_action(ModelAction *curr)
 
 /**
  * Initialize the current action by performing one or more of the following
- * actions, as appropriate: merging RMWR and RMWC/RMW actions, stepping forward
- * in the NodeStack, manipulating backtracking sets, allocating and
+ * actions, as appropriate: merging RMWR and RMWC/RMW actions,
+ * manipulating backtracking sets, allocating and
  * initializing clock vectors, and computing the promises to fulfill.
  *
  * @param curr The current action, as passed from the user context; may be
@@ -540,38 +695,16 @@ bool ModelExecution::process_thread_action(ModelAction *curr)
  */
 bool ModelExecution::initialize_curr_action(ModelAction **curr)
 {
-       ModelAction *newcurr;
-
        if ((*curr)->is_rmwc() || (*curr)->is_rmw()) {
-               newcurr = process_rmw(*curr);
+               ModelAction *newcurr = process_rmw(*curr);
                delete *curr;
 
                *curr = newcurr;
                return false;
-       }
-
-       (*curr)->set_seq_number(get_next_seq_num());
-
-       newcurr = node_stack->explore_action(*curr);
-       if (newcurr) {
-               /* First restore type and order in case of RMW operation */
-               if ((*curr)->is_rmwr())
-                       newcurr->copy_typeandorder(*curr);
-
-               ASSERT((*curr)->get_location() == newcurr->get_location());
-               newcurr->copy_from_new(*curr);
-
-               /* Discard duplicate ModelAction; use action from NodeStack */
-               delete *curr;
-
-               /* Always compute new clock vector */
-               newcurr->create_cv(get_parent_action(newcurr->get_tid()));
-
-               *curr = newcurr;
-               return false;   /* Action was explored previously */
        } else {
-               newcurr = *curr;
+               ModelAction *newcurr = *curr;
 
+               newcurr->set_seq_number(get_next_seq_num());
                /* Always compute new clock vector */
                newcurr->create_cv(get_parent_action(newcurr->get_tid()));
 
@@ -594,22 +727,18 @@ bool ModelExecution::initialize_curr_action(ModelAction **curr)
  * @return True if this read established synchronization
  */
 
-bool ModelExecution::read_from(ModelAction *act, const ModelAction *rf)
+void ModelExecution::read_from(ModelAction *act, ModelAction *rf)
 {
        ASSERT(rf);
        ASSERT(rf->is_write());
 
        act->set_read_from(rf);
        if (act->is_acquire()) {
-               rel_heads_list_t release_heads;
-               get_release_seq_heads(act, act, &release_heads);
-               int num_heads = release_heads.size();
-               for (unsigned int i = 0;i < release_heads.size();i++)
-                       if (!synchronize(release_heads[i], act))
-                               num_heads--;
-               return num_heads > 0;
+               ClockVector *cv = get_hb_from_write(rf);
+               if (cv == NULL)
+                       return;
+               act->get_cv()->merge(cv);
        }
-       return false;
 }
 
 /**
@@ -685,6 +814,9 @@ ModelAction * ModelExecution::check_current_action(ModelAction *curr)
        if (!second_part_of_rmw && curr->get_type() != NOOP)
                add_action_to_lists(curr);
 
+       if (curr->is_write())
+               add_write_to_lists(curr);
+
        SnapVector<ModelAction *> * rf_set = NULL;
        /* Build may_read_from set for newly-created actions */
        if (newly_explored && curr->is_read())
@@ -774,7 +906,7 @@ ModelAction * ModelExecution::process_rmw(ModelAction *act) {
  * @return True if modification order edges were added; false otherwise
  */
 
-bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *rf, SnapVector<const ModelAction *> * priorset)
+bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *rf, SnapVector<const ModelAction *> * priorset, bool * canprune)
 {
        SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
        unsigned int i;
@@ -782,24 +914,30 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *
 
        /* Last SC fence in the current thread */
        ModelAction *last_sc_fence_local = get_last_seq_cst_fence(curr->get_tid(), NULL);
-       ModelAction *last_sc_write = NULL;
-       if (curr->is_seqcst())
-               last_sc_write = get_last_seq_cst_write(curr);
 
+       int tid = curr->get_tid();
+       ModelAction *prev_same_thread = NULL;
        /* Iterate over all threads */
-       for (i = 0;i < thrd_lists->size();i++) {
-               /* Last SC fence in thread i */
+       for (i = 0;i < thrd_lists->size();i++, tid = (((unsigned int)(tid+1)) == thrd_lists->size()) ? 0 : tid + 1) {
+               /* Last SC fence in thread tid */
                ModelAction *last_sc_fence_thread_local = NULL;
-               if (int_to_id((int)i) != curr->get_tid())
-                       last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(i), NULL);
+               if (i != 0)
+                       last_sc_fence_thread_local = get_last_seq_cst_fence(int_to_id(tid), NULL);
 
-               /* Last SC fence in thread i, before last SC fence in current thread */
+               /* Last SC fence in thread tid, before last SC fence in current thread */
                ModelAction *last_sc_fence_thread_before = NULL;
                if (last_sc_fence_local)
-                       last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(i), last_sc_fence_local);
+                       last_sc_fence_thread_before = get_last_seq_cst_fence(int_to_id(tid), last_sc_fence_local);
+
+               //Only need to iterate if either hb has changed for thread in question or SC fence after last operation...
+               if (prev_same_thread != NULL &&
+                               (prev_same_thread->get_cv()->getClock(tid) == curr->get_cv()->getClock(tid)) &&
+                               (last_sc_fence_thread_local == NULL || *last_sc_fence_thread_local < *prev_same_thread)) {
+                       continue;
+               }
 
                /* Iterate over actions in thread, starting from most recent */
-               action_list_t *list = &(*thrd_lists)[i];
+               action_list_t *list = &(*thrd_lists)[tid];
                action_list_t::reverse_iterator rit;
                for (rit = list->rbegin();rit != list->rend();rit++) {
                        ModelAction *act = *rit;
@@ -847,6 +985,12 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *
                         * before" curr
                         */
                        if (act->happens_before(curr)) {
+                               if (i==0) {
+                                       if (last_sc_fence_local == NULL ||
+                                                       (*last_sc_fence_local < *act)) {
+                                               prev_same_thread = act;
+                                       }
+                               }
                                if (act->is_write()) {
                                        if (mo_graph->checkReachable(rf, act))
                                                return false;
@@ -857,6 +1001,11 @@ bool ModelExecution::r_modification_order(ModelAction *curr, const ModelAction *
                                                if (mo_graph->checkReachable(rf, prevrf))
                                                        return false;
                                                priorset->push_back(prevrf);
+                                       } else {
+                                               if (act->get_tid() == curr->get_tid()) {
+                                                       //Can prune curr from obj list
+                                                       *canprune = true;
+                                               }
                                        }
                                }
                                break;
@@ -968,21 +1117,6 @@ void ModelExecution::w_modification_order(ModelAction *curr)
                                        mo_graph->addEdge(act->get_reads_from(), curr, force_edge);
                                }
                                break;
-                       } else if (act->is_read() && !act->could_synchronize_with(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.  Note that
-                                  these checks are overly conservative at this point, we'll
-                                  do more checks before actually removing the
-                                  pendingfuturevalue.
-
-                                */
-
                        }
                }
        }
@@ -1026,91 +1160,57 @@ bool ModelExecution::mo_may_allow(const ModelAction *writer, const ModelAction *
 }
 
 /**
- * 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.
+ * Computes the clock vector that happens before propagates from this write.
  *
  * @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 with certainty
- * @return true, if the ModelExecution is certain that release_heads is complete;
- * false otherwise
+ * @return ClockVector of happens before relation.
  */
-bool ModelExecution::release_seq_heads(const ModelAction *rf, rel_heads_list_t *release_heads) const
-{
 
+ClockVector * ModelExecution::get_hb_from_write(ModelAction *rf) const {
+       SnapVector<ModelAction *> * processset = NULL;
        for ( ;rf != NULL;rf = rf->get_reads_from()) {
                ASSERT(rf->is_write());
+               if (!rf->is_rmw() || (rf->is_acquire() && rf->is_release()) || rf->get_rfcv() != NULL)
+                       break;
+               if (processset == NULL)
+                       processset = new SnapVector<ModelAction *>();
+               processset->push_back(rf);
+       }
 
-               if (rf->is_release())
-                       release_heads->push_back(rf);
-               else if (rf->get_last_fence_release())
-                       release_heads->push_back(rf->get_last_fence_release());
-               if (!rf->is_rmw())
-                       break;/* End of RMW chain */
-
-               /** @todo Need to be smarter here...  In the linux lock
-                * example, this will run to the beginning of the program for
-                * every acquire. */
-               /** @todo The way to be smarter here is to keep going until 1
-                * thread has a release preceded by an acquire and you've seen
-                *       both. */
-
-               /* acq_rel RMW is a sufficient stopping condition */
-               if (rf->is_acquire() && rf->is_release())
-                       return true;/* complete */
-       };
-       ASSERT(rf);     // Needs to be real write
-
-       if (rf->is_release())
-               return true;/* complete */
-
-       /* else relaxed write
-        * - check for fence-release in the same thread (29.8, stmt. 3)
-        * - check modification order for contiguous subsequence
-        *   -> rf must be same thread as release */
-
-       const ModelAction *fence_release = rf->get_last_fence_release();
-       /* Synchronize with a fence-release unconditionally; we don't need to
-        * find any more "contiguous subsequence..." for it */
-       if (fence_release)
-               release_heads->push_back(fence_release);
-
-       return true;    /* complete */
-}
-
-/**
- * An 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 ModelExecution so that it will handle
- * the release sequence at a later time, causing @a acquire to update its
- * synchronization at some later point in execution.
- *
- * @param acquire The 'acquire' action that may synchronize with a release
- * sequence
- * @param read The read action that may read from a release sequence; this may
- * be the same as acquire, or else an earlier action in the same thread (i.e.,
- * when 'acquire' is a fence-acquire)
- * @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 ModelExecution::release_seq_heads
- */
-void ModelExecution::get_release_seq_heads(ModelAction *acquire,
-                ModelAction *read, rel_heads_list_t *release_heads)
-{
-       const ModelAction *rf = read->get_reads_from();
+       int i = (processset == NULL) ? 0 : processset->size();
 
-       release_seq_heads(rf, release_heads);
+       ClockVector * vec = NULL;
+       while(true) {
+               if (rf->get_rfcv() != NULL) {
+                       vec = rf->get_rfcv();
+               } else if (rf->is_acquire() && rf->is_release()) {
+                       vec = rf->get_cv();
+               } else if (rf->is_release() && !rf->is_rmw()) {
+                       vec = rf->get_cv();
+               } else if (rf->is_release()) {
+                       //have rmw that is release and doesn't have a rfcv
+                       (vec = new ClockVector(vec, NULL))->merge(rf->get_cv());
+                       rf->set_rfcv(vec);
+               } else {
+                       //operation that isn't release
+                       if (rf->get_last_fence_release()) {
+                               if (vec == NULL)
+                                       vec = rf->get_last_fence_release()->get_cv();
+                               else
+                                       (vec=new ClockVector(vec, NULL))->merge(rf->get_last_fence_release()->get_cv());
+                       }
+                       rf->set_rfcv(vec);
+               }
+               i--;
+               if (i >= 0) {
+                       rf = (*processset)[i];
+               } else
+                       break;
+       }
+       if (processset != NULL)
+               delete processset;
+       return vec;
 }
 
 /**
@@ -1130,13 +1230,19 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
                uninit = get_uninitialized_action(act);
                uninit_id = id_to_int(uninit->get_tid());
                list->push_front(uninit);
+               SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_wr_thrd_map, act->get_location());
+               if (uninit_id >= (int)vec->size())
+                       vec->resize(uninit_id + 1);
+               (*vec)[uninit_id].push_front(uninit);
        }
        list->push_back(act);
 
+       // Update action trace, a total order of all actions
        action_trace.push_back(act);
        if (uninit)
                action_trace.push_front(uninit);
 
+       // Update obj_thrd_map, a per location, per thread, order of actions
        SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
        if (tid >= (int)vec->size())
                vec->resize(priv->next_thread_id);
@@ -1144,12 +1250,14 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
        if (uninit)
                (*vec)[uninit_id].push_front(uninit);
 
+       // Update thrd_last_action, the last action taken by each thrad
        if ((int)thrd_last_action.size() <= tid)
                thrd_last_action.resize(get_num_threads());
        thrd_last_action[tid] = act;
        if (uninit)
                thrd_last_action[uninit_id] = uninit;
 
+       // Update thrd_last_fence_release, the last release fence taken by each thread
        if (act->is_fence() && act->is_release()) {
                if ((int)thrd_last_fence_release.size() <= tid)
                        thrd_last_fence_release.resize(get_num_threads());
@@ -1167,6 +1275,82 @@ void ModelExecution::add_action_to_lists(ModelAction *act)
        }
 }
 
+void insertIntoActionList(action_list_t *list, ModelAction *act) {
+       action_list_t::reverse_iterator rit = list->rbegin();
+       modelclock_t next_seq = act->get_seq_number();
+       if (rit == list->rend() || (*rit)->get_seq_number() == next_seq)
+               list->push_back(act);
+       else {
+               for(;rit != list->rend();rit++) {
+                       if ((*rit)->get_seq_number() == next_seq) {
+                               action_list_t::iterator it = rit.base();
+                               list->insert(it, act);
+                               break;
+                       }
+               }
+       }
+}
+
+void insertIntoActionListAndSetCV(action_list_t *list, ModelAction *act) {
+       action_list_t::reverse_iterator rit = list->rbegin();
+       modelclock_t next_seq = act->get_seq_number();
+       if (rit == list->rend()) {
+               act->create_cv(NULL);
+       } else if ((*rit)->get_seq_number() == next_seq) {
+               act->create_cv((*rit));
+               list->push_back(act);
+       } else {
+               for(;rit != list->rend();rit++) {
+                       if ((*rit)->get_seq_number() == next_seq) {
+                               act->create_cv((*rit));
+                               action_list_t::iterator it = rit.base();
+                               list->insert(it, act);
+                               break;
+                       }
+               }
+       }
+}
+
+/**
+ * Performs various bookkeeping operations for a normal write.  The
+ * complication is that we are typically inserting a normal write
+ * lazily, so we need to insert it into the middle of lists.
+ *
+ * @param act is the ModelAction to add.
+ */
+
+void ModelExecution::add_normal_write_to_lists(ModelAction *act)
+{
+       int tid = id_to_int(act->get_tid());
+       insertIntoActionListAndSetCV(&action_trace, act);
+
+       action_list_t *list = get_safe_ptr_action(&obj_map, act->get_location());
+       insertIntoActionList(list, act);
+
+       // Update obj_thrd_map, a per location, per thread, order of actions
+       SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_thrd_map, act->get_location());
+       if (tid >= (int)vec->size())
+               vec->resize(priv->next_thread_id);
+       insertIntoActionList(&(*vec)[tid],act);
+
+       // Update thrd_last_action, the last action taken by each thrad
+       if (thrd_last_action[tid]->get_seq_number() == act->get_seq_number())
+               thrd_last_action[tid] = act;
+}
+
+
+void ModelExecution::add_write_to_lists(ModelAction *write) {
+       // Update seq_cst map
+       if (write->is_seqcst())
+               obj_last_sc_map.put(write->get_location(), write);
+
+       SnapVector<action_list_t> *vec = get_safe_ptr_vect_action(&obj_wr_thrd_map, write->get_location());
+       int tid = id_to_int(write->get_tid());
+       if (tid >= (int)vec->size())
+               vec->resize(priv->next_thread_id);
+       (*vec)[tid].push_back(write);
+}
+
 /**
  * @brief Get the last action performed by a particular Thread
  * @param tid The thread ID of the Thread in question
@@ -1206,16 +1390,7 @@ ModelAction * ModelExecution::get_last_fence_release(thread_id_t tid) const
 ModelAction * ModelExecution::get_last_seq_cst_write(ModelAction *curr) const
 {
        void *location = curr->get_location();
-       action_list_t *list = obj_map.get(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) != curr;rit++)
-               ;
-       rit++;  /* Skip past curr */
-       for ( ;rit != list->rend();rit++)
-               if ((*rit)->is_write() && (*rit)->is_seqcst())
-                       return *rit;
-       return NULL;
+       return obj_last_sc_map.get(location);
 }
 
 /**
@@ -1287,7 +1462,8 @@ ModelAction * ModelExecution::get_parent_action(thread_id_t tid) const
  */
 ClockVector * ModelExecution::get_cv(thread_id_t tid) const
 {
-       return get_parent_action(tid)->get_cv();
+       ModelAction *firstaction=get_parent_action(tid);
+       return firstaction != NULL ? firstaction->get_cv() : NULL;
 }
 
 bool valequals(uint64_t val1, uint64_t val2, int size) {
@@ -1315,7 +1491,7 @@ bool valequals(uint64_t val1, uint64_t val2, int size) {
  */
 SnapVector<ModelAction *> *  ModelExecution::build_may_read_from(ModelAction *curr)
 {
-       SnapVector<action_list_t> *thrd_lists = obj_thrd_map.get(curr->get_location());
+       SnapVector<action_list_t> *thrd_lists = obj_wr_thrd_map.get(curr->get_location());
        unsigned int i;
        ASSERT(curr->is_read());
 
@@ -1334,8 +1510,7 @@ SnapVector<ModelAction *> *  ModelExecution::build_may_read_from(ModelAction *cu
                for (rit = list->rbegin();rit != list->rend();rit++) {
                        ModelAction *act = *rit;
 
-                       /* Only consider 'write' actions */
-                       if (!act->is_write() || act == curr)
+                       if (act == curr)
                                continue;
 
                        /* Don't consider more than one seq_cst write if we are a seq_cst read. */
@@ -1380,18 +1555,17 @@ SnapVector<ModelAction *> *  ModelExecution::build_may_read_from(ModelAction *cu
 /**
  * @brief Get an action representing an uninitialized atomic
  *
- * This function may create a new one or try to retrieve one from the NodeStack
+ * This function may create a new one.
  *
  * @param curr The current action, which prompts the creation of an UNINIT action
  * @return A pointer to the UNINIT ModelAction
  */
-ModelAction * ModelExecution::get_uninitialized_action(const ModelAction *curr) const
+ModelAction * ModelExecution::get_uninitialized_action(ModelAction *curr) const
 {
-       Node *node = curr->get_node();
-       ModelAction *act = node->get_uninit_action();
+       ModelAction *act = curr->get_uninit_action();
        if (!act) {
                act = new ModelAction(ATOMIC_UNINIT, std::memory_order_relaxed, curr->get_location(), params->uninitvalue, model_thread);
-               node->set_uninit_action(act);
+               curr->set_uninit_action(act);
        }
        act->create_cv(NULL);
        return act;
@@ -1567,18 +1741,6 @@ Thread * ModelExecution::action_select_next_thread(const ModelAction *curr) cons
        /* Do not split atomic RMW */
        if (curr->is_rmwr())
                return get_thread(curr);
-       if (curr->is_write()) {
-               std::memory_order order = curr->get_mo();
-               switch(order) {
-               case std::memory_order_relaxed:
-                       return get_thread(curr);
-               case std::memory_order_release:
-                       return get_thread(curr);
-               default:
-                       return NULL;
-               }
-       }
-
        /* Follow CREATE with the created thread */
        /* which is not needed, because model.cc takes care of this */
        if (curr->get_type() == THREAD_CREATE)
@@ -1604,6 +1766,9 @@ Thread * ModelExecution::take_step(ModelAction *curr)
        curr = check_current_action(curr);
        ASSERT(curr);
 
+       /* Process this action in ModelHistory for records*/
+       model->get_history()->process_action( curr, curr_thrd->get_id() );
+
        if (curr_thrd->is_blocked() || curr_thrd->is_complete())
                scheduler->remove_thread(curr_thrd);