Logo AND Algorithmique Numérique Distribuée

Public GIT Repository
model-checker : deal with the soundness of DPOR only if DPOR is enabled
[simgrid.git] / src / mc / mc_dpor.c
index a3e2949..6c22b82 100644 (file)
@@ -1,4 +1,4 @@
-/* Copyright (c) 2008-2013. The SimGrid Team.
+/* Copyright (c) 2008-2014. The SimGrid Team.
  * All rights reserved.                                                     */
 
 /* This program is free software; you can redistribute it and/or modify it
@@ -14,6 +14,7 @@ XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_dpor, mc,
 xbt_dynar_t visited_states;
 xbt_dict_t first_enabled_state;
 xbt_dynar_t initial_communications_pattern;
+xbt_dynar_t incomplete_communications_pattern;
 xbt_dynar_t communications_pattern;
 int nb_comm_pattern;
 
@@ -82,6 +83,8 @@ static void deterministic_pattern(xbt_dynar_t initial_pattern, xbt_dynar_t patte
         comm_comparison = compare_comm_pattern(comm1, comm2);
         if(comm_comparison == 1){
           initial_state_safety->comm_deterministic = 0;
+          if(!_sg_mc_send_determinism)
+            return;
         }
         recv_index++;
       }
@@ -92,24 +95,28 @@ static void deterministic_pattern(xbt_dynar_t initial_pattern, xbt_dynar_t patte
     recv_index = 0;
     current_process++;
   }
-  // XBT_DEBUG("Communication-deterministic : %d, Send-deterministic : %d", initial_state_safety->comm_deterministic, initial_state_safety->send_deterministic);
 }
 
 static int complete_comm_pattern(xbt_dynar_t list, mc_comm_pattern_t pattern){
   mc_comm_pattern_t current_pattern;
   unsigned int cursor = 0;
-  xbt_dynar_foreach(list, cursor, current_pattern){
+  int index;
+  xbt_dynar_foreach(incomplete_communications_pattern, cursor, index){
+    current_pattern = (mc_comm_pattern_t)xbt_dynar_get_as(list, index, mc_comm_pattern_t);
     if(current_pattern->comm == pattern->comm){
-      if(!current_pattern->completed){
-        current_pattern->src_proc = pattern->comm->comm.src_proc->pid;
-        current_pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+      current_pattern->src_proc = pattern->comm->comm.src_proc->pid;
+      current_pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+      current_pattern->src_host = simcall_host_get_name(pattern->comm->comm.src_proc->smx_host);
+      current_pattern->dst_host = simcall_host_get_name(pattern->comm->comm.dst_proc->smx_host);
+      if(current_pattern->data_size == -1){
         current_pattern->data_size = pattern->comm->comm.src_buff_size;
         current_pattern->data = xbt_malloc0(current_pattern->data_size);
-        current_pattern->matched_comm = pattern->num;
         memcpy(current_pattern->data, current_pattern->comm->comm.src_buff, current_pattern->data_size);
-        current_pattern->completed = 1;
-        return current_pattern->num;
       }
+      current_pattern->matched_comm = pattern->num;
+      current_pattern->completed = 1;
+      xbt_dynar_remove_at(incomplete_communications_pattern, cursor, NULL);
+      return current_pattern->num;
     }
   }
   return -1;
@@ -120,16 +127,18 @@ void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, int call){
   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
   pattern->num = ++nb_comm_pattern;
   pattern->completed = 0;
+  pattern->data_size = -1;
   if(call == 1){ // ISEND
     pattern->comm = simcall_comm_isend__get__result(request);
     pattern->type = SIMIX_COMM_SEND;
-    if(pattern->comm->comm.dst_proc != NULL){
+    if(pattern->comm->comm.dst_proc != NULL){ 
       pattern->matched_comm = complete_comm_pattern(list, pattern);
       pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+      pattern->dst_host = simcall_host_get_name(pattern->comm->comm.dst_proc->smx_host);
       pattern->completed = 1;
     }
     pattern->src_proc = pattern->comm->comm.src_proc->pid;
+    pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
     pattern->data_size = pattern->comm->comm.src_buff_size;
     pattern->data=xbt_malloc0(pattern->data_size);
     memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
@@ -139,25 +148,36 @@ void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, int call){
     if(pattern->comm->comm.src_proc != NULL){
       pattern->matched_comm = complete_comm_pattern(list, pattern);
       pattern->src_proc = pattern->comm->comm.src_proc->pid;
+      pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
       pattern->completed = 1;
       pattern->data_size = pattern->comm->comm.src_buff_size;
       pattern->data=xbt_malloc0(pattern->data_size);
       memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
     }
     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
+    pattern->dst_host = simcall_host_get_name(pattern->comm->comm.dst_proc->smx_host);
   }
+  
   if(pattern->comm->comm.rdv != NULL)
     pattern->rdv = strdup(pattern->comm->comm.rdv->name);
   else
     pattern->rdv = strdup(pattern->comm->comm.rdv_cpy->name);
+  
   xbt_dynar_push(list, &pattern);
+
+  if(!pattern->completed)
+    xbt_dynar_push_as(incomplete_communications_pattern, int, xbt_dynar_length(list) - 1);
+
 }
 
 static void print_communications_pattern(xbt_dynar_t comms_pattern){
   unsigned int cursor = 0;
   mc_comm_pattern_t current_comm;
   xbt_dynar_foreach(comms_pattern, cursor, current_comm){
-    // fprintf(stderr, "%s (%d - comm %p, src : %lu, dst %lu, rdv name %s, data %p, matched with %d)\n", current_comm->type == SIMIX_COMM_SEND ? "iSend" : "iRecv", current_comm->num, current_comm->comm, current_comm->src_proc, current_comm->dst_proc, current_comm->rdv, current_comm->data, current_comm->matched_comm);
+    if(current_comm->type == SIMIX_COMM_SEND)
+      XBT_INFO("[(%lu) %s -> %s] %s ", current_comm->src_proc, current_comm->src_host, current_comm->dst_host, "iSend");
+    else
+      XBT_INFO("[(%lu) %s <- %s] %s ", current_comm->dst_proc, current_comm->dst_host, current_comm->src_host, "iRecv");
   }
 }
 
@@ -172,6 +192,10 @@ static void visited_state_free_voidp(void *s){
   visited_state_free((mc_visited_state_t) * (void **) s);
 }
 
+/** \brief Save the current state
+ *
+ *  \return Snapshot of the current state.
+ */
 static mc_visited_state_t visited_state_new(){
 
   mc_visited_state_t new_state = NULL;
@@ -186,7 +210,25 @@ static mc_visited_state_t visited_state_new(){
   
 }
 
+/** \brief Find a suitable subrange of candidate duplicates for a given state
+ *
+ *  \param all_ pairs dynamic array of states with candidate duplicates of the current state;
+ *  \param pair current state;
+ *  \param min (output) index of the beginning of the the subrange
+ *  \param max (output) index of the enf of the subrange
+ *
+ *  Given a suitably ordered array of state, this function extracts a subrange
+ *  (with index *min <= i <= *max) with candidate duplicates of the given state.
+ *  This function uses only fast discriminating criterions and does not use the
+ *  full state comparison algorithms.
+ *
+ *  The states in all_pairs MUST be ordered using a (given) weak order
+ *  (based on nb_processes and heap_bytes_used).
+ *  The subrange is the subrange of "equivalence" of the given state.
+ */
 static int get_search_interval(xbt_dynar_t all_states, mc_visited_state_t state, int *min, int *max){
+  XBT_VERB("Searching interval for state %i: nd_processes=%zu heap_bytes_used=%zu",
+    state->num, (size_t)state->nb_processes, (size_t)state->heap_bytes_used);
 
   int raw_mem_set = (mmalloc_get_current_heap() == raw_heap);
 
@@ -240,6 +282,10 @@ static int get_search_interval(xbt_dynar_t all_states, mc_visited_state_t state,
   return cursor;
 }
 
+/** \brief Take a snapshot the current state and process it.
+ *
+ *  \return number of the duplicate state or -1 (not visited)
+ */
 static int is_visited_state(){
 
   if(_sg_mc_visited == 0)
@@ -270,6 +316,8 @@ static int is_visited_state(){
     index = get_search_interval(visited_states, new_state, &min, &max);
 
     if(min != -1 && max != -1){
+
+      // Parallell implementation
       /*res = xbt_parmap_mc_apply(parmap, snapshot_compare, xbt_dynar_get_ptr(visited_states, min), (max-min)+1, new_state);
       if(res != -1){
         state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, (min+res)-1, mc_visited_state_t);
@@ -287,10 +335,13 @@ static int is_visited_state(){
           MC_UNSET_RAW_MEM;
         return new_state->other_num;
         }*/
+
       cursor = min;
       while(cursor <= max){
         state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, cursor, mc_visited_state_t);
         if(snapshot_compare(state_test, new_state) == 0){
+          // The state has been visited:
+
           if(state_test->other_num == -1)
             new_state->other_num = state_test->num;
           else
@@ -299,16 +350,24 @@ static int is_visited_state(){
             XBT_DEBUG("State %d already visited ! (equal to state %d)", new_state->num, state_test->num);
           else
             XBT_DEBUG("State %d already visited ! (equal to state %d (state %d in dot_output))", new_state->num, state_test->num, new_state->other_num);
+
+          // Replace the old state with the new one (why?):
           xbt_dynar_remove_at(visited_states, cursor, NULL);
           xbt_dynar_insert_at(visited_states, cursor, &new_state);
+
           if(!raw_mem_set)
             MC_UNSET_RAW_MEM;
           return new_state->other_num;
         }
         cursor++;
       }
+
+      // The state has not been visited, add it to the list:
       xbt_dynar_insert_at(visited_states, min, &new_state);
+
     }else{
+
+      // The state has not been visited: insert the state in the dynamic array.
       state_test = (mc_visited_state_t)xbt_dynar_get_as(visited_states, index, mc_visited_state_t);
       if(state_test->nb_processes < new_state->nb_processes){
         xbt_dynar_insert_at(visited_states, index+1, &new_state);
@@ -318,9 +377,13 @@ static int is_visited_state(){
         else
           xbt_dynar_insert_at(visited_states, index, &new_state);
       }
+
     }
 
+    // We have reached the maximum number of stored states;
     if(xbt_dynar_length(visited_states) > _sg_mc_visited){
+
+      // Find the (index of the) older state:
       int min2 = mc_stats->expanded_states;
       unsigned int cursor2 = 0;
       unsigned int index2 = 0;
@@ -330,6 +393,8 @@ static int is_visited_state(){
           min2 = state_test->num;
         }
       }
+
+      // and drop it:
       xbt_dynar_remove_at(visited_states, index2, NULL);
     }
 
@@ -358,11 +423,15 @@ void MC_dpor_init()
   if(_sg_mc_visited > 0)
     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 
-  first_enabled_state = xbt_dict_new_homogeneous(&xbt_free_f);
+  if(mc_reduce_kind == e_mc_reduce_dpor)
+    first_enabled_state = xbt_dict_new_homogeneous(&xbt_free_f);
 
-  initial_communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
-  communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
-  nb_comm_pattern = 0;
+  if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
+    initial_communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
+    communications_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
+    incomplete_communications_pattern = xbt_dynar_new(sizeof(int), NULL);
+    nb_comm_pattern = 0;
+  }
 
   initial_state = MC_state_new();
 
@@ -390,16 +459,18 @@ void MC_dpor_init()
 
   xbt_fifo_unshift(mc_stack_safety, initial_state);
 
-  /* To ensure the soundness of DPOR, we have to keep a list of 
-     processes which are still enabled at each step of the exploration. 
-     If max depth is reached, we interleave them in the state in which they have 
-     been enabled for the first time. */
-  xbt_swag_foreach(process, simix_global->process_list){
-    if(MC_process_is_enabled(process)){
-      char *key = bprintf("%lu", process->pid);
-      char *data = bprintf("%d", xbt_fifo_size(mc_stack_safety));
-      xbt_dict_set(first_enabled_state, key, data, NULL);
-      xbt_free(key);
+  if(mc_reduce_kind == e_mc_reduce_dpor){
+    /* To ensure the soundness of DPOR, we have to keep a list of 
+       processes which are still enabled at each step of the exploration. 
+       If max depth is reached, we interleave them in the state in which they have 
+       been enabled for the first time. */
+    xbt_swag_foreach(process, simix_global->process_list){
+      if(MC_process_is_enabled(process)){
+        char *key = bprintf("%lu", process->pid);
+        char *data = bprintf("%d", xbt_fifo_size(mc_stack_safety));
+        xbt_dict_set(first_enabled_state, key, data, NULL);
+        xbt_free(key);
+      }
     }
   }
 
@@ -413,9 +484,8 @@ void MC_dpor_init()
 }
 
 
-/**
- *   \brief Perform the model-checking operation using a depth-first search exploration
- *         with Dynamic Partial Order Reductions
+/** \brief Model-check the application using a DFS exploration
+ *         with DPOR (Dynamic Partial Order Reductions)
  */
 void MC_dpor(void)
 {
@@ -469,13 +539,15 @@ void MC_dpor(void)
       MC_state_set_executed_request(state, req, value);
       mc_stats->executed_transitions++;
 
-      MC_SET_RAW_MEM;
-      char *key = bprintf("%lu", req->issuer->pid);
-      xbt_dict_remove(first_enabled_state, key); 
-      xbt_free(key);
-      MC_UNSET_RAW_MEM;
-      
-      if(_sg_mc_comms_determinism){
+      if(mc_reduce_kind ==  e_mc_reduce_dpor){
+        MC_SET_RAW_MEM;
+        char *key = bprintf("%lu", req->issuer->pid);
+        xbt_dict_remove(first_enabled_state, key); 
+        xbt_free(key);
+        MC_UNSET_RAW_MEM;
+      }
+
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
         if(req->call == SIMCALL_COMM_ISEND)
           comm_pattern = 1;
         else if(req->call == SIMCALL_COMM_IRECV)
@@ -485,7 +557,7 @@ void MC_dpor(void)
       /* Answer the request */
       SIMIX_simcall_pre(req, value); /* After this call req is no longer usefull */
 
-      if(_sg_mc_comms_determinism){
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
         MC_SET_RAW_MEM;
         if(comm_pattern != 0){
           if(!initial_state_safety->initial_communications_pattern_done)
@@ -532,15 +604,17 @@ void MC_dpor(void)
 
       xbt_fifo_unshift(mc_stack_safety, next_state);
 
-      /* Insert in dict all enabled processes, if not included yet */
-      xbt_swag_foreach(process, simix_global->process_list){
-        if(MC_process_is_enabled(process)){
-          char *key = bprintf("%lu", process->pid);
-          if(xbt_dict_get_or_null(first_enabled_state, key) == NULL){
-            char *data = bprintf("%d", xbt_fifo_size(mc_stack_safety));
-            xbt_dict_set(first_enabled_state, key, data, NULL); 
+      if(mc_reduce_kind ==  e_mc_reduce_dpor){
+        /* Insert in dict all enabled processes, if not included yet */
+        xbt_swag_foreach(process, simix_global->process_list){
+          if(MC_process_is_enabled(process)){
+            char *key = bprintf("%lu", process->pid);
+            if(xbt_dict_get_or_null(first_enabled_state, key) == NULL){
+              char *data = bprintf("%d", xbt_fifo_size(mc_stack_safety));
+              xbt_dict_set(first_enabled_state, key, data, NULL); 
+            }
+            xbt_free(key);
           }
-          xbt_free(key);
         }
       }
       
@@ -563,19 +637,21 @@ void MC_dpor(void)
 
         visited_state = -1;
 
-        /* Interleave enabled processes in the state in which they have been enabled for the first time */
-        xbt_swag_foreach(process, simix_global->process_list){
-          if(MC_process_is_enabled(process)){
-            char *key = bprintf("%lu", process->pid);
-            enabled = (int)strtoul(xbt_dict_get_or_null(first_enabled_state, key), 0, 10);
-            xbt_free(key);
-            int cursor = xbt_fifo_size(mc_stack_safety);
-            xbt_fifo_foreach(mc_stack_safety, item, state_test, mc_state_t){
-              if(cursor-- == enabled){ 
-                if(!MC_state_process_is_done(state_test, process) && state_test->num != state->num){ 
-                  XBT_DEBUG("Interleave process %lu in state %d", process->pid, state_test->num);
-                  MC_state_interleave_process(state_test, process);
-                  break;
+        if(mc_reduce_kind ==  e_mc_reduce_dpor){
+          /* Interleave enabled processes in the state in which they have been enabled for the first time */
+          xbt_swag_foreach(process, simix_global->process_list){
+            if(MC_process_is_enabled(process)){
+              char *key = bprintf("%lu", process->pid);
+              enabled = (int)strtoul(xbt_dict_get_or_null(first_enabled_state, key), 0, 10);
+              xbt_free(key);
+              int cursor = xbt_fifo_size(mc_stack_safety);
+              xbt_fifo_foreach(mc_stack_safety, item, state_test, mc_state_t){
+                if(cursor-- == enabled){ 
+                  if(!MC_state_process_is_done(state_test, process) && state_test->num != state->num){ 
+                    XBT_DEBUG("Interleave process %lu in state %d", process->pid, state_test->num);
+                    MC_state_interleave_process(state_test, process);
+                    break;
+                  }
                 }
               }
             }
@@ -590,16 +666,36 @@ void MC_dpor(void)
 
       MC_SET_RAW_MEM;
 
-      if(_sg_mc_comms_determinism){
-        if(!initial_state_safety->initial_communications_pattern_done){
-          //print_communications_pattern(initial_communications_pattern);
-        }else{
+      if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
+        if(initial_state_safety->initial_communications_pattern_done){
           if(interleave_size == 0){ /* if (interleave_size > 0), process interleaved but not enabled => "incorrect" path, determinism not evaluated */
             //print_communications_pattern(communications_pattern);
             deterministic_pattern(initial_communications_pattern, communications_pattern);
+            if(initial_state_safety->comm_deterministic == 0 && _sg_mc_comms_determinism){
+              XBT_INFO("****************************************************");
+              XBT_INFO("***** Non-deterministic communications pattern *****");
+              XBT_INFO("****************************************************");
+              XBT_INFO("Initial communications pattern:");
+              print_communications_pattern(initial_communications_pattern);
+              XBT_INFO("Communications pattern counter-example:");
+              print_communications_pattern(communications_pattern);
+              MC_print_statistics(mc_stats);
+              return;
+            }else if(initial_state_safety->send_deterministic == 0 && _sg_mc_send_determinism){
+              XBT_INFO("****************************************************");
+              XBT_INFO("***** Non-send-deterministic communications pattern *****");
+              XBT_INFO("****************************************************");
+              XBT_INFO("Initial communications pattern:");
+              print_communications_pattern(initial_communications_pattern);
+              XBT_INFO("Communications pattern counter-example:");
+              print_communications_pattern(communications_pattern);
+              MC_print_statistics(mc_stats);
+              return;
+            }
           }
+        }else{
+          initial_state_safety->initial_communications_pattern_done = 1;
         }
-        initial_state_safety->initial_communications_pattern_done = 1;
       }
 
       /* Trash the current state, no longer needed */
@@ -693,13 +789,13 @@ void MC_dpor(void)
           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack_safety));
           break;
         } else {
-          req = MC_state_get_internal_request(state);
-          if(_sg_mc_comms_determinism){
+          /*req = MC_state_get_internal_request(state);
+          if(_sg_mc_comms_determinism || _sg_mc_send_determinism){
             if(req->call == SIMCALL_COMM_ISEND || req->call == SIMCALL_COMM_IRECV){
               if(!xbt_dynar_is_empty(communications_pattern))
                 xbt_dynar_remove_at(communications_pattern, xbt_dynar_length(communications_pattern) - 1, NULL);
             }
-          }
+            }*/
           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack_safety) + 1); 
           MC_state_delete(state);
         }