src/mc/mc_comm_determinism.c

   1 /* Copyright (c) 2008-2014. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include "mc_state.h"
   8 #include "mc_comm_pattern.h"
   9 #include "mc_request.h"
  10 #include "mc_safety.h"
  11 #include "mc_private.h"
  12 #include "mc_record.h"
  13
  14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  15                                 "Logging specific to MC communication determinism detection");
  16
  17 /********** Global variables **********/
  18
  19 xbt_dynar_t initial_communications_pattern;
  20 xbt_dynar_t incomplete_communications_pattern;
  21
  22 /********** Static functions ***********/
  23
  24 static void comm_pattern_free(mc_comm_pattern_t p)
  25 {
  26   xbt_free(p->rdv);
  27   xbt_free(p->data);
  28   xbt_free(p);
  29   p = NULL;
  30 }
  31
  32 static void list_comm_pattern_free(mc_list_comm_pattern_t l)
  33 {
  34   xbt_dynar_free(&(l->list));
  35   xbt_free(l);
  36   l = NULL;
  37 }
  38
  39 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  40   if(comm1->type != comm2->type)
  41     return TYPE_DIFF;
  42   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  43     return RDV_DIFF;
  44   if (comm1->src_proc != comm2->src_proc)
  45     return SRC_PROC_DIFF;
  46   if (comm1->dst_proc != comm2->dst_proc)
  47     return DST_PROC_DIFF;
  48   if (comm1->tag != comm2->tag)
  49     return TAG_DIFF;
  50   if (comm1->data_size != comm2->data_size)
  51     return DATA_SIZE_DIFF;
  52   if(comm1->data == NULL && comm2->data == NULL)
  53     return 0;
  54   if(comm1->data != NULL && comm2->data !=NULL) {
  55     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  56       return 0;
  57     return DATA_DIFF;
  58   }else{
  59     return DATA_DIFF;
  60   }
  61   return 0;
  62 }
  63
  64 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  65   char *type, *res;
  66
  67   if(comm->type == SIMIX_COMM_SEND)
  68     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  69   else
  70     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  71
  72   switch(diff) {
  73   case TYPE_DIFF:
  74     res = bprintf("%s Different type for communication #%d", type, cursor);
  75     break;
  76   case RDV_DIFF:
  77     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  78     break;
  79   case TAG_DIFF:
  80     res = bprintf("%s Different tag for communication #%d", type, cursor);
  81     break;
  82   case SRC_PROC_DIFF:
  83       res = bprintf("%s Different source for communication #%d", type, cursor);
  84     break;
  85   case DST_PROC_DIFF:
  86       res = bprintf("%s Different destination for communication #%d", type, cursor);
  87     break;
  88   case DATA_SIZE_DIFF:
  89     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  90     break;
  91   case DATA_DIFF:
  92     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  93     break;
  94   default:
  95     res = NULL;
  96     break;
  97   }
  98
  99   return res;
 100 }
 101
 102 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
 103 {
 104   void *addr_pointed;
 105   comm_pattern->src_proc = comm->comm.src_proc->pid;
 106   comm_pattern->dst_proc = comm->comm.dst_proc->pid;
 107   comm_pattern->src_host = simcall_host_get_name(comm->comm.src_proc->smx_host);
 108   comm_pattern->dst_host = simcall_host_get_name(comm->comm.dst_proc->smx_host);
 109   if (comm_pattern->data_size == -1 && comm->comm.src_buff != NULL) {
 110     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 111     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 112     addr_pointed = *(void **) comm->comm.src_buff;
 113     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 114       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 115     else
 116       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 117   }
 118 }
 119
 120 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 121
 122   mc_list_comm_pattern_t list_comm_pattern = (mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, process, mc_list_comm_pattern_t);
 123
 124   if(!backtracking){
 125     mc_comm_pattern_t initial_comm = xbt_dynar_get_as(list_comm_pattern->list, list_comm_pattern->index_comm, mc_comm_pattern_t);
 126     e_mc_comm_pattern_difference_t diff;
 127
 128     if((diff = compare_comm_pattern(initial_comm, comm)) != NONE_DIFF){
 129       if (comm->type == SIMIX_COMM_SEND){
 130         initial_global_state->send_deterministic = 0;
 131         if(initial_global_state->send_diff != NULL)
 132           xbt_free(initial_global_state->send_diff);
 133         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 134       }else{
 135         initial_global_state->recv_deterministic = 0;
 136         if(initial_global_state->recv_diff != NULL)
 137           xbt_free(initial_global_state->recv_diff);
 138         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 139       }
 140       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 141         XBT_INFO("*********************************************************");
 142         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 143         XBT_INFO("*********************************************************");
 144         XBT_INFO("%s", initial_global_state->send_diff);
 145         xbt_free(initial_global_state->send_diff);
 146         initial_global_state->send_diff = NULL;
 147         MC_print_statistics(mc_stats);
 148         xbt_abort();
 149       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 150         XBT_INFO("****************************************************");
 151         XBT_INFO("***** Non-deterministic communications pattern *****");
 152         XBT_INFO("****************************************************");
 153         XBT_INFO("%s", initial_global_state->send_diff);
 154         XBT_INFO("%s", initial_global_state->recv_diff);
 155         xbt_free(initial_global_state->send_diff);
 156         initial_global_state->send_diff = NULL;
 157         xbt_free(initial_global_state->recv_diff);
 158         initial_global_state->recv_diff = NULL;
 159         MC_print_statistics(mc_stats);
 160         xbt_abort();
 161       }
 162     }
 163   }
 164
 165   comm_pattern_free(comm);
 166
 167 }
 168
 169 /********** Non Static functions ***********/
 170
 171 void comm_pattern_free_voidp(void *p) {
 172   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
 173 }
 174
 175 void list_comm_pattern_free_voidp(void *p) {
 176   list_comm_pattern_free((mc_list_comm_pattern_t) * (void **) p);
 177 }
 178
 179 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 180 {
 181
 182   mc_comm_pattern_t pattern = NULL;
 183   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 184   pattern->data_size = -1;
 185   pattern->data = NULL;
 186
 187   void *addr_pointed;
 188
 189   if (call_type == MC_CALL_TYPE_SEND) {
 190     /* Create comm pattern */
 191     pattern->type = SIMIX_COMM_SEND;
 192     pattern->comm = simcall_comm_isend__get__result(request);
 193     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 194     pattern->src_proc = pattern->comm->comm.src_proc->pid;
 195     pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
 196     pattern->tag = ((MPI_Request)simcall_comm_isend__get__data(request))->tag;
 197     if(pattern->comm->comm.src_buff != NULL){
 198       pattern->data_size = pattern->comm->comm.src_buff_size;
 199       pattern->data = xbt_malloc0(pattern->data_size);
 200       addr_pointed = *(void **) pattern->comm->comm.src_buff;
 201       if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 202         memcpy(pattern->data, addr_pointed, pattern->data_size);
 203       else
 204         memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 205     }
 206     if(((MPI_Request)simcall_comm_isend__get__data(request))->detached){
 207       if (!initial_global_state->initial_communications_pattern_done) {
 208         /* Store comm pattern */
 209         xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->list, &pattern);
 210       } else {
 211         /* Evaluate comm determinism */
 212         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 213         ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->index_comm++;
 214       }
 215       return;
 216     }
 217   } else if (call_type == MC_CALL_TYPE_RECV) {
 218     pattern->type = SIMIX_COMM_RECEIVE;
 219     pattern->comm = simcall_comm_irecv__get__result(request);
 220     pattern->tag = ((MPI_Request)simcall_comm_irecv__get__data(request))->tag;
 221     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 222     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
 223     pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
 224   } else {
 225     xbt_die("Unexpected call_type %i", (int) call_type);
 226   }
 227
 228   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, request->issuer->pid, xbt_dynar_t), &pattern);
 229
 230   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, request->issuer->pid);
 231 }
 232
 233 void complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm, unsigned int issuer, int backtracking) {
 234
 235   mc_comm_pattern_t current_comm_pattern;
 236   unsigned int cursor = 0;
 237   mc_comm_pattern_t comm_pattern;
 238   int completed = 0;
 239
 240   /* Complete comm pattern */
 241   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 242     if (current_comm_pattern-> comm == comm) {
 243       update_comm_pattern(current_comm_pattern, comm);
 244       completed = 1;
 245       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &comm_pattern);
 246       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 247       break;
 248     }
 249   }
 250   if(!completed)
 251     xbt_die("Corresponding communication not found!");
 252
 253   if (!initial_global_state->initial_communications_pattern_done) {
 254     /* Store comm pattern */
 255     xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->list, &comm_pattern);
 256   } else {
 257     /* Evaluate comm determinism */
 258     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 259     ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->index_comm++;
 260   }
 261 }
 262
 263
 264 /************************ Main algorithm ************************/
 265
 266 void MC_pre_modelcheck_comm_determinism(void)
 267 {
 268
 269   int mc_mem_set = (mmalloc_get_current_heap() == mc_heap);
 270
 271   mc_state_t initial_state = NULL;
 272   smx_process_t process;
 273   int i;
 274
 275   if (!mc_mem_set)
 276     MC_SET_MC_HEAP;
 277
 278   if (_sg_mc_visited > 0)
 279     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 280
 281   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), list_comm_pattern_free_voidp);
 282   for (i=0; i<simix_process_maxpid; i++){
 283     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 284     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 285     process_list_pattern->index_comm = 0;
 286     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 287   }
 288   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 289   for (i=0; i<simix_process_maxpid; i++){
 290     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 291     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 292   }
 293
 294   initial_state = MC_state_new();
 295   MC_SET_STD_HEAP;
 296
 297   XBT_DEBUG("********* Start communication determinism verification *********");
 298
 299   /* Wait for requests (schedules processes) */
 300   MC_wait_for_requests();
 301
 302   MC_SET_MC_HEAP;
 303
 304   /* Get an enabled process and insert it in the interleave set of the initial state */
 305   xbt_swag_foreach(process, simix_global->process_list) {
 306     if (MC_process_is_enabled(process)) {
 307       MC_state_interleave_process(initial_state, process);
 308     }
 309   }
 310
 311   xbt_fifo_unshift(mc_stack, initial_state);
 312
 313   MC_SET_STD_HEAP;
 314
 315 }
 316
 317 void MC_modelcheck_comm_determinism(void)
 318 {
 319
 320   char *req_str = NULL;
 321   int value;
 322   mc_visited_state_t visited_state = NULL;
 323   smx_simcall_t req = NULL;
 324   smx_process_t process = NULL;
 325   mc_state_t state = NULL, next_state = NULL;
 326
 327   while (xbt_fifo_size(mc_stack) > 0) {
 328
 329     /* Get current state */
 330     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 331
 332     XBT_DEBUG("**************************************************");
 333     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 334               xbt_fifo_size(mc_stack), state->num,
 335               MC_state_interleave_size(state));
 336
 337     /* Update statistics */
 338     mc_stats->visited_states++;
 339
 340     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 341         && (req = MC_state_get_request(state, &value))
 342         && (visited_state == NULL)) {
 343
 344       req_str = MC_request_to_string(req, value);
 345       XBT_DEBUG("Execute: %s", req_str);
 346       xbt_free(req_str);
 347
 348       if (dot_output != NULL) {
 349         MC_SET_MC_HEAP;
 350         req_str = MC_request_get_dot_output(req, value);
 351         MC_SET_STD_HEAP;
 352       }
 353
 354       MC_state_set_executed_request(state, req, value);
 355       mc_stats->executed_transitions++;
 356
 357       /* TODO : handle test and testany simcalls */
 358       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 359       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 360         call = mc_get_call_type(req);
 361       }
 362
 363       /* Answer the request */
 364       SIMIX_simcall_handle(req, value);    /* After this call req is no longer useful */
 365
 366       MC_SET_MC_HEAP;
 367       if(!initial_global_state->initial_communications_pattern_done)
 368         handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 369       else
 370         handle_comm_pattern(call, req, value, NULL, 0);
 371       MC_SET_STD_HEAP;
 372
 373       /* Wait for requests (schedules processes) */
 374       MC_wait_for_requests();
 375
 376       /* Create the new expanded state */
 377       MC_SET_MC_HEAP;
 378
 379       next_state = MC_state_new();
 380
 381       if ((visited_state = is_visited_state(next_state)) == NULL) {
 382
 383         /* Get enabled processes and insert them in the interleave set of the next state */
 384         xbt_swag_foreach(process, simix_global->process_list) {
 385           if (MC_process_is_enabled(process)) {
 386             MC_state_interleave_process(next_state, process);
 387           }
 388         }
 389
 390         if (dot_output != NULL)
 391           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 392
 393       } else {
 394
 395         if (dot_output != NULL)
 396           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 397
 398       }
 399
 400       xbt_fifo_unshift(mc_stack, next_state);
 401
 402       if (dot_output != NULL)
 403         xbt_free(req_str);
 404
 405       MC_SET_STD_HEAP;
 406
 407     } else {
 408
 409       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 410         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 411       } else if (visited_state != NULL) {
 412         XBT_DEBUG("State already visited (equal to state %d), exploration stopped on this path.", visited_state->other_num == -1 ? visited_state->num : visited_state->other_num);
 413       } else {
 414         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 415       }
 416
 417       MC_SET_MC_HEAP;
 418
 419       if (!initial_global_state->initial_communications_pattern_done)
 420         initial_global_state->initial_communications_pattern_done = 1;
 421
 422       /* Trash the current state, no longer needed */
 423       xbt_fifo_shift(mc_stack);
 424       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 425       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 426
 427       MC_SET_STD_HEAP;
 428
 429       visited_state = NULL;
 430
 431       /* Check for deadlocks */
 432       if (MC_deadlock_check()) {
 433         MC_show_deadlock(NULL);
 434         return;
 435       }
 436
 437       MC_SET_MC_HEAP;
 438
 439       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 440         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 441           /* We found a back-tracking point, let's loop */
 442           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 443           xbt_fifo_unshift(mc_stack, state);
 444           MC_SET_STD_HEAP;
 445
 446           MC_replay(mc_stack);
 447
 448           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 449
 450           break;
 451         } else {
 452           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 453           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 454         }
 455       }
 456
 457       MC_SET_STD_HEAP;
 458     }
 459   }
 460
 461   MC_print_statistics(mc_stats);
 462   MC_SET_STD_HEAP;
 463
 464   return;
 465 }