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 #include "mc_smx.h"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  16                                 "Logging specific to MC communication determinism detection");
  17
  18 /********** Global variables **********/
  19
  20 xbt_dynar_t initial_communications_pattern;
  21 xbt_dynar_t incomplete_communications_pattern;
  22
  23 /********** Static functions ***********/
  24
  25 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  26   if(comm1->type != comm2->type)
  27     return TYPE_DIFF;
  28   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  29     return RDV_DIFF;
  30   if (comm1->src_proc != comm2->src_proc)
  31     return SRC_PROC_DIFF;
  32   if (comm1->dst_proc != comm2->dst_proc)
  33     return DST_PROC_DIFF;
  34   if (comm1->tag != comm2->tag)
  35     return TAG_DIFF;
  36   if (comm1->data_size != comm2->data_size)
  37     return DATA_SIZE_DIFF;
  38   if(comm1->data == NULL && comm2->data == NULL)
  39     return 0;
  40   if(comm1->data != NULL && comm2->data !=NULL) {
  41     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  42       return 0;
  43     return DATA_DIFF;
  44   }else{
  45     return DATA_DIFF;
  46   }
  47   return 0;
  48 }
  49
  50 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  51   char *type, *res;
  52
  53   if(comm->type == SIMIX_COMM_SEND)
  54     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  55   else
  56     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  57
  58   switch(diff) {
  59   case TYPE_DIFF:
  60     res = bprintf("%s Different type for communication #%d", type, cursor);
  61     break;
  62   case RDV_DIFF:
  63     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  64     break;
  65   case TAG_DIFF:
  66     res = bprintf("%s Different tag for communication #%d", type, cursor);
  67     break;
  68   case SRC_PROC_DIFF:
  69       res = bprintf("%s Different source for communication #%d", type, cursor);
  70     break;
  71   case DST_PROC_DIFF:
  72       res = bprintf("%s Different destination for communication #%d", type, cursor);
  73     break;
  74   case DATA_SIZE_DIFF:
  75     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  76     break;
  77   case DATA_DIFF:
  78     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  79     break;
  80   default:
  81     res = NULL;
  82     break;
  83   }
  84
  85   return res;
  86 }
  87
  88 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
  89 {
  90   smx_process_t src_proc = MC_smx_resolve_process(comm->comm.src_proc);
  91   smx_process_t dst_proc = MC_smx_resolve_process(comm->comm.dst_proc);
  92   comm_pattern->src_proc = src_proc->pid;
  93   comm_pattern->dst_proc = dst_proc->pid;
  94   comm_pattern->src_host = MC_smx_process_get_host_name(src_proc);
  95   comm_pattern->dst_host = MC_smx_process_get_host_name(dst_proc);
  96   if (comm_pattern->data_size == -1 && comm->comm.src_buff != NULL) {
  97     comm_pattern->data_size = *(comm->comm.dst_buff_size);
  98     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
  99     MC_process_read_simple(&mc_model_checker->process,
 100       comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 101   }
 102 }
 103
 104 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 105
 106   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);
 107
 108   if(!backtracking){
 109     mc_comm_pattern_t initial_comm = xbt_dynar_get_as(list_comm_pattern->list, list_comm_pattern->index_comm, mc_comm_pattern_t);
 110     e_mc_comm_pattern_difference_t diff;
 111
 112     if((diff = compare_comm_pattern(initial_comm, comm)) != NONE_DIFF){
 113       if (comm->type == SIMIX_COMM_SEND){
 114         initial_global_state->send_deterministic = 0;
 115         if(initial_global_state->send_diff != NULL)
 116           xbt_free(initial_global_state->send_diff);
 117         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 118       }else{
 119         initial_global_state->recv_deterministic = 0;
 120         if(initial_global_state->recv_diff != NULL)
 121           xbt_free(initial_global_state->recv_diff);
 122         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 123       }
 124       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 125         XBT_INFO("*********************************************************");
 126         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 127         XBT_INFO("*********************************************************");
 128         XBT_INFO("%s", initial_global_state->send_diff);
 129         xbt_free(initial_global_state->send_diff);
 130         initial_global_state->send_diff = NULL;
 131         MC_print_statistics(mc_stats);
 132         xbt_abort();
 133       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 134         XBT_INFO("****************************************************");
 135         XBT_INFO("***** Non-deterministic communications pattern *****");
 136         XBT_INFO("****************************************************");
 137         XBT_INFO("%s", initial_global_state->send_diff);
 138         XBT_INFO("%s", initial_global_state->recv_diff);
 139         xbt_free(initial_global_state->send_diff);
 140         initial_global_state->send_diff = NULL;
 141         xbt_free(initial_global_state->recv_diff);
 142         initial_global_state->recv_diff = NULL;
 143         MC_print_statistics(mc_stats);
 144         xbt_abort();
 145       }
 146     }
 147   }
 148
 149   MC_comm_pattern_free(comm);
 150
 151 }
 152
 153 /********** Non Static functions ***********/
 154
 155 void MC_get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 156 {
 157   mc_comm_pattern_t pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 158   pattern->data_size = -1;
 159   pattern->data = NULL;
 160
 161   // Fill initial_pattern->index_comm:
 162   const smx_process_t issuer = MC_smx_simcall_get_issuer(request);
 163   mc_list_comm_pattern_t initial_pattern =
 164     (mc_list_comm_pattern_t) xbt_dynar_get_as(initial_communications_pattern, issuer->pid, mc_list_comm_pattern_t);
 165   xbt_dynar_t incomplete_pattern =
 166     (xbt_dynar_t) xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
 167   pattern->index =
 168     initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
 169
 170
 171   if (call_type == MC_CALL_TYPE_SEND) {
 172     /* Create comm pattern */
 173     pattern->type = SIMIX_COMM_SEND;
 174     pattern->comm = simcall_comm_isend__get__result(request);
 175
 176     s_smx_synchro_t synchro;
 177     MC_process_read_simple(&mc_model_checker->process,
 178       &synchro, pattern->comm, sizeof(synchro));
 179
 180     char* remote_name;
 181     MC_process_read_simple(&mc_model_checker->process, &remote_name,
 182       synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name,
 183       sizeof(remote_name));
 184     pattern->rdv =
 185       MC_process_read_string(&mc_model_checker->process, remote_name);
 186     pattern->src_proc = MC_smx_resolve_process(synchro.comm.src_proc)->pid;
 187     pattern->src_host = MC_smx_process_get_host_name(issuer);
 188
 189     struct s_smpi_mpi_request mpi_request;
 190     MC_process_read_simple(&mc_model_checker->process,
 191       &mpi_request, (MPI_Request) simcall_comm_isend__get__data(request),
 192       sizeof(mpi_request));
 193     pattern->tag = mpi_request.tag;
 194
 195     if(synchro.comm.src_buff != NULL){
 196       pattern->data_size = synchro.comm.src_buff_size;
 197       pattern->data = xbt_malloc0(pattern->data_size);
 198       MC_process_read_simple(&mc_model_checker->process,
 199         pattern->data, synchro.comm.src_buff, pattern->data_size);
 200     }
 201     if(mpi_request.detached){
 202       if (!initial_global_state->initial_communications_pattern_done) {
 203         /* Store comm pattern */
 204         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);
 205       } else {
 206         /* Evaluate comm determinism */
 207         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 208         ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->index_comm++;
 209       }
 210       return;
 211     }
 212   } else if (call_type == MC_CALL_TYPE_RECV) {
 213     pattern->type = SIMIX_COMM_RECEIVE;
 214     pattern->comm = simcall_comm_irecv__get__result(request);
 215
 216     struct s_smpi_mpi_request mpi_request;
 217     MC_process_read_simple(&mc_model_checker->process,
 218       &mpi_request, (MPI_Request) simcall_comm_irecv__get__data(request),
 219       sizeof(mpi_request));
 220     pattern->tag = mpi_request.tag;
 221
 222     s_smx_synchro_t synchro;
 223     MC_process_read_simple(&mc_model_checker->process,
 224       &synchro, pattern->comm, sizeof(synchro));
 225
 226     char* remote_name;
 227     MC_process_read_simple(&mc_model_checker->process, &remote_name,
 228       synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name,
 229       sizeof(remote_name));
 230     pattern->rdv =
 231       MC_process_read_string(&mc_model_checker->process, remote_name);
 232     pattern->dst_proc = MC_smx_resolve_process(synchro.comm.dst_proc)->pid;
 233     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 234   } else {
 235     xbt_die("Unexpected call_type %i", (int) call_type);
 236   }
 237
 238   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t), &pattern);
 239
 240   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 241 }
 242
 243 void MC_complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm, unsigned int issuer, int backtracking) {
 244   mc_comm_pattern_t current_comm_pattern;
 245   unsigned int cursor = 0;
 246   mc_comm_pattern_t comm_pattern;
 247   int completed = 0;
 248
 249   /* Complete comm pattern */
 250   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 251     if (current_comm_pattern-> comm == comm) {
 252       update_comm_pattern(current_comm_pattern, comm);
 253       completed = 1;
 254       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &comm_pattern);
 255       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 256       break;
 257     }
 258   }
 259   if(!completed)
 260     xbt_die("Corresponding communication not found!");
 261
 262   if (!initial_global_state->initial_communications_pattern_done) {
 263     /* Store comm pattern */
 264     xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->list, &comm_pattern);
 265   } else {
 266     /* Evaluate comm determinism */
 267     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 268     ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->index_comm++;
 269   }
 270 }
 271
 272
 273 /************************ Main algorithm ************************/
 274
 275 void MC_pre_modelcheck_comm_determinism(void)
 276 {
 277   MC_SET_MC_HEAP;
 278
 279   mc_state_t initial_state = NULL;
 280   smx_process_t process;
 281   int i;
 282
 283   if (_sg_mc_visited > 0)
 284     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 285
 286   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), MC_list_comm_pattern_free_voidp);
 287   for (i=0; i < MC_smx_get_maxpid(); i++){
 288     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 289     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), MC_comm_pattern_free_voidp);
 290     process_list_pattern->index_comm = 0;
 291     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 292   }
 293   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 294   for (i=0; i < MC_smx_get_maxpid(); i++){
 295     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 296     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 297   }
 298
 299   initial_state = MC_state_new();
 300   MC_SET_STD_HEAP;
 301
 302   XBT_DEBUG("********* Start communication determinism verification *********");
 303
 304   /* Wait for requests (schedules processes) */
 305   MC_wait_for_requests();
 306
 307   MC_SET_MC_HEAP;
 308
 309   /* Get an enabled process and insert it in the interleave set of the initial state */
 310   MC_EACH_SIMIX_PROCESS(process,
 311     if (MC_process_is_enabled(process)) {
 312       MC_state_interleave_process(initial_state, process);
 313     }
 314   );
 315
 316   xbt_fifo_unshift(mc_stack, initial_state);
 317
 318   MC_SET_STD_HEAP;
 319
 320 }
 321
 322 void MC_modelcheck_comm_determinism(void)
 323 {
 324
 325   char *req_str = NULL;
 326   int value;
 327   mc_visited_state_t visited_state = NULL;
 328   smx_simcall_t req = NULL;
 329   smx_process_t process = NULL;
 330   mc_state_t state = NULL, next_state = NULL;
 331
 332   while (xbt_fifo_size(mc_stack) > 0) {
 333
 334     /* Get current state */
 335     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 336
 337     XBT_DEBUG("**************************************************");
 338     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 339               xbt_fifo_size(mc_stack), state->num,
 340               MC_state_interleave_size(state));
 341
 342     /* Update statistics */
 343     mc_stats->visited_states++;
 344
 345     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 346         && (req = MC_state_get_request(state, &value))
 347         && (visited_state == NULL)) {
 348
 349       req_str = MC_request_to_string(req, value, MC_REQUEST_SIMIX);
 350       XBT_DEBUG("Execute: %s", req_str);
 351       xbt_free(req_str);
 352
 353       if (dot_output != NULL) {
 354         MC_SET_MC_HEAP;
 355         req_str = MC_request_get_dot_output(req, value);
 356         MC_SET_STD_HEAP;
 357       }
 358
 359       MC_state_set_executed_request(state, req, value);
 360       mc_stats->executed_transitions++;
 361
 362       /* TODO : handle test and testany simcalls */
 363       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 364       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 365         call = MC_get_call_type(req);
 366       }
 367
 368       /* Answer the request */
 369       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 370
 371       MC_SET_MC_HEAP;
 372       if(!initial_global_state->initial_communications_pattern_done)
 373         MC_handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 374       else
 375         MC_handle_comm_pattern(call, req, value, NULL, 0);
 376       MC_SET_STD_HEAP;
 377
 378       /* Wait for requests (schedules processes) */
 379       MC_wait_for_requests();
 380
 381       /* Create the new expanded state */
 382       MC_SET_MC_HEAP;
 383
 384       next_state = MC_state_new();
 385
 386       if ((visited_state = is_visited_state(next_state)) == NULL) {
 387
 388         /* Get enabled processes and insert them in the interleave set of the next state */
 389         MC_EACH_SIMIX_PROCESS(process,
 390           if (MC_process_is_enabled(process)) {
 391             MC_state_interleave_process(next_state, process);
 392           }
 393         );
 394
 395         if (dot_output != NULL)
 396           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 397
 398       } else {
 399
 400         if (dot_output != NULL)
 401           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 402
 403       }
 404
 405       xbt_fifo_unshift(mc_stack, next_state);
 406
 407       if (dot_output != NULL)
 408         xbt_free(req_str);
 409
 410       MC_SET_STD_HEAP;
 411
 412     } else {
 413
 414       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 415         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 416       } else if (visited_state != NULL) {
 417         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);
 418       } else {
 419         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 420       }
 421
 422       MC_SET_MC_HEAP;
 423
 424       if (!initial_global_state->initial_communications_pattern_done)
 425         initial_global_state->initial_communications_pattern_done = 1;
 426
 427       /* Trash the current state, no longer needed */
 428       xbt_fifo_shift(mc_stack);
 429       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 430       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 431
 432       MC_SET_STD_HEAP;
 433
 434       visited_state = NULL;
 435
 436       /* Check for deadlocks */
 437       if (MC_deadlock_check()) {
 438         MC_show_deadlock(NULL);
 439         return;
 440       }
 441
 442       MC_SET_MC_HEAP;
 443
 444       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 445         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 446           /* We found a back-tracking point, let's loop */
 447           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 448           xbt_fifo_unshift(mc_stack, state);
 449           MC_SET_STD_HEAP;
 450
 451           MC_replay(mc_stack);
 452
 453           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 454
 455           break;
 456         } else {
 457           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 458           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 459         }
 460       }
 461
 462       MC_SET_STD_HEAP;
 463     }
 464   }
 465
 466   MC_print_statistics(mc_stats);
 467   MC_SET_STD_HEAP;
 468
 469   return;
 470 }