src/mc/mc_comm_determinism.cpp

   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 #include "mc_client.h"
  15
  16 extern "C" {
  17
  18 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  19                                 "Logging specific to MC communication determinism detection");
  20
  21 /********** Global variables **********/
  22
  23 xbt_dynar_t initial_communications_pattern;
  24 xbt_dynar_t incomplete_communications_pattern;
  25
  26 /********** Static functions ***********/
  27
  28 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  29   if(comm1->type != comm2->type)
  30     return TYPE_DIFF;
  31   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  32     return RDV_DIFF;
  33   if (comm1->src_proc != comm2->src_proc)
  34     return SRC_PROC_DIFF;
  35   if (comm1->dst_proc != comm2->dst_proc)
  36     return DST_PROC_DIFF;
  37   if (comm1->tag != comm2->tag)
  38     return TAG_DIFF;
  39   if (comm1->data_size != comm2->data_size)
  40     return DATA_SIZE_DIFF;
  41   if(comm1->data == NULL && comm2->data == NULL)
  42     return NONE_DIFF;
  43   if(comm1->data != NULL && comm2->data !=NULL) {
  44     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  45       return NONE_DIFF;
  46     return DATA_DIFF;
  47   }else{
  48     return DATA_DIFF;
  49   }
  50   return NONE_DIFF;
  51 }
  52
  53 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  54   char *type, *res;
  55
  56   if(comm->type == SIMIX_COMM_SEND)
  57     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  58   else
  59     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  60
  61   switch(diff) {
  62   case TYPE_DIFF:
  63     res = bprintf("%s Different type for communication #%d", type, cursor);
  64     break;
  65   case RDV_DIFF:
  66     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  67     break;
  68   case TAG_DIFF:
  69     res = bprintf("%s Different tag for communication #%d", type, cursor);
  70     break;
  71   case SRC_PROC_DIFF:
  72       res = bprintf("%s Different source for communication #%d", type, cursor);
  73     break;
  74   case DST_PROC_DIFF:
  75       res = bprintf("%s Different destination for communication #%d", type, cursor);
  76     break;
  77   case DATA_SIZE_DIFF:
  78     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  79     break;
  80   case DATA_DIFF:
  81     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  82     break;
  83   default:
  84     res = NULL;
  85     break;
  86   }
  87
  88   return res;
  89 }
  90
  91 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm_addr)
  92 {
  93   s_smx_synchro_t comm;
  94   MC_process_read_simple(&mc_model_checker->process,
  95     &comm, comm_addr, sizeof(comm));
  96
  97   smx_process_t src_proc = MC_smx_resolve_process(comm.comm.src_proc);
  98   smx_process_t dst_proc = MC_smx_resolve_process(comm.comm.dst_proc);
  99   comm_pattern->src_proc = src_proc->pid;
 100   comm_pattern->dst_proc = dst_proc->pid;
 101   comm_pattern->src_host = MC_smx_process_get_host_name(src_proc);
 102   comm_pattern->dst_host = MC_smx_process_get_host_name(dst_proc);
 103   if (comm_pattern->data_size == -1 && comm.comm.src_buff != NULL) {
 104     size_t buff_size;
 105     MC_process_read_simple(&mc_model_checker->process,
 106       &buff_size, comm.comm.dst_buff_size, sizeof(buff_size));
 107     comm_pattern->data_size = buff_size;
 108     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 109     MC_process_read_simple(&mc_model_checker->process,
 110       comm_pattern->data, comm.comm.src_buff, comm_pattern->data_size);
 111   }
 112 }
 113
 114 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 115
 116   mc_list_comm_pattern_t list =
 117     xbt_dynar_get_as(initial_communications_pattern, process, mc_list_comm_pattern_t);
 118
 119   if(!backtracking){
 120     mc_comm_pattern_t initial_comm =
 121       xbt_dynar_get_as(list->list, list->index_comm, mc_comm_pattern_t);
 122     e_mc_comm_pattern_difference_t diff =
 123       compare_comm_pattern(initial_comm, comm);
 124
 125     if (diff != NONE_DIFF) {
 126       if (comm->type == SIMIX_COMM_SEND){
 127         initial_global_state->send_deterministic = 0;
 128         if(initial_global_state->send_diff != NULL)
 129           xbt_free(initial_global_state->send_diff);
 130         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
 131       }else{
 132         initial_global_state->recv_deterministic = 0;
 133         if(initial_global_state->recv_diff != NULL)
 134           xbt_free(initial_global_state->recv_diff);
 135         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
 136       }
 137       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 138         XBT_INFO("*********************************************************");
 139         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 140         XBT_INFO("*********************************************************");
 141         XBT_INFO("%s", initial_global_state->send_diff);
 142         xbt_free(initial_global_state->send_diff);
 143         initial_global_state->send_diff = NULL;
 144         MC_print_statistics(mc_stats);
 145         xbt_abort();
 146       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 147         XBT_INFO("****************************************************");
 148         XBT_INFO("***** Non-deterministic communications pattern *****");
 149         XBT_INFO("****************************************************");
 150         XBT_INFO("%s", initial_global_state->send_diff);
 151         XBT_INFO("%s", initial_global_state->recv_diff);
 152         xbt_free(initial_global_state->send_diff);
 153         initial_global_state->send_diff = NULL;
 154         xbt_free(initial_global_state->recv_diff);
 155         initial_global_state->recv_diff = NULL;
 156         MC_print_statistics(mc_stats);
 157         xbt_abort();
 158       }
 159     }
 160   }
 161
 162   MC_comm_pattern_free(comm);
 163
 164 }
 165
 166 /********** Non Static functions ***********/
 167
 168 void MC_get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 169 {
 170   const smx_process_t issuer = MC_smx_simcall_get_issuer(request);
 171   mc_list_comm_pattern_t initial_pattern = xbt_dynar_get_as(
 172     initial_communications_pattern, issuer->pid, mc_list_comm_pattern_t);
 173   xbt_dynar_t incomplete_pattern = xbt_dynar_get_as(
 174     incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
 175
 176   mc_comm_pattern_t pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 177   pattern->data_size = -1;
 178   pattern->data = NULL;
 179   pattern->index =
 180     initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
 181
 182   if (call_type == MC_CALL_TYPE_SEND) {
 183     /* Create comm pattern */
 184     pattern->type = SIMIX_COMM_SEND;
 185     pattern->comm_addr = simcall_comm_isend__get__result(request);
 186
 187     s_smx_synchro_t synchro;
 188     MC_process_read_simple(&mc_model_checker->process,
 189       &synchro, pattern->comm_addr, sizeof(synchro));
 190
 191     char* remote_name;
 192     MC_process_read_simple(&mc_model_checker->process, &remote_name,
 193       synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name,
 194       sizeof(remote_name));
 195     pattern->rdv =
 196       MC_process_read_string(&mc_model_checker->process, remote_name);
 197     pattern->src_proc = MC_smx_resolve_process(synchro.comm.src_proc)->pid;
 198     pattern->src_host = MC_smx_process_get_host_name(issuer);
 199
 200     struct s_smpi_mpi_request mpi_request;
 201     MC_process_read_simple(&mc_model_checker->process,
 202       &mpi_request, (MPI_Request) simcall_comm_isend__get__data(request),
 203       sizeof(mpi_request));
 204     pattern->tag = mpi_request.tag;
 205
 206     if(synchro.comm.src_buff != NULL){
 207       pattern->data_size = synchro.comm.src_buff_size;
 208       pattern->data = xbt_malloc0(pattern->data_size);
 209       MC_process_read_simple(&mc_model_checker->process,
 210         pattern->data, synchro.comm.src_buff, pattern->data_size);
 211     }
 212     if(mpi_request.detached){
 213       if (!initial_global_state->initial_communications_pattern_done) {
 214         /* Store comm pattern */
 215         xbt_dynar_push(
 216           xbt_dynar_get_as(
 217             initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 218           )->list,
 219           &pattern);
 220       } else {
 221         /* Evaluate comm determinism */
 222         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 223         xbt_dynar_get_as(
 224           initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 225         )->index_comm++;
 226       }
 227       return;
 228     }
 229   } else if (call_type == MC_CALL_TYPE_RECV) {
 230     pattern->type = SIMIX_COMM_RECEIVE;
 231     pattern->comm_addr = simcall_comm_irecv__get__result(request);
 232
 233     struct s_smpi_mpi_request mpi_request;
 234     MC_process_read_simple(&mc_model_checker->process,
 235       &mpi_request, (MPI_Request) simcall_comm_irecv__get__data(request),
 236       sizeof(mpi_request));
 237     pattern->tag = mpi_request.tag;
 238
 239     s_smx_synchro_t synchro;
 240     MC_process_read_simple(&mc_model_checker->process,
 241       &synchro, pattern->comm_addr, sizeof(synchro));
 242
 243     char* remote_name;
 244     MC_process_read_simple(&mc_model_checker->process, &remote_name,
 245       synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name,
 246       sizeof(remote_name));
 247     pattern->rdv =
 248       MC_process_read_string(&mc_model_checker->process, remote_name);
 249     pattern->dst_proc = MC_smx_resolve_process(synchro.comm.dst_proc)->pid;
 250     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 251   } else {
 252     xbt_die("Unexpected call_type %i", (int) call_type);
 253   }
 254
 255   xbt_dynar_push(
 256     xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t),
 257     &pattern);
 258
 259   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 260 }
 261
 262 void MC_complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm_addr, unsigned int issuer, int backtracking) {
 263   mc_comm_pattern_t current_comm_pattern;
 264   unsigned int cursor = 0;
 265   mc_comm_pattern_t comm_pattern;
 266   int completed = 0;
 267
 268   /* Complete comm pattern */
 269   xbt_dynar_foreach(xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 270     if (current_comm_pattern->comm_addr == comm_addr) {
 271       update_comm_pattern(current_comm_pattern, comm_addr);
 272       completed = 1;
 273       xbt_dynar_remove_at(
 274         xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t),
 275         cursor, &comm_pattern);
 276       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 277       break;
 278     }
 279   }
 280   if(!completed)
 281     xbt_die("Corresponding communication not found!");
 282
 283   mc_list_comm_pattern_t pattern = xbt_dynar_get_as(
 284     initial_communications_pattern, issuer, mc_list_comm_pattern_t);
 285
 286   if (!initial_global_state->initial_communications_pattern_done) {
 287     /* Store comm pattern */
 288     xbt_dynar_push(pattern->list, &comm_pattern);
 289   } else {
 290     /* Evaluate comm determinism */
 291     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 292     pattern->index_comm++;
 293   }
 294 }
 295
 296
 297 /************************ Main algorithm ************************/
 298
 299 static void MC_modelcheck_comm_determinism_main(void);
 300
 301 static void MC_pre_modelcheck_comm_determinism(void)
 302 {
 303   MC_SET_MC_HEAP;
 304
 305   mc_state_t initial_state = NULL;
 306   smx_process_t process;
 307   int i;
 308   const int maxpid = MC_smx_get_maxpid();
 309
 310   if (_sg_mc_visited > 0)
 311     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 312
 313   // Create initial_communications_pattern elements:
 314   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), MC_list_comm_pattern_free_voidp);
 315   for (i=0; i < maxpid; i++){
 316     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 317     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), MC_comm_pattern_free_voidp);
 318     process_list_pattern->index_comm = 0;
 319     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 320   }
 321
 322   // Create incomplete_communications_pattern elements:
 323   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 324   for (i=0; i < maxpid; i++){
 325     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 326     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 327   }
 328
 329   initial_state = MC_state_new();
 330   MC_SET_STD_HEAP;
 331
 332   XBT_DEBUG("********* Start communication determinism verification *********");
 333
 334   /* Wait for requests (schedules processes) */
 335   MC_wait_for_requests();
 336
 337   MC_SET_MC_HEAP;
 338
 339   /* Get an enabled process and insert it in the interleave set of the initial state */
 340   MC_EACH_SIMIX_PROCESS(process,
 341     if (MC_process_is_enabled(process)) {
 342       MC_state_interleave_process(initial_state, process);
 343     }
 344   );
 345
 346   xbt_fifo_unshift(mc_stack, initial_state);
 347
 348   MC_SET_STD_HEAP;
 349
 350 }
 351
 352 static void MC_modelcheck_comm_determinism_main(void)
 353 {
 354
 355   char *req_str = NULL;
 356   int value;
 357   mc_visited_state_t visited_state = NULL;
 358   smx_simcall_t req = NULL;
 359   smx_process_t process = NULL;
 360   mc_state_t state = NULL, next_state = NULL;
 361
 362   while (xbt_fifo_size(mc_stack) > 0) {
 363
 364     /* Get current state */
 365     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 366
 367     XBT_DEBUG("**************************************************");
 368     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 369               xbt_fifo_size(mc_stack), state->num,
 370               MC_state_interleave_size(state));
 371
 372     /* Update statistics */
 373     mc_stats->visited_states++;
 374
 375     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 376         && (req = MC_state_get_request(state, &value))
 377         && (visited_state == NULL)) {
 378
 379       req_str = MC_request_to_string(req, value, MC_REQUEST_SIMIX);
 380       XBT_DEBUG("Execute: %s", req_str);
 381       xbt_free(req_str);
 382
 383       if (dot_output != NULL) {
 384         MC_SET_MC_HEAP;
 385         req_str = MC_request_get_dot_output(req, value);
 386         MC_SET_STD_HEAP;
 387       }
 388
 389       MC_state_set_executed_request(state, req, value);
 390       mc_stats->executed_transitions++;
 391
 392       /* TODO : handle test and testany simcalls */
 393       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 394       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 395         call = MC_get_call_type(req);
 396       }
 397
 398       /* Answer the request */
 399       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 400
 401       MC_SET_MC_HEAP;
 402       if(!initial_global_state->initial_communications_pattern_done)
 403         MC_handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 404       else
 405         MC_handle_comm_pattern(call, req, value, NULL, 0);
 406       MC_SET_STD_HEAP;
 407
 408       /* Wait for requests (schedules processes) */
 409       MC_wait_for_requests();
 410
 411       /* Create the new expanded state */
 412       MC_SET_MC_HEAP;
 413
 414       next_state = MC_state_new();
 415
 416       if ((visited_state = is_visited_state(next_state)) == NULL) {
 417
 418         /* Get enabled processes and insert them in the interleave set of the next state */
 419         MC_EACH_SIMIX_PROCESS(process,
 420           if (MC_process_is_enabled(process)) {
 421             MC_state_interleave_process(next_state, process);
 422           }
 423         );
 424
 425         if (dot_output != NULL)
 426           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 427
 428       } else {
 429
 430         if (dot_output != NULL)
 431           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 432
 433       }
 434
 435       xbt_fifo_unshift(mc_stack, next_state);
 436
 437       if (dot_output != NULL)
 438         xbt_free(req_str);
 439
 440       MC_SET_STD_HEAP;
 441
 442     } else {
 443
 444       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 445         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 446       } else if (visited_state != NULL) {
 447         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);
 448       } else {
 449         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 450       }
 451
 452       MC_SET_MC_HEAP;
 453
 454       if (!initial_global_state->initial_communications_pattern_done)
 455         initial_global_state->initial_communications_pattern_done = 1;
 456
 457       /* Trash the current state, no longer needed */
 458       xbt_fifo_shift(mc_stack);
 459       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 460       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 461
 462       MC_SET_STD_HEAP;
 463
 464       visited_state = NULL;
 465
 466       /* Check for deadlocks */
 467       if (MC_deadlock_check()) {
 468         MC_show_deadlock(NULL);
 469         return;
 470       }
 471
 472       MC_SET_MC_HEAP;
 473
 474       while ((state = (mc_state_t) xbt_fifo_shift(mc_stack)) != NULL) {
 475         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 476           /* We found a back-tracking point, let's loop */
 477           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 478           xbt_fifo_unshift(mc_stack, state);
 479           MC_SET_STD_HEAP;
 480
 481           MC_replay(mc_stack);
 482
 483           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 484
 485           break;
 486         } else {
 487           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 488           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 489         }
 490       }
 491
 492       MC_SET_STD_HEAP;
 493     }
 494   }
 495
 496   MC_print_statistics(mc_stats);
 497   MC_SET_STD_HEAP;
 498
 499   exit(0);
 500 }
 501
 502 void MC_modelcheck_comm_determinism(void)
 503 {
 504   if (mc_mode == MC_MODE_CLIENT) {
 505     // This will move somehwere else:
 506     MC_client_handle_messages();
 507   }
 508
 509   xbt_mheap_t heap = mmalloc_set_current_heap(mc_heap);
 510
 511   /* Create exploration stack */
 512   mc_stack = xbt_fifo_new();
 513
 514   MC_SET_STD_HEAP;
 515
 516   MC_pre_modelcheck_comm_determinism();
 517
 518   MC_SET_MC_HEAP;
 519   initial_global_state = xbt_new0(s_mc_global_t, 1);
 520   initial_global_state->snapshot = MC_take_snapshot(0);
 521   initial_global_state->initial_communications_pattern_done = 0;
 522   initial_global_state->recv_deterministic = 1;
 523   initial_global_state->send_deterministic = 1;
 524   initial_global_state->recv_diff = NULL;
 525   initial_global_state->send_diff = NULL;
 526
 527   MC_SET_STD_HEAP;
 528
 529   MC_modelcheck_comm_determinism_main();
 530
 531   mmalloc_set_current_heap(heap);
 532 }
 533
 534 }