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 using simgrid::mc::remote;
  17
  18 extern "C" {
  19
  20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  21                                 "Logging specific to MC communication determinism detection");
  22
  23 /********** Global variables **********/
  24
  25 xbt_dynar_t initial_communications_pattern;
  26 xbt_dynar_t incomplete_communications_pattern;
  27
  28 /********** Static functions ***********/
  29
  30 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  31   if(comm1->type != comm2->type)
  32     return TYPE_DIFF;
  33   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  34     return RDV_DIFF;
  35   if (comm1->src_proc != comm2->src_proc)
  36     return SRC_PROC_DIFF;
  37   if (comm1->dst_proc != comm2->dst_proc)
  38     return DST_PROC_DIFF;
  39   if (comm1->tag != comm2->tag)
  40     return TAG_DIFF;
  41   if (comm1->data_size != comm2->data_size)
  42     return DATA_SIZE_DIFF;
  43   if(comm1->data == NULL && comm2->data == NULL)
  44     return NONE_DIFF;
  45   if(comm1->data != NULL && comm2->data !=NULL) {
  46     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  47       return NONE_DIFF;
  48     return DATA_DIFF;
  49   }else{
  50     return DATA_DIFF;
  51   }
  52   return NONE_DIFF;
  53 }
  54
  55 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  56   char *type, *res;
  57
  58   if(comm->type == SIMIX_COMM_SEND)
  59     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  60   else
  61     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  62
  63   switch(diff) {
  64   case TYPE_DIFF:
  65     res = bprintf("%s Different type for communication #%d", type, cursor);
  66     break;
  67   case RDV_DIFF:
  68     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  69     break;
  70   case TAG_DIFF:
  71     res = bprintf("%s Different tag for communication #%d", type, cursor);
  72     break;
  73   case SRC_PROC_DIFF:
  74       res = bprintf("%s Different source for communication #%d", type, cursor);
  75     break;
  76   case DST_PROC_DIFF:
  77       res = bprintf("%s Different destination for communication #%d", type, cursor);
  78     break;
  79   case DATA_SIZE_DIFF:
  80     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  81     break;
  82   case DATA_DIFF:
  83     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  84     break;
  85   default:
  86     res = NULL;
  87     break;
  88   }
  89
  90   return res;
  91 }
  92
  93 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm_addr)
  94 {
  95   s_smx_synchro_t comm;
  96   mc_model_checker->process().read(&comm, remote(comm_addr));
  97
  98   smx_process_t src_proc = MC_smx_resolve_process(comm.comm.src_proc);
  99   smx_process_t dst_proc = MC_smx_resolve_process(comm.comm.dst_proc);
 100   comm_pattern->src_proc = src_proc->pid;
 101   comm_pattern->dst_proc = dst_proc->pid;
 102   comm_pattern->src_host = MC_smx_process_get_host_name(src_proc);
 103   comm_pattern->dst_host = MC_smx_process_get_host_name(dst_proc);
 104   if (comm_pattern->data_size == -1 && comm.comm.src_buff != NULL) {
 105     size_t buff_size;
 106     mc_model_checker->process().read(
 107       &buff_size, remote(comm.comm.dst_buff_size));
 108     comm_pattern->data_size = buff_size;
 109     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 110     mc_model_checker->process().read_bytes(
 111       comm_pattern->data, comm_pattern->data_size,
 112       remote(comm.comm.src_buff));
 113   }
 114 }
 115
 116 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 117
 118   mc_list_comm_pattern_t list =
 119     xbt_dynar_get_as(initial_communications_pattern, process, mc_list_comm_pattern_t);
 120
 121   if(!backtracking){
 122     mc_comm_pattern_t initial_comm =
 123       xbt_dynar_get_as(list->list, list->index_comm, mc_comm_pattern_t);
 124     e_mc_comm_pattern_difference_t diff =
 125       compare_comm_pattern(initial_comm, comm);
 126
 127     if (diff != NONE_DIFF) {
 128       if (comm->type == SIMIX_COMM_SEND){
 129         initial_global_state->send_deterministic = 0;
 130         if(initial_global_state->send_diff != NULL)
 131           xbt_free(initial_global_state->send_diff);
 132         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
 133       }else{
 134         initial_global_state->recv_deterministic = 0;
 135         if(initial_global_state->recv_diff != NULL)
 136           xbt_free(initial_global_state->recv_diff);
 137         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list->index_comm + 1);
 138       }
 139       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 140         XBT_INFO("*********************************************************");
 141         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 142         XBT_INFO("*********************************************************");
 143         XBT_INFO("%s", initial_global_state->send_diff);
 144         xbt_free(initial_global_state->send_diff);
 145         initial_global_state->send_diff = NULL;
 146         MC_print_statistics(mc_stats);
 147         xbt_abort();
 148       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 149         XBT_INFO("****************************************************");
 150         XBT_INFO("***** Non-deterministic communications pattern *****");
 151         XBT_INFO("****************************************************");
 152         XBT_INFO("%s", initial_global_state->send_diff);
 153         XBT_INFO("%s", initial_global_state->recv_diff);
 154         xbt_free(initial_global_state->send_diff);
 155         initial_global_state->send_diff = NULL;
 156         xbt_free(initial_global_state->recv_diff);
 157         initial_global_state->recv_diff = NULL;
 158         MC_print_statistics(mc_stats);
 159         xbt_abort();
 160       }
 161     }
 162   }
 163
 164   MC_comm_pattern_free(comm);
 165
 166 }
 167
 168 /********** Non Static functions ***********/
 169
 170 void MC_get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 171 {
 172   const smx_process_t issuer = MC_smx_simcall_get_issuer(request);
 173   mc_list_comm_pattern_t initial_pattern = xbt_dynar_get_as(
 174     initial_communications_pattern, issuer->pid, mc_list_comm_pattern_t);
 175   xbt_dynar_t incomplete_pattern = xbt_dynar_get_as(
 176     incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
 177
 178   mc_comm_pattern_t pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 179   pattern->data_size = -1;
 180   pattern->data = NULL;
 181   pattern->index =
 182     initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
 183
 184   if (call_type == MC_CALL_TYPE_SEND) {
 185     /* Create comm pattern */
 186     pattern->type = SIMIX_COMM_SEND;
 187     pattern->comm_addr = simcall_comm_isend__get__result(request);
 188
 189     s_smx_synchro_t synchro = mc_model_checker->process().read<s_smx_synchro_t>(
 190       (std::uint64_t) pattern->comm_addr);
 191
 192     char* remote_name = mc_model_checker->process().read<char*>(
 193       (std::uint64_t)(synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name));
 194     pattern->rdv =
 195       MC_process_read_string(&mc_model_checker->process(), remote_name);
 196     pattern->src_proc = MC_smx_resolve_process(synchro.comm.src_proc)->pid;
 197     pattern->src_host = MC_smx_process_get_host_name(issuer);
 198
 199     struct s_smpi_mpi_request mpi_request =
 200       mc_model_checker->process().read<s_smpi_mpi_request>(
 201         (std::uint64_t) simcall_comm_isend__get__data(request));
 202     pattern->tag = mpi_request.tag;
 203
 204     if(synchro.comm.src_buff != NULL){
 205       pattern->data_size = synchro.comm.src_buff_size;
 206       pattern->data = xbt_malloc0(pattern->data_size);
 207       mc_model_checker->process().read_bytes(
 208         pattern->data, pattern->data_size, remote(synchro.comm.src_buff));
 209     }
 210     if(mpi_request.detached){
 211       if (!initial_global_state->initial_communications_pattern_done) {
 212         /* Store comm pattern */
 213         xbt_dynar_push(
 214           xbt_dynar_get_as(
 215             initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 216           )->list,
 217           &pattern);
 218       } else {
 219         /* Evaluate comm determinism */
 220         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 221         xbt_dynar_get_as(
 222           initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 223         )->index_comm++;
 224       }
 225       return;
 226     }
 227   } else if (call_type == MC_CALL_TYPE_RECV) {
 228     pattern->type = SIMIX_COMM_RECEIVE;
 229     pattern->comm_addr = simcall_comm_irecv__get__result(request);
 230
 231     struct s_smpi_mpi_request mpi_request;
 232     mc_model_checker->process().read(
 233       &mpi_request, remote((struct s_smpi_mpi_request*)simcall_comm_irecv__get__data(request)));
 234     pattern->tag = mpi_request.tag;
 235
 236     s_smx_synchro_t synchro;
 237     mc_model_checker->process().read(&synchro, remote(pattern->comm_addr));
 238
 239     char* remote_name;
 240     mc_model_checker->process().read(&remote_name,
 241       remote(synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name));
 242     pattern->rdv =
 243       MC_process_read_string(&mc_model_checker->process(), remote_name);
 244     pattern->dst_proc = MC_smx_resolve_process(synchro.comm.dst_proc)->pid;
 245     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 246   } else {
 247     xbt_die("Unexpected call_type %i", (int) call_type);
 248   }
 249
 250   xbt_dynar_push(
 251     xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t),
 252     &pattern);
 253
 254   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 255 }
 256
 257 void MC_complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm_addr, unsigned int issuer, int backtracking) {
 258   mc_comm_pattern_t current_comm_pattern;
 259   unsigned int cursor = 0;
 260   mc_comm_pattern_t comm_pattern;
 261   int completed = 0;
 262
 263   /* Complete comm pattern */
 264   xbt_dynar_foreach(xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 265     if (current_comm_pattern->comm_addr == comm_addr) {
 266       update_comm_pattern(current_comm_pattern, comm_addr);
 267       completed = 1;
 268       xbt_dynar_remove_at(
 269         xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t),
 270         cursor, &comm_pattern);
 271       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 272       break;
 273     }
 274   }
 275   if(!completed)
 276     xbt_die("Corresponding communication not found!");
 277
 278   mc_list_comm_pattern_t pattern = xbt_dynar_get_as(
 279     initial_communications_pattern, issuer, mc_list_comm_pattern_t);
 280
 281   if (!initial_global_state->initial_communications_pattern_done) {
 282     /* Store comm pattern */
 283     xbt_dynar_push(pattern->list, &comm_pattern);
 284   } else {
 285     /* Evaluate comm determinism */
 286     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 287     pattern->index_comm++;
 288   }
 289 }
 290
 291
 292 /************************ Main algorithm ************************/
 293
 294 static void MC_modelcheck_comm_determinism_main(void);
 295
 296 static void MC_pre_modelcheck_comm_determinism(void)
 297 {
 298   mc_state_t initial_state = NULL;
 299   smx_process_t process;
 300   int i;
 301   const int maxpid = MC_smx_get_maxpid();
 302
 303   if (_sg_mc_visited > 0)
 304     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 305
 306   // Create initial_communications_pattern elements:
 307   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), MC_list_comm_pattern_free_voidp);
 308   for (i=0; i < maxpid; i++){
 309     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 310     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), MC_comm_pattern_free_voidp);
 311     process_list_pattern->index_comm = 0;
 312     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 313   }
 314
 315   // Create incomplete_communications_pattern elements:
 316   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 317   for (i=0; i < maxpid; i++){
 318     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 319     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 320   }
 321
 322   initial_state = MC_state_new();
 323
 324   XBT_DEBUG("********* Start communication determinism verification *********");
 325
 326   /* Wait for requests (schedules processes) */
 327   MC_wait_for_requests();
 328
 329   /* Get an enabled process and insert it in the interleave set of the initial state */
 330   MC_EACH_SIMIX_PROCESS(process,
 331     if (MC_process_is_enabled(process)) {
 332       MC_state_interleave_process(initial_state, process);
 333     }
 334   );
 335
 336   xbt_fifo_unshift(mc_stack, initial_state);
 337 }
 338
 339 static void MC_modelcheck_comm_determinism_main(void)
 340 {
 341
 342   char *req_str = NULL;
 343   int value;
 344   mc_visited_state_t visited_state = NULL;
 345   smx_simcall_t req = NULL;
 346   smx_process_t process = NULL;
 347   mc_state_t state = NULL, next_state = NULL;
 348
 349   while (xbt_fifo_size(mc_stack) > 0) {
 350
 351     /* Get current state */
 352     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 353
 354     XBT_DEBUG("**************************************************");
 355     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 356               xbt_fifo_size(mc_stack), state->num,
 357               MC_state_interleave_size(state));
 358
 359     /* Update statistics */
 360     mc_stats->visited_states++;
 361
 362     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 363         && (req = MC_state_get_request(state, &value))
 364         && (visited_state == NULL)) {
 365
 366       req_str = MC_request_to_string(req, value, MC_REQUEST_SIMIX);
 367       XBT_DEBUG("Execute: %s", req_str);
 368       xbt_free(req_str);
 369
 370       if (dot_output != NULL) {
 371         req_str = MC_request_get_dot_output(req, value);
 372       }
 373
 374       MC_state_set_executed_request(state, req, value);
 375       mc_stats->executed_transitions++;
 376
 377       /* TODO : handle test and testany simcalls */
 378       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 379       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 380         call = MC_get_call_type(req);
 381       }
 382
 383       /* Answer the request */
 384       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 385
 386       if(!initial_global_state->initial_communications_pattern_done)
 387         MC_handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 388       else
 389         MC_handle_comm_pattern(call, req, value, NULL, 0);
 390
 391       /* Wait for requests (schedules processes) */
 392       MC_wait_for_requests();
 393
 394       /* Create the new expanded state */
 395       next_state = MC_state_new();
 396
 397       if ((visited_state = is_visited_state(next_state)) == NULL) {
 398
 399         /* Get enabled processes and insert them in the interleave set of the next state */
 400         MC_EACH_SIMIX_PROCESS(process,
 401           if (MC_process_is_enabled(process)) {
 402             MC_state_interleave_process(next_state, process);
 403           }
 404         );
 405
 406         if (dot_output != NULL)
 407           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 408
 409       } else {
 410
 411         if (dot_output != NULL)
 412           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 413
 414       }
 415
 416       xbt_fifo_unshift(mc_stack, next_state);
 417
 418       if (dot_output != NULL)
 419         xbt_free(req_str);
 420
 421     } else {
 422
 423       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 424         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 425       } else if (visited_state != NULL) {
 426         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);
 427       } else {
 428         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 429       }
 430
 431       if (!initial_global_state->initial_communications_pattern_done)
 432         initial_global_state->initial_communications_pattern_done = 1;
 433
 434       /* Trash the current state, no longer needed */
 435       xbt_fifo_shift(mc_stack);
 436       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 437       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 438
 439       visited_state = NULL;
 440
 441       /* Check for deadlocks */
 442       if (MC_deadlock_check()) {
 443         MC_show_deadlock(NULL);
 444         return;
 445       }
 446
 447       while ((state = (mc_state_t) xbt_fifo_shift(mc_stack)) != NULL) {
 448         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 449           /* We found a back-tracking point, let's loop */
 450           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 451           xbt_fifo_unshift(mc_stack, state);
 452
 453           MC_replay(mc_stack);
 454
 455           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 456
 457           break;
 458         } else {
 459           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 460           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 461         }
 462       }
 463     }
 464   }
 465
 466   MC_print_statistics(mc_stats);
 467   exit(0);
 468 }
 469
 470 void MC_modelcheck_comm_determinism(void)
 471 {
 472   XBT_INFO("Check communication determinism");
 473   mc_reduce_kind = e_mc_reduce_none;
 474   MC_wait_for_requests();
 475
 476   if (mc_mode == MC_MODE_CLIENT) {
 477     // This will move somehwere else:
 478     MC_client_handle_messages();
 479   }
 480
 481   /* Create exploration stack */
 482   mc_stack = xbt_fifo_new();
 483
 484   MC_pre_modelcheck_comm_determinism();
 485
 486   initial_global_state = xbt_new0(s_mc_global_t, 1);
 487   initial_global_state->snapshot = MC_take_snapshot(0);
 488   initial_global_state->initial_communications_pattern_done = 0;
 489   initial_global_state->recv_deterministic = 1;
 490   initial_global_state->send_deterministic = 1;
 491   initial_global_state->recv_diff = NULL;
 492   initial_global_state->send_diff = NULL;
 493
 494   MC_modelcheck_comm_determinism_main();
 495 }
 496
 497 }