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 = mc_model_checker->process().read_string(remote_name);
 195     pattern->src_proc = MC_smx_resolve_process(synchro.comm.src_proc)->pid;
 196     pattern->src_host = MC_smx_process_get_host_name(issuer);
 197
 198     struct s_smpi_mpi_request mpi_request =
 199       mc_model_checker->process().read<s_smpi_mpi_request>(
 200         (std::uint64_t) simcall_comm_isend__get__data(request));
 201     pattern->tag = mpi_request.tag;
 202
 203     if(synchro.comm.src_buff != NULL){
 204       pattern->data_size = synchro.comm.src_buff_size;
 205       pattern->data = xbt_malloc0(pattern->data_size);
 206       mc_model_checker->process().read_bytes(
 207         pattern->data, pattern->data_size, remote(synchro.comm.src_buff));
 208     }
 209     if(mpi_request.detached){
 210       if (!initial_global_state->initial_communications_pattern_done) {
 211         /* Store comm pattern */
 212         xbt_dynar_push(
 213           xbt_dynar_get_as(
 214             initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 215           )->list,
 216           &pattern);
 217       } else {
 218         /* Evaluate comm determinism */
 219         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 220         xbt_dynar_get_as(
 221           initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t
 222         )->index_comm++;
 223       }
 224       return;
 225     }
 226   } else if (call_type == MC_CALL_TYPE_RECV) {
 227     pattern->type = SIMIX_COMM_RECEIVE;
 228     pattern->comm_addr = simcall_comm_irecv__get__result(request);
 229
 230     struct s_smpi_mpi_request mpi_request;
 231     mc_model_checker->process().read(
 232       &mpi_request, remote((struct s_smpi_mpi_request*)simcall_comm_irecv__get__data(request)));
 233     pattern->tag = mpi_request.tag;
 234
 235     s_smx_synchro_t synchro;
 236     mc_model_checker->process().read(&synchro, remote(pattern->comm_addr));
 237
 238     char* remote_name;
 239     mc_model_checker->process().read(&remote_name,
 240       remote(synchro.comm.rdv ? &synchro.comm.rdv->name : &synchro.comm.rdv_cpy->name));
 241     pattern->rdv = mc_model_checker->process().read_string(remote_name);
 242     pattern->dst_proc = MC_smx_resolve_process(synchro.comm.dst_proc)->pid;
 243     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 244   } else {
 245     xbt_die("Unexpected call_type %i", (int) call_type);
 246   }
 247
 248   xbt_dynar_push(
 249     xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t),
 250     &pattern);
 251
 252   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 253 }
 254
 255 void MC_complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm_addr, unsigned int issuer, int backtracking) {
 256   mc_comm_pattern_t current_comm_pattern;
 257   unsigned int cursor = 0;
 258   mc_comm_pattern_t comm_pattern;
 259   int completed = 0;
 260
 261   /* Complete comm pattern */
 262   xbt_dynar_foreach(xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 263     if (current_comm_pattern->comm_addr == comm_addr) {
 264       update_comm_pattern(current_comm_pattern, comm_addr);
 265       completed = 1;
 266       xbt_dynar_remove_at(
 267         xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t),
 268         cursor, &comm_pattern);
 269       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 270       break;
 271     }
 272   }
 273   if(!completed)
 274     xbt_die("Corresponding communication not found!");
 275
 276   mc_list_comm_pattern_t pattern = xbt_dynar_get_as(
 277     initial_communications_pattern, issuer, mc_list_comm_pattern_t);
 278
 279   if (!initial_global_state->initial_communications_pattern_done) {
 280     /* Store comm pattern */
 281     xbt_dynar_push(pattern->list, &comm_pattern);
 282   } else {
 283     /* Evaluate comm determinism */
 284     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 285     pattern->index_comm++;
 286   }
 287 }
 288
 289
 290 /************************ Main algorithm ************************/
 291
 292 static void MC_modelcheck_comm_determinism_main(void);
 293
 294 static void MC_pre_modelcheck_comm_determinism(void)
 295 {
 296   mc_state_t initial_state = NULL;
 297   smx_process_t process;
 298   int i;
 299   const int maxpid = MC_smx_get_maxpid();
 300
 301   if (_sg_mc_visited > 0)
 302     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 303
 304   // Create initial_communications_pattern elements:
 305   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), MC_list_comm_pattern_free_voidp);
 306   for (i=0; i < maxpid; i++){
 307     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 308     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), MC_comm_pattern_free_voidp);
 309     process_list_pattern->index_comm = 0;
 310     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 311   }
 312
 313   // Create incomplete_communications_pattern elements:
 314   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 315   for (i=0; i < maxpid; i++){
 316     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 317     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 318   }
 319
 320   initial_state = MC_state_new();
 321
 322   XBT_DEBUG("********* Start communication determinism verification *********");
 323
 324   /* Wait for requests (schedules processes) */
 325   MC_wait_for_requests();
 326
 327   /* Get an enabled process and insert it in the interleave set of the initial state */
 328   MC_EACH_SIMIX_PROCESS(process,
 329     if (MC_process_is_enabled(process)) {
 330       MC_state_interleave_process(initial_state, process);
 331     }
 332   );
 333
 334   xbt_fifo_unshift(mc_stack, initial_state);
 335 }
 336
 337 static void MC_modelcheck_comm_determinism_main(void)
 338 {
 339
 340   char *req_str = NULL;
 341   int value;
 342   mc_visited_state_t visited_state = NULL;
 343   smx_simcall_t req = NULL;
 344   smx_process_t process = NULL;
 345   mc_state_t state = NULL, next_state = NULL;
 346
 347   while (xbt_fifo_size(mc_stack) > 0) {
 348
 349     /* Get current state */
 350     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 351
 352     XBT_DEBUG("**************************************************");
 353     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 354               xbt_fifo_size(mc_stack), state->num,
 355               MC_state_interleave_size(state));
 356
 357     /* Update statistics */
 358     mc_stats->visited_states++;
 359
 360     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 361         && (req = MC_state_get_request(state, &value))
 362         && (visited_state == NULL)) {
 363
 364       req_str = MC_request_to_string(req, value, MC_REQUEST_SIMIX);
 365       XBT_DEBUG("Execute: %s", req_str);
 366       xbt_free(req_str);
 367
 368       if (dot_output != NULL) {
 369         req_str = MC_request_get_dot_output(req, value);
 370       }
 371
 372       MC_state_set_executed_request(state, req, value);
 373       mc_stats->executed_transitions++;
 374
 375       /* TODO : handle test and testany simcalls */
 376       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 377       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 378         call = MC_get_call_type(req);
 379       }
 380
 381       /* Answer the request */
 382       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 383
 384       if(!initial_global_state->initial_communications_pattern_done)
 385         MC_handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 386       else
 387         MC_handle_comm_pattern(call, req, value, NULL, 0);
 388
 389       /* Wait for requests (schedules processes) */
 390       MC_wait_for_requests();
 391
 392       /* Create the new expanded state */
 393       next_state = MC_state_new();
 394
 395       if ((visited_state = is_visited_state(next_state)) == NULL) {
 396
 397         /* Get enabled processes and insert them in the interleave set of the next state */
 398         MC_EACH_SIMIX_PROCESS(process,
 399           if (MC_process_is_enabled(process)) {
 400             MC_state_interleave_process(next_state, process);
 401           }
 402         );
 403
 404         if (dot_output != NULL)
 405           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 406
 407       } else {
 408
 409         if (dot_output != NULL)
 410           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 411
 412       }
 413
 414       xbt_fifo_unshift(mc_stack, next_state);
 415
 416       if (dot_output != NULL)
 417         xbt_free(req_str);
 418
 419     } else {
 420
 421       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 422         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 423       } else if (visited_state != NULL) {
 424         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);
 425       } else {
 426         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 427       }
 428
 429       if (!initial_global_state->initial_communications_pattern_done)
 430         initial_global_state->initial_communications_pattern_done = 1;
 431
 432       /* Trash the current state, no longer needed */
 433       xbt_fifo_shift(mc_stack);
 434       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 435       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 436
 437       visited_state = NULL;
 438
 439       /* Check for deadlocks */
 440       if (MC_deadlock_check()) {
 441         MC_show_deadlock(NULL);
 442         return;
 443       }
 444
 445       while ((state = (mc_state_t) xbt_fifo_shift(mc_stack)) != NULL) {
 446         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 447           /* We found a back-tracking point, let's loop */
 448           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 449           xbt_fifo_unshift(mc_stack, state);
 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   }
 463
 464   MC_print_statistics(mc_stats);
 465   exit(0);
 466 }
 467
 468 void MC_modelcheck_comm_determinism(void)
 469 {
 470   XBT_INFO("Check communication determinism");
 471   mc_reduce_kind = e_mc_reduce_none;
 472   MC_wait_for_requests();
 473
 474   if (mc_mode == MC_MODE_CLIENT) {
 475     // This will move somehwere else:
 476     MC_client_handle_messages();
 477   }
 478
 479   /* Create exploration stack */
 480   mc_stack = xbt_fifo_new();
 481
 482   MC_pre_modelcheck_comm_determinism();
 483
 484   initial_global_state = xbt_new0(s_mc_global_t, 1);
 485   initial_global_state->snapshot = MC_take_snapshot(0);
 486   initial_global_state->initial_communications_pattern_done = 0;
 487   initial_global_state->recv_deterministic = 1;
 488   initial_global_state->send_deterministic = 1;
 489   initial_global_state->recv_diff = NULL;
 490   initial_global_state->send_diff = NULL;
 491
 492   MC_modelcheck_comm_determinism_main();
 493 }
 494
 495 }