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_private.h"
   8 #include "mc_record.h"
   9
  10 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  11                                 "Logging specific to MC communication determinism detection");
  12
  13 /********** Global variables **********/
  14
  15 xbt_dynar_t initial_communications_pattern;
  16 xbt_dynar_t incomplete_communications_pattern;
  17 xbt_dynar_t communications_pattern;
  18 int nb_comm_pattern;
  19
  20 /********** Static functions ***********/
  21
  22 static void comm_pattern_free(mc_comm_pattern_t p)
  23 {
  24   xbt_free(p->rdv);
  25   xbt_free(p->data);
  26   xbt_free(p);
  27   p = NULL;
  28 }
  29
  30 static void comm_pattern_free_voidp(void *p)
  31 {
  32   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
  33 }
  34
  35 static mc_comm_pattern_t get_comm_pattern_from_idx(xbt_dynar_t pattern,
  36                                                    unsigned int *idx,
  37                                                    e_smx_comm_type_t type,
  38                                                    unsigned long proc)
  39 {
  40   mc_comm_pattern_t current_comm;
  41   while (*idx < xbt_dynar_length(pattern)) {
  42     current_comm =
  43         (mc_comm_pattern_t) xbt_dynar_get_as(pattern, *idx, mc_comm_pattern_t);
  44     if (current_comm->type == type && type == SIMIX_COMM_SEND) {
  45       if (current_comm->src_proc == proc)
  46         return current_comm;
  47     } else if (current_comm->type == type && type == SIMIX_COMM_RECEIVE) {
  48       if (current_comm->dst_proc == proc)
  49         return current_comm;
  50     }
  51     (*idx)++;
  52   }
  53   return NULL;
  54 }
  55
  56 static int compare_comm_pattern(mc_comm_pattern_t comm1,
  57                                 mc_comm_pattern_t comm2)
  58 {
  59   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  60     return 1;
  61   if (comm1->src_proc != comm2->src_proc)
  62     return 1;
  63   if (comm1->dst_proc != comm2->dst_proc)
  64     return 1;
  65   if (comm1->data_size != comm2->data_size)
  66     return 1;
  67   if (memcmp(comm1->data, comm2->data, comm1->data_size) != 0)
  68     return 1;
  69   return 0;
  70 }
  71
  72 static void deterministic_pattern(xbt_dynar_t pattern, int partial)
  73 {
  74
  75   unsigned int cursor = 0, send_index = 0, recv_index = 0;
  76   mc_comm_pattern_t comm1, comm2;
  77   unsigned int current_process = 1; /* Process 0 corresponds to maestro */
  78   unsigned int nb_comms1, nb_comms2;
  79   xbt_dynar_t process_comms_pattern1, process_comms_pattern2;
  80
  81   while (current_process < simix_process_maxpid) {
  82     process_comms_pattern1 = (xbt_dynar_t)xbt_dynar_get_as(initial_communications_pattern, current_process, xbt_dynar_t);
  83     process_comms_pattern2 = (xbt_dynar_t)xbt_dynar_get_as(pattern, current_process, xbt_dynar_t);
  84     nb_comms1 = xbt_dynar_length(process_comms_pattern1);
  85     nb_comms2 = xbt_dynar_length(process_comms_pattern2);
  86     if(!xbt_dynar_is_empty((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, current_process, xbt_dynar_t)))
  87       xbt_die("Damn ! Some communications from the process %u are incomplete (%lu)! That means one or several simcalls are not handle.", current_process, xbt_dynar_length((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, current_process, xbt_dynar_t)));
  88     if (!partial && (nb_comms1 != nb_comms2)) {
  89       XBT_INFO("The total number of communications is different between the compared patterns for the process %u.\n Communication determinism verification for this process cannot be performed.", current_process);
  90       initial_global_state->send_deterministic = -1;
  91       initial_global_state->comm_deterministic = -1;
  92     } else {
  93       while (cursor < nb_comms2) {
  94         comm1 = (mc_comm_pattern_t)xbt_dynar_get_as(process_comms_pattern1, cursor, mc_comm_pattern_t);
  95         if (comm1->type == SIMIX_COMM_SEND) {
  96           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &send_index, comm1->type, current_process);
  97           if (compare_comm_pattern(comm1, comm2)) {
  98             XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
  99             initial_global_state->send_deterministic = 0;
 100             initial_global_state->comm_deterministic = 0;
 101             return;
 102           }
 103           send_index++;
 104         } else if (comm1->type == SIMIX_COMM_RECEIVE) {
 105           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &recv_index, comm1->type, current_process);
 106           if (compare_comm_pattern(comm1, comm2)) {
 107             initial_global_state->comm_deterministic = 0;
 108             if (!_sg_mc_send_determinism){
 109               XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
 110               return;
 111             }
 112           }
 113           recv_index++;
 114         }
 115         cursor++;
 116       }
 117     }
 118     current_process++;
 119     cursor = 0;
 120     send_index = 0;
 121     recv_index = 0;
 122   }
 123 }
 124
 125 static void print_communications_pattern(xbt_dynar_t comms_pattern)
 126 {
 127   unsigned int cursor = 0;
 128   mc_comm_pattern_t current_comm;
 129   unsigned int current_process = 1;
 130   xbt_dynar_t current_pattern;
 131   while (current_process < simix_process_maxpid) {
 132     current_pattern = (xbt_dynar_t)xbt_dynar_get_as(comms_pattern, current_process, xbt_dynar_t);
 133     XBT_INFO("Communications from the process %u:", current_process);
 134     xbt_dynar_foreach(current_pattern, cursor, current_comm) {
 135       if (current_comm->type == SIMIX_COMM_SEND) {
 136         XBT_INFO("[(%lu) %s -> (%lu) %s] %s ", current_comm->src_proc,
 137                  current_comm->src_host, current_comm->dst_proc,
 138                  current_comm->dst_host, "iSend");
 139       } else {
 140         XBT_INFO("[(%lu) %s <- (%lu) %s] %s ", current_comm->dst_proc,
 141                  current_comm->dst_host, current_comm->src_proc,
 142                  current_comm->src_host, "iRecv");
 143       }
 144     }
 145     current_process++;
 146     cursor = 0;
 147   }
 148 }
 149
 150 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
 151 {
 152   void *addr_pointed;
 153   comm_pattern->src_proc = comm->comm.src_proc->pid;
 154   comm_pattern->dst_proc = comm->comm.dst_proc->pid;
 155   comm_pattern->src_host =
 156     simcall_host_get_name(comm->comm.src_proc->smx_host);
 157   comm_pattern->dst_host =
 158     simcall_host_get_name(comm->comm.dst_proc->smx_host);
 159   if (comm_pattern->data_size == -1) {
 160     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 161     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 162     addr_pointed = *(void **) comm->comm.src_buff;
 163     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 164       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 165     else
 166       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 167   }
 168 }
 169
 170 /********** Non Static functions ***********/
 171
 172 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, mc_call_type call_type)
 173 {
 174   mc_comm_pattern_t pattern = NULL;
 175   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 176   pattern->num = ++nb_comm_pattern;
 177   pattern->data_size = -1;
 178   void *addr_pointed;
 179   if (call_type == MC_CALL_TYPE_SEND) {              // ISEND
 180     pattern->type = SIMIX_COMM_SEND;
 181     pattern->comm = simcall_comm_isend__get__result(request);
 182     pattern->src_proc = pattern->comm->comm.src_proc->pid;
 183     pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
 184     pattern->data_size = pattern->comm->comm.src_buff_size;
 185     pattern->data = xbt_malloc0(pattern->data_size);
 186     addr_pointed = *(void **) pattern->comm->comm.src_buff;
 187     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 188       memcpy(pattern->data, addr_pointed, pattern->data_size);
 189     else
 190       memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 191   } else if (call_type == MC_CALL_TYPE_RECV) {                      // IRECV
 192     pattern->type = SIMIX_COMM_RECEIVE;
 193     pattern->comm = simcall_comm_irecv__get__result(request);
 194     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
 195     pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
 196   } else {
 197     xbt_die("Unexpected call_type %i", (int) call_type);
 198   }
 199
 200   if (pattern->comm->comm.rdv != NULL)
 201     pattern->rdv = strdup(pattern->comm->comm.rdv->name);
 202   else
 203     pattern->rdv = strdup(pattern->comm->comm.rdv_cpy->name);
 204
 205   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(list, request->issuer->pid, xbt_dynar_t), &pattern);
 206
 207   xbt_dynar_push_as((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, request->issuer->pid, xbt_dynar_t), int, xbt_dynar_length((xbt_dynar_t)xbt_dynar_get_as(list, request->issuer->pid, xbt_dynar_t)) - 1);
 208
 209 }
 210
 211 void complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm)
 212 {
 213   mc_comm_pattern_t current_comm_pattern;
 214   unsigned int cursor = 0;
 215   int index;
 216   unsigned int src = comm->comm.src_proc->pid;
 217   unsigned int dst = comm->comm.dst_proc->pid;
 218   int src_completed = 0, dst_completed = 0;
 219
 220   /* Looking for the corresponding communication in the comm pattern list of the src process */
 221   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, index){
 222     current_comm_pattern = (mc_comm_pattern_t) xbt_dynar_get_as((xbt_dynar_t)xbt_dynar_get_as(list, src, xbt_dynar_t), index, mc_comm_pattern_t);
 223     if(current_comm_pattern->comm == comm){
 224       update_comm_pattern(current_comm_pattern, comm);
 225       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, NULL);
 226       src_completed = 1;
 227       break;
 228     }
 229   }
 230
 231   if(!src_completed)
 232     xbt_die("Corresponding communication for the source process not found!");
 233
 234   cursor = 0;
 235
 236   /* Looking for the corresponding communication in the comm pattern list of the dst process */
 237   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, index){
 238     current_comm_pattern = (mc_comm_pattern_t) xbt_dynar_get_as((xbt_dynar_t)xbt_dynar_get_as(list, dst, xbt_dynar_t), index, mc_comm_pattern_t);
 239     if(current_comm_pattern->comm == comm){
 240       update_comm_pattern(current_comm_pattern, comm);
 241       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, NULL);
 242       dst_completed = 1;
 243       break;
 244     }
 245   }
 246
 247   if(!dst_completed)
 248     xbt_die("Corresponding communication for the dest process not found!");
 249
 250
 251 }
 252
 253 /************************ Main algorithm ************************/
 254
 255 void MC_pre_modelcheck_comm_determinism(void)
 256 {
 257
 258   int mc_mem_set = (mmalloc_get_current_heap() == mc_heap);
 259
 260   mc_state_t initial_state = NULL;
 261   smx_process_t process;
 262   int i;
 263
 264   if (!mc_mem_set)
 265     MC_SET_MC_HEAP;
 266
 267   if (_sg_mc_visited > 0)
 268     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 269
 270   initial_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 271   for (i=0; i<simix_process_maxpid; i++){
 272     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 273     xbt_dynar_insert_at(initial_communications_pattern, i, &process_pattern);
 274   }
 275   communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 276   for (i=0; i<simix_process_maxpid; i++){
 277     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 278     xbt_dynar_insert_at(communications_pattern, i, &process_pattern);
 279   }
 280   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 281   for (i=0; i<simix_process_maxpid; i++){
 282     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(int), NULL);
 283     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 284   }
 285
 286   nb_comm_pattern = 0;
 287
 288   initial_state = MC_state_new();
 289
 290   MC_SET_STD_HEAP;
 291
 292   XBT_DEBUG("********* Start communication determinism verification *********");
 293
 294   /* Wait for requests (schedules processes) */
 295   MC_wait_for_requests();
 296
 297   MC_SET_MC_HEAP;
 298
 299   /* Get an enabled process and insert it in the interleave set of the initial state */
 300   xbt_swag_foreach(process, simix_global->process_list) {
 301     if (MC_process_is_enabled(process)) {
 302       MC_state_interleave_process(initial_state, process);
 303     }
 304   }
 305
 306   xbt_fifo_unshift(mc_stack, initial_state);
 307
 308   MC_SET_STD_HEAP;
 309
 310 }
 311
 312 void MC_modelcheck_comm_determinism(void)
 313 {
 314
 315   char *req_str = NULL;
 316   int value;
 317   mc_visited_state_t visited_state = NULL;
 318   smx_simcall_t req = NULL;
 319   smx_process_t process = NULL;
 320   mc_state_t state = NULL, next_state = NULL;
 321   xbt_dynar_t current_pattern;
 322
 323   while (xbt_fifo_size(mc_stack) > 0) {
 324
 325     /* Get current state */
 326     state =
 327         (mc_state_t)
 328         xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 329
 330     XBT_DEBUG("**************************************************");
 331     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 332               xbt_fifo_size(mc_stack), state->num,
 333               MC_state_interleave_size(state));
 334
 335     /* Update statistics */
 336     mc_stats->visited_states++;
 337
 338     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 339         && (req = MC_state_get_request(state, &value))
 340         && (visited_state == NULL)) {
 341
 342       MC_LOG_REQUEST(mc_comm_determinism, req, value);
 343
 344       if (dot_output != NULL) {
 345         MC_SET_MC_HEAP;
 346         req_str = MC_request_get_dot_output(req, value);
 347         MC_SET_STD_HEAP;
 348       }
 349
 350       MC_state_set_executed_request(state, req, value);
 351       mc_stats->executed_transitions++;
 352
 353       /* TODO : handle test and testany simcalls */
 354       mc_call_type call = MC_CALL_TYPE_NONE;
 355       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 356         call = mc_get_call_type(req);
 357       }
 358
 359       /* Answer the request */
 360       SIMIX_simcall_handle(req, value);    /* After this call req is no longer useful */
 361
 362       MC_SET_MC_HEAP;
 363       current_pattern = !initial_global_state->initial_communications_pattern_done ? initial_communications_pattern : communications_pattern;
 364       mc_update_comm_pattern(call, req, value, current_pattern);
 365       MC_SET_STD_HEAP;
 366
 367       /* Wait for requests (schedules processes) */
 368       MC_wait_for_requests();
 369
 370       /* Create the new expanded state */
 371       MC_SET_MC_HEAP;
 372
 373       next_state = MC_state_new();
 374
 375       if ((visited_state = is_visited_state()) == NULL) {
 376
 377         /* Get enabled processes and insert them in the interleave set of the next state */
 378         xbt_swag_foreach(process, simix_global->process_list) {
 379           if (MC_process_is_enabled(process)) {
 380             MC_state_interleave_process(next_state, process);
 381           }
 382         }
 383
 384         if (dot_output != NULL)
 385           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 386                   next_state->num, req_str);
 387
 388       } else {
 389
 390         if (dot_output != NULL)
 391           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 392                   visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 393
 394       }
 395
 396       xbt_fifo_unshift(mc_stack, next_state);
 397
 398       if (dot_output != NULL)
 399         xbt_free(req_str);
 400
 401       MC_SET_STD_HEAP;
 402
 403     } else {
 404
 405       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 406         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 407       } else if (visited_state != NULL) {
 408         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);
 409       } else {
 410         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 411       }
 412
 413       MC_SET_MC_HEAP;
 414
 415       if (initial_global_state->initial_communications_pattern_done) {
 416         if (!visited_state) {
 417           deterministic_pattern(communications_pattern, 0);
 418         } else {
 419           deterministic_pattern(communications_pattern, 1);
 420         }
 421
 422         if (_sg_mc_comms_determinism && !initial_global_state->comm_deterministic) {
 423             XBT_INFO("****************************************************");
 424             XBT_INFO("***** Non-deterministic communications pattern *****");
 425             XBT_INFO("****************************************************");
 426             XBT_INFO("** Initial communications pattern (per process): **");
 427             print_communications_pattern(initial_communications_pattern);
 428             XBT_INFO("** Communications pattern counter-example (per process): **");
 429             print_communications_pattern(communications_pattern);
 430             MC_print_statistics(mc_stats);
 431             MC_SET_STD_HEAP;
 432             return;
 433           } else if (_sg_mc_send_determinism && !initial_global_state->send_deterministic) {
 434             XBT_INFO
 435                 ("*********************************************************");
 436             XBT_INFO
 437                 ("***** Non-send-deterministic communications pattern *****");
 438             XBT_INFO
 439                 ("*********************************************************");
 440             XBT_INFO("** Initial communications pattern: **");
 441             print_communications_pattern(initial_communications_pattern);
 442             XBT_INFO("** Communications pattern counter-example: **");
 443             print_communications_pattern(communications_pattern);
 444             MC_print_statistics(mc_stats);
 445             MC_SET_STD_HEAP;
 446             return;
 447         }
 448
 449       } else {
 450         initial_global_state->initial_communications_pattern_done = 1;
 451       }
 452
 453       /* Trash the current state, no longer needed */
 454       xbt_fifo_shift(mc_stack);
 455       MC_state_delete(state);
 456       XBT_DEBUG("Delete state %d at depth %d", state->num,
 457                 xbt_fifo_size(mc_stack) + 1);
 458
 459       MC_SET_STD_HEAP;
 460
 461       visited_state = NULL;
 462
 463       /* Check for deadlocks */
 464       if (MC_deadlock_check()) {
 465         MC_show_deadlock(NULL);
 466         return;
 467       }
 468
 469       MC_SET_MC_HEAP;
 470
 471       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 472         if (MC_state_interleave_size(state)
 473             && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 474           /* We found a back-tracking point, let's loop */
 475           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num,
 476                     xbt_fifo_size(mc_stack) + 1);
 477           xbt_fifo_unshift(mc_stack, state);
 478           MC_SET_STD_HEAP;
 479
 480           MC_replay(mc_stack, -1);
 481
 482           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num,
 483                     xbt_fifo_size(mc_stack));
 484           break;
 485         } else {
 486           XBT_DEBUG("Delete state %d at depth %d", state->num,
 487                     xbt_fifo_size(mc_stack) + 1);
 488           MC_state_delete(state);
 489         }
 490       }
 491
 492       MC_SET_STD_HEAP;
 493     }
 494   }
 495
 496   MC_print_statistics(mc_stats);
 497   MC_SET_STD_HEAP;
 498
 499   return;
 500 }