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
   9 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  10                                 "Logging specific to MC communication determinism detection");
  11
  12 /********** Global variables **********/
  13
  14 xbt_dynar_t initial_communications_pattern;
  15 xbt_dynar_t incomplete_communications_pattern;
  16 xbt_dynar_t communications_pattern;
  17 int nb_comm_pattern;
  18
  19 /********** Static functions ***********/
  20
  21 static void comm_pattern_free(mc_comm_pattern_t p)
  22 {
  23   xbt_free(p->rdv);
  24   xbt_free(p->data);
  25   xbt_free(p);
  26   p = NULL;
  27 }
  28
  29 static void comm_pattern_free_voidp(void *p)
  30 {
  31   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
  32 }
  33
  34 static mc_comm_pattern_t get_comm_pattern_from_idx(xbt_dynar_t pattern,
  35                                                    unsigned int *idx,
  36                                                    e_smx_comm_type_t type,
  37                                                    unsigned long proc)
  38 {
  39   mc_comm_pattern_t current_comm;
  40   while (*idx < xbt_dynar_length(pattern)) {
  41     current_comm =
  42         (mc_comm_pattern_t) xbt_dynar_get_as(pattern, *idx, mc_comm_pattern_t);
  43     if (current_comm->type == type && type == SIMIX_COMM_SEND) {
  44       if (current_comm->src_proc == proc)
  45         return current_comm;
  46     } else if (current_comm->type == type && type == SIMIX_COMM_RECEIVE) {
  47       if (current_comm->dst_proc == proc)
  48         return current_comm;
  49     }
  50     (*idx)++;
  51   }
  52   return NULL;
  53 }
  54
  55 static int compare_comm_pattern(mc_comm_pattern_t comm1,
  56                                 mc_comm_pattern_t comm2)
  57 {
  58   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  59     return 1;
  60   if (comm1->src_proc != comm2->src_proc)
  61     return 1;
  62   if (comm1->dst_proc != comm2->dst_proc)
  63     return 1;
  64   if (comm1->data_size != comm2->data_size)
  65     return 1;
  66   if (memcmp(comm1->data, comm2->data, comm1->data_size) != 0)
  67     return 1;
  68   return 0;
  69 }
  70
  71 static void deterministic_pattern(xbt_dynar_t pattern, int partial)
  72 {
  73
  74   unsigned int cursor = 0, send_index = 0, recv_index = 0;
  75   mc_comm_pattern_t comm1, comm2;
  76   unsigned int current_process = 1; /* Process 0 corresponds to maestro */
  77   unsigned int nb_comms1, nb_comms2;
  78   xbt_dynar_t process_comms_pattern1, process_comms_pattern2;
  79
  80   while (current_process < simix_process_maxpid) {
  81     process_comms_pattern1 = (xbt_dynar_t)xbt_dynar_get_as(initial_communications_pattern, current_process, xbt_dynar_t);
  82     process_comms_pattern2 = (xbt_dynar_t)xbt_dynar_get_as(pattern, current_process, xbt_dynar_t);
  83     nb_comms1 = xbt_dynar_length(process_comms_pattern1);
  84     nb_comms2 = xbt_dynar_length(process_comms_pattern2);
  85     if(!xbt_dynar_is_empty((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, current_process, xbt_dynar_t)))
  86       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)));
  87     if (!partial && (nb_comms1 != nb_comms2)) {
  88       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);
  89       initial_global_state->send_deterministic = -1;
  90       initial_global_state->comm_deterministic = -1;
  91     } else {
  92       while (cursor < nb_comms2) {
  93         comm1 = (mc_comm_pattern_t)xbt_dynar_get_as(process_comms_pattern1, cursor, mc_comm_pattern_t);
  94         if (comm1->type == SIMIX_COMM_SEND) {
  95           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &send_index, comm1->type, current_process);
  96           if (compare_comm_pattern(comm1, comm2)) {
  97             XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
  98             initial_global_state->send_deterministic = 0;
  99             initial_global_state->comm_deterministic = 0;
 100             return;
 101           }
 102           send_index++;
 103         } else if (comm1->type == SIMIX_COMM_RECEIVE) {
 104           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &recv_index, comm1->type, current_process);
 105           if (compare_comm_pattern(comm1, comm2)) {
 106             initial_global_state->comm_deterministic = 0;
 107             if (!_sg_mc_send_determinism){
 108               XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
 109               return;
 110             }
 111           }
 112           recv_index++;
 113         }
 114         cursor++;
 115       }
 116     }
 117     current_process++;
 118     cursor = 0;
 119     send_index = 0;
 120     recv_index = 0;
 121   }
 122 }
 123
 124 static void print_communications_pattern(xbt_dynar_t comms_pattern)
 125 {
 126   unsigned int cursor = 0;
 127   mc_comm_pattern_t current_comm;
 128   unsigned int current_process = 1;
 129   xbt_dynar_t current_pattern;
 130   while (current_process < simix_process_maxpid) {
 131     current_pattern = (xbt_dynar_t)xbt_dynar_get_as(comms_pattern, current_process, xbt_dynar_t);
 132     XBT_INFO("Communications from the process %u:", current_process);
 133     xbt_dynar_foreach(current_pattern, cursor, current_comm) {
 134       if (current_comm->type == SIMIX_COMM_SEND) {
 135         XBT_INFO("[(%lu) %s -> (%lu) %s] %s ", current_comm->src_proc,
 136                  current_comm->src_host, current_comm->dst_proc,
 137                  current_comm->dst_host, "iSend");
 138       } else {
 139         XBT_INFO("[(%lu) %s <- (%lu) %s] %s ", current_comm->dst_proc,
 140                  current_comm->dst_host, current_comm->src_proc,
 141                  current_comm->src_host, "iRecv");
 142       }
 143     }
 144     current_process++;
 145     cursor = 0;
 146   }
 147 }
 148
 149 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_action_t comm)
 150 {
 151   void *addr_pointed;
 152   comm_pattern->src_proc = comm->comm.src_proc->pid;
 153   comm_pattern->dst_proc = comm->comm.dst_proc->pid;
 154   comm_pattern->src_host =
 155     simcall_host_get_name(comm->comm.src_proc->smx_host);
 156   comm_pattern->dst_host =
 157     simcall_host_get_name(comm->comm.dst_proc->smx_host);
 158   if (comm_pattern->data_size == -1) {
 159     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 160     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 161     addr_pointed = *(void **) comm->comm.src_buff;
 162     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 163       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 164     else
 165       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 166   }
 167 }
 168
 169 /********** Non Static functions ***********/
 170
 171 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, mc_call_type call_type)
 172 {
 173   mc_comm_pattern_t pattern = NULL;
 174   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 175   pattern->num = ++nb_comm_pattern;
 176   pattern->data_size = -1;
 177   void *addr_pointed;
 178   if (call_type == MC_CALL_TYPE_SEND) {              // ISEND
 179     pattern->type = SIMIX_COMM_SEND;
 180     pattern->comm = simcall_comm_isend__get__result(request);
 181     pattern->src_proc = pattern->comm->comm.src_proc->pid;
 182     pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
 183     pattern->data_size = pattern->comm->comm.src_buff_size;
 184     pattern->data = xbt_malloc0(pattern->data_size);
 185     addr_pointed = *(void **) pattern->comm->comm.src_buff;
 186     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 187       memcpy(pattern->data, addr_pointed, pattern->data_size);
 188     else
 189       memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 190   } else if (call_type == MC_CALL_TYPE_RECV) {                      // IRECV
 191     pattern->type = SIMIX_COMM_RECEIVE;
 192     pattern->comm = simcall_comm_irecv__get__result(request);
 193     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
 194     pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
 195   } else {
 196     xbt_die("Unexpected call_type %i", (int) call_type);
 197   }
 198
 199   if (pattern->comm->comm.rdv != NULL)
 200     pattern->rdv = strdup(pattern->comm->comm.rdv->name);
 201   else
 202     pattern->rdv = strdup(pattern->comm->comm.rdv_cpy->name);
 203
 204   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(list, request->issuer->pid, xbt_dynar_t), &pattern);
 205
 206   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);
 207
 208 }
 209
 210 void complete_comm_pattern(xbt_dynar_t list, smx_action_t comm)
 211 {
 212   mc_comm_pattern_t current_comm_pattern;
 213   unsigned int cursor = 0;
 214   int index;
 215   unsigned int src = comm->comm.src_proc->pid;
 216   unsigned int dst = comm->comm.dst_proc->pid;
 217   int src_completed = 0, dst_completed = 0;
 218
 219   /* Looking for the corresponding communication in the comm pattern list of the src process */
 220   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, index){
 221     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);
 222     if(current_comm_pattern->comm == comm){
 223       update_comm_pattern(current_comm_pattern, comm);
 224       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, NULL);
 225       src_completed = 1;
 226       break;
 227     }
 228   }
 229
 230   if(!src_completed)
 231     xbt_die("Corresponding communication for the source process not found!");
 232
 233   cursor = 0;
 234
 235   /* Looking for the corresponding communication in the comm pattern list of the dst process */
 236   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, index){
 237     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);
 238     if(current_comm_pattern->comm == comm){
 239       update_comm_pattern(current_comm_pattern, comm);
 240       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, NULL);
 241       dst_completed = 1;
 242       break;
 243     }
 244   }
 245
 246   if(!dst_completed)
 247     xbt_die("Corresponding communication for the dest process not found!");
 248
 249
 250 }
 251
 252 /************************ Main algorithm ************************/
 253
 254 void MC_pre_modelcheck_comm_determinism(void)
 255 {
 256
 257   int mc_mem_set = (mmalloc_get_current_heap() == mc_heap);
 258
 259   mc_state_t initial_state = NULL;
 260   smx_process_t process;
 261   int i;
 262
 263   if (!mc_mem_set)
 264     MC_SET_MC_HEAP;
 265
 266   if (_sg_mc_visited > 0)
 267     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 268
 269   initial_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 270   for (i=0; i<simix_process_maxpid; i++){
 271     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 272     xbt_dynar_insert_at(initial_communications_pattern, i, &process_pattern);
 273   }
 274   communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 275   for (i=0; i<simix_process_maxpid; i++){
 276     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 277     xbt_dynar_insert_at(communications_pattern, i, &process_pattern);
 278   }
 279   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 280   for (i=0; i<simix_process_maxpid; i++){
 281     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(int), NULL);
 282     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 283   }
 284
 285   nb_comm_pattern = 0;
 286
 287   initial_state = MC_state_new();
 288
 289   MC_SET_STD_HEAP;
 290
 291   XBT_DEBUG("********* Start communication determinism verification *********");
 292
 293   /* Wait for requests (schedules processes) */
 294   MC_wait_for_requests();
 295
 296   MC_SET_MC_HEAP;
 297
 298   /* Get an enabled process and insert it in the interleave set of the initial state */
 299   xbt_swag_foreach(process, simix_global->process_list) {
 300     if (MC_process_is_enabled(process)) {
 301       MC_state_interleave_process(initial_state, process);
 302     }
 303   }
 304
 305   xbt_fifo_unshift(mc_stack, initial_state);
 306
 307   MC_SET_STD_HEAP;
 308
 309 }
 310
 311 void MC_modelcheck_comm_determinism(void)
 312 {
 313
 314   char *req_str = NULL;
 315   int value;
 316   mc_visited_state_t visited_state = NULL;
 317   smx_simcall_t req = NULL;
 318   smx_process_t process = NULL;
 319   mc_state_t state = NULL, next_state = NULL;
 320   smx_action_t current_comm;
 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_enter(req, value);    /* After this call req is no longer usefull */
 361
 362       MC_SET_MC_HEAP;
 363       current_pattern = !initial_global_state->initial_communications_pattern_done ? initial_communications_pattern : communications_pattern;
 364       if (call == MC_CALL_TYPE_SEND) { /* Send */
 365         get_comm_pattern(current_pattern, req, call);
 366       } else if (call == MC_CALL_TYPE_RECV) { /* Recv */
 367         get_comm_pattern(current_pattern, req, call);
 368       } else if (call == MC_CALL_TYPE_WAIT) { /* Wait */
 369         current_comm = simcall_comm_wait__get__comm(req);
 370         if (current_comm->comm.refcount == 1)  /* First wait only must be considered */
 371           complete_comm_pattern(current_pattern, current_comm);
 372       } else if (call == MC_CALL_TYPE_WAITANY) { /* WaitAny */
 373         current_comm = xbt_dynar_get_as(simcall_comm_waitany__get__comms(req), value, smx_action_t);
 374         if (current_comm->comm.refcount == 1) /* First wait only must be considered */
 375           complete_comm_pattern(current_pattern, current_comm);
 376       }
 377       MC_SET_STD_HEAP;
 378
 379       /* Wait for requests (schedules processes) */
 380       MC_wait_for_requests();
 381
 382       /* Create the new expanded state */
 383       MC_SET_MC_HEAP;
 384
 385       next_state = MC_state_new();
 386
 387       if ((visited_state = is_visited_state()) == NULL) {
 388
 389         /* Get enabled processes and insert them in the interleave set of the next state */
 390         xbt_swag_foreach(process, simix_global->process_list) {
 391           if (MC_process_is_enabled(process)) {
 392             MC_state_interleave_process(next_state, process);
 393           }
 394         }
 395
 396         if (dot_output != NULL)
 397           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 398                   next_state->num, req_str);
 399
 400       } else {
 401
 402         if (dot_output != NULL)
 403           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 404                   visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 405
 406       }
 407
 408       xbt_fifo_unshift(mc_stack, next_state);
 409
 410       if (dot_output != NULL)
 411         xbt_free(req_str);
 412
 413       MC_SET_STD_HEAP;
 414
 415     } else {
 416
 417       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 418         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 419       } else if (visited_state != NULL) {
 420         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);
 421       } else {
 422         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 423       }
 424
 425       MC_SET_MC_HEAP;
 426
 427       if (initial_global_state->initial_communications_pattern_done) {
 428         if (!visited_state) {
 429           deterministic_pattern(communications_pattern, 0);
 430         } else {
 431           deterministic_pattern(communications_pattern, 1);
 432         }
 433
 434         if (_sg_mc_comms_determinism && !initial_global_state->comm_deterministic) {
 435             XBT_INFO("****************************************************");
 436             XBT_INFO("***** Non-deterministic communications pattern *****");
 437             XBT_INFO("****************************************************");
 438             XBT_INFO("** Initial communications pattern (per process): **");
 439             print_communications_pattern(initial_communications_pattern);
 440             XBT_INFO("** Communications pattern counter-example (per process): **");
 441             print_communications_pattern(communications_pattern);
 442             MC_print_statistics(mc_stats);
 443             MC_SET_STD_HEAP;
 444             return;
 445           } else if (_sg_mc_send_determinism && !initial_global_state->send_deterministic) {
 446             XBT_INFO
 447                 ("*********************************************************");
 448             XBT_INFO
 449                 ("***** Non-send-deterministic communications pattern *****");
 450             XBT_INFO
 451                 ("*********************************************************");
 452             XBT_INFO("** Initial communications pattern: **");
 453             print_communications_pattern(initial_communications_pattern);
 454             XBT_INFO("** Communications pattern counter-example: **");
 455             print_communications_pattern(communications_pattern);
 456             MC_print_statistics(mc_stats);
 457             MC_SET_STD_HEAP;
 458             return;
 459         }
 460
 461       } else {
 462         initial_global_state->initial_communications_pattern_done = 1;
 463       }
 464
 465       /* Trash the current state, no longer needed */
 466       xbt_fifo_shift(mc_stack);
 467       MC_state_delete(state);
 468       XBT_DEBUG("Delete state %d at depth %d", state->num,
 469                 xbt_fifo_size(mc_stack) + 1);
 470
 471       MC_SET_STD_HEAP;
 472
 473       visited_state = NULL;
 474
 475       /* Check for deadlocks */
 476       if (MC_deadlock_check()) {
 477         MC_show_deadlock(NULL);
 478         return;
 479       }
 480
 481       MC_SET_MC_HEAP;
 482
 483       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 484         if (MC_state_interleave_size(state)
 485             && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 486           /* We found a back-tracking point, let's loop */
 487           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num,
 488                     xbt_fifo_size(mc_stack) + 1);
 489           xbt_fifo_unshift(mc_stack, state);
 490           MC_SET_STD_HEAP;
 491
 492           MC_replay(mc_stack, -1);
 493
 494           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num,
 495                     xbt_fifo_size(mc_stack));
 496           break;
 497         } else {
 498           XBT_DEBUG("Delete state %d at depth %d", state->num,
 499                     xbt_fifo_size(mc_stack) + 1);
 500           MC_state_delete(state);
 501         }
 502       }
 503
 504       MC_SET_STD_HEAP;
 505     }
 506   }
 507
 508   MC_print_statistics(mc_stats);
 509   MC_SET_STD_HEAP;
 510
 511   return;
 512 }