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_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
  14 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  15                                 "Logging specific to MC communication determinism detection");
  16
  17 /********** Global variables **********/
  18
  19 xbt_dynar_t initial_communications_pattern;
  20 xbt_dynar_t incomplete_communications_pattern;
  21 xbt_dynar_t communications_pattern;
  22 int nb_comm_pattern;
  23
  24 /********** Static functions ***********/
  25
  26 static void comm_pattern_free(mc_comm_pattern_t p)
  27 {
  28   xbt_free(p->rdv);
  29   xbt_free(p->data);
  30   xbt_free(p);
  31   p = NULL;
  32 }
  33
  34 static void comm_pattern_free_voidp(void *p)
  35 {
  36   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
  37 }
  38
  39 static mc_comm_pattern_t get_comm_pattern_from_idx(xbt_dynar_t pattern,
  40                                                    unsigned int *idx,
  41                                                    e_smx_comm_type_t type,
  42                                                    unsigned long proc)
  43 {
  44   mc_comm_pattern_t current_comm;
  45   while (*idx < xbt_dynar_length(pattern)) {
  46     current_comm =
  47         (mc_comm_pattern_t) xbt_dynar_get_as(pattern, *idx, mc_comm_pattern_t);
  48     if (current_comm->type == type && type == SIMIX_COMM_SEND) {
  49       if (current_comm->src_proc == proc)
  50         return current_comm;
  51     } else if (current_comm->type == type && type == SIMIX_COMM_RECEIVE) {
  52       if (current_comm->dst_proc == proc)
  53         return current_comm;
  54     }
  55     (*idx)++;
  56   }
  57   return NULL;
  58 }
  59
  60 static int compare_comm_pattern(mc_comm_pattern_t comm1,
  61                                 mc_comm_pattern_t comm2)
  62 {
  63   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  64     return 1;
  65   if (comm1->src_proc != comm2->src_proc)
  66     return 1;
  67   if (comm1->dst_proc != comm2->dst_proc)
  68     return 1;
  69   if (comm1->data_size != comm2->data_size)
  70     return 1;
  71   if (memcmp(comm1->data, comm2->data, comm1->data_size) != 0)
  72     return 1;
  73   return 0;
  74 }
  75
  76 static void deterministic_pattern(xbt_dynar_t pattern, int partial)
  77 {
  78
  79   unsigned int cursor = 0, send_index = 0, recv_index = 0;
  80   mc_comm_pattern_t comm1, comm2;
  81   unsigned int current_process = 1; /* Process 0 corresponds to maestro */
  82   unsigned int nb_comms1, nb_comms2;
  83   xbt_dynar_t process_comms_pattern1, process_comms_pattern2;
  84
  85   while (current_process < simix_process_maxpid) {
  86     process_comms_pattern1 = (xbt_dynar_t)xbt_dynar_get_as(initial_communications_pattern, current_process, xbt_dynar_t);
  87     process_comms_pattern2 = (xbt_dynar_t)xbt_dynar_get_as(pattern, current_process, xbt_dynar_t);
  88     nb_comms1 = xbt_dynar_length(process_comms_pattern1);
  89     nb_comms2 = xbt_dynar_length(process_comms_pattern2);
  90     if(!xbt_dynar_is_empty((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, current_process, xbt_dynar_t)))
  91       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)));
  92     if (!partial && (nb_comms1 != nb_comms2)) {
  93       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);
  94       initial_global_state->send_deterministic = -1;
  95       initial_global_state->comm_deterministic = -1;
  96     } else {
  97       while (cursor < nb_comms2) {
  98         comm1 = (mc_comm_pattern_t)xbt_dynar_get_as(process_comms_pattern1, cursor, mc_comm_pattern_t);
  99         if (comm1->type == SIMIX_COMM_SEND) {
 100           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &send_index, comm1->type, current_process);
 101           if (compare_comm_pattern(comm1, comm2)) {
 102             XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
 103             initial_global_state->send_deterministic = 0;
 104             initial_global_state->comm_deterministic = 0;
 105             return;
 106           }
 107           send_index++;
 108         } else if (comm1->type == SIMIX_COMM_RECEIVE) {
 109           comm2 = get_comm_pattern_from_idx(process_comms_pattern2, &recv_index, comm1->type, current_process);
 110           if (compare_comm_pattern(comm1, comm2)) {
 111             initial_global_state->comm_deterministic = 0;
 112             if (!_sg_mc_send_determinism){
 113               XBT_INFO("The communications pattern of the process %u is different! (Different communication : %u)", current_process, cursor+1);
 114               return;
 115             }
 116           }
 117           recv_index++;
 118         }
 119         cursor++;
 120       }
 121     }
 122     current_process++;
 123     cursor = 0;
 124     send_index = 0;
 125     recv_index = 0;
 126   }
 127 }
 128
 129 static void print_communications_pattern(xbt_dynar_t comms_pattern)
 130 {
 131   unsigned int cursor = 0;
 132   mc_comm_pattern_t current_comm;
 133   unsigned int current_process = 1;
 134   xbt_dynar_t current_pattern;
 135   while (current_process < simix_process_maxpid) {
 136     current_pattern = (xbt_dynar_t)xbt_dynar_get_as(comms_pattern, current_process, xbt_dynar_t);
 137     XBT_INFO("Communications from the process %u:", current_process);
 138     xbt_dynar_foreach(current_pattern, cursor, current_comm) {
 139       if (current_comm->type == SIMIX_COMM_SEND) {
 140         XBT_INFO("[(%lu) %s -> (%lu) %s] %s ", current_comm->src_proc,
 141                  current_comm->src_host, current_comm->dst_proc,
 142                  current_comm->dst_host, "iSend");
 143       } else {
 144         XBT_INFO("[(%lu) %s <- (%lu) %s] %s ", current_comm->dst_proc,
 145                  current_comm->dst_host, current_comm->src_proc,
 146                  current_comm->src_host, "iRecv");
 147       }
 148     }
 149     current_process++;
 150     cursor = 0;
 151   }
 152 }
 153
 154 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
 155 {
 156   void *addr_pointed;
 157   comm_pattern->src_proc = comm->comm.src_proc->pid;
 158   comm_pattern->dst_proc = comm->comm.dst_proc->pid;
 159   comm_pattern->src_host =
 160     simcall_host_get_name(comm->comm.src_proc->smx_host);
 161   comm_pattern->dst_host =
 162     simcall_host_get_name(comm->comm.dst_proc->smx_host);
 163   if (comm_pattern->data_size == -1) {
 164     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 165     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 166     addr_pointed = *(void **) comm->comm.src_buff;
 167     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 168       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 169     else
 170       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 171   }
 172 }
 173
 174 /********** Non Static functions ***********/
 175
 176 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, mc_call_type call_type)
 177 {
 178   mc_comm_pattern_t pattern = NULL;
 179   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 180   pattern->num = ++nb_comm_pattern;
 181   pattern->data_size = -1;
 182   void *addr_pointed;
 183   if (call_type == MC_CALL_TYPE_SEND) {              // ISEND
 184     pattern->type = SIMIX_COMM_SEND;
 185     pattern->comm = simcall_comm_isend__get__result(request);
 186     pattern->src_proc = pattern->comm->comm.src_proc->pid;
 187     pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
 188     pattern->data_size = pattern->comm->comm.src_buff_size;
 189     pattern->data = xbt_malloc0(pattern->data_size);
 190     addr_pointed = *(void **) pattern->comm->comm.src_buff;
 191     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 192       memcpy(pattern->data, addr_pointed, pattern->data_size);
 193     else
 194       memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 195   } else if (call_type == MC_CALL_TYPE_RECV) {                      // IRECV
 196     pattern->type = SIMIX_COMM_RECEIVE;
 197     pattern->comm = simcall_comm_irecv__get__result(request);
 198     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
 199     pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
 200   } else {
 201     xbt_die("Unexpected call_type %i", (int) call_type);
 202   }
 203
 204   if (pattern->comm->comm.rdv != NULL)
 205     pattern->rdv = strdup(pattern->comm->comm.rdv->name);
 206   else
 207     pattern->rdv = strdup(pattern->comm->comm.rdv_cpy->name);
 208
 209   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(list, request->issuer->pid, xbt_dynar_t), &pattern);
 210
 211   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);
 212
 213 }
 214
 215 void complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm)
 216 {
 217   mc_comm_pattern_t current_comm_pattern;
 218   unsigned int cursor = 0;
 219   int index;
 220   unsigned int src = comm->comm.src_proc->pid;
 221   unsigned int dst = comm->comm.dst_proc->pid;
 222   int src_completed = 0, dst_completed = 0;
 223
 224   /* Looking for the corresponding communication in the comm pattern list of the src process */
 225   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, index){
 226     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);
 227     if(current_comm_pattern->comm == comm){
 228       update_comm_pattern(current_comm_pattern, comm);
 229       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, src, xbt_dynar_t), cursor, NULL);
 230       src_completed = 1;
 231       break;
 232     }
 233   }
 234
 235   if(!src_completed)
 236     xbt_die("Corresponding communication for the source process not found!");
 237
 238   cursor = 0;
 239
 240   /* Looking for the corresponding communication in the comm pattern list of the dst process */
 241   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, index){
 242     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);
 243     if(current_comm_pattern->comm == comm){
 244       update_comm_pattern(current_comm_pattern, comm);
 245       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, dst, xbt_dynar_t), cursor, NULL);
 246       dst_completed = 1;
 247       break;
 248     }
 249   }
 250
 251   if(!dst_completed)
 252     xbt_die("Corresponding communication for the dest process not found!");
 253
 254
 255 }
 256
 257 /************************ Main algorithm ************************/
 258
 259 void MC_pre_modelcheck_comm_determinism(void)
 260 {
 261
 262   int mc_mem_set = (mmalloc_get_current_heap() == mc_heap);
 263
 264   mc_state_t initial_state = NULL;
 265   smx_process_t process;
 266   int i;
 267
 268   if (!mc_mem_set)
 269     MC_SET_MC_HEAP;
 270
 271   if (_sg_mc_visited > 0)
 272     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 273
 274   initial_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(initial_communications_pattern, i, &process_pattern);
 278   }
 279   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(mc_comm_pattern_t), comm_pattern_free_voidp);
 282     xbt_dynar_insert_at(communications_pattern, i, &process_pattern);
 283   }
 284   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 285   for (i=0; i<simix_process_maxpid; i++){
 286     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(int), NULL);
 287     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 288   }
 289
 290   nb_comm_pattern = 0;
 291
 292   initial_state = MC_state_new();
 293
 294   MC_SET_STD_HEAP;
 295
 296   XBT_DEBUG("********* Start communication determinism verification *********");
 297
 298   /* Wait for requests (schedules processes) */
 299   MC_wait_for_requests();
 300
 301   MC_SET_MC_HEAP;
 302
 303   /* Get an enabled process and insert it in the interleave set of the initial state */
 304   xbt_swag_foreach(process, simix_global->process_list) {
 305     if (MC_process_is_enabled(process)) {
 306       MC_state_interleave_process(initial_state, process);
 307     }
 308   }
 309
 310   xbt_fifo_unshift(mc_stack, initial_state);
 311
 312   MC_SET_STD_HEAP;
 313
 314 }
 315
 316 void MC_modelcheck_comm_determinism(void)
 317 {
 318
 319   char *req_str = NULL;
 320   int value;
 321   mc_visited_state_t visited_state = NULL;
 322   smx_simcall_t req = NULL;
 323   smx_process_t process = NULL;
 324   mc_state_t state = NULL, next_state = NULL;
 325   xbt_dynar_t current_pattern;
 326
 327   while (xbt_fifo_size(mc_stack) > 0) {
 328
 329     /* Get current state */
 330     state =
 331         (mc_state_t)
 332         xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 333
 334     XBT_DEBUG("**************************************************");
 335     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 336               xbt_fifo_size(mc_stack), state->num,
 337               MC_state_interleave_size(state));
 338
 339     /* Update statistics */
 340     mc_stats->visited_states++;
 341
 342     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 343         && (req = MC_state_get_request(state, &value))
 344         && (visited_state == NULL)) {
 345
 346       MC_LOG_REQUEST(mc_comm_determinism, req, value);
 347
 348       if (dot_output != NULL) {
 349         MC_SET_MC_HEAP;
 350         req_str = MC_request_get_dot_output(req, value);
 351         MC_SET_STD_HEAP;
 352       }
 353
 354       MC_state_set_executed_request(state, req, value);
 355       mc_stats->executed_transitions++;
 356
 357       /* TODO : handle test and testany simcalls */
 358       mc_call_type call = MC_CALL_TYPE_NONE;
 359       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 360         call = mc_get_call_type(req);
 361       }
 362
 363       /* Answer the request */
 364       SIMIX_simcall_handle(req, value);    /* After this call req is no longer useful */
 365
 366       MC_SET_MC_HEAP;
 367       current_pattern = !initial_global_state->initial_communications_pattern_done ? initial_communications_pattern : communications_pattern;
 368       mc_update_comm_pattern(call, req, value, current_pattern);
 369       MC_SET_STD_HEAP;
 370
 371       /* Wait for requests (schedules processes) */
 372       MC_wait_for_requests();
 373
 374       /* Create the new expanded state */
 375       MC_SET_MC_HEAP;
 376
 377       next_state = MC_state_new();
 378
 379       if ((visited_state = is_visited_state()) == NULL) {
 380
 381         /* Get enabled processes and insert them in the interleave set of the next state */
 382         xbt_swag_foreach(process, simix_global->process_list) {
 383           if (MC_process_is_enabled(process)) {
 384             MC_state_interleave_process(next_state, process);
 385           }
 386         }
 387
 388         if (dot_output != NULL)
 389           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 390                   next_state->num, req_str);
 391
 392       } else {
 393
 394         if (dot_output != NULL)
 395           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,
 396                   visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 397
 398       }
 399
 400       xbt_fifo_unshift(mc_stack, next_state);
 401
 402       if (dot_output != NULL)
 403         xbt_free(req_str);
 404
 405       MC_SET_STD_HEAP;
 406
 407     } else {
 408
 409       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 410         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 411       } else if (visited_state != NULL) {
 412         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);
 413       } else {
 414         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 415       }
 416
 417       MC_SET_MC_HEAP;
 418
 419       if (initial_global_state->initial_communications_pattern_done) {
 420         if (!visited_state) {
 421           deterministic_pattern(communications_pattern, 0);
 422         } else {
 423           deterministic_pattern(communications_pattern, 1);
 424         }
 425
 426         if (_sg_mc_comms_determinism && !initial_global_state->comm_deterministic) {
 427             XBT_INFO("****************************************************");
 428             XBT_INFO("***** Non-deterministic communications pattern *****");
 429             XBT_INFO("****************************************************");
 430             XBT_INFO("** Initial communications pattern (per process): **");
 431             print_communications_pattern(initial_communications_pattern);
 432             XBT_INFO("** Communications pattern counter-example (per process): **");
 433             print_communications_pattern(communications_pattern);
 434             MC_print_statistics(mc_stats);
 435             MC_SET_STD_HEAP;
 436             return;
 437           } else if (_sg_mc_send_determinism && !initial_global_state->send_deterministic) {
 438             XBT_INFO
 439                 ("*********************************************************");
 440             XBT_INFO
 441                 ("***** Non-send-deterministic communications pattern *****");
 442             XBT_INFO
 443                 ("*********************************************************");
 444             XBT_INFO("** Initial communications pattern: **");
 445             print_communications_pattern(initial_communications_pattern);
 446             XBT_INFO("** Communications pattern counter-example: **");
 447             print_communications_pattern(communications_pattern);
 448             MC_print_statistics(mc_stats);
 449             MC_SET_STD_HEAP;
 450             return;
 451         }
 452
 453       } else {
 454         initial_global_state->initial_communications_pattern_done = 1;
 455       }
 456
 457       /* Trash the current state, no longer needed */
 458       xbt_fifo_shift(mc_stack);
 459       MC_state_delete(state);
 460       XBT_DEBUG("Delete state %d at depth %d", state->num,
 461                 xbt_fifo_size(mc_stack) + 1);
 462
 463       MC_SET_STD_HEAP;
 464
 465       visited_state = NULL;
 466
 467       /* Check for deadlocks */
 468       if (MC_deadlock_check()) {
 469         MC_show_deadlock(NULL);
 470         return;
 471       }
 472
 473       MC_SET_MC_HEAP;
 474
 475       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 476         if (MC_state_interleave_size(state)
 477             && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 478           /* We found a back-tracking point, let's loop */
 479           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num,
 480                     xbt_fifo_size(mc_stack) + 1);
 481           xbt_fifo_unshift(mc_stack, state);
 482           MC_SET_STD_HEAP;
 483
 484           MC_replay(mc_stack, -1);
 485
 486           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num,
 487                     xbt_fifo_size(mc_stack));
 488           break;
 489         } else {
 490           XBT_DEBUG("Delete state %d at depth %d", state->num,
 491                     xbt_fifo_size(mc_stack) + 1);
 492           MC_state_delete(state);
 493         }
 494       }
 495
 496       MC_SET_STD_HEAP;
 497     }
 498   }
 499
 500   MC_print_statistics(mc_stats);
 501   MC_SET_STD_HEAP;
 502
 503   return;
 504 }