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
  22 /********** Static functions ***********/
  23
  24 static void comm_pattern_free(mc_comm_pattern_t p)
  25 {
  26   xbt_free(p->rdv);
  27   xbt_free(p->data);
  28   xbt_free(p);
  29   p = NULL;
  30 }
  31
  32 static void list_comm_pattern_free(mc_list_comm_pattern_t l)
  33 {
  34   xbt_dynar_free(&(l->list));
  35   xbt_free(l);
  36   l = NULL;
  37 }
  38
  39 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  40   if(comm1->type != comm2->type)
  41     return TYPE_DIFF;
  42   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  43     return RDV_DIFF;
  44   if (comm1->src_proc != comm2->src_proc)
  45     return SRC_PROC_DIFF;
  46   if (comm1->dst_proc != comm2->dst_proc)
  47     return DST_PROC_DIFF;
  48   if (comm1->tag != comm2->tag)
  49     return TAG_DIFF;
  50   if (comm1->data_size != comm2->data_size)
  51     return DATA_SIZE_DIFF;
  52   if(comm1->data == NULL && comm2->data == NULL)
  53     return 0;
  54   if(comm1->data != NULL && comm2->data !=NULL) {
  55     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  56       return 0;
  57     return DATA_DIFF;
  58   }else{
  59     return DATA_DIFF;
  60   }
  61   return 0;
  62 }
  63
  64 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  65   char *type, *res;
  66
  67   if(comm->type == SIMIX_COMM_SEND)
  68     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  69   else
  70     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  71
  72   switch(diff) {
  73   case TYPE_DIFF:
  74     res = bprintf("%s Different type for communication #%d", type, cursor);
  75     break;
  76   case RDV_DIFF:
  77     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  78     break;
  79   case TAG_DIFF:
  80     res = bprintf("%s Different tag for communication #%d", type, cursor);
  81     break;
  82   case SRC_PROC_DIFF:
  83       res = bprintf("%s Different source for communication #%d", type, cursor);
  84     break;
  85   case DST_PROC_DIFF:
  86       res = bprintf("%s Different destination for communication #%d", type, cursor);
  87     break;
  88   case DATA_SIZE_DIFF:
  89     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  90     break;
  91   case DATA_DIFF:
  92     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  93     break;
  94   default:
  95     break;
  96   }
  97
  98   return res;
  99 }
 100
 101 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
 102 {
 103   void *addr_pointed;
 104   comm_pattern->src_proc = comm->comm.src_proc->pid;
 105   comm_pattern->dst_proc = comm->comm.dst_proc->pid;
 106   comm_pattern->src_host = simcall_host_get_name(comm->comm.src_proc->smx_host);
 107   comm_pattern->dst_host = simcall_host_get_name(comm->comm.dst_proc->smx_host);
 108   if (comm_pattern->data_size == -1 && comm->comm.src_buff != NULL) {
 109     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 110     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 111     addr_pointed = *(void **) comm->comm.src_buff;
 112     if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 113       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 114     else
 115       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 116   }
 117 }
 118
 119 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 120
 121   mc_list_comm_pattern_t list_comm_pattern = (mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, process, mc_list_comm_pattern_t);
 122
 123   if(!backtracking){
 124     mc_comm_pattern_t initial_comm = xbt_dynar_get_as(list_comm_pattern->list, list_comm_pattern->index_comm, mc_comm_pattern_t);
 125     e_mc_comm_pattern_difference_t diff;
 126
 127     if((diff = compare_comm_pattern(initial_comm, comm)) != 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_comm_pattern->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_comm_pattern->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   comm_pattern_free(comm);
 165
 166 }
 167
 168 /********** Non Static functions ***********/
 169
 170 void comm_pattern_free_voidp(void *p) {
 171   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
 172 }
 173
 174 void list_comm_pattern_free_voidp(void *p) {
 175   list_comm_pattern_free((mc_list_comm_pattern_t) * (void **) p);
 176 }
 177
 178 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 179 {
 180
 181   mc_comm_pattern_t pattern = NULL;
 182   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 183   pattern->data_size = -1;
 184   pattern->data = NULL;
 185
 186   void *addr_pointed;
 187
 188   if (call_type == MC_CALL_TYPE_SEND) {
 189     /* Create comm pattern */
 190     pattern->type = SIMIX_COMM_SEND;
 191     pattern->comm = simcall_comm_isend__get__result(request);
 192     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 193     pattern->src_proc = pattern->comm->comm.src_proc->pid;
 194     pattern->src_host = simcall_host_get_name(request->issuer->smx_host);
 195     pattern->tag = ((MPI_Request)simcall_comm_isend__get__data(request))->tag;
 196     if(pattern->comm->comm.src_buff != NULL){
 197       pattern->data_size = pattern->comm->comm.src_buff_size;
 198       pattern->data = xbt_malloc0(pattern->data_size);
 199       addr_pointed = *(void **) pattern->comm->comm.src_buff;
 200       if (addr_pointed > (void*) std_heap && addr_pointed < std_heap->breakval)
 201         memcpy(pattern->data, addr_pointed, pattern->data_size);
 202       else
 203         memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 204     }
 205     if(((MPI_Request)simcall_comm_isend__get__data(request))->detached){
 206       if (!initial_global_state->initial_communications_pattern_done) {
 207         /* Store comm pattern */
 208         xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->list, &pattern);
 209       } else {
 210         /* Evaluate comm determinism */
 211         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 212         ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->index_comm++;
 213       }
 214       return;
 215     }
 216   } else if (call_type == MC_CALL_TYPE_RECV) {
 217     pattern->type = SIMIX_COMM_RECEIVE;
 218     pattern->comm = simcall_comm_irecv__get__result(request);
 219     pattern->tag = ((MPI_Request)simcall_comm_irecv__get__data(request))->tag;
 220     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 221     pattern->dst_proc = pattern->comm->comm.dst_proc->pid;
 222     pattern->dst_host = simcall_host_get_name(request->issuer->smx_host);
 223   } else {
 224     xbt_die("Unexpected call_type %i", (int) call_type);
 225   }
 226
 227   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, request->issuer->pid, xbt_dynar_t), &pattern);
 228
 229   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, request->issuer->pid);
 230 }
 231
 232 void complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm, unsigned int issuer, int backtracking) {
 233
 234   mc_comm_pattern_t current_comm_pattern;
 235   unsigned int cursor = 0;
 236   mc_comm_pattern_t comm_pattern;
 237   int completed = 0;
 238
 239   /* Complete comm pattern */
 240   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 241     if (current_comm_pattern-> comm == comm) {
 242       update_comm_pattern(current_comm_pattern, comm);
 243       completed = 1;
 244       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &comm_pattern);
 245       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 246       break;
 247     }
 248   }
 249   if(!completed)
 250     xbt_die("Corresponding communication not found!");
 251
 252   if (!initial_global_state->initial_communications_pattern_done) {
 253     /* Store comm pattern */
 254     xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->list, &comm_pattern);
 255   } else {
 256     /* Evaluate comm determinism */
 257     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 258     ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->index_comm++;
 259   }
 260 }
 261
 262
 263 /************************ Main algorithm ************************/
 264
 265 void MC_pre_modelcheck_comm_determinism(void)
 266 {
 267
 268   int mc_mem_set = (mmalloc_get_current_heap() == mc_heap);
 269
 270   mc_state_t initial_state = NULL;
 271   smx_process_t process;
 272   int i;
 273
 274   if (!mc_mem_set)
 275     MC_SET_MC_HEAP;
 276
 277   if (_sg_mc_visited > 0)
 278     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 279
 280   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), list_comm_pattern_free_voidp);
 281   for (i=0; i<simix_process_maxpid; i++){
 282     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 283     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 284     process_list_pattern->index_comm = 0;
 285     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 286   }
 287   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 288   for (i=0; i<simix_process_maxpid; i++){
 289     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 290     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 291   }
 292
 293   initial_state = MC_state_new();
 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
 326   while (xbt_fifo_size(mc_stack) > 0) {
 327
 328     /* Get current state */
 329     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 330
 331     XBT_DEBUG("**************************************************");
 332     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 333               xbt_fifo_size(mc_stack), state->num,
 334               MC_state_interleave_size(state));
 335
 336     /* Update statistics */
 337     mc_stats->visited_states++;
 338
 339     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 340         && (req = MC_state_get_request(state, &value))
 341         && (visited_state == NULL)) {
 342
 343       req_str = MC_request_to_string(req, value);
 344       XBT_DEBUG("Execute: %s", req_str);
 345       xbt_free(req_str);
 346
 347       if (dot_output != NULL) {
 348         MC_SET_MC_HEAP;
 349         req_str = MC_request_get_dot_output(req, value);
 350         MC_SET_STD_HEAP;
 351       }
 352
 353       MC_state_set_executed_request(state, req, value);
 354       mc_stats->executed_transitions++;
 355
 356       /* TODO : handle test and testany simcalls */
 357       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 358       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 359         call = mc_get_call_type(req);
 360       }
 361
 362       /* Answer the request */
 363       SIMIX_simcall_handle(req, value);    /* After this call req is no longer useful */
 364
 365       MC_SET_MC_HEAP;
 366       if(!initial_global_state->initial_communications_pattern_done)
 367         handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 368       else
 369         handle_comm_pattern(call, req, value, NULL, 0);
 370       MC_SET_STD_HEAP;
 371
 372       /* Wait for requests (schedules processes) */
 373       MC_wait_for_requests();
 374
 375       /* Create the new expanded state */
 376       MC_SET_MC_HEAP;
 377
 378       next_state = MC_state_new();
 379
 380       if ((visited_state = is_visited_state(next_state)) == NULL) {
 381
 382         /* Get enabled processes and insert them in the interleave set of the next state */
 383         xbt_swag_foreach(process, simix_global->process_list) {
 384           if (MC_process_is_enabled(process)) {
 385             MC_state_interleave_process(next_state, process);
 386           }
 387         }
 388
 389         if (dot_output != NULL)
 390           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 391
 392       } else {
 393
 394         if (dot_output != NULL)
 395           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 396
 397       }
 398
 399       xbt_fifo_unshift(mc_stack, next_state);
 400
 401       if (dot_output != NULL)
 402         xbt_free(req_str);
 403
 404       MC_SET_STD_HEAP;
 405
 406     } else {
 407
 408       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 409         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 410       } else if (visited_state != NULL) {
 411         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);
 412       } else {
 413         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 414       }
 415
 416       MC_SET_MC_HEAP;
 417
 418       if (!initial_global_state->initial_communications_pattern_done)
 419         initial_global_state->initial_communications_pattern_done = 1;
 420
 421       /* Trash the current state, no longer needed */
 422       xbt_fifo_shift(mc_stack);
 423       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 424       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 425
 426       MC_SET_STD_HEAP;
 427
 428       visited_state = NULL;
 429
 430       /* Check for deadlocks */
 431       if (MC_deadlock_check()) {
 432         MC_show_deadlock(NULL);
 433         return;
 434       }
 435
 436       MC_SET_MC_HEAP;
 437
 438       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 439         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 440           /* We found a back-tracking point, let's loop */
 441           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 442           xbt_fifo_unshift(mc_stack, state);
 443           MC_SET_STD_HEAP;
 444
 445           MC_replay(mc_stack);
 446
 447           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 448
 449           break;
 450         } else {
 451           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 452           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 453         }
 454       }
 455
 456       MC_SET_STD_HEAP;
 457     }
 458   }
 459
 460   MC_print_statistics(mc_stats);
 461   MC_SET_STD_HEAP;
 462
 463   return;
 464 }