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 #include "mc_smx.h"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_comm_determinism, mc,
  16                                 "Logging specific to MC communication determinism detection");
  17
  18 /********** Global variables **********/
  19
  20 xbt_dynar_t initial_communications_pattern;
  21 xbt_dynar_t incomplete_communications_pattern;
  22
  23 /********** Static functions ***********/
  24
  25 static void comm_pattern_free(mc_comm_pattern_t p)
  26 {
  27   xbt_free(p->rdv);
  28   xbt_free(p->data);
  29   xbt_free(p);
  30   p = NULL;
  31 }
  32
  33 static void list_comm_pattern_free(mc_list_comm_pattern_t l)
  34 {
  35   xbt_dynar_free(&(l->list));
  36   xbt_free(l);
  37   l = NULL;
  38 }
  39
  40 static e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  41   if(comm1->type != comm2->type)
  42     return TYPE_DIFF;
  43   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  44     return RDV_DIFF;
  45   if (comm1->src_proc != comm2->src_proc)
  46     return SRC_PROC_DIFF;
  47   if (comm1->dst_proc != comm2->dst_proc)
  48     return DST_PROC_DIFF;
  49   if (comm1->tag != comm2->tag)
  50     return TAG_DIFF;
  51   if (comm1->data_size != comm2->data_size)
  52     return DATA_SIZE_DIFF;
  53   if(comm1->data == NULL && comm2->data == NULL)
  54     return 0;
  55   if(comm1->data != NULL && comm2->data !=NULL) {
  56     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  57       return 0;
  58     return DATA_DIFF;
  59   }else{
  60     return DATA_DIFF;
  61   }
  62   return 0;
  63 }
  64
  65 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  66   char *type, *res;
  67
  68   if(comm->type == SIMIX_COMM_SEND)
  69     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  70   else
  71     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  72
  73   switch(diff) {
  74   case TYPE_DIFF:
  75     res = bprintf("%s Different type for communication #%d", type, cursor);
  76     break;
  77   case RDV_DIFF:
  78     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  79     break;
  80   case TAG_DIFF:
  81     res = bprintf("%s Different tag for communication #%d", type, cursor);
  82     break;
  83   case SRC_PROC_DIFF:
  84       res = bprintf("%s Different source for communication #%d", type, cursor);
  85     break;
  86   case DST_PROC_DIFF:
  87       res = bprintf("%s Different destination for communication #%d", type, cursor);
  88     break;
  89   case DATA_SIZE_DIFF:
  90     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  91     break;
  92   case DATA_DIFF:
  93     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  94     break;
  95   default:
  96     res = NULL;
  97     break;
  98   }
  99
 100   return res;
 101 }
 102
 103 // FIXME, remote comm
 104 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
 105 {
 106   mc_process_t process = &mc_model_checker->process;
 107   void *addr_pointed;
 108   smx_process_t src_proc = MC_smx_resolve_process(comm->comm.src_proc);
 109   smx_process_t dst_proc = MC_smx_resolve_process(comm->comm.dst_proc);
 110   comm_pattern->src_proc = src_proc->pid;
 111   comm_pattern->dst_proc = dst_proc->pid;
 112   // TODO, resolve host name
 113   comm_pattern->src_host = MC_smx_process_get_host_name(src_proc);
 114   comm_pattern->dst_host = MC_smx_process_get_host_name(dst_proc);
 115   if (comm_pattern->data_size == -1 && comm->comm.src_buff != NULL) {
 116     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 117     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 118     addr_pointed = *(void **) comm->comm.src_buff;
 119     if (addr_pointed > (void*) process->heap_address
 120         && addr_pointed < MC_process_get_heap(process)->breakval)
 121       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 122     else
 123       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 124   }
 125 }
 126
 127 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 128
 129   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);
 130
 131   if(!backtracking){
 132     mc_comm_pattern_t initial_comm = xbt_dynar_get_as(list_comm_pattern->list, list_comm_pattern->index_comm, mc_comm_pattern_t);
 133     e_mc_comm_pattern_difference_t diff;
 134
 135     if((diff = compare_comm_pattern(initial_comm, comm)) != NONE_DIFF){
 136       if (comm->type == SIMIX_COMM_SEND){
 137         initial_global_state->send_deterministic = 0;
 138         if(initial_global_state->send_diff != NULL)
 139           xbt_free(initial_global_state->send_diff);
 140         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 141       }else{
 142         initial_global_state->recv_deterministic = 0;
 143         if(initial_global_state->recv_diff != NULL)
 144           xbt_free(initial_global_state->recv_diff);
 145         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 146       }
 147       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 148         XBT_INFO("*********************************************************");
 149         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 150         XBT_INFO("*********************************************************");
 151         XBT_INFO("%s", initial_global_state->send_diff);
 152         xbt_free(initial_global_state->send_diff);
 153         initial_global_state->send_diff = NULL;
 154         MC_print_statistics(mc_stats);
 155         xbt_abort();
 156       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 157         XBT_INFO("****************************************************");
 158         XBT_INFO("***** Non-deterministic communications pattern *****");
 159         XBT_INFO("****************************************************");
 160         XBT_INFO("%s", initial_global_state->send_diff);
 161         XBT_INFO("%s", initial_global_state->recv_diff);
 162         xbt_free(initial_global_state->send_diff);
 163         initial_global_state->send_diff = NULL;
 164         xbt_free(initial_global_state->recv_diff);
 165         initial_global_state->recv_diff = NULL;
 166         MC_print_statistics(mc_stats);
 167         xbt_abort();
 168       }
 169     }
 170   }
 171
 172   comm_pattern_free(comm);
 173
 174 }
 175
 176 /********** Non Static functions ***********/
 177
 178 void comm_pattern_free_voidp(void *p) {
 179   comm_pattern_free((mc_comm_pattern_t) * (void **) p);
 180 }
 181
 182 void list_comm_pattern_free_voidp(void *p) {
 183   list_comm_pattern_free((mc_list_comm_pattern_t) * (void **) p);
 184 }
 185
 186 void get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 187 {
 188   mc_process_t process = &mc_model_checker->process;
 189   mc_comm_pattern_t pattern = NULL;
 190   pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 191   pattern->data_size = -1;
 192   pattern->data = NULL;
 193
 194   const smx_process_t issuer = MC_smx_simcall_get_issuer(request);
 195   mc_list_comm_pattern_t initial_pattern =
 196     (mc_list_comm_pattern_t) xbt_dynar_get_as(initial_communications_pattern, issuer->pid, mc_list_comm_pattern_t);
 197   xbt_dynar_t incomplete_pattern =
 198     (xbt_dynar_t) xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
 199   pattern->index =
 200     initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
 201
 202   void *addr_pointed;
 203
 204   if (call_type == MC_CALL_TYPE_SEND) {
 205     /* Create comm pattern */
 206     pattern->type = SIMIX_COMM_SEND;
 207     pattern->comm = simcall_comm_isend__get__result(request);
 208     // FIXME, remote access to rdv->name
 209     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 210     pattern->src_proc = MC_smx_resolve_process(pattern->comm->comm.src_proc)->pid;
 211     pattern->src_host = MC_smx_process_get_host_name(issuer);
 212     pattern->tag = ((MPI_Request)simcall_comm_isend__get__data(request))->tag;
 213     if(pattern->comm->comm.src_buff != NULL){
 214       pattern->data_size = pattern->comm->comm.src_buff_size;
 215       pattern->data = xbt_malloc0(pattern->data_size);
 216       addr_pointed = *(void **) pattern->comm->comm.src_buff;
 217       if (addr_pointed > (void*) process->heap_address
 218           && addr_pointed < MC_process_get_heap(process)->breakval)
 219         memcpy(pattern->data, addr_pointed, pattern->data_size);
 220       else
 221         memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 222     }
 223     if(((MPI_Request)simcall_comm_isend__get__data(request))->detached){
 224       if (!initial_global_state->initial_communications_pattern_done) {
 225         /* Store comm pattern */
 226         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);
 227       } else {
 228         /* Evaluate comm determinism */
 229         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 230         ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->index_comm++;
 231       }
 232       return;
 233     }
 234   } else if (call_type == MC_CALL_TYPE_RECV) {
 235     pattern->type = SIMIX_COMM_RECEIVE;
 236     pattern->comm = simcall_comm_irecv__get__result(request);
 237     // TODO, remote access
 238     pattern->tag = ((MPI_Request)simcall_comm_irecv__get__data(request))->tag;
 239     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 240     pattern->dst_proc = MC_smx_resolve_process(pattern->comm->comm.dst_proc)->pid;
 241     // FIXME, remote process access
 242     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 243   } else {
 244     xbt_die("Unexpected call_type %i", (int) call_type);
 245   }
 246
 247   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t), &pattern);
 248
 249   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 250 }
 251
 252 void complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm, unsigned int issuer, int backtracking) {
 253   mc_comm_pattern_t current_comm_pattern;
 254   unsigned int cursor = 0;
 255   mc_comm_pattern_t comm_pattern;
 256   int completed = 0;
 257
 258   /* Complete comm pattern */
 259   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 260     if (current_comm_pattern-> comm == comm) {
 261       update_comm_pattern(current_comm_pattern, comm);
 262       completed = 1;
 263       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &comm_pattern);
 264       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 265       break;
 266     }
 267   }
 268   if(!completed)
 269     xbt_die("Corresponding communication not found!");
 270
 271   if (!initial_global_state->initial_communications_pattern_done) {
 272     /* Store comm pattern */
 273     xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->list, &comm_pattern);
 274   } else {
 275     /* Evaluate comm determinism */
 276     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 277     ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->index_comm++;
 278   }
 279 }
 280
 281
 282 /************************ Main algorithm ************************/
 283
 284 void MC_pre_modelcheck_comm_determinism(void)
 285 {
 286   MC_SET_MC_HEAP;
 287
 288   mc_state_t initial_state = NULL;
 289   smx_process_t process;
 290   int i;
 291
 292   if (_sg_mc_visited > 0)
 293     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 294
 295   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), list_comm_pattern_free_voidp);
 296   for (i=0; i < MC_smx_get_maxpid(); i++){
 297     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 298     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), comm_pattern_free_voidp);
 299     process_list_pattern->index_comm = 0;
 300     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 301   }
 302   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 303   for (i=0; i < MC_smx_get_maxpid(); i++){
 304     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 305     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 306   }
 307
 308   initial_state = MC_state_new();
 309   MC_SET_STD_HEAP;
 310
 311   XBT_DEBUG("********* Start communication determinism verification *********");
 312
 313   /* Wait for requests (schedules processes) */
 314   MC_wait_for_requests();
 315
 316   MC_SET_MC_HEAP;
 317
 318   /* Get an enabled process and insert it in the interleave set of the initial state */
 319   MC_EACH_SIMIX_PROCESS(process,
 320     if (MC_process_is_enabled(process)) {
 321       MC_state_interleave_process(initial_state, process);
 322     }
 323   );
 324
 325   xbt_fifo_unshift(mc_stack, initial_state);
 326
 327   MC_SET_STD_HEAP;
 328
 329 }
 330
 331 void MC_modelcheck_comm_determinism(void)
 332 {
 333
 334   char *req_str = NULL;
 335   int value;
 336   mc_visited_state_t visited_state = NULL;
 337   smx_simcall_t req = NULL;
 338   smx_process_t process = NULL;
 339   mc_state_t state = NULL, next_state = NULL;
 340
 341   while (xbt_fifo_size(mc_stack) > 0) {
 342
 343     /* Get current state */
 344     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 345
 346     XBT_DEBUG("**************************************************");
 347     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 348               xbt_fifo_size(mc_stack), state->num,
 349               MC_state_interleave_size(state));
 350
 351     /* Update statistics */
 352     mc_stats->visited_states++;
 353
 354     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 355         && (req = MC_state_get_request(state, &value))
 356         && (visited_state == NULL)) {
 357
 358       req_str = MC_request_to_string(req, value);
 359       XBT_DEBUG("Execute: %s", req_str);
 360       xbt_free(req_str);
 361
 362       if (dot_output != NULL) {
 363         MC_SET_MC_HEAP;
 364         req_str = MC_request_get_dot_output(req, value);
 365         MC_SET_STD_HEAP;
 366       }
 367
 368       MC_state_set_executed_request(state, req, value);
 369       mc_stats->executed_transitions++;
 370
 371       /* TODO : handle test and testany simcalls */
 372       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 373       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 374         call = mc_get_call_type(req);
 375       }
 376
 377       /* Answer the request */
 378       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 379
 380       MC_SET_MC_HEAP;
 381       if(!initial_global_state->initial_communications_pattern_done)
 382         handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 383       else
 384         handle_comm_pattern(call, req, value, NULL, 0);
 385       MC_SET_STD_HEAP;
 386
 387       /* Wait for requests (schedules processes) */
 388       MC_wait_for_requests();
 389
 390       /* Create the new expanded state */
 391       MC_SET_MC_HEAP;
 392
 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       MC_SET_STD_HEAP;
 420
 421     } else {
 422
 423       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 424         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 425       } else if (visited_state != NULL) {
 426         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);
 427       } else {
 428         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 429       }
 430
 431       MC_SET_MC_HEAP;
 432
 433       if (!initial_global_state->initial_communications_pattern_done)
 434         initial_global_state->initial_communications_pattern_done = 1;
 435
 436       /* Trash the current state, no longer needed */
 437       xbt_fifo_shift(mc_stack);
 438       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 439       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 440
 441       MC_SET_STD_HEAP;
 442
 443       visited_state = NULL;
 444
 445       /* Check for deadlocks */
 446       if (MC_deadlock_check()) {
 447         MC_show_deadlock(NULL);
 448         return;
 449       }
 450
 451       MC_SET_MC_HEAP;
 452
 453       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 454         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 455           /* We found a back-tracking point, let's loop */
 456           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 457           xbt_fifo_unshift(mc_stack, state);
 458           MC_SET_STD_HEAP;
 459
 460           MC_replay(mc_stack);
 461
 462           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 463
 464           break;
 465         } else {
 466           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 467           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 468         }
 469       }
 470
 471       MC_SET_STD_HEAP;
 472     }
 473   }
 474
 475   MC_print_statistics(mc_stats);
 476   MC_SET_STD_HEAP;
 477
 478   return;
 479 }