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 e_mc_comm_pattern_difference_t compare_comm_pattern(mc_comm_pattern_t comm1, mc_comm_pattern_t comm2) {
  26   if(comm1->type != comm2->type)
  27     return TYPE_DIFF;
  28   if (strcmp(comm1->rdv, comm2->rdv) != 0)
  29     return RDV_DIFF;
  30   if (comm1->src_proc != comm2->src_proc)
  31     return SRC_PROC_DIFF;
  32   if (comm1->dst_proc != comm2->dst_proc)
  33     return DST_PROC_DIFF;
  34   if (comm1->tag != comm2->tag)
  35     return TAG_DIFF;
  36   if (comm1->data_size != comm2->data_size)
  37     return DATA_SIZE_DIFF;
  38   if(comm1->data == NULL && comm2->data == NULL)
  39     return 0;
  40   if(comm1->data != NULL && comm2->data !=NULL) {
  41     if (!memcmp(comm1->data, comm2->data, comm1->data_size))
  42       return 0;
  43     return DATA_DIFF;
  44   }else{
  45     return DATA_DIFF;
  46   }
  47   return 0;
  48 }
  49
  50 static char* print_determinism_result(e_mc_comm_pattern_difference_t diff, int process, mc_comm_pattern_t comm, unsigned int cursor) {
  51   char *type, *res;
  52
  53   if(comm->type == SIMIX_COMM_SEND)
  54     type = bprintf("The send communications pattern of the process %d is different!", process - 1);
  55   else
  56     type = bprintf("The recv communications pattern of the process %d is different!", process - 1);
  57
  58   switch(diff) {
  59   case TYPE_DIFF:
  60     res = bprintf("%s Different type for communication #%d", type, cursor);
  61     break;
  62   case RDV_DIFF:
  63     res = bprintf("%s Different rdv for communication #%d", type, cursor);
  64     break;
  65   case TAG_DIFF:
  66     res = bprintf("%s Different tag for communication #%d", type, cursor);
  67     break;
  68   case SRC_PROC_DIFF:
  69       res = bprintf("%s Different source for communication #%d", type, cursor);
  70     break;
  71   case DST_PROC_DIFF:
  72       res = bprintf("%s Different destination for communication #%d", type, cursor);
  73     break;
  74   case DATA_SIZE_DIFF:
  75     res = bprintf("%s\n Different data size for communication #%d", type, cursor);
  76     break;
  77   case DATA_DIFF:
  78     res = bprintf("%s\n Different data for communication #%d", type, cursor);
  79     break;
  80   default:
  81     res = NULL;
  82     break;
  83   }
  84
  85   return res;
  86 }
  87
  88 // FIXME, remote comm
  89 static void update_comm_pattern(mc_comm_pattern_t comm_pattern, smx_synchro_t comm)
  90 {
  91   mc_process_t process = &mc_model_checker->process;
  92   void *addr_pointed;
  93   smx_process_t src_proc = MC_smx_resolve_process(comm->comm.src_proc);
  94   smx_process_t dst_proc = MC_smx_resolve_process(comm->comm.dst_proc);
  95   comm_pattern->src_proc = src_proc->pid;
  96   comm_pattern->dst_proc = dst_proc->pid;
  97   // TODO, resolve host name
  98   comm_pattern->src_host = MC_smx_process_get_host_name(src_proc);
  99   comm_pattern->dst_host = MC_smx_process_get_host_name(dst_proc);
 100   if (comm_pattern->data_size == -1 && comm->comm.src_buff != NULL) {
 101     comm_pattern->data_size = *(comm->comm.dst_buff_size);
 102     comm_pattern->data = xbt_malloc0(comm_pattern->data_size);
 103     addr_pointed = *(void **) comm->comm.src_buff;
 104     if (addr_pointed > (void*) process->heap_address
 105         && addr_pointed < MC_process_get_heap(process)->breakval)
 106       memcpy(comm_pattern->data, addr_pointed, comm_pattern->data_size);
 107     else
 108       memcpy(comm_pattern->data, comm->comm.src_buff, comm_pattern->data_size);
 109   }
 110 }
 111
 112 static void deterministic_comm_pattern(int process, mc_comm_pattern_t comm, int backtracking) {
 113
 114   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);
 115
 116   if(!backtracking){
 117     mc_comm_pattern_t initial_comm = xbt_dynar_get_as(list_comm_pattern->list, list_comm_pattern->index_comm, mc_comm_pattern_t);
 118     e_mc_comm_pattern_difference_t diff;
 119
 120     if((diff = compare_comm_pattern(initial_comm, comm)) != NONE_DIFF){
 121       if (comm->type == SIMIX_COMM_SEND){
 122         initial_global_state->send_deterministic = 0;
 123         if(initial_global_state->send_diff != NULL)
 124           xbt_free(initial_global_state->send_diff);
 125         initial_global_state->send_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 126       }else{
 127         initial_global_state->recv_deterministic = 0;
 128         if(initial_global_state->recv_diff != NULL)
 129           xbt_free(initial_global_state->recv_diff);
 130         initial_global_state->recv_diff = print_determinism_result(diff, process, comm, list_comm_pattern->index_comm + 1);
 131       }
 132       if(_sg_mc_send_determinism && !initial_global_state->send_deterministic){
 133         XBT_INFO("*********************************************************");
 134         XBT_INFO("***** Non-send-deterministic communications pattern *****");
 135         XBT_INFO("*********************************************************");
 136         XBT_INFO("%s", initial_global_state->send_diff);
 137         xbt_free(initial_global_state->send_diff);
 138         initial_global_state->send_diff = NULL;
 139         MC_print_statistics(mc_stats);
 140         xbt_abort();
 141       }else if(_sg_mc_comms_determinism && (!initial_global_state->send_deterministic && !initial_global_state->recv_deterministic)) {
 142         XBT_INFO("****************************************************");
 143         XBT_INFO("***** Non-deterministic communications pattern *****");
 144         XBT_INFO("****************************************************");
 145         XBT_INFO("%s", initial_global_state->send_diff);
 146         XBT_INFO("%s", initial_global_state->recv_diff);
 147         xbt_free(initial_global_state->send_diff);
 148         initial_global_state->send_diff = NULL;
 149         xbt_free(initial_global_state->recv_diff);
 150         initial_global_state->recv_diff = NULL;
 151         MC_print_statistics(mc_stats);
 152         xbt_abort();
 153       }
 154     }
 155   }
 156
 157   MC_comm_pattern_free(comm);
 158
 159 }
 160
 161 /********** Non Static functions ***********/
 162
 163 void MC_get_comm_pattern(xbt_dynar_t list, smx_simcall_t request, e_mc_call_type_t call_type, int backtracking)
 164 {
 165   mc_process_t process = &mc_model_checker->process;
 166
 167   mc_comm_pattern_t pattern = xbt_new0(s_mc_comm_pattern_t, 1);
 168   pattern->data_size = -1;
 169   pattern->data = NULL;
 170
 171   // Fill initial_pattern->index_comm:
 172   const smx_process_t issuer = MC_smx_simcall_get_issuer(request);
 173   mc_list_comm_pattern_t initial_pattern =
 174     (mc_list_comm_pattern_t) xbt_dynar_get_as(initial_communications_pattern, issuer->pid, mc_list_comm_pattern_t);
 175   xbt_dynar_t incomplete_pattern =
 176     (xbt_dynar_t) xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t);
 177   pattern->index =
 178     initial_pattern->index_comm + xbt_dynar_length(incomplete_pattern);
 179
 180   void *addr_pointed;
 181
 182   if (call_type == MC_CALL_TYPE_SEND) {
 183     /* Create comm pattern */
 184     pattern->type = SIMIX_COMM_SEND;
 185     pattern->comm = simcall_comm_isend__get__result(request);
 186     // FIXME, remote access to rdv->name
 187     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 188     pattern->src_proc = MC_smx_resolve_process(pattern->comm->comm.src_proc)->pid;
 189     pattern->src_host = MC_smx_process_get_host_name(issuer);
 190     pattern->tag = ((MPI_Request)simcall_comm_isend__get__data(request))->tag;
 191     if(pattern->comm->comm.src_buff != NULL){
 192       pattern->data_size = pattern->comm->comm.src_buff_size;
 193       pattern->data = xbt_malloc0(pattern->data_size);
 194       addr_pointed = *(void **) pattern->comm->comm.src_buff;
 195       if (addr_pointed > (void*) process->heap_address
 196           && addr_pointed < MC_process_get_heap(process)->breakval)
 197         memcpy(pattern->data, addr_pointed, pattern->data_size);
 198       else
 199         memcpy(pattern->data, pattern->comm->comm.src_buff, pattern->data_size);
 200     }
 201     if(((MPI_Request)simcall_comm_isend__get__data(request))->detached){
 202       if (!initial_global_state->initial_communications_pattern_done) {
 203         /* Store comm pattern */
 204         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);
 205       } else {
 206         /* Evaluate comm determinism */
 207         deterministic_comm_pattern(pattern->src_proc, pattern, backtracking);
 208         ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, pattern->src_proc, mc_list_comm_pattern_t))->index_comm++;
 209       }
 210       return;
 211     }
 212   } else if (call_type == MC_CALL_TYPE_RECV) {
 213     pattern->type = SIMIX_COMM_RECEIVE;
 214     pattern->comm = simcall_comm_irecv__get__result(request);
 215     // TODO, remote access
 216     pattern->tag = ((MPI_Request)simcall_comm_irecv__get__data(request))->tag;
 217     pattern->rdv = (pattern->comm->comm.rdv != NULL) ? strdup(pattern->comm->comm.rdv->name) : strdup(pattern->comm->comm.rdv_cpy->name);
 218     pattern->dst_proc = MC_smx_resolve_process(pattern->comm->comm.dst_proc)->pid;
 219     // FIXME, remote process access
 220     pattern->dst_host = MC_smx_process_get_host_name(issuer);
 221   } else {
 222     xbt_die("Unexpected call_type %i", (int) call_type);
 223   }
 224
 225   xbt_dynar_push((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer->pid, xbt_dynar_t), &pattern);
 226
 227   XBT_DEBUG("Insert incomplete comm pattern %p for process %lu", pattern, issuer->pid);
 228 }
 229
 230 void MC_complete_comm_pattern(xbt_dynar_t list, smx_synchro_t comm, unsigned int issuer, int backtracking) {
 231   mc_comm_pattern_t current_comm_pattern;
 232   unsigned int cursor = 0;
 233   mc_comm_pattern_t comm_pattern;
 234   int completed = 0;
 235
 236   /* Complete comm pattern */
 237   xbt_dynar_foreach((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, current_comm_pattern) {
 238     if (current_comm_pattern-> comm == comm) {
 239       update_comm_pattern(current_comm_pattern, comm);
 240       completed = 1;
 241       xbt_dynar_remove_at((xbt_dynar_t)xbt_dynar_get_as(incomplete_communications_pattern, issuer, xbt_dynar_t), cursor, &comm_pattern);
 242       XBT_DEBUG("Remove incomplete comm pattern for process %u at cursor %u", issuer, cursor);
 243       break;
 244     }
 245   }
 246   if(!completed)
 247     xbt_die("Corresponding communication not found!");
 248
 249   if (!initial_global_state->initial_communications_pattern_done) {
 250     /* Store comm pattern */
 251     xbt_dynar_push(((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->list, &comm_pattern);
 252   } else {
 253     /* Evaluate comm determinism */
 254     deterministic_comm_pattern(issuer, comm_pattern, backtracking);
 255     ((mc_list_comm_pattern_t)xbt_dynar_get_as(initial_communications_pattern, issuer, mc_list_comm_pattern_t))->index_comm++;
 256   }
 257 }
 258
 259
 260 /************************ Main algorithm ************************/
 261
 262 void MC_pre_modelcheck_comm_determinism(void)
 263 {
 264   MC_SET_MC_HEAP;
 265
 266   mc_state_t initial_state = NULL;
 267   smx_process_t process;
 268   int i;
 269
 270   if (_sg_mc_visited > 0)
 271     visited_states = xbt_dynar_new(sizeof(mc_visited_state_t), visited_state_free_voidp);
 272
 273   initial_communications_pattern = xbt_dynar_new(sizeof(mc_list_comm_pattern_t), MC_list_comm_pattern_free_voidp);
 274   for (i=0; i < MC_smx_get_maxpid(); i++){
 275     mc_list_comm_pattern_t process_list_pattern = xbt_new0(s_mc_list_comm_pattern_t, 1);
 276     process_list_pattern->list = xbt_dynar_new(sizeof(mc_comm_pattern_t), MC_comm_pattern_free_voidp);
 277     process_list_pattern->index_comm = 0;
 278     xbt_dynar_insert_at(initial_communications_pattern, i, &process_list_pattern);
 279   }
 280   incomplete_communications_pattern = xbt_dynar_new(sizeof(xbt_dynar_t), xbt_dynar_free_voidp);
 281   for (i=0; i < MC_smx_get_maxpid(); i++){
 282     xbt_dynar_t process_pattern = xbt_dynar_new(sizeof(mc_comm_pattern_t), NULL);
 283     xbt_dynar_insert_at(incomplete_communications_pattern, i, &process_pattern);
 284   }
 285
 286   initial_state = MC_state_new();
 287   MC_SET_STD_HEAP;
 288
 289   XBT_DEBUG("********* Start communication determinism verification *********");
 290
 291   /* Wait for requests (schedules processes) */
 292   MC_wait_for_requests();
 293
 294   MC_SET_MC_HEAP;
 295
 296   /* Get an enabled process and insert it in the interleave set of the initial state */
 297   MC_EACH_SIMIX_PROCESS(process,
 298     if (MC_process_is_enabled(process)) {
 299       MC_state_interleave_process(initial_state, process);
 300     }
 301   );
 302
 303   xbt_fifo_unshift(mc_stack, initial_state);
 304
 305   MC_SET_STD_HEAP;
 306
 307 }
 308
 309 void MC_modelcheck_comm_determinism(void)
 310 {
 311
 312   char *req_str = NULL;
 313   int value;
 314   mc_visited_state_t visited_state = NULL;
 315   smx_simcall_t req = NULL;
 316   smx_process_t process = NULL;
 317   mc_state_t state = NULL, next_state = NULL;
 318
 319   while (xbt_fifo_size(mc_stack) > 0) {
 320
 321     /* Get current state */
 322     state = (mc_state_t) xbt_fifo_get_item_content(xbt_fifo_get_first_item(mc_stack));
 323
 324     XBT_DEBUG("**************************************************");
 325     XBT_DEBUG("Exploration depth = %d (state = %d, interleaved processes = %d)",
 326               xbt_fifo_size(mc_stack), state->num,
 327               MC_state_interleave_size(state));
 328
 329     /* Update statistics */
 330     mc_stats->visited_states++;
 331
 332     if ((xbt_fifo_size(mc_stack) <= _sg_mc_max_depth)
 333         && (req = MC_state_get_request(state, &value))
 334         && (visited_state == NULL)) {
 335
 336       req_str = MC_request_to_string(req, value);
 337       XBT_DEBUG("Execute: %s", req_str);
 338       xbt_free(req_str);
 339
 340       if (dot_output != NULL) {
 341         MC_SET_MC_HEAP;
 342         req_str = MC_request_get_dot_output(req, value);
 343         MC_SET_STD_HEAP;
 344       }
 345
 346       MC_state_set_executed_request(state, req, value);
 347       mc_stats->executed_transitions++;
 348
 349       /* TODO : handle test and testany simcalls */
 350       e_mc_call_type_t call = MC_CALL_TYPE_NONE;
 351       if (_sg_mc_comms_determinism || _sg_mc_send_determinism) {
 352         call = MC_get_call_type(req);
 353       }
 354
 355       /* Answer the request */
 356       MC_simcall_handle(req, value);    /* After this call req is no longer useful */
 357
 358       MC_SET_MC_HEAP;
 359       if(!initial_global_state->initial_communications_pattern_done)
 360         MC_handle_comm_pattern(call, req, value, initial_communications_pattern, 0);
 361       else
 362         MC_handle_comm_pattern(call, req, value, NULL, 0);
 363       MC_SET_STD_HEAP;
 364
 365       /* Wait for requests (schedules processes) */
 366       MC_wait_for_requests();
 367
 368       /* Create the new expanded state */
 369       MC_SET_MC_HEAP;
 370
 371       next_state = MC_state_new();
 372
 373       if ((visited_state = is_visited_state(next_state)) == NULL) {
 374
 375         /* Get enabled processes and insert them in the interleave set of the next state */
 376         MC_EACH_SIMIX_PROCESS(process,
 377           if (MC_process_is_enabled(process)) {
 378             MC_state_interleave_process(next_state, process);
 379           }
 380         );
 381
 382         if (dot_output != NULL)
 383           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num,  next_state->num, req_str);
 384
 385       } else {
 386
 387         if (dot_output != NULL)
 388           fprintf(dot_output, "\"%d\" -> \"%d\" [%s];\n", state->num, visited_state->other_num == -1 ? visited_state->num : visited_state->other_num, req_str);
 389
 390       }
 391
 392       xbt_fifo_unshift(mc_stack, next_state);
 393
 394       if (dot_output != NULL)
 395         xbt_free(req_str);
 396
 397       MC_SET_STD_HEAP;
 398
 399     } else {
 400
 401       if (xbt_fifo_size(mc_stack) > _sg_mc_max_depth) {
 402         XBT_WARN("/!\\ Max depth reached ! /!\\ ");
 403       } else if (visited_state != NULL) {
 404         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);
 405       } else {
 406         XBT_DEBUG("There are no more processes to interleave. (depth %d)", xbt_fifo_size(mc_stack));
 407       }
 408
 409       MC_SET_MC_HEAP;
 410
 411       if (!initial_global_state->initial_communications_pattern_done)
 412         initial_global_state->initial_communications_pattern_done = 1;
 413
 414       /* Trash the current state, no longer needed */
 415       xbt_fifo_shift(mc_stack);
 416       MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 417       XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 418
 419       MC_SET_STD_HEAP;
 420
 421       visited_state = NULL;
 422
 423       /* Check for deadlocks */
 424       if (MC_deadlock_check()) {
 425         MC_show_deadlock(NULL);
 426         return;
 427       }
 428
 429       MC_SET_MC_HEAP;
 430
 431       while ((state = xbt_fifo_shift(mc_stack)) != NULL) {
 432         if (MC_state_interleave_size(state) && xbt_fifo_size(mc_stack) < _sg_mc_max_depth) {
 433           /* We found a back-tracking point, let's loop */
 434           XBT_DEBUG("Back-tracking to state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 435           xbt_fifo_unshift(mc_stack, state);
 436           MC_SET_STD_HEAP;
 437
 438           MC_replay(mc_stack);
 439
 440           XBT_DEBUG("Back-tracking to state %d at depth %d done", state->num, xbt_fifo_size(mc_stack));
 441
 442           break;
 443         } else {
 444           XBT_DEBUG("Delete state %d at depth %d", state->num, xbt_fifo_size(mc_stack) + 1);
 445           MC_state_delete(state, !state->in_visited_states ? 1 : 0);
 446         }
 447       }
 448
 449       MC_SET_STD_HEAP;
 450     }
 451   }
 452
 453   MC_print_statistics(mc_stats);
 454   MC_SET_STD_HEAP;
 455
 456   return;
 457 }