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