1 /* Copyright (c) 2008-2014. The SimGrid Team.
2 * All rights reserved. */
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. */
15 #include "internal_config.h"
16 #include "mc_private.h"
17 #include "xbt/module.h"
18 #include <xbt/mmalloc.h>
19 #include "../smpi/private.h"
22 #include "xbt/mmalloc/mmprivate.h"
24 #include "../simix/smx_private.h"
26 #define UNW_LOCAL_ONLY
27 #include <libunwind.h>
30 #include "mc_private.h"
35 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc,
36 "Logging specific to mc_checkpoint");
38 char *libsimgrid_path;
40 /************************************ Free functions **************************************/
41 /*****************************************************************************************/
43 static void MC_snapshot_stack_free(mc_snapshot_stack_t s)
46 xbt_dynar_free(&(s->local_variables));
47 xbt_dynar_free(&(s->stack_frames));
52 static void MC_snapshot_stack_free_voidp(void *s)
54 MC_snapshot_stack_free((mc_snapshot_stack_t) * (void **) s);
57 static void local_variable_free(local_variable_t v)
63 static void local_variable_free_voidp(void *v)
65 local_variable_free((local_variable_t) * (void **) v);
68 void MC_region_destroy(mc_mem_region_t reg)
73 //munmap(reg->data, reg->size);
75 if (reg->page_numbers) {
76 mc_free_page_snapshot_region(reg->page_numbers, mc_page_count(reg->size));
81 void MC_free_snapshot(mc_snapshot_t snapshot)
84 for (i = 0; i < NB_REGIONS; i++) {
85 MC_region_destroy(snapshot->regions[i]);
88 xbt_free(snapshot->stack_sizes);
89 xbt_dynar_free(&(snapshot->stacks));
90 xbt_dynar_free(&(snapshot->to_ignore));
91 xbt_dynar_free(&snapshot->ignored_data);
93 if (snapshot->privatization_regions) {
94 size_t n = xbt_dynar_length(snapshot->enabled_processes);
95 for (i = 0; i != n; ++i) {
96 MC_region_destroy(snapshot->privatization_regions[i]);
98 xbt_free(snapshot->privatization_regions);
104 /******************************* Snapshot regions ********************************/
105 /*********************************************************************************/
107 static mc_mem_region_t mc_region_new_dense(int type, void *start_addr, void* permanent_addr, size_t size, mc_mem_region_t ref_reg)
109 mc_mem_region_t new_reg = xbt_new(s_mc_mem_region_t, 1);
110 new_reg->start_addr = start_addr;
111 new_reg->permanent_addr = permanent_addr;
112 new_reg->data = NULL;
113 new_reg->size = size;
114 new_reg->page_numbers = NULL;
115 new_reg->data = xbt_malloc(size);
116 memcpy(new_reg->data, permanent_addr, size);
117 XBT_DEBUG("New region : type : %d, data : %p (real addr %p), size : %zu",
118 type, new_reg->data, permanent_addr, size);
123 /** @brief Take a snapshot of a given region
126 * @param start_addr Address of the region in the simulated process
127 * @param permanent_addr Permanent address of this data (for privatized variables, this is the virtual address of the privatized mapping)
128 * @param size Size of the data*
129 * @param ref_reg Reference corresponding region
131 static mc_mem_region_t MC_region_new(int type, void *start_addr, void* permanent_addr, size_t size, mc_mem_region_t ref_reg)
133 if (_sg_mc_sparse_checkpoint) {
134 return mc_region_new_sparse(type, start_addr, permanent_addr, size, ref_reg);
136 return mc_region_new_dense(type, start_addr, permanent_addr, size, ref_reg);
140 /** @brief Restore a region from a snapshot
142 * If we are using per page snapshots, it is possible to use the reference
143 * region in order to do an incremental restoration of the region: the
144 * softclean pages which are shared between the two snapshots do not need
147 * @param reg Target region
148 * @param reg_reg Current region (if not NULL), used for lazy per page restoration
150 static void MC_region_restore(mc_mem_region_t reg, mc_mem_region_t ref_reg)
152 /*FIXME: check if start_addr is still mapped, if it is not, then map it
153 before copying the data */
154 if (!reg->page_numbers) {
155 memcpy(reg->permanent_addr, reg->data, reg->size);
157 mc_region_restore_sparse(reg, ref_reg);
162 static void MC_snapshot_add_region(mc_snapshot_t snapshot, int type,
163 void *start_addr, void* permanent_addr, size_t size)
166 mc_mem_region_t ref_reg =
167 mc_model_checker->parent_snapshot ? mc_model_checker->parent_snapshot->regions[type] : NULL;
168 mc_mem_region_t new_reg = MC_region_new(type, start_addr, permanent_addr, size, ref_reg);
169 snapshot->regions[type] = new_reg;
173 static void MC_get_memory_regions(mc_snapshot_t snapshot)
176 void *start_heap = std_heap->base;
177 void *end_heap = std_heap->breakval;
178 MC_snapshot_add_region(snapshot, 0, start_heap, start_heap,
179 (char *) end_heap - (char *) start_heap);
180 snapshot->heap_bytes_used = mmalloc_get_bytes_used(std_heap);
181 snapshot->privatization_regions = NULL;
183 MC_snapshot_add_region(snapshot, 1,
184 mc_libsimgrid_info->start_rw, mc_libsimgrid_info->start_rw,
185 mc_libsimgrid_info->end_rw - mc_libsimgrid_info->start_rw);
189 if (smpi_privatize_global_variables && smpi_process_count()) {
190 // Snapshot the global variable of the application separately for each
191 // simulated process:
192 snapshot->privatization_regions =
193 xbt_new(mc_mem_region_t, smpi_process_count());
194 for (i = 0; i < smpi_process_count(); i++) {
195 mc_mem_region_t ref_reg =
196 mc_model_checker->parent_snapshot ? mc_model_checker->parent_snapshot->privatization_regions[i] : NULL;
197 snapshot->privatization_regions[i] =
198 MC_region_new(-1, mc_binary_info->start_rw, smpi_privatisation_regions[i].address, size_data_exe, ref_reg);
200 snapshot->privatization_index = smpi_loaded_page;
201 snapshot->regions[2] = NULL;
205 MC_snapshot_add_region(snapshot, 2,
206 mc_binary_info->start_rw, mc_binary_info->start_rw,
207 mc_binary_info->end_rw - mc_binary_info->start_rw);
208 snapshot->privatization_regions = NULL;
209 snapshot->privatization_index = -1;
213 /** @brief Finds the range of the different memory segments and binary paths */
214 void MC_init_memory_map_info()
219 memory_map_t maps = MC_get_memory_map();
221 maestro_stack_start = NULL;
222 maestro_stack_end = NULL;
223 libsimgrid_path = NULL;
225 while (i < maps->mapsize) {
226 reg = maps->regions[i];
227 if (maps->regions[i].pathname == NULL) {
229 } else if ((reg.prot & PROT_WRITE)
230 && !memcmp(maps->regions[i].pathname, "[stack]", 7)) {
231 maestro_stack_start = reg.start_addr;
232 maestro_stack_end = reg.end_addr;
233 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)
234 && !memcmp(basename(maps->regions[i].pathname), "libsimgrid",
236 if (libsimgrid_path == NULL)
237 libsimgrid_path = strdup(maps->regions[i].pathname);
242 xbt_assert(maestro_stack_start, "maestro_stack_start");
243 xbt_assert(maestro_stack_end, "maestro_stack_end");
244 xbt_assert(libsimgrid_path, "libsimgrid_path&");
246 MC_free_memory_map(maps);
250 /** \brief Fills the position of the segments (executable, read-only, read/write).
252 * TODO, use dl_iterate_phdr to be more robust
254 void MC_find_object_address(memory_map_t maps, mc_object_info_t result)
259 const char *name = basename(result->file_name);
260 while (i < maps->mapsize) {
261 reg = maps->regions[i];
262 if (maps->regions[i].pathname == NULL
263 || strcmp(basename(maps->regions[i].pathname), name)) {
265 } else if ((reg.prot & PROT_WRITE)) {
266 xbt_assert(!result->start_rw,
267 "Multiple read-write segments for %s, not supported",
268 maps->regions[i].pathname);
269 result->start_rw = reg.start_addr;
270 result->end_rw = reg.end_addr;
271 // .bss is usually after the .data:
272 s_map_region_t *next = &(maps->regions[i + 1]);
273 if (next->pathname == NULL && (next->prot & PROT_WRITE)
274 && next->start_addr == reg.end_addr) {
275 result->end_rw = maps->regions[i + 1].end_addr;
277 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) {
278 xbt_assert(!result->start_exec,
279 "Multiple executable segments for %s, not supported",
280 maps->regions[i].pathname);
281 result->start_exec = reg.start_addr;
282 result->end_exec = reg.end_addr;
283 } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) {
284 xbt_assert(!result->start_ro,
285 "Multiple read only segments for %s, not supported",
286 maps->regions[i].pathname);
287 result->start_ro = reg.start_addr;
288 result->end_ro = reg.end_addr;
293 xbt_assert(result->file_name);
294 xbt_assert(result->start_rw);
295 xbt_assert(result->start_exec);
298 /************************************* Take Snapshot ************************************/
299 /****************************************************************************************/
301 /** \brief Checks whether the variable is in scope for a given IP.
303 * A variable may be defined only from a given value of IP.
305 * \param var Variable description
306 * \param frame Scope description
307 * \param ip Instruction pointer
308 * \return true if the variable is valid
310 static bool mc_valid_variable(dw_variable_t var, dw_frame_t scope,
313 // The variable is not yet valid:
314 if ((const void *) ((const char *) scope->low_pc + var->start_scope) > ip)
320 static void mc_fill_local_variables_values(mc_stack_frame_t stack_frame,
321 dw_frame_t scope, int process_index, xbt_dynar_t result)
323 void *ip = (void *) stack_frame->ip;
324 if (ip < scope->low_pc || ip >= scope->high_pc)
328 dw_variable_t current_variable;
329 xbt_dynar_foreach(scope->variables, cursor, current_variable) {
331 if (!mc_valid_variable(current_variable, scope, (void *) stack_frame->ip))
335 if ((long) stack_frame->ip > (long) mc_libsimgrid_info->start_exec)
340 local_variable_t new_var = xbt_new0(s_local_variable_t, 1);
341 new_var->subprogram = stack_frame->frame;
342 new_var->ip = stack_frame->ip;
343 new_var->name = xbt_strdup(current_variable->name);
344 new_var->type = current_variable->type;
345 new_var->region = region_type;
347 if (current_variable->address != NULL) {
348 new_var->address = current_variable->address;
349 } else if (current_variable->locations.size != 0) {
350 s_mc_location_t location;
351 mc_dwarf_resolve_locations(&location, ¤t_variable->locations,
352 current_variable->object_info,
353 &(stack_frame->unw_cursor),
354 (void *) stack_frame->frame_base,
355 NULL, process_index);
357 switch(mc_get_location_type(&location)) {
358 case MC_LOCATION_TYPE_ADDRESS:
359 new_var->address = location.memory_location;
361 case MC_LOCATION_TYPE_REGISTER:
363 xbt_die("Cannot handle non-address variable");
367 xbt_die("No address");
370 xbt_dynar_push(result, &new_var);
373 // Recursive processing of nested scopes:
374 dw_frame_t nested_scope = NULL;
375 xbt_dynar_foreach(scope->scopes, cursor, nested_scope) {
376 mc_fill_local_variables_values(stack_frame, nested_scope, process_index, result);
380 static xbt_dynar_t MC_get_local_variables_values(xbt_dynar_t stack_frames, int process_index)
383 unsigned cursor1 = 0;
384 mc_stack_frame_t stack_frame;
385 xbt_dynar_t variables =
386 xbt_dynar_new(sizeof(local_variable_t), local_variable_free_voidp);
388 xbt_dynar_foreach(stack_frames, cursor1, stack_frame) {
389 mc_fill_local_variables_values(stack_frame, stack_frame->frame, process_index, variables);
395 static void MC_stack_frame_free_voipd(void *s)
397 mc_stack_frame_t stack_frame = *(mc_stack_frame_t *) s;
399 xbt_free(stack_frame->frame_name);
400 xbt_free(stack_frame);
404 static xbt_dynar_t MC_unwind_stack_frames(void *stack_context)
407 xbt_dynar_new(sizeof(mc_stack_frame_t), MC_stack_frame_free_voipd);
411 // TODO, check condition check (unw_init_local==0 means end of frame)
412 if (unw_init_local(&c, (unw_context_t *) stack_context) != 0) {
414 xbt_die("Could not initialize stack unwinding");
419 mc_stack_frame_t stack_frame = xbt_new(s_mc_stack_frame_t, 1);
420 xbt_dynar_push(result, &stack_frame);
422 stack_frame->unw_cursor = c;
426 unw_get_reg(&c, UNW_REG_IP, &ip);
427 unw_get_reg(&c, UNW_REG_SP, &sp);
429 stack_frame->ip = ip;
430 stack_frame->sp = sp;
432 // TODO, use real addresses in frame_t instead of fixing it here
434 dw_frame_t frame = MC_find_function_by_ip((void *) ip);
435 stack_frame->frame = frame;
438 stack_frame->frame_name = xbt_strdup(frame->name);
439 stack_frame->frame_base =
440 (unw_word_t) mc_find_frame_base(frame, frame->object_info, &c);
442 stack_frame->frame_base = 0;
443 stack_frame->frame_name = NULL;
446 /* Stop before context switch with maestro */
447 if (frame != NULL && frame->name != NULL
448 && !strcmp(frame->name, "smx_ctx_sysv_wrapper"))
451 int ret = ret = unw_step(&c);
453 xbt_die("Unexpected end of stack.");
454 } else if (ret < 0) {
455 xbt_die("Error while unwinding stack.");
459 if (xbt_dynar_length(result) == 0) {
460 XBT_INFO("unw_init_local failed");
467 static xbt_dynar_t MC_take_snapshot_stacks(mc_snapshot_t * snapshot)
471 xbt_dynar_new(sizeof(s_mc_snapshot_stack_t),
472 MC_snapshot_stack_free_voidp);
474 unsigned int cursor = 0;
475 stack_region_t current_stack;
477 xbt_dynar_foreach(stacks_areas, cursor, current_stack) {
478 mc_snapshot_stack_t st = xbt_new(s_mc_snapshot_stack_t, 1);
479 st->stack_frames = MC_unwind_stack_frames(current_stack->context);
480 st->local_variables = MC_get_local_variables_values(st->stack_frames, current_stack->process_index);
481 st->process_index = current_stack->process_index;
483 unw_word_t sp = xbt_dynar_get_as(st->stack_frames, 0, mc_stack_frame_t)->sp;
485 xbt_dynar_push(res, &st);
486 (*snapshot)->stack_sizes =
487 xbt_realloc((*snapshot)->stack_sizes, (cursor + 1) * sizeof(size_t));
488 (*snapshot)->stack_sizes[cursor] =
489 (char*) current_stack->address + current_stack->size - (char*) sp;
496 static xbt_dynar_t MC_take_snapshot_ignore()
499 if (mc_heap_comparison_ignore == NULL)
503 xbt_dynar_new(sizeof(mc_heap_ignore_region_t),
504 heap_ignore_region_free_voidp);
506 unsigned int cursor = 0;
507 mc_heap_ignore_region_t current_region;
509 xbt_dynar_foreach(mc_heap_comparison_ignore, cursor, current_region) {
510 mc_heap_ignore_region_t new_region = NULL;
511 new_region = xbt_new0(s_mc_heap_ignore_region_t, 1);
512 new_region->address = current_region->address;
513 new_region->size = current_region->size;
514 new_region->block = current_region->block;
515 new_region->fragment = current_region->fragment;
516 xbt_dynar_push(cpy, &new_region);
523 static void mc_free_snapshot_ignored_data_pvoid(void* data) {
524 mc_snapshot_ignored_data_t ignored_data = (mc_snapshot_ignored_data_t) data;
525 free(ignored_data->data);
528 static void MC_snapshot_handle_ignore(mc_snapshot_t snapshot)
530 snapshot->ignored_data = xbt_dynar_new(sizeof(s_mc_snapshot_ignored_data_t), mc_free_snapshot_ignored_data_pvoid);
533 unsigned int cursor = 0;
534 mc_checkpoint_ignore_region_t region;
535 xbt_dynar_foreach (mc_checkpoint_ignore, cursor, region) {
536 s_mc_snapshot_ignored_data_t ignored_data;
537 ignored_data.start = region->addr;
538 ignored_data.size = region->size;
539 ignored_data.data = malloc(region->size);
540 memcpy(ignored_data.data, region->addr, region->size);
541 xbt_dynar_push(snapshot->ignored_data, &ignored_data);
545 xbt_dynar_foreach (mc_checkpoint_ignore, cursor, region) {
546 memset(region->addr, 0, region->size);
551 static void MC_snapshot_ignore_restore(mc_snapshot_t snapshot)
553 unsigned int cursor = 0;
554 s_mc_snapshot_ignored_data_t ignored_data;
555 xbt_dynar_foreach (snapshot->ignored_data, cursor, ignored_data) {
556 memcpy(ignored_data.start, ignored_data.data, ignored_data.size);
560 /** @brief Can we remove this snapshot?
562 * Some snapshots cannot be removed (yet) because we need them
567 int mc_important_snapshot(mc_snapshot_t snapshot)
569 // We need this snapshot in order to know which
570 // pages needs to be stored in the next snapshot.
571 // This field is only non-NULL when using soft-dirty
573 if (snapshot == mc_model_checker->parent_snapshot)
579 static void MC_get_current_fd(mc_snapshot_t snapshot){
581 struct dirent *fd_number;
589 snapshot->total_fd = 0;
591 fd_dir = opendir ("/proc/self/fd");
592 if (fd_dir == NULL) {
593 printf ("Cannot open directory '/proc/self/fd'\n");
597 while ((fd_number = readdir(fd_dir)) != NULL) {
599 fd_value = atoi(fd_number->d_name);
604 command = bprintf("readlink /proc/self/fd/%s", fd_number->d_name);
605 fp = popen(command, "r");
607 perror("popen failed");
610 xbt_getline(&line, &n, fp);
611 if(line && strncmp(line, "pipe:", 5) != 0 && strncmp(line, "socket:", 7) != 0){
612 fd_infos_t fd = xbt_new0(s_fd_infos_t, 1);
613 fd->filename = strdup(line);
614 fd->filename[strlen(line)-1] = '\0';
615 fd->number = fd_value;
616 fd->flags = fcntl(fd_value, F_GETFL) | fcntl(fd_value, F_GETFD) ;
617 fd->current_position = lseek(fd_value, 0, SEEK_CUR);
618 snapshot->current_fd = xbt_realloc(snapshot->current_fd, (snapshot->total_fd + 1) * sizeof(fd_infos_t));
619 snapshot->current_fd[snapshot->total_fd] = fd;
620 snapshot->total_fd++;
632 mc_snapshot_t MC_take_snapshot(int num_state)
635 mc_snapshot_t snapshot = xbt_new0(s_mc_snapshot_t, 1);
636 snapshot->enabled_processes = xbt_dynar_new(sizeof(int), NULL);
637 smx_process_t process;
638 xbt_swag_foreach(process, simix_global->process_list) {
639 xbt_dynar_push_as(snapshot->enabled_processes, int, (int)process->pid);
642 MC_snapshot_handle_ignore(snapshot);
644 MC_get_current_fd(snapshot);
646 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
647 MC_get_memory_regions(snapshot);
648 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
649 mc_softdirty_reset();
652 snapshot->to_ignore = MC_take_snapshot_ignore();
654 if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) {
656 MC_take_snapshot_stacks(&snapshot);
657 if (_sg_mc_hash && snapshot->stacks != NULL) {
658 snapshot->hash = mc_hash_processes_state(num_state, snapshot->stacks);
666 MC_snapshot_ignore_restore(snapshot);
667 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
668 mc_model_checker->parent_snapshot = snapshot;
673 void MC_restore_snapshot(mc_snapshot_t snapshot)
675 mc_snapshot_t parent_snapshot = mc_model_checker->parent_snapshot;
679 for (i = 0; i < NB_REGIONS; i++) {
680 // For privatized, variables we decided it was not necessary to take the snapshot:
681 if (snapshot->regions[i])
682 MC_region_restore(snapshot->regions[i],
683 parent_snapshot ? parent_snapshot->regions[i] : NULL);
687 if (snapshot->privatization_regions) {
688 // Restore the global variables of the application separately for each
689 // simulated process:
690 for (i = 0; i < smpi_process_count(); i++) {
691 if (snapshot->privatization_regions[i]) {
692 MC_region_restore(snapshot->privatization_regions[i],
693 parent_snapshot ? parent_snapshot->privatization_regions[i] : NULL);
697 if(snapshot->privatization_index >= 0) {
698 // We just rewrote the global variables.
699 // The privatisation segment SMPI thinks
700 // is mapped might be inconsistent with the segment which
701 // is really mapped in memory (kernel state).
702 // We ask politely SMPI to map the segment anyway,
703 // even if it thinks it is the current one:
704 smpi_really_switch_data_segment(snapshot->privatization_index);
708 for(i=0; i < snapshot->total_fd; i++){
710 new_fd = open(snapshot->current_fd[i]->filename, snapshot->current_fd[i]->flags);
711 if(new_fd != -1 && new_fd != snapshot->current_fd[i]->number){
712 dup2(new_fd, snapshot->current_fd[i]->number);
713 //fprintf(stderr, "%p\n", fdopen(snapshot->current_fd[i]->number, "rw"));
716 lseek(snapshot->current_fd[i]->number, snapshot->current_fd[i]->current_position, SEEK_SET);
719 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
720 mc_softdirty_reset();
723 MC_snapshot_ignore_restore(snapshot);
724 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
725 mc_model_checker->parent_snapshot = snapshot;
730 mc_snapshot_t simcall_HANDLER_mc_snapshot(smx_simcall_t simcall)
732 return MC_take_snapshot(1);
735 void *MC_snapshot(void)
737 return simcall_mc_snapshot();