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 #include <libunwind.h>
29 #include "mc_private.h"
34 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(mc_checkpoint, mc,
35 "Logging specific to mc_checkpoint");
37 char *libsimgrid_path;
39 /************************************ Free functions **************************************/
40 /*****************************************************************************************/
42 static void MC_snapshot_stack_free(mc_snapshot_stack_t s)
45 xbt_dynar_free(&(s->local_variables));
46 xbt_dynar_free(&(s->stack_frames));
51 static void MC_snapshot_stack_free_voidp(void *s)
53 MC_snapshot_stack_free((mc_snapshot_stack_t) * (void **) s);
56 static void local_variable_free(local_variable_t v)
62 static void local_variable_free_voidp(void *v)
64 local_variable_free((local_variable_t) * (void **) v);
67 void MC_region_destroy(mc_mem_region_t reg)
72 //munmap(reg->data, reg->size);
74 if (reg->page_numbers) {
75 mc_free_page_snapshot_region(reg->page_numbers, mc_page_count(reg->size));
80 void MC_free_snapshot(mc_snapshot_t snapshot)
83 for (i = 0; i < NB_REGIONS; i++) {
84 MC_region_destroy(snapshot->regions[i]);
87 xbt_free(snapshot->stack_sizes);
88 xbt_dynar_free(&(snapshot->stacks));
89 xbt_dynar_free(&(snapshot->to_ignore));
90 xbt_dynar_free(&snapshot->ignored_data);
92 if (snapshot->privatization_regions) {
93 size_t n = xbt_dynar_length(snapshot->enabled_processes);
94 for (i = 0; i != n; ++i) {
95 MC_region_destroy(snapshot->privatization_regions[i]);
97 xbt_free(snapshot->privatization_regions);
103 /******************************* Snapshot regions ********************************/
104 /*********************************************************************************/
106 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)
108 mc_mem_region_t new_reg = xbt_new(s_mc_mem_region_t, 1);
109 new_reg->start_addr = start_addr;
110 new_reg->permanent_addr = permanent_addr;
111 new_reg->data = NULL;
112 new_reg->size = size;
113 new_reg->page_numbers = NULL;
114 new_reg->data = xbt_malloc(size);
115 memcpy(new_reg->data, permanent_addr, size);
116 XBT_DEBUG("New region : type : %d, data : %p (real addr %p), size : %zu",
117 type, new_reg->data, permanent_addr, size);
122 /** @brief Take a snapshot of a given region
125 * @param start_addr Address of the region in the simulated process
126 * @param permanent_addr Permanent address of this data (for privatized variables, this is the virtual address of the privatized mapping)
127 * @param size Size of the data*
128 * @param ref_reg Reference corresponding region
130 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)
132 if (_sg_mc_sparse_checkpoint) {
133 return mc_region_new_sparse(type, start_addr, permanent_addr, size, ref_reg);
135 return mc_region_new_dense(type, start_addr, permanent_addr, size, ref_reg);
139 /** @brief Restore a region from a snapshot
141 * If we are using per page snapshots, it is possible to use the reference
142 * region in order to do an incremental restoration of the region: the
143 * softclean pages which are shared between the two snapshots do not need
146 * @param reg Target region
147 * @param reg_reg Current region (if not NULL), used for lazy per page restoration
149 static void MC_region_restore(mc_mem_region_t reg, mc_mem_region_t ref_reg)
151 /*FIXME: check if start_addr is still mapped, if it is not, then map it
152 before copying the data */
153 if (!reg->page_numbers) {
154 memcpy(reg->permanent_addr, reg->data, reg->size);
156 mc_region_restore_sparse(reg, ref_reg);
161 static void MC_snapshot_add_region(mc_snapshot_t snapshot, int type,
162 void *start_addr, void* permanent_addr, size_t size)
165 mc_mem_region_t ref_reg =
166 mc_model_checker->parent_snapshot ? mc_model_checker->parent_snapshot->regions[type] : NULL;
167 mc_mem_region_t new_reg = MC_region_new(type, start_addr, permanent_addr, size, ref_reg);
168 snapshot->regions[type] = new_reg;
172 static void MC_get_memory_regions(mc_snapshot_t snapshot)
175 void *start_heap = std_heap->base;
176 void *end_heap = std_heap->breakval;
177 MC_snapshot_add_region(snapshot, 0, start_heap, start_heap,
178 (char *) end_heap - (char *) start_heap);
179 snapshot->heap_bytes_used = mmalloc_get_bytes_used(std_heap);
180 snapshot->privatization_regions = NULL;
182 MC_snapshot_add_region(snapshot, 1,
183 mc_libsimgrid_info->start_rw, mc_libsimgrid_info->start_rw,
184 mc_libsimgrid_info->end_rw - mc_libsimgrid_info->start_rw);
188 if (smpi_privatize_global_variables && smpi_process_count()) {
189 // Snapshot the global variable of the application separately for each
190 // simulated process:
191 snapshot->privatization_regions =
192 xbt_new(mc_mem_region_t, smpi_process_count());
193 for (i = 0; i < smpi_process_count(); i++) {
194 mc_mem_region_t ref_reg =
195 mc_model_checker->parent_snapshot ? mc_model_checker->parent_snapshot->privatization_regions[i] : NULL;
196 snapshot->privatization_regions[i] =
197 MC_region_new(-1, mc_binary_info->start_rw, smpi_privatisation_regions[i].address, size_data_exe, ref_reg);
199 snapshot->privatization_index = smpi_loaded_page;
200 snapshot->regions[2] = NULL;
204 MC_snapshot_add_region(snapshot, 2,
205 mc_binary_info->start_rw, mc_binary_info->start_rw,
206 mc_binary_info->end_rw - mc_binary_info->start_rw);
207 snapshot->privatization_regions = NULL;
208 snapshot->privatization_index = -1;
212 /** @brief Finds the range of the different memory segments and binary paths */
213 void MC_init_memory_map_info()
218 memory_map_t maps = MC_get_memory_map();
220 maestro_stack_start = NULL;
221 maestro_stack_end = NULL;
222 libsimgrid_path = NULL;
224 while (i < maps->mapsize) {
225 reg = maps->regions[i];
226 if (maps->regions[i].pathname == NULL) {
228 } else if ((reg.prot & PROT_WRITE)
229 && !memcmp(maps->regions[i].pathname, "[stack]", 7)) {
230 maestro_stack_start = reg.start_addr;
231 maestro_stack_end = reg.end_addr;
232 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)
233 && !memcmp(basename(maps->regions[i].pathname), "libsimgrid",
235 if (libsimgrid_path == NULL)
236 libsimgrid_path = strdup(maps->regions[i].pathname);
241 xbt_assert(maestro_stack_start, "maestro_stack_start");
242 xbt_assert(maestro_stack_end, "maestro_stack_end");
243 xbt_assert(libsimgrid_path, "libsimgrid_path&");
245 MC_free_memory_map(maps);
249 /** \brief Fills the position of the segments (executable, read-only, read/write).
251 * TODO, use dl_iterate_phdr to be more robust
253 void MC_find_object_address(memory_map_t maps, mc_object_info_t result)
258 const char *name = basename(result->file_name);
259 while (i < maps->mapsize) {
260 reg = maps->regions[i];
261 if (maps->regions[i].pathname == NULL
262 || strcmp(basename(maps->regions[i].pathname), name)) {
264 } else if ((reg.prot & PROT_WRITE)) {
265 xbt_assert(!result->start_rw,
266 "Multiple read-write segments for %s, not supported",
267 maps->regions[i].pathname);
268 result->start_rw = reg.start_addr;
269 result->end_rw = reg.end_addr;
270 // .bss is usually after the .data:
271 s_map_region_t *next = &(maps->regions[i + 1]);
272 if (next->pathname == NULL && (next->prot & PROT_WRITE)
273 && next->start_addr == reg.end_addr) {
274 result->end_rw = maps->regions[i + 1].end_addr;
276 } else if ((reg.prot & PROT_READ) && (reg.prot & PROT_EXEC)) {
277 xbt_assert(!result->start_exec,
278 "Multiple executable segments for %s, not supported",
279 maps->regions[i].pathname);
280 result->start_exec = reg.start_addr;
281 result->end_exec = reg.end_addr;
282 } else if ((reg.prot & PROT_READ) && !(reg.prot & PROT_EXEC)) {
283 xbt_assert(!result->start_ro,
284 "Multiple read only segments for %s, not supported",
285 maps->regions[i].pathname);
286 result->start_ro = reg.start_addr;
287 result->end_ro = reg.end_addr;
292 xbt_assert(result->file_name);
293 xbt_assert(result->start_rw);
294 xbt_assert(result->start_exec);
297 /************************************* Take Snapshot ************************************/
298 /****************************************************************************************/
300 /** \brief Checks whether the variable is in scope for a given IP.
302 * A variable may be defined only from a given value of IP.
304 * \param var Variable description
305 * \param frame Scope description
306 * \param ip Instruction pointer
307 * \return true if the variable is valid
309 static bool mc_valid_variable(dw_variable_t var, dw_frame_t scope,
312 // The variable is not yet valid:
313 if ((const void *) ((const char *) scope->low_pc + var->start_scope) > ip)
319 static void mc_fill_local_variables_values(mc_stack_frame_t stack_frame,
320 dw_frame_t scope, int process_index, xbt_dynar_t result)
322 void *ip = (void *) stack_frame->ip;
323 if (ip < scope->low_pc || ip >= scope->high_pc)
327 dw_variable_t current_variable;
328 xbt_dynar_foreach(scope->variables, cursor, current_variable) {
330 if (!mc_valid_variable(current_variable, scope, (void *) stack_frame->ip))
334 if ((long) stack_frame->ip > (long) mc_libsimgrid_info->start_exec)
339 local_variable_t new_var = xbt_new0(s_local_variable_t, 1);
340 new_var->subprogram = stack_frame->frame;
341 new_var->ip = stack_frame->ip;
342 new_var->name = xbt_strdup(current_variable->name);
343 new_var->type = current_variable->type;
344 new_var->region = region_type;
346 if (current_variable->address != NULL) {
347 new_var->address = current_variable->address;
348 } else if (current_variable->locations.size != 0) {
349 s_mc_location_t location;
350 mc_dwarf_resolve_locations(&location, ¤t_variable->locations,
351 current_variable->object_info,
352 &(stack_frame->unw_cursor),
353 (void *) stack_frame->frame_base,
354 NULL, process_index);
356 switch(mc_get_location_type(&location)) {
357 case MC_LOCATION_TYPE_ADDRESS:
358 new_var->address = location.memory_location;
360 case MC_LOCATION_TYPE_REGISTER:
362 xbt_die("Cannot handle non-address variable");
366 xbt_die("No address");
369 xbt_dynar_push(result, &new_var);
372 // Recursive processing of nested scopes:
373 dw_frame_t nested_scope = NULL;
374 xbt_dynar_foreach(scope->scopes, cursor, nested_scope) {
375 mc_fill_local_variables_values(stack_frame, nested_scope, process_index, result);
379 static xbt_dynar_t MC_get_local_variables_values(xbt_dynar_t stack_frames, int process_index)
382 unsigned cursor1 = 0;
383 mc_stack_frame_t stack_frame;
384 xbt_dynar_t variables =
385 xbt_dynar_new(sizeof(local_variable_t), local_variable_free_voidp);
387 xbt_dynar_foreach(stack_frames, cursor1, stack_frame) {
388 mc_fill_local_variables_values(stack_frame, stack_frame->frame, process_index, variables);
394 static void MC_stack_frame_free_voipd(void *s)
396 mc_stack_frame_t stack_frame = *(mc_stack_frame_t *) s;
398 xbt_free(stack_frame->frame_name);
399 xbt_free(stack_frame);
403 static xbt_dynar_t MC_unwind_stack_frames(void *stack_context)
406 xbt_dynar_new(sizeof(mc_stack_frame_t), MC_stack_frame_free_voipd);
410 // TODO, check condition check (unw_init_local==0 means end of frame)
411 if (unw_init_local(&c, (unw_context_t *) stack_context) != 0) {
413 xbt_die("Could not initialize stack unwinding");
418 mc_stack_frame_t stack_frame = xbt_new(s_mc_stack_frame_t, 1);
419 xbt_dynar_push(result, &stack_frame);
421 stack_frame->unw_cursor = c;
425 unw_get_reg(&c, UNW_REG_IP, &ip);
426 unw_get_reg(&c, UNW_REG_SP, &sp);
428 stack_frame->ip = ip;
429 stack_frame->sp = sp;
431 // TODO, use real addresses in frame_t instead of fixing it here
433 dw_frame_t frame = MC_find_function_by_ip((void *) ip);
434 stack_frame->frame = frame;
437 stack_frame->frame_name = xbt_strdup(frame->name);
438 stack_frame->frame_base =
439 (unw_word_t) mc_find_frame_base(frame, frame->object_info, &c);
441 stack_frame->frame_base = 0;
442 stack_frame->frame_name = NULL;
445 /* Stop before context switch with maestro */
446 if (frame != NULL && frame->name != NULL
447 && !strcmp(frame->name, "smx_ctx_sysv_wrapper"))
450 int ret = ret = unw_step(&c);
452 xbt_die("Unexpected end of stack.");
453 } else if (ret < 0) {
454 xbt_die("Error while unwinding stack.");
458 if (xbt_dynar_length(result) == 0) {
459 XBT_INFO("unw_init_local failed");
466 static xbt_dynar_t MC_take_snapshot_stacks(mc_snapshot_t * snapshot)
470 xbt_dynar_new(sizeof(s_mc_snapshot_stack_t),
471 MC_snapshot_stack_free_voidp);
473 unsigned int cursor = 0;
474 stack_region_t current_stack;
476 xbt_dynar_foreach(stacks_areas, cursor, current_stack) {
477 mc_snapshot_stack_t st = xbt_new(s_mc_snapshot_stack_t, 1);
478 st->stack_frames = MC_unwind_stack_frames(current_stack->context);
479 st->local_variables = MC_get_local_variables_values(st->stack_frames, current_stack->process_index);
480 st->process_index = current_stack->process_index;
482 unw_word_t sp = xbt_dynar_get_as(st->stack_frames, 0, mc_stack_frame_t)->sp;
484 xbt_dynar_push(res, &st);
485 (*snapshot)->stack_sizes =
486 xbt_realloc((*snapshot)->stack_sizes, (cursor + 1) * sizeof(size_t));
487 (*snapshot)->stack_sizes[cursor] =
488 (char*) current_stack->address + current_stack->size - (char*) sp;
495 static xbt_dynar_t MC_take_snapshot_ignore()
498 if (mc_heap_comparison_ignore == NULL)
502 xbt_dynar_new(sizeof(mc_heap_ignore_region_t),
503 heap_ignore_region_free_voidp);
505 unsigned int cursor = 0;
506 mc_heap_ignore_region_t current_region;
508 xbt_dynar_foreach(mc_heap_comparison_ignore, cursor, current_region) {
509 mc_heap_ignore_region_t new_region = NULL;
510 new_region = xbt_new0(s_mc_heap_ignore_region_t, 1);
511 new_region->address = current_region->address;
512 new_region->size = current_region->size;
513 new_region->block = current_region->block;
514 new_region->fragment = current_region->fragment;
515 xbt_dynar_push(cpy, &new_region);
522 static void mc_free_snapshot_ignored_data_pvoid(void* data) {
523 mc_snapshot_ignored_data_t ignored_data = (mc_snapshot_ignored_data_t) data;
524 free(ignored_data->data);
527 static void MC_snapshot_handle_ignore(mc_snapshot_t snapshot)
529 snapshot->ignored_data = xbt_dynar_new(sizeof(s_mc_snapshot_ignored_data_t), mc_free_snapshot_ignored_data_pvoid);
532 unsigned int cursor = 0;
533 mc_checkpoint_ignore_region_t region;
534 xbt_dynar_foreach (mc_checkpoint_ignore, cursor, region) {
535 s_mc_snapshot_ignored_data_t ignored_data;
536 ignored_data.start = region->addr;
537 ignored_data.size = region->size;
538 ignored_data.data = malloc(region->size);
539 memcpy(ignored_data.data, region->addr, region->size);
540 xbt_dynar_push(snapshot->ignored_data, &ignored_data);
544 xbt_dynar_foreach (mc_checkpoint_ignore, cursor, region) {
545 memset(region->addr, 0, region->size);
550 static void MC_snapshot_ignore_restore(mc_snapshot_t snapshot)
552 unsigned int cursor = 0;
553 s_mc_snapshot_ignored_data_t ignored_data;
554 xbt_dynar_foreach (snapshot->ignored_data, cursor, ignored_data) {
555 memcpy(ignored_data.start, ignored_data.data, ignored_data.size);
559 /** @brief Can we remove this snapshot?
561 * Some snapshots cannot be removed (yet) because we need them
566 int mc_important_snapshot(mc_snapshot_t snapshot)
568 // We need this snapshot in order to know which
569 // pages needs to be stored in the next snapshot.
570 // This field is only non-NULL when using soft-dirty
572 if (snapshot == mc_model_checker->parent_snapshot)
578 static void MC_get_current_fd(mc_snapshot_t snapshot){
580 struct dirent *fd_number;
588 snapshot->total_fd = 0;
590 fd_dir = opendir ("/proc/self/fd");
591 if (fd_dir == NULL) {
592 printf ("Cannot open directory '/proc/self/fd'\n");
596 while ((fd_number = readdir(fd_dir)) != NULL) {
598 fd_value = atoi(fd_number->d_name);
603 command = bprintf("readlink /proc/self/fd/%s", fd_number->d_name);
604 fp = popen(command, "r");
606 perror("popen failed");
609 xbt_getline(&line, &n, fp);
610 if(line && strncmp(line, "pipe:", 5) != 0 && strncmp(line, "socket:", 7) != 0){
611 fd_infos_t fd = xbt_new0(s_fd_infos_t, 1);
612 fd->filename = strdup(line);
613 fd->filename[strlen(line)-1] = '\0';
614 fd->number = fd_value;
615 fd->flags = fcntl(fd_value, F_GETFD);
616 fd->current_position = lseek(fd_value, 0, SEEK_CUR);
617 snapshot->current_fd = xbt_realloc(snapshot->current_fd, (snapshot->total_fd + 1) * sizeof(fd_infos_t));
618 snapshot->current_fd[snapshot->total_fd] = fd;
619 snapshot->total_fd++;
631 mc_snapshot_t MC_take_snapshot(int num_state)
634 mc_snapshot_t snapshot = xbt_new0(s_mc_snapshot_t, 1);
635 snapshot->enabled_processes = xbt_dynar_new(sizeof(int), NULL);
636 smx_process_t process;
637 xbt_swag_foreach(process, simix_global->process_list) {
638 xbt_dynar_push_as(snapshot->enabled_processes, int, (int)process->pid);
641 MC_snapshot_handle_ignore(snapshot);
643 MC_get_current_fd(snapshot);
645 /* Save the std heap and the writable mapped pages of libsimgrid and binary */
646 MC_get_memory_regions(snapshot);
647 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
648 mc_softdirty_reset();
651 snapshot->to_ignore = MC_take_snapshot_ignore();
653 if (_sg_mc_visited > 0 || strcmp(_sg_mc_property_file, "")) {
655 MC_take_snapshot_stacks(&snapshot);
656 if (_sg_mc_hash && snapshot->stacks != NULL) {
657 snapshot->hash = mc_hash_processes_state(num_state, snapshot->stacks);
665 MC_snapshot_ignore_restore(snapshot);
666 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
667 mc_model_checker->parent_snapshot = snapshot;
672 void MC_restore_snapshot(mc_snapshot_t snapshot)
674 mc_snapshot_t parent_snapshot = mc_model_checker->parent_snapshot;
678 for (i = 0; i < NB_REGIONS; i++) {
679 // For privatized, variables we decided it was not necessary to take the snapshot:
680 if (snapshot->regions[i])
681 MC_region_restore(snapshot->regions[i],
682 parent_snapshot ? parent_snapshot->regions[i] : NULL);
686 if (snapshot->privatization_regions) {
687 // Restore the global variables of the application separately for each
688 // simulated process:
689 for (i = 0; i < smpi_process_count(); i++) {
690 if (snapshot->privatization_regions[i]) {
691 MC_region_restore(snapshot->privatization_regions[i],
692 parent_snapshot ? parent_snapshot->privatization_regions[i] : NULL);
696 if(snapshot->privatization_index >= 0) {
697 // We just rewrote the global variables.
698 // The privatisation segment SMPI thinks
699 // is mapped might be inconsistent with the segment which
700 // is really mapped in memory (kernel state).
701 // We ask politely SMPI to map the segment anyway,
702 // even if it thinks it is the current one:
703 smpi_really_switch_data_segment(snapshot->privatization_index);
707 for(i=0; i < snapshot->total_fd; i++){
709 new_fd = open(snapshot->current_fd[i]->filename, snapshot->current_fd[i]->flags);
710 if(new_fd != -1 && new_fd != snapshot->current_fd[i]->number){
711 dup2(new_fd, snapshot->current_fd[i]->number);
712 fprintf(stderr, "%p\n", fdopen(snapshot->current_fd[i]->number, "rw"));
713 lseek(snapshot->current_fd[i]->number, snapshot->current_fd[i]->current_position, SEEK_SET);
717 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
718 mc_softdirty_reset();
721 MC_snapshot_ignore_restore(snapshot);
722 if (_sg_mc_sparse_checkpoint && _sg_mc_soft_dirty) {
723 mc_model_checker->parent_snapshot = snapshot;
728 mc_snapshot_t simcall_HANDLER_mc_snapshot(smx_simcall_t simcall)
730 return MC_take_snapshot(1);
733 void *MC_snapshot(void)
735 return simcall_mc_snapshot();