13 #include <symbol/kallsyms.h>
15 #include "linux/hash.h"
17 static void machine__remove_thread(struct machine *machine, struct thread *th);
19 static void dsos__init(struct dsos *dsos)
21 INIT_LIST_HEAD(&dsos->head);
25 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
27 map_groups__init(&machine->kmaps, machine);
28 RB_CLEAR_NODE(&machine->rb_node);
29 dsos__init(&machine->user_dsos);
30 dsos__init(&machine->kernel_dsos);
32 machine->threads = RB_ROOT;
33 INIT_LIST_HEAD(&machine->dead_threads);
34 machine->last_match = NULL;
36 machine->vdso_info = NULL;
40 machine->symbol_filter = NULL;
41 machine->id_hdr_size = 0;
42 machine->comm_exec = false;
43 machine->kernel_start = 0;
45 machine->root_dir = strdup(root_dir);
46 if (machine->root_dir == NULL)
49 if (pid != HOST_KERNEL_ID) {
50 struct thread *thread = machine__findnew_thread(machine, -1,
57 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
58 thread__set_comm(thread, comm, 0);
61 machine->current_tid = NULL;
66 struct machine *machine__new_host(void)
68 struct machine *machine = malloc(sizeof(*machine));
70 if (machine != NULL) {
71 machine__init(machine, "", HOST_KERNEL_ID);
73 if (machine__create_kernel_maps(machine) < 0)
83 static void dsos__delete(struct dsos *dsos)
87 list_for_each_entry_safe(pos, n, &dsos->head, node) {
88 RB_CLEAR_NODE(&pos->rb_node);
94 void machine__delete_threads(struct machine *machine)
96 struct rb_node *nd = rb_first(&machine->threads);
99 struct thread *t = rb_entry(nd, struct thread, rb_node);
102 machine__remove_thread(machine, t);
106 void machine__exit(struct machine *machine)
108 map_groups__exit(&machine->kmaps);
109 dsos__delete(&machine->user_dsos);
110 dsos__delete(&machine->kernel_dsos);
112 zfree(&machine->root_dir);
113 zfree(&machine->current_tid);
116 void machine__delete(struct machine *machine)
118 machine__exit(machine);
122 void machines__init(struct machines *machines)
124 machine__init(&machines->host, "", HOST_KERNEL_ID);
125 machines->guests = RB_ROOT;
126 machines->symbol_filter = NULL;
129 void machines__exit(struct machines *machines)
131 machine__exit(&machines->host);
135 struct machine *machines__add(struct machines *machines, pid_t pid,
136 const char *root_dir)
138 struct rb_node **p = &machines->guests.rb_node;
139 struct rb_node *parent = NULL;
140 struct machine *pos, *machine = malloc(sizeof(*machine));
145 if (machine__init(machine, root_dir, pid) != 0) {
150 machine->symbol_filter = machines->symbol_filter;
154 pos = rb_entry(parent, struct machine, rb_node);
161 rb_link_node(&machine->rb_node, parent, p);
162 rb_insert_color(&machine->rb_node, &machines->guests);
167 void machines__set_symbol_filter(struct machines *machines,
168 symbol_filter_t symbol_filter)
172 machines->symbol_filter = symbol_filter;
173 machines->host.symbol_filter = symbol_filter;
175 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
176 struct machine *machine = rb_entry(nd, struct machine, rb_node);
178 machine->symbol_filter = symbol_filter;
182 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
186 machines->host.comm_exec = comm_exec;
188 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
189 struct machine *machine = rb_entry(nd, struct machine, rb_node);
191 machine->comm_exec = comm_exec;
195 struct machine *machines__find(struct machines *machines, pid_t pid)
197 struct rb_node **p = &machines->guests.rb_node;
198 struct rb_node *parent = NULL;
199 struct machine *machine;
200 struct machine *default_machine = NULL;
202 if (pid == HOST_KERNEL_ID)
203 return &machines->host;
207 machine = rb_entry(parent, struct machine, rb_node);
208 if (pid < machine->pid)
210 else if (pid > machine->pid)
215 default_machine = machine;
218 return default_machine;
221 struct machine *machines__findnew(struct machines *machines, pid_t pid)
224 const char *root_dir = "";
225 struct machine *machine = machines__find(machines, pid);
227 if (machine && (machine->pid == pid))
230 if ((pid != HOST_KERNEL_ID) &&
231 (pid != DEFAULT_GUEST_KERNEL_ID) &&
232 (symbol_conf.guestmount)) {
233 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
234 if (access(path, R_OK)) {
235 static struct strlist *seen;
238 seen = strlist__new(true, NULL);
240 if (!strlist__has_entry(seen, path)) {
241 pr_err("Can't access file %s\n", path);
242 strlist__add(seen, path);
250 machine = machines__add(machines, pid, root_dir);
255 void machines__process_guests(struct machines *machines,
256 machine__process_t process, void *data)
260 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
261 struct machine *pos = rb_entry(nd, struct machine, rb_node);
266 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
268 if (machine__is_host(machine))
269 snprintf(bf, size, "[%s]", "kernel.kallsyms");
270 else if (machine__is_default_guest(machine))
271 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
273 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
280 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
282 struct rb_node *node;
283 struct machine *machine;
285 machines->host.id_hdr_size = id_hdr_size;
287 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
288 machine = rb_entry(node, struct machine, rb_node);
289 machine->id_hdr_size = id_hdr_size;
295 static void machine__update_thread_pid(struct machine *machine,
296 struct thread *th, pid_t pid)
298 struct thread *leader;
300 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
305 if (th->pid_ == th->tid)
308 leader = machine__findnew_thread(machine, th->pid_, th->pid_);
313 leader->mg = map_groups__new(machine);
318 if (th->mg == leader->mg)
323 * Maps are created from MMAP events which provide the pid and
324 * tid. Consequently there never should be any maps on a thread
325 * with an unknown pid. Just print an error if there are.
327 if (!map_groups__empty(th->mg))
328 pr_err("Discarding thread maps for %d:%d\n",
330 map_groups__delete(th->mg);
333 th->mg = map_groups__get(leader->mg);
338 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
341 static struct thread *__machine__findnew_thread(struct machine *machine,
342 pid_t pid, pid_t tid,
345 struct rb_node **p = &machine->threads.rb_node;
346 struct rb_node *parent = NULL;
350 * Front-end cache - TID lookups come in blocks,
351 * so most of the time we dont have to look up
354 th = machine->last_match;
356 if (th->tid == tid) {
357 machine__update_thread_pid(machine, th, pid);
361 thread__zput(machine->last_match);
366 th = rb_entry(parent, struct thread, rb_node);
368 if (th->tid == tid) {
369 machine->last_match = thread__get(th);
370 machine__update_thread_pid(machine, th, pid);
383 th = thread__new(pid, tid);
385 rb_link_node(&th->rb_node, parent, p);
386 rb_insert_color(&th->rb_node, &machine->threads);
389 * We have to initialize map_groups separately
390 * after rb tree is updated.
392 * The reason is that we call machine__findnew_thread
393 * within thread__init_map_groups to find the thread
394 * leader and that would screwed the rb tree.
396 if (thread__init_map_groups(th, machine)) {
397 rb_erase(&th->rb_node, &machine->threads);
402 * It is now in the rbtree, get a ref
405 machine->last_match = thread__get(th);
411 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
414 return __machine__findnew_thread(machine, pid, tid, true);
417 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
420 return __machine__findnew_thread(machine, pid, tid, false);
423 struct comm *machine__thread_exec_comm(struct machine *machine,
424 struct thread *thread)
426 if (machine->comm_exec)
427 return thread__exec_comm(thread);
429 return thread__comm(thread);
432 int machine__process_comm_event(struct machine *machine, union perf_event *event,
433 struct perf_sample *sample)
435 struct thread *thread = machine__findnew_thread(machine,
438 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
441 machine->comm_exec = true;
444 perf_event__fprintf_comm(event, stdout);
446 if (thread == NULL ||
447 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
448 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
455 int machine__process_lost_event(struct machine *machine __maybe_unused,
456 union perf_event *event, struct perf_sample *sample __maybe_unused)
458 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
459 event->lost.id, event->lost.lost);
464 machine__module_dso(struct machine *machine, struct kmod_path *m,
465 const char *filename)
469 dso = dsos__find(&machine->kernel_dsos, m->name, true);
471 dso = dsos__addnew(&machine->kernel_dsos, m->name);
475 if (machine__is_host(machine))
476 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
478 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
480 /* _KMODULE_COMP should be next to _KMODULE */
481 if (m->kmod && m->comp)
484 dso__set_short_name(dso, strdup(m->name), true);
485 dso__set_long_name(dso, strdup(filename), true);
491 struct map *machine__new_module(struct machine *machine, u64 start,
492 const char *filename)
494 struct map *map = NULL;
498 if (kmod_path__parse_name(&m, filename))
501 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
506 dso = machine__module_dso(machine, &m, filename);
510 map = map__new2(start, dso, MAP__FUNCTION);
514 map_groups__insert(&machine->kmaps, map);
521 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
524 size_t ret = __dsos__fprintf(&machines->host.kernel_dsos.head, fp) +
525 __dsos__fprintf(&machines->host.user_dsos.head, fp);
527 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
528 struct machine *pos = rb_entry(nd, struct machine, rb_node);
529 ret += __dsos__fprintf(&pos->kernel_dsos.head, fp);
530 ret += __dsos__fprintf(&pos->user_dsos.head, fp);
536 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
537 bool (skip)(struct dso *dso, int parm), int parm)
539 return __dsos__fprintf_buildid(&m->kernel_dsos.head, fp, skip, parm) +
540 __dsos__fprintf_buildid(&m->user_dsos.head, fp, skip, parm);
543 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
544 bool (skip)(struct dso *dso, int parm), int parm)
547 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
549 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
550 struct machine *pos = rb_entry(nd, struct machine, rb_node);
551 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
556 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
560 struct dso *kdso = machine->vmlinux_maps[MAP__FUNCTION]->dso;
562 if (kdso->has_build_id) {
563 char filename[PATH_MAX];
564 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
565 printed += fprintf(fp, "[0] %s\n", filename);
568 for (i = 0; i < vmlinux_path__nr_entries; ++i)
569 printed += fprintf(fp, "[%d] %s\n",
570 i + kdso->has_build_id, vmlinux_path[i]);
575 size_t machine__fprintf(struct machine *machine, FILE *fp)
580 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
581 struct thread *pos = rb_entry(nd, struct thread, rb_node);
583 ret += thread__fprintf(pos, fp);
589 static struct dso *machine__get_kernel(struct machine *machine)
591 const char *vmlinux_name = NULL;
594 if (machine__is_host(machine)) {
595 vmlinux_name = symbol_conf.vmlinux_name;
597 vmlinux_name = "[kernel.kallsyms]";
599 kernel = dso__kernel_findnew(machine, vmlinux_name,
605 if (machine__is_default_guest(machine))
606 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
608 vmlinux_name = machine__mmap_name(machine, bf,
611 kernel = dso__kernel_findnew(machine, vmlinux_name,
613 DSO_TYPE_GUEST_KERNEL);
616 if (kernel != NULL && (!kernel->has_build_id))
617 dso__read_running_kernel_build_id(kernel, machine);
622 struct process_args {
626 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
629 if (machine__is_default_guest(machine))
630 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
632 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
635 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
637 /* Figure out the start address of kernel map from /proc/kallsyms.
638 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
639 * symbol_name if it's not that important.
641 static u64 machine__get_running_kernel_start(struct machine *machine,
642 const char **symbol_name)
644 char filename[PATH_MAX];
649 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
651 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
654 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
655 addr = kallsyms__get_function_start(filename, name);
666 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
669 u64 start = machine__get_running_kernel_start(machine, NULL);
671 for (type = 0; type < MAP__NR_TYPES; ++type) {
674 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
675 if (machine->vmlinux_maps[type] == NULL)
678 machine->vmlinux_maps[type]->map_ip =
679 machine->vmlinux_maps[type]->unmap_ip =
681 kmap = map__kmap(machine->vmlinux_maps[type]);
685 kmap->kmaps = &machine->kmaps;
686 map_groups__insert(&machine->kmaps,
687 machine->vmlinux_maps[type]);
693 void machine__destroy_kernel_maps(struct machine *machine)
697 for (type = 0; type < MAP__NR_TYPES; ++type) {
700 if (machine->vmlinux_maps[type] == NULL)
703 kmap = map__kmap(machine->vmlinux_maps[type]);
704 map_groups__remove(&machine->kmaps,
705 machine->vmlinux_maps[type]);
706 if (kmap && kmap->ref_reloc_sym) {
708 * ref_reloc_sym is shared among all maps, so free just
711 if (type == MAP__FUNCTION) {
712 zfree((char **)&kmap->ref_reloc_sym->name);
713 zfree(&kmap->ref_reloc_sym);
715 kmap->ref_reloc_sym = NULL;
718 map__delete(machine->vmlinux_maps[type]);
719 machine->vmlinux_maps[type] = NULL;
723 int machines__create_guest_kernel_maps(struct machines *machines)
726 struct dirent **namelist = NULL;
732 if (symbol_conf.default_guest_vmlinux_name ||
733 symbol_conf.default_guest_modules ||
734 symbol_conf.default_guest_kallsyms) {
735 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
738 if (symbol_conf.guestmount) {
739 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
742 for (i = 0; i < items; i++) {
743 if (!isdigit(namelist[i]->d_name[0])) {
744 /* Filter out . and .. */
747 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
748 if ((*endp != '\0') ||
749 (endp == namelist[i]->d_name) ||
751 pr_debug("invalid directory (%s). Skipping.\n",
752 namelist[i]->d_name);
755 sprintf(path, "%s/%s/proc/kallsyms",
756 symbol_conf.guestmount,
757 namelist[i]->d_name);
758 ret = access(path, R_OK);
760 pr_debug("Can't access file %s\n", path);
763 machines__create_kernel_maps(machines, pid);
772 void machines__destroy_kernel_maps(struct machines *machines)
774 struct rb_node *next = rb_first(&machines->guests);
776 machine__destroy_kernel_maps(&machines->host);
779 struct machine *pos = rb_entry(next, struct machine, rb_node);
781 next = rb_next(&pos->rb_node);
782 rb_erase(&pos->rb_node, &machines->guests);
783 machine__delete(pos);
787 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
789 struct machine *machine = machines__findnew(machines, pid);
794 return machine__create_kernel_maps(machine);
797 int machine__load_kallsyms(struct machine *machine, const char *filename,
798 enum map_type type, symbol_filter_t filter)
800 struct map *map = machine->vmlinux_maps[type];
801 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
804 dso__set_loaded(map->dso, type);
806 * Since /proc/kallsyms will have multiple sessions for the
807 * kernel, with modules between them, fixup the end of all
810 __map_groups__fixup_end(&machine->kmaps, type);
816 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
817 symbol_filter_t filter)
819 struct map *map = machine->vmlinux_maps[type];
820 int ret = dso__load_vmlinux_path(map->dso, map, filter);
823 dso__set_loaded(map->dso, type);
828 static void map_groups__fixup_end(struct map_groups *mg)
831 for (i = 0; i < MAP__NR_TYPES; ++i)
832 __map_groups__fixup_end(mg, i);
835 static char *get_kernel_version(const char *root_dir)
837 char version[PATH_MAX];
840 const char *prefix = "Linux version ";
842 sprintf(version, "%s/proc/version", root_dir);
843 file = fopen(version, "r");
848 tmp = fgets(version, sizeof(version), file);
851 name = strstr(version, prefix);
854 name += strlen(prefix);
855 tmp = strchr(name, ' ');
862 static bool is_kmod_dso(struct dso *dso)
864 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
865 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
868 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
874 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
878 long_name = strdup(path);
879 if (long_name == NULL)
882 dso__set_long_name(map->dso, long_name, true);
883 dso__kernel_module_get_build_id(map->dso, "");
886 * Full name could reveal us kmod compression, so
887 * we need to update the symtab_type if needed.
889 if (m->comp && is_kmod_dso(map->dso))
890 map->dso->symtab_type++;
895 static int map_groups__set_modules_path_dir(struct map_groups *mg,
896 const char *dir_name, int depth)
899 DIR *dir = opendir(dir_name);
903 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
907 while ((dent = readdir(dir)) != NULL) {
911 /*sshfs might return bad dent->d_type, so we have to stat*/
912 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
916 if (S_ISDIR(st.st_mode)) {
917 if (!strcmp(dent->d_name, ".") ||
918 !strcmp(dent->d_name, ".."))
921 /* Do not follow top-level source and build symlinks */
923 if (!strcmp(dent->d_name, "source") ||
924 !strcmp(dent->d_name, "build"))
928 ret = map_groups__set_modules_path_dir(mg, path,
935 ret = kmod_path__parse_name(&m, dent->d_name);
940 ret = map_groups__set_module_path(mg, path, &m);
954 static int machine__set_modules_path(struct machine *machine)
957 char modules_path[PATH_MAX];
959 version = get_kernel_version(machine->root_dir);
963 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
964 machine->root_dir, version);
967 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
970 static int machine__create_module(void *arg, const char *name, u64 start)
972 struct machine *machine = arg;
975 map = machine__new_module(machine, start, name);
979 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
984 static int machine__create_modules(struct machine *machine)
989 if (machine__is_default_guest(machine)) {
990 modules = symbol_conf.default_guest_modules;
992 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
996 if (symbol__restricted_filename(modules, "/proc/modules"))
999 if (modules__parse(modules, machine, machine__create_module))
1002 if (!machine__set_modules_path(machine))
1005 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1010 int machine__create_kernel_maps(struct machine *machine)
1012 struct dso *kernel = machine__get_kernel(machine);
1014 u64 addr = machine__get_running_kernel_start(machine, &name);
1018 if (kernel == NULL ||
1019 __machine__create_kernel_maps(machine, kernel) < 0)
1022 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1023 if (machine__is_host(machine))
1024 pr_debug("Problems creating module maps, "
1025 "continuing anyway...\n");
1027 pr_debug("Problems creating module maps for guest %d, "
1028 "continuing anyway...\n", machine->pid);
1032 * Now that we have all the maps created, just set the ->end of them:
1034 map_groups__fixup_end(&machine->kmaps);
1036 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
1038 machine__destroy_kernel_maps(machine);
1045 static void machine__set_kernel_mmap_len(struct machine *machine,
1046 union perf_event *event)
1050 for (i = 0; i < MAP__NR_TYPES; i++) {
1051 machine->vmlinux_maps[i]->start = event->mmap.start;
1052 machine->vmlinux_maps[i]->end = (event->mmap.start +
1055 * Be a bit paranoid here, some perf.data file came with
1056 * a zero sized synthesized MMAP event for the kernel.
1058 if (machine->vmlinux_maps[i]->end == 0)
1059 machine->vmlinux_maps[i]->end = ~0ULL;
1063 static bool machine__uses_kcore(struct machine *machine)
1067 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1068 if (dso__is_kcore(dso))
1075 static int machine__process_kernel_mmap_event(struct machine *machine,
1076 union perf_event *event)
1079 char kmmap_prefix[PATH_MAX];
1080 enum dso_kernel_type kernel_type;
1081 bool is_kernel_mmap;
1083 /* If we have maps from kcore then we do not need or want any others */
1084 if (machine__uses_kcore(machine))
1087 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1088 if (machine__is_host(machine))
1089 kernel_type = DSO_TYPE_KERNEL;
1091 kernel_type = DSO_TYPE_GUEST_KERNEL;
1093 is_kernel_mmap = memcmp(event->mmap.filename,
1095 strlen(kmmap_prefix) - 1) == 0;
1096 if (event->mmap.filename[0] == '/' ||
1097 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1098 map = machine__new_module(machine, event->mmap.start,
1099 event->mmap.filename);
1103 map->end = map->start + event->mmap.len;
1104 } else if (is_kernel_mmap) {
1105 const char *symbol_name = (event->mmap.filename +
1106 strlen(kmmap_prefix));
1108 * Should be there already, from the build-id table in
1111 struct dso *kernel = NULL;
1114 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1115 if (is_kernel_module(dso->long_name))
1123 kernel = __dsos__findnew(&machine->kernel_dsos,
1128 kernel->kernel = kernel_type;
1129 if (__machine__create_kernel_maps(machine, kernel) < 0)
1132 if (strstr(kernel->long_name, "vmlinux"))
1133 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1135 machine__set_kernel_mmap_len(machine, event);
1138 * Avoid using a zero address (kptr_restrict) for the ref reloc
1139 * symbol. Effectively having zero here means that at record
1140 * time /proc/sys/kernel/kptr_restrict was non zero.
1142 if (event->mmap.pgoff != 0) {
1143 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1148 if (machine__is_default_guest(machine)) {
1150 * preload dso of guest kernel and modules
1152 dso__load(kernel, machine->vmlinux_maps[MAP__FUNCTION],
1161 int machine__process_mmap2_event(struct machine *machine,
1162 union perf_event *event,
1163 struct perf_sample *sample __maybe_unused)
1165 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1166 struct thread *thread;
1172 perf_event__fprintf_mmap2(event, stdout);
1174 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1175 cpumode == PERF_RECORD_MISC_KERNEL) {
1176 ret = machine__process_kernel_mmap_event(machine, event);
1182 thread = machine__findnew_thread(machine, event->mmap2.pid,
1187 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1188 type = MAP__VARIABLE;
1190 type = MAP__FUNCTION;
1192 map = map__new(machine, event->mmap2.start,
1193 event->mmap2.len, event->mmap2.pgoff,
1194 event->mmap2.pid, event->mmap2.maj,
1195 event->mmap2.min, event->mmap2.ino,
1196 event->mmap2.ino_generation,
1199 event->mmap2.filename, type, thread);
1204 thread__insert_map(thread, map);
1208 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1212 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1213 struct perf_sample *sample __maybe_unused)
1215 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1216 struct thread *thread;
1222 perf_event__fprintf_mmap(event, stdout);
1224 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1225 cpumode == PERF_RECORD_MISC_KERNEL) {
1226 ret = machine__process_kernel_mmap_event(machine, event);
1232 thread = machine__findnew_thread(machine, event->mmap.pid,
1237 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1238 type = MAP__VARIABLE;
1240 type = MAP__FUNCTION;
1242 map = map__new(machine, event->mmap.start,
1243 event->mmap.len, event->mmap.pgoff,
1244 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1245 event->mmap.filename,
1251 thread__insert_map(thread, map);
1255 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1259 static void machine__remove_thread(struct machine *machine, struct thread *th)
1261 if (machine->last_match == th)
1262 thread__zput(machine->last_match);
1264 rb_erase(&th->rb_node, &machine->threads);
1266 * Move it first to the dead_threads list, then drop the reference,
1267 * if this is the last reference, then the thread__delete destructor
1268 * will be called and we will remove it from the dead_threads list.
1270 list_add_tail(&th->node, &machine->dead_threads);
1274 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1275 struct perf_sample *sample)
1277 struct thread *thread = machine__find_thread(machine,
1280 struct thread *parent = machine__findnew_thread(machine,
1284 /* if a thread currently exists for the thread id remove it */
1286 machine__remove_thread(machine, thread);
1288 thread = machine__findnew_thread(machine, event->fork.pid,
1291 perf_event__fprintf_task(event, stdout);
1293 if (thread == NULL || parent == NULL ||
1294 thread__fork(thread, parent, sample->time) < 0) {
1295 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1302 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1303 struct perf_sample *sample __maybe_unused)
1305 struct thread *thread = machine__find_thread(machine,
1310 perf_event__fprintf_task(event, stdout);
1313 thread__exited(thread);
1318 int machine__process_event(struct machine *machine, union perf_event *event,
1319 struct perf_sample *sample)
1323 switch (event->header.type) {
1324 case PERF_RECORD_COMM:
1325 ret = machine__process_comm_event(machine, event, sample); break;
1326 case PERF_RECORD_MMAP:
1327 ret = machine__process_mmap_event(machine, event, sample); break;
1328 case PERF_RECORD_MMAP2:
1329 ret = machine__process_mmap2_event(machine, event, sample); break;
1330 case PERF_RECORD_FORK:
1331 ret = machine__process_fork_event(machine, event, sample); break;
1332 case PERF_RECORD_EXIT:
1333 ret = machine__process_exit_event(machine, event, sample); break;
1334 case PERF_RECORD_LOST:
1335 ret = machine__process_lost_event(machine, event, sample); break;
1344 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1346 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1351 static void ip__resolve_ams(struct thread *thread,
1352 struct addr_map_symbol *ams,
1355 struct addr_location al;
1357 memset(&al, 0, sizeof(al));
1359 * We cannot use the header.misc hint to determine whether a
1360 * branch stack address is user, kernel, guest, hypervisor.
1361 * Branches may straddle the kernel/user/hypervisor boundaries.
1362 * Thus, we have to try consecutively until we find a match
1363 * or else, the symbol is unknown
1365 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1368 ams->al_addr = al.addr;
1373 static void ip__resolve_data(struct thread *thread,
1374 u8 m, struct addr_map_symbol *ams, u64 addr)
1376 struct addr_location al;
1378 memset(&al, 0, sizeof(al));
1380 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1381 if (al.map == NULL) {
1383 * some shared data regions have execute bit set which puts
1384 * their mapping in the MAP__FUNCTION type array.
1385 * Check there as a fallback option before dropping the sample.
1387 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1391 ams->al_addr = al.addr;
1396 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1397 struct addr_location *al)
1399 struct mem_info *mi = zalloc(sizeof(*mi));
1404 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1405 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1406 mi->data_src.val = sample->data_src;
1411 static int add_callchain_ip(struct thread *thread,
1412 struct symbol **parent,
1413 struct addr_location *root_al,
1417 struct addr_location al;
1422 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1425 if (ip >= PERF_CONTEXT_MAX) {
1427 case PERF_CONTEXT_HV:
1428 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1430 case PERF_CONTEXT_KERNEL:
1431 *cpumode = PERF_RECORD_MISC_KERNEL;
1433 case PERF_CONTEXT_USER:
1434 *cpumode = PERF_RECORD_MISC_USER;
1437 pr_debug("invalid callchain context: "
1438 "%"PRId64"\n", (s64) ip);
1440 * It seems the callchain is corrupted.
1443 callchain_cursor_reset(&callchain_cursor);
1448 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1452 if (al.sym != NULL) {
1453 if (sort__has_parent && !*parent &&
1454 symbol__match_regex(al.sym, &parent_regex))
1456 else if (have_ignore_callees && root_al &&
1457 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1458 /* Treat this symbol as the root,
1459 forgetting its callees. */
1461 callchain_cursor_reset(&callchain_cursor);
1465 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1468 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1469 struct addr_location *al)
1472 const struct branch_stack *bs = sample->branch_stack;
1473 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1478 for (i = 0; i < bs->nr; i++) {
1479 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1480 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1481 bi[i].flags = bs->entries[i].flags;
1488 #define NO_ENTRY 0xff
1490 #define PERF_MAX_BRANCH_DEPTH 127
1493 static int remove_loops(struct branch_entry *l, int nr)
1496 unsigned char chash[CHASHSZ];
1498 memset(chash, NO_ENTRY, sizeof(chash));
1500 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1502 for (i = 0; i < nr; i++) {
1503 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1505 /* no collision handling for now */
1506 if (chash[h] == NO_ENTRY) {
1508 } else if (l[chash[h]].from == l[i].from) {
1509 bool is_loop = true;
1510 /* check if it is a real loop */
1512 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1513 if (l[j].from != l[i + off].from) {
1518 memmove(l + i, l + i + off,
1519 (nr - (i + off)) * sizeof(*l));
1528 * Recolve LBR callstack chain sample
1530 * 1 on success get LBR callchain information
1531 * 0 no available LBR callchain information, should try fp
1532 * negative error code on other errors.
1534 static int resolve_lbr_callchain_sample(struct thread *thread,
1535 struct perf_sample *sample,
1536 struct symbol **parent,
1537 struct addr_location *root_al,
1540 struct ip_callchain *chain = sample->callchain;
1541 int chain_nr = min(max_stack, (int)chain->nr);
1542 u8 cpumode = PERF_RECORD_MISC_USER;
1546 for (i = 0; i < chain_nr; i++) {
1547 if (chain->ips[i] == PERF_CONTEXT_USER)
1551 /* LBR only affects the user callchain */
1552 if (i != chain_nr) {
1553 struct branch_stack *lbr_stack = sample->branch_stack;
1554 int lbr_nr = lbr_stack->nr;
1556 * LBR callstack can only get user call chain.
1557 * The mix_chain_nr is kernel call chain
1558 * number plus LBR user call chain number.
1559 * i is kernel call chain number,
1560 * 1 is PERF_CONTEXT_USER,
1561 * lbr_nr + 1 is the user call chain number.
1562 * For details, please refer to the comments
1563 * in callchain__printf
1565 int mix_chain_nr = i + 1 + lbr_nr + 1;
1567 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1568 pr_warning("corrupted callchain. skipping...\n");
1572 for (j = 0; j < mix_chain_nr; j++) {
1573 if (callchain_param.order == ORDER_CALLEE) {
1577 ip = lbr_stack->entries[j - i - 2].from;
1579 ip = lbr_stack->entries[0].to;
1582 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1583 else if (j > lbr_nr)
1584 ip = chain->ips[i + 1 - (j - lbr_nr)];
1586 ip = lbr_stack->entries[0].to;
1589 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1591 return (err < 0) ? err : 0;
1599 static int thread__resolve_callchain_sample(struct thread *thread,
1600 struct perf_evsel *evsel,
1601 struct perf_sample *sample,
1602 struct symbol **parent,
1603 struct addr_location *root_al,
1606 struct branch_stack *branch = sample->branch_stack;
1607 struct ip_callchain *chain = sample->callchain;
1608 int chain_nr = min(max_stack, (int)chain->nr);
1609 u8 cpumode = PERF_RECORD_MISC_USER;
1614 callchain_cursor_reset(&callchain_cursor);
1616 if (has_branch_callstack(evsel)) {
1617 err = resolve_lbr_callchain_sample(thread, sample, parent,
1618 root_al, max_stack);
1620 return (err < 0) ? err : 0;
1624 * Based on DWARF debug information, some architectures skip
1625 * a callchain entry saved by the kernel.
1627 if (chain->nr < PERF_MAX_STACK_DEPTH)
1628 skip_idx = arch_skip_callchain_idx(thread, chain);
1631 * Add branches to call stack for easier browsing. This gives
1632 * more context for a sample than just the callers.
1634 * This uses individual histograms of paths compared to the
1635 * aggregated histograms the normal LBR mode uses.
1637 * Limitations for now:
1638 * - No extra filters
1639 * - No annotations (should annotate somehow)
1642 if (branch && callchain_param.branch_callstack) {
1643 int nr = min(max_stack, (int)branch->nr);
1644 struct branch_entry be[nr];
1646 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1647 pr_warning("corrupted branch chain. skipping...\n");
1651 for (i = 0; i < nr; i++) {
1652 if (callchain_param.order == ORDER_CALLEE) {
1653 be[i] = branch->entries[i];
1655 * Check for overlap into the callchain.
1656 * The return address is one off compared to
1657 * the branch entry. To adjust for this
1658 * assume the calling instruction is not longer
1661 if (i == skip_idx ||
1662 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1664 else if (be[i].from < chain->ips[first_call] &&
1665 be[i].from >= chain->ips[first_call] - 8)
1668 be[i] = branch->entries[branch->nr - i - 1];
1671 nr = remove_loops(be, nr);
1673 for (i = 0; i < nr; i++) {
1674 err = add_callchain_ip(thread, parent, root_al,
1677 err = add_callchain_ip(thread, parent, root_al,
1688 if (chain->nr > PERF_MAX_STACK_DEPTH) {
1689 pr_warning("corrupted callchain. skipping...\n");
1693 for (i = first_call; i < chain_nr; i++) {
1696 if (callchain_param.order == ORDER_CALLEE)
1699 j = chain->nr - i - 1;
1701 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1707 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1710 return (err < 0) ? err : 0;
1716 static int unwind_entry(struct unwind_entry *entry, void *arg)
1718 struct callchain_cursor *cursor = arg;
1719 return callchain_cursor_append(cursor, entry->ip,
1720 entry->map, entry->sym);
1723 int thread__resolve_callchain(struct thread *thread,
1724 struct perf_evsel *evsel,
1725 struct perf_sample *sample,
1726 struct symbol **parent,
1727 struct addr_location *root_al,
1730 int ret = thread__resolve_callchain_sample(thread, evsel,
1732 root_al, max_stack);
1736 /* Can we do dwarf post unwind? */
1737 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1738 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1741 /* Bail out if nothing was captured. */
1742 if ((!sample->user_regs.regs) ||
1743 (!sample->user_stack.size))
1746 return unwind__get_entries(unwind_entry, &callchain_cursor,
1747 thread, sample, max_stack);
1751 int machine__for_each_thread(struct machine *machine,
1752 int (*fn)(struct thread *thread, void *p),
1756 struct thread *thread;
1759 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1760 thread = rb_entry(nd, struct thread, rb_node);
1761 rc = fn(thread, priv);
1766 list_for_each_entry(thread, &machine->dead_threads, node) {
1767 rc = fn(thread, priv);
1774 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1775 struct target *target, struct thread_map *threads,
1776 perf_event__handler_t process, bool data_mmap)
1778 if (target__has_task(target))
1779 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap);
1780 else if (target__has_cpu(target))
1781 return perf_event__synthesize_threads(tool, process, machine, data_mmap);
1782 /* command specified */
1786 pid_t machine__get_current_tid(struct machine *machine, int cpu)
1788 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
1791 return machine->current_tid[cpu];
1794 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
1797 struct thread *thread;
1802 if (!machine->current_tid) {
1805 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
1806 if (!machine->current_tid)
1808 for (i = 0; i < MAX_NR_CPUS; i++)
1809 machine->current_tid[i] = -1;
1812 if (cpu >= MAX_NR_CPUS) {
1813 pr_err("Requested CPU %d too large. ", cpu);
1814 pr_err("Consider raising MAX_NR_CPUS\n");
1818 machine->current_tid[cpu] = tid;
1820 thread = machine__findnew_thread(machine, pid, tid);
1829 int machine__get_kernel_start(struct machine *machine)
1831 struct map *map = machine__kernel_map(machine, MAP__FUNCTION);
1835 * The only addresses above 2^63 are kernel addresses of a 64-bit
1836 * kernel. Note that addresses are unsigned so that on a 32-bit system
1837 * all addresses including kernel addresses are less than 2^32. In
1838 * that case (32-bit system), if the kernel mapping is unknown, all
1839 * addresses will be assumed to be in user space - see
1840 * machine__kernel_ip().
1842 machine->kernel_start = 1ULL << 63;
1844 err = map__load(map, machine->symbol_filter);
1846 machine->kernel_start = map->start;