13 #include <symbol/kallsyms.h>
15 #include "linux/hash.h"
17 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
19 static void dsos__init(struct dsos *dsos)
21 INIT_LIST_HEAD(&dsos->head);
23 pthread_rwlock_init(&dsos->lock, NULL);
26 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
28 map_groups__init(&machine->kmaps, machine);
29 RB_CLEAR_NODE(&machine->rb_node);
30 dsos__init(&machine->dsos);
32 machine->threads = RB_ROOT;
33 pthread_rwlock_init(&machine->threads_lock, NULL);
34 INIT_LIST_HEAD(&machine->dead_threads);
35 machine->last_match = NULL;
37 machine->vdso_info = NULL;
42 machine->symbol_filter = NULL;
43 machine->id_hdr_size = 0;
44 machine->comm_exec = false;
45 machine->kernel_start = 0;
47 machine->root_dir = strdup(root_dir);
48 if (machine->root_dir == NULL)
51 if (pid != HOST_KERNEL_ID) {
52 struct thread *thread = machine__findnew_thread(machine, -1,
59 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
60 thread__set_comm(thread, comm, 0);
64 machine->current_tid = NULL;
69 struct machine *machine__new_host(void)
71 struct machine *machine = malloc(sizeof(*machine));
73 if (machine != NULL) {
74 machine__init(machine, "", HOST_KERNEL_ID);
76 if (machine__create_kernel_maps(machine) < 0)
86 static void dsos__purge(struct dsos *dsos)
90 pthread_rwlock_wrlock(&dsos->lock);
92 list_for_each_entry_safe(pos, n, &dsos->head, node) {
93 RB_CLEAR_NODE(&pos->rb_node);
94 list_del_init(&pos->node);
98 pthread_rwlock_unlock(&dsos->lock);
101 static void dsos__exit(struct dsos *dsos)
104 pthread_rwlock_destroy(&dsos->lock);
107 void machine__delete_threads(struct machine *machine)
111 pthread_rwlock_wrlock(&machine->threads_lock);
112 nd = rb_first(&machine->threads);
114 struct thread *t = rb_entry(nd, struct thread, rb_node);
117 __machine__remove_thread(machine, t, false);
119 pthread_rwlock_unlock(&machine->threads_lock);
122 void machine__exit(struct machine *machine)
124 map_groups__exit(&machine->kmaps);
125 dsos__exit(&machine->dsos);
126 machine__exit_vdso(machine);
127 zfree(&machine->root_dir);
128 zfree(&machine->current_tid);
129 pthread_rwlock_destroy(&machine->threads_lock);
132 void machine__delete(struct machine *machine)
134 machine__exit(machine);
138 void machines__init(struct machines *machines)
140 machine__init(&machines->host, "", HOST_KERNEL_ID);
141 machines->guests = RB_ROOT;
142 machines->symbol_filter = NULL;
145 void machines__exit(struct machines *machines)
147 machine__exit(&machines->host);
151 struct machine *machines__add(struct machines *machines, pid_t pid,
152 const char *root_dir)
154 struct rb_node **p = &machines->guests.rb_node;
155 struct rb_node *parent = NULL;
156 struct machine *pos, *machine = malloc(sizeof(*machine));
161 if (machine__init(machine, root_dir, pid) != 0) {
166 machine->symbol_filter = machines->symbol_filter;
170 pos = rb_entry(parent, struct machine, rb_node);
177 rb_link_node(&machine->rb_node, parent, p);
178 rb_insert_color(&machine->rb_node, &machines->guests);
183 void machines__set_symbol_filter(struct machines *machines,
184 symbol_filter_t symbol_filter)
188 machines->symbol_filter = symbol_filter;
189 machines->host.symbol_filter = symbol_filter;
191 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
192 struct machine *machine = rb_entry(nd, struct machine, rb_node);
194 machine->symbol_filter = symbol_filter;
198 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
202 machines->host.comm_exec = comm_exec;
204 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
205 struct machine *machine = rb_entry(nd, struct machine, rb_node);
207 machine->comm_exec = comm_exec;
211 struct machine *machines__find(struct machines *machines, pid_t pid)
213 struct rb_node **p = &machines->guests.rb_node;
214 struct rb_node *parent = NULL;
215 struct machine *machine;
216 struct machine *default_machine = NULL;
218 if (pid == HOST_KERNEL_ID)
219 return &machines->host;
223 machine = rb_entry(parent, struct machine, rb_node);
224 if (pid < machine->pid)
226 else if (pid > machine->pid)
231 default_machine = machine;
234 return default_machine;
237 struct machine *machines__findnew(struct machines *machines, pid_t pid)
240 const char *root_dir = "";
241 struct machine *machine = machines__find(machines, pid);
243 if (machine && (machine->pid == pid))
246 if ((pid != HOST_KERNEL_ID) &&
247 (pid != DEFAULT_GUEST_KERNEL_ID) &&
248 (symbol_conf.guestmount)) {
249 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
250 if (access(path, R_OK)) {
251 static struct strlist *seen;
254 seen = strlist__new(NULL, NULL);
256 if (!strlist__has_entry(seen, path)) {
257 pr_err("Can't access file %s\n", path);
258 strlist__add(seen, path);
266 machine = machines__add(machines, pid, root_dir);
271 void machines__process_guests(struct machines *machines,
272 machine__process_t process, void *data)
276 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
277 struct machine *pos = rb_entry(nd, struct machine, rb_node);
282 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
284 if (machine__is_host(machine))
285 snprintf(bf, size, "[%s]", "kernel.kallsyms");
286 else if (machine__is_default_guest(machine))
287 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
289 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
296 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
298 struct rb_node *node;
299 struct machine *machine;
301 machines->host.id_hdr_size = id_hdr_size;
303 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
304 machine = rb_entry(node, struct machine, rb_node);
305 machine->id_hdr_size = id_hdr_size;
311 static void machine__update_thread_pid(struct machine *machine,
312 struct thread *th, pid_t pid)
314 struct thread *leader;
316 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
321 if (th->pid_ == th->tid)
324 leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
329 leader->mg = map_groups__new(machine);
334 if (th->mg == leader->mg)
339 * Maps are created from MMAP events which provide the pid and
340 * tid. Consequently there never should be any maps on a thread
341 * with an unknown pid. Just print an error if there are.
343 if (!map_groups__empty(th->mg))
344 pr_err("Discarding thread maps for %d:%d\n",
346 map_groups__put(th->mg);
349 th->mg = map_groups__get(leader->mg);
354 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
357 static struct thread *____machine__findnew_thread(struct machine *machine,
358 pid_t pid, pid_t tid,
361 struct rb_node **p = &machine->threads.rb_node;
362 struct rb_node *parent = NULL;
366 * Front-end cache - TID lookups come in blocks,
367 * so most of the time we dont have to look up
370 th = machine->last_match;
372 if (th->tid == tid) {
373 machine__update_thread_pid(machine, th, pid);
377 machine->last_match = NULL;
382 th = rb_entry(parent, struct thread, rb_node);
384 if (th->tid == tid) {
385 machine->last_match = th;
386 machine__update_thread_pid(machine, th, pid);
399 th = thread__new(pid, tid);
401 rb_link_node(&th->rb_node, parent, p);
402 rb_insert_color(&th->rb_node, &machine->threads);
405 * We have to initialize map_groups separately
406 * after rb tree is updated.
408 * The reason is that we call machine__findnew_thread
409 * within thread__init_map_groups to find the thread
410 * leader and that would screwed the rb tree.
412 if (thread__init_map_groups(th, machine)) {
413 rb_erase_init(&th->rb_node, &machine->threads);
414 RB_CLEAR_NODE(&th->rb_node);
419 * It is now in the rbtree, get a ref
422 machine->last_match = th;
428 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
430 return ____machine__findnew_thread(machine, pid, tid, true);
433 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
438 pthread_rwlock_wrlock(&machine->threads_lock);
439 th = thread__get(__machine__findnew_thread(machine, pid, tid));
440 pthread_rwlock_unlock(&machine->threads_lock);
444 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
448 pthread_rwlock_rdlock(&machine->threads_lock);
449 th = thread__get(____machine__findnew_thread(machine, pid, tid, false));
450 pthread_rwlock_unlock(&machine->threads_lock);
454 struct comm *machine__thread_exec_comm(struct machine *machine,
455 struct thread *thread)
457 if (machine->comm_exec)
458 return thread__exec_comm(thread);
460 return thread__comm(thread);
463 int machine__process_comm_event(struct machine *machine, union perf_event *event,
464 struct perf_sample *sample)
466 struct thread *thread = machine__findnew_thread(machine,
469 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
473 machine->comm_exec = true;
476 perf_event__fprintf_comm(event, stdout);
478 if (thread == NULL ||
479 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
480 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
489 int machine__process_lost_event(struct machine *machine __maybe_unused,
490 union perf_event *event, struct perf_sample *sample __maybe_unused)
492 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
493 event->lost.id, event->lost.lost);
497 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
498 union perf_event *event, struct perf_sample *sample)
500 dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
501 sample->id, event->lost_samples.lost);
505 static struct dso *machine__findnew_module_dso(struct machine *machine,
507 const char *filename)
511 pthread_rwlock_wrlock(&machine->dsos.lock);
513 dso = __dsos__find(&machine->dsos, m->name, true);
515 dso = __dsos__addnew(&machine->dsos, m->name);
519 if (machine__is_host(machine))
520 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
522 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
524 /* _KMODULE_COMP should be next to _KMODULE */
525 if (m->kmod && m->comp)
528 dso__set_short_name(dso, strdup(m->name), true);
529 dso__set_long_name(dso, strdup(filename), true);
534 pthread_rwlock_unlock(&machine->dsos.lock);
538 int machine__process_aux_event(struct machine *machine __maybe_unused,
539 union perf_event *event)
542 perf_event__fprintf_aux(event, stdout);
546 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
547 union perf_event *event)
550 perf_event__fprintf_itrace_start(event, stdout);
554 int machine__process_switch_event(struct machine *machine __maybe_unused,
555 union perf_event *event)
558 perf_event__fprintf_switch(event, stdout);
562 struct map *machine__findnew_module_map(struct machine *machine, u64 start,
563 const char *filename)
565 struct map *map = NULL;
569 if (kmod_path__parse_name(&m, filename))
572 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
577 dso = machine__findnew_module_dso(machine, &m, filename);
581 map = map__new2(start, dso, MAP__FUNCTION);
585 map_groups__insert(&machine->kmaps, map);
592 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
595 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
597 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
598 struct machine *pos = rb_entry(nd, struct machine, rb_node);
599 ret += __dsos__fprintf(&pos->dsos.head, fp);
605 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
606 bool (skip)(struct dso *dso, int parm), int parm)
608 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
611 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
612 bool (skip)(struct dso *dso, int parm), int parm)
615 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
617 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
618 struct machine *pos = rb_entry(nd, struct machine, rb_node);
619 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
624 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
628 struct dso *kdso = machine__kernel_map(machine)->dso;
630 if (kdso->has_build_id) {
631 char filename[PATH_MAX];
632 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
633 printed += fprintf(fp, "[0] %s\n", filename);
636 for (i = 0; i < vmlinux_path__nr_entries; ++i)
637 printed += fprintf(fp, "[%d] %s\n",
638 i + kdso->has_build_id, vmlinux_path[i]);
643 size_t machine__fprintf(struct machine *machine, FILE *fp)
648 pthread_rwlock_rdlock(&machine->threads_lock);
650 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
651 struct thread *pos = rb_entry(nd, struct thread, rb_node);
653 ret += thread__fprintf(pos, fp);
656 pthread_rwlock_unlock(&machine->threads_lock);
661 static struct dso *machine__get_kernel(struct machine *machine)
663 const char *vmlinux_name = NULL;
666 if (machine__is_host(machine)) {
667 vmlinux_name = symbol_conf.vmlinux_name;
669 vmlinux_name = "[kernel.kallsyms]";
671 kernel = machine__findnew_kernel(machine, vmlinux_name,
672 "[kernel]", DSO_TYPE_KERNEL);
676 if (machine__is_default_guest(machine))
677 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
679 vmlinux_name = machine__mmap_name(machine, bf,
682 kernel = machine__findnew_kernel(machine, vmlinux_name,
684 DSO_TYPE_GUEST_KERNEL);
687 if (kernel != NULL && (!kernel->has_build_id))
688 dso__read_running_kernel_build_id(kernel, machine);
693 struct process_args {
697 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
700 if (machine__is_default_guest(machine))
701 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
703 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
706 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
708 /* Figure out the start address of kernel map from /proc/kallsyms.
709 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
710 * symbol_name if it's not that important.
712 static u64 machine__get_running_kernel_start(struct machine *machine,
713 const char **symbol_name)
715 char filename[PATH_MAX];
720 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
722 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
725 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
726 addr = kallsyms__get_function_start(filename, name);
737 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
740 u64 start = machine__get_running_kernel_start(machine, NULL);
742 for (type = 0; type < MAP__NR_TYPES; ++type) {
746 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
747 if (machine->vmlinux_maps[type] == NULL)
750 machine->vmlinux_maps[type]->map_ip =
751 machine->vmlinux_maps[type]->unmap_ip =
753 map = __machine__kernel_map(machine, type);
754 kmap = map__kmap(map);
758 kmap->kmaps = &machine->kmaps;
759 map_groups__insert(&machine->kmaps, map);
765 void machine__destroy_kernel_maps(struct machine *machine)
769 for (type = 0; type < MAP__NR_TYPES; ++type) {
771 struct map *map = __machine__kernel_map(machine, type);
776 kmap = map__kmap(map);
777 map_groups__remove(&machine->kmaps, map);
778 if (kmap && kmap->ref_reloc_sym) {
780 * ref_reloc_sym is shared among all maps, so free just
783 if (type == MAP__FUNCTION) {
784 zfree((char **)&kmap->ref_reloc_sym->name);
785 zfree(&kmap->ref_reloc_sym);
787 kmap->ref_reloc_sym = NULL;
790 machine->vmlinux_maps[type] = NULL;
794 int machines__create_guest_kernel_maps(struct machines *machines)
797 struct dirent **namelist = NULL;
803 if (symbol_conf.default_guest_vmlinux_name ||
804 symbol_conf.default_guest_modules ||
805 symbol_conf.default_guest_kallsyms) {
806 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
809 if (symbol_conf.guestmount) {
810 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
813 for (i = 0; i < items; i++) {
814 if (!isdigit(namelist[i]->d_name[0])) {
815 /* Filter out . and .. */
818 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
819 if ((*endp != '\0') ||
820 (endp == namelist[i]->d_name) ||
822 pr_debug("invalid directory (%s). Skipping.\n",
823 namelist[i]->d_name);
826 sprintf(path, "%s/%s/proc/kallsyms",
827 symbol_conf.guestmount,
828 namelist[i]->d_name);
829 ret = access(path, R_OK);
831 pr_debug("Can't access file %s\n", path);
834 machines__create_kernel_maps(machines, pid);
843 void machines__destroy_kernel_maps(struct machines *machines)
845 struct rb_node *next = rb_first(&machines->guests);
847 machine__destroy_kernel_maps(&machines->host);
850 struct machine *pos = rb_entry(next, struct machine, rb_node);
852 next = rb_next(&pos->rb_node);
853 rb_erase(&pos->rb_node, &machines->guests);
854 machine__delete(pos);
858 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
860 struct machine *machine = machines__findnew(machines, pid);
865 return machine__create_kernel_maps(machine);
868 int machine__load_kallsyms(struct machine *machine, const char *filename,
869 enum map_type type, symbol_filter_t filter)
871 struct map *map = machine__kernel_map(machine);
872 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
875 dso__set_loaded(map->dso, type);
877 * Since /proc/kallsyms will have multiple sessions for the
878 * kernel, with modules between them, fixup the end of all
881 __map_groups__fixup_end(&machine->kmaps, type);
887 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
888 symbol_filter_t filter)
890 struct map *map = machine__kernel_map(machine);
891 int ret = dso__load_vmlinux_path(map->dso, map, filter);
894 dso__set_loaded(map->dso, type);
899 static void map_groups__fixup_end(struct map_groups *mg)
902 for (i = 0; i < MAP__NR_TYPES; ++i)
903 __map_groups__fixup_end(mg, i);
906 static char *get_kernel_version(const char *root_dir)
908 char version[PATH_MAX];
911 const char *prefix = "Linux version ";
913 sprintf(version, "%s/proc/version", root_dir);
914 file = fopen(version, "r");
919 tmp = fgets(version, sizeof(version), file);
922 name = strstr(version, prefix);
925 name += strlen(prefix);
926 tmp = strchr(name, ' ');
933 static bool is_kmod_dso(struct dso *dso)
935 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
936 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
939 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
945 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
949 long_name = strdup(path);
950 if (long_name == NULL)
953 dso__set_long_name(map->dso, long_name, true);
954 dso__kernel_module_get_build_id(map->dso, "");
957 * Full name could reveal us kmod compression, so
958 * we need to update the symtab_type if needed.
960 if (m->comp && is_kmod_dso(map->dso))
961 map->dso->symtab_type++;
966 static int map_groups__set_modules_path_dir(struct map_groups *mg,
967 const char *dir_name, int depth)
970 DIR *dir = opendir(dir_name);
974 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
978 while ((dent = readdir(dir)) != NULL) {
982 /*sshfs might return bad dent->d_type, so we have to stat*/
983 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
987 if (S_ISDIR(st.st_mode)) {
988 if (!strcmp(dent->d_name, ".") ||
989 !strcmp(dent->d_name, ".."))
992 /* Do not follow top-level source and build symlinks */
994 if (!strcmp(dent->d_name, "source") ||
995 !strcmp(dent->d_name, "build"))
999 ret = map_groups__set_modules_path_dir(mg, path,
1006 ret = kmod_path__parse_name(&m, dent->d_name);
1011 ret = map_groups__set_module_path(mg, path, &m);
1025 static int machine__set_modules_path(struct machine *machine)
1028 char modules_path[PATH_MAX];
1030 version = get_kernel_version(machine->root_dir);
1034 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1035 machine->root_dir, version);
1038 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1041 static int machine__create_module(void *arg, const char *name, u64 start)
1043 struct machine *machine = arg;
1046 map = machine__findnew_module_map(machine, start, name);
1050 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1055 static int machine__create_modules(struct machine *machine)
1057 const char *modules;
1058 char path[PATH_MAX];
1060 if (machine__is_default_guest(machine)) {
1061 modules = symbol_conf.default_guest_modules;
1063 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1067 if (symbol__restricted_filename(modules, "/proc/modules"))
1070 if (modules__parse(modules, machine, machine__create_module))
1073 if (!machine__set_modules_path(machine))
1076 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1081 int machine__create_kernel_maps(struct machine *machine)
1083 struct dso *kernel = machine__get_kernel(machine);
1085 u64 addr = machine__get_running_kernel_start(machine, &name);
1089 if (kernel == NULL ||
1090 __machine__create_kernel_maps(machine, kernel) < 0)
1093 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1094 if (machine__is_host(machine))
1095 pr_debug("Problems creating module maps, "
1096 "continuing anyway...\n");
1098 pr_debug("Problems creating module maps for guest %d, "
1099 "continuing anyway...\n", machine->pid);
1103 * Now that we have all the maps created, just set the ->end of them:
1105 map_groups__fixup_end(&machine->kmaps);
1107 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
1109 machine__destroy_kernel_maps(machine);
1116 static void machine__set_kernel_mmap_len(struct machine *machine,
1117 union perf_event *event)
1121 for (i = 0; i < MAP__NR_TYPES; i++) {
1122 machine->vmlinux_maps[i]->start = event->mmap.start;
1123 machine->vmlinux_maps[i]->end = (event->mmap.start +
1126 * Be a bit paranoid here, some perf.data file came with
1127 * a zero sized synthesized MMAP event for the kernel.
1129 if (machine->vmlinux_maps[i]->end == 0)
1130 machine->vmlinux_maps[i]->end = ~0ULL;
1134 static bool machine__uses_kcore(struct machine *machine)
1138 list_for_each_entry(dso, &machine->dsos.head, node) {
1139 if (dso__is_kcore(dso))
1146 static int machine__process_kernel_mmap_event(struct machine *machine,
1147 union perf_event *event)
1150 char kmmap_prefix[PATH_MAX];
1151 enum dso_kernel_type kernel_type;
1152 bool is_kernel_mmap;
1154 /* If we have maps from kcore then we do not need or want any others */
1155 if (machine__uses_kcore(machine))
1158 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1159 if (machine__is_host(machine))
1160 kernel_type = DSO_TYPE_KERNEL;
1162 kernel_type = DSO_TYPE_GUEST_KERNEL;
1164 is_kernel_mmap = memcmp(event->mmap.filename,
1166 strlen(kmmap_prefix) - 1) == 0;
1167 if (event->mmap.filename[0] == '/' ||
1168 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1169 map = machine__findnew_module_map(machine, event->mmap.start,
1170 event->mmap.filename);
1174 map->end = map->start + event->mmap.len;
1175 } else if (is_kernel_mmap) {
1176 const char *symbol_name = (event->mmap.filename +
1177 strlen(kmmap_prefix));
1179 * Should be there already, from the build-id table in
1182 struct dso *kernel = NULL;
1185 pthread_rwlock_rdlock(&machine->dsos.lock);
1187 list_for_each_entry(dso, &machine->dsos.head, node) {
1190 * The cpumode passed to is_kernel_module is not the
1191 * cpumode of *this* event. If we insist on passing
1192 * correct cpumode to is_kernel_module, we should
1193 * record the cpumode when we adding this dso to the
1196 * However we don't really need passing correct
1197 * cpumode. We know the correct cpumode must be kernel
1198 * mode (if not, we should not link it onto kernel_dsos
1201 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1202 * is_kernel_module() treats it as a kernel cpumode.
1206 is_kernel_module(dso->long_name,
1207 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1215 pthread_rwlock_unlock(&machine->dsos.lock);
1218 kernel = machine__findnew_dso(machine, kmmap_prefix);
1222 kernel->kernel = kernel_type;
1223 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1228 if (strstr(kernel->long_name, "vmlinux"))
1229 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1231 machine__set_kernel_mmap_len(machine, event);
1234 * Avoid using a zero address (kptr_restrict) for the ref reloc
1235 * symbol. Effectively having zero here means that at record
1236 * time /proc/sys/kernel/kptr_restrict was non zero.
1238 if (event->mmap.pgoff != 0) {
1239 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1244 if (machine__is_default_guest(machine)) {
1246 * preload dso of guest kernel and modules
1248 dso__load(kernel, machine__kernel_map(machine), NULL);
1256 int machine__process_mmap2_event(struct machine *machine,
1257 union perf_event *event,
1258 struct perf_sample *sample __maybe_unused)
1260 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1261 struct thread *thread;
1267 perf_event__fprintf_mmap2(event, stdout);
1269 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1270 cpumode == PERF_RECORD_MISC_KERNEL) {
1271 ret = machine__process_kernel_mmap_event(machine, event);
1277 thread = machine__findnew_thread(machine, event->mmap2.pid,
1282 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1283 type = MAP__VARIABLE;
1285 type = MAP__FUNCTION;
1287 map = map__new(machine, event->mmap2.start,
1288 event->mmap2.len, event->mmap2.pgoff,
1289 event->mmap2.pid, event->mmap2.maj,
1290 event->mmap2.min, event->mmap2.ino,
1291 event->mmap2.ino_generation,
1294 event->mmap2.filename, type, thread);
1297 goto out_problem_map;
1299 thread__insert_map(thread, map);
1300 thread__put(thread);
1305 thread__put(thread);
1307 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1311 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1312 struct perf_sample *sample __maybe_unused)
1314 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1315 struct thread *thread;
1321 perf_event__fprintf_mmap(event, stdout);
1323 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1324 cpumode == PERF_RECORD_MISC_KERNEL) {
1325 ret = machine__process_kernel_mmap_event(machine, event);
1331 thread = machine__findnew_thread(machine, event->mmap.pid,
1336 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1337 type = MAP__VARIABLE;
1339 type = MAP__FUNCTION;
1341 map = map__new(machine, event->mmap.start,
1342 event->mmap.len, event->mmap.pgoff,
1343 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1344 event->mmap.filename,
1348 goto out_problem_map;
1350 thread__insert_map(thread, map);
1351 thread__put(thread);
1356 thread__put(thread);
1358 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1362 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1364 if (machine->last_match == th)
1365 machine->last_match = NULL;
1367 BUG_ON(atomic_read(&th->refcnt) == 0);
1369 pthread_rwlock_wrlock(&machine->threads_lock);
1370 rb_erase_init(&th->rb_node, &machine->threads);
1371 RB_CLEAR_NODE(&th->rb_node);
1373 * Move it first to the dead_threads list, then drop the reference,
1374 * if this is the last reference, then the thread__delete destructor
1375 * will be called and we will remove it from the dead_threads list.
1377 list_add_tail(&th->node, &machine->dead_threads);
1379 pthread_rwlock_unlock(&machine->threads_lock);
1383 void machine__remove_thread(struct machine *machine, struct thread *th)
1385 return __machine__remove_thread(machine, th, true);
1388 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1389 struct perf_sample *sample)
1391 struct thread *thread = machine__find_thread(machine,
1394 struct thread *parent = machine__findnew_thread(machine,
1400 perf_event__fprintf_task(event, stdout);
1403 * There may be an existing thread that is not actually the parent,
1404 * either because we are processing events out of order, or because the
1405 * (fork) event that would have removed the thread was lost. Assume the
1406 * latter case and continue on as best we can.
1408 if (parent->pid_ != (pid_t)event->fork.ppid) {
1409 dump_printf("removing erroneous parent thread %d/%d\n",
1410 parent->pid_, parent->tid);
1411 machine__remove_thread(machine, parent);
1412 thread__put(parent);
1413 parent = machine__findnew_thread(machine, event->fork.ppid,
1417 /* if a thread currently exists for the thread id remove it */
1418 if (thread != NULL) {
1419 machine__remove_thread(machine, thread);
1420 thread__put(thread);
1423 thread = machine__findnew_thread(machine, event->fork.pid,
1426 if (thread == NULL || parent == NULL ||
1427 thread__fork(thread, parent, sample->time) < 0) {
1428 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1431 thread__put(thread);
1432 thread__put(parent);
1437 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1438 struct perf_sample *sample __maybe_unused)
1440 struct thread *thread = machine__find_thread(machine,
1445 perf_event__fprintf_task(event, stdout);
1447 if (thread != NULL) {
1448 thread__exited(thread);
1449 thread__put(thread);
1455 int machine__process_event(struct machine *machine, union perf_event *event,
1456 struct perf_sample *sample)
1460 switch (event->header.type) {
1461 case PERF_RECORD_COMM:
1462 ret = machine__process_comm_event(machine, event, sample); break;
1463 case PERF_RECORD_MMAP:
1464 ret = machine__process_mmap_event(machine, event, sample); break;
1465 case PERF_RECORD_MMAP2:
1466 ret = machine__process_mmap2_event(machine, event, sample); break;
1467 case PERF_RECORD_FORK:
1468 ret = machine__process_fork_event(machine, event, sample); break;
1469 case PERF_RECORD_EXIT:
1470 ret = machine__process_exit_event(machine, event, sample); break;
1471 case PERF_RECORD_LOST:
1472 ret = machine__process_lost_event(machine, event, sample); break;
1473 case PERF_RECORD_AUX:
1474 ret = machine__process_aux_event(machine, event); break;
1475 case PERF_RECORD_ITRACE_START:
1476 ret = machine__process_itrace_start_event(machine, event); break;
1477 case PERF_RECORD_LOST_SAMPLES:
1478 ret = machine__process_lost_samples_event(machine, event, sample); break;
1479 case PERF_RECORD_SWITCH:
1480 case PERF_RECORD_SWITCH_CPU_WIDE:
1481 ret = machine__process_switch_event(machine, event); break;
1490 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1492 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1497 static void ip__resolve_ams(struct thread *thread,
1498 struct addr_map_symbol *ams,
1501 struct addr_location al;
1503 memset(&al, 0, sizeof(al));
1505 * We cannot use the header.misc hint to determine whether a
1506 * branch stack address is user, kernel, guest, hypervisor.
1507 * Branches may straddle the kernel/user/hypervisor boundaries.
1508 * Thus, we have to try consecutively until we find a match
1509 * or else, the symbol is unknown
1511 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1514 ams->al_addr = al.addr;
1519 static void ip__resolve_data(struct thread *thread,
1520 u8 m, struct addr_map_symbol *ams, u64 addr)
1522 struct addr_location al;
1524 memset(&al, 0, sizeof(al));
1526 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1527 if (al.map == NULL) {
1529 * some shared data regions have execute bit set which puts
1530 * their mapping in the MAP__FUNCTION type array.
1531 * Check there as a fallback option before dropping the sample.
1533 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1537 ams->al_addr = al.addr;
1542 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1543 struct addr_location *al)
1545 struct mem_info *mi = zalloc(sizeof(*mi));
1550 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1551 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1552 mi->data_src.val = sample->data_src;
1557 static int add_callchain_ip(struct thread *thread,
1558 struct symbol **parent,
1559 struct addr_location *root_al,
1563 struct addr_location al;
1568 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1571 if (ip >= PERF_CONTEXT_MAX) {
1573 case PERF_CONTEXT_HV:
1574 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1576 case PERF_CONTEXT_KERNEL:
1577 *cpumode = PERF_RECORD_MISC_KERNEL;
1579 case PERF_CONTEXT_USER:
1580 *cpumode = PERF_RECORD_MISC_USER;
1583 pr_debug("invalid callchain context: "
1584 "%"PRId64"\n", (s64) ip);
1586 * It seems the callchain is corrupted.
1589 callchain_cursor_reset(&callchain_cursor);
1594 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1598 if (al.sym != NULL) {
1599 if (sort__has_parent && !*parent &&
1600 symbol__match_regex(al.sym, &parent_regex))
1602 else if (have_ignore_callees && root_al &&
1603 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1604 /* Treat this symbol as the root,
1605 forgetting its callees. */
1607 callchain_cursor_reset(&callchain_cursor);
1611 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1614 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1615 struct addr_location *al)
1618 const struct branch_stack *bs = sample->branch_stack;
1619 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1624 for (i = 0; i < bs->nr; i++) {
1625 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1626 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1627 bi[i].flags = bs->entries[i].flags;
1634 #define NO_ENTRY 0xff
1636 #define PERF_MAX_BRANCH_DEPTH 127
1639 static int remove_loops(struct branch_entry *l, int nr)
1642 unsigned char chash[CHASHSZ];
1644 memset(chash, NO_ENTRY, sizeof(chash));
1646 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1648 for (i = 0; i < nr; i++) {
1649 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1651 /* no collision handling for now */
1652 if (chash[h] == NO_ENTRY) {
1654 } else if (l[chash[h]].from == l[i].from) {
1655 bool is_loop = true;
1656 /* check if it is a real loop */
1658 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1659 if (l[j].from != l[i + off].from) {
1664 memmove(l + i, l + i + off,
1665 (nr - (i + off)) * sizeof(*l));
1674 * Recolve LBR callstack chain sample
1676 * 1 on success get LBR callchain information
1677 * 0 no available LBR callchain information, should try fp
1678 * negative error code on other errors.
1680 static int resolve_lbr_callchain_sample(struct thread *thread,
1681 struct perf_sample *sample,
1682 struct symbol **parent,
1683 struct addr_location *root_al,
1686 struct ip_callchain *chain = sample->callchain;
1687 int chain_nr = min(max_stack, (int)chain->nr);
1688 u8 cpumode = PERF_RECORD_MISC_USER;
1692 for (i = 0; i < chain_nr; i++) {
1693 if (chain->ips[i] == PERF_CONTEXT_USER)
1697 /* LBR only affects the user callchain */
1698 if (i != chain_nr) {
1699 struct branch_stack *lbr_stack = sample->branch_stack;
1700 int lbr_nr = lbr_stack->nr;
1702 * LBR callstack can only get user call chain.
1703 * The mix_chain_nr is kernel call chain
1704 * number plus LBR user call chain number.
1705 * i is kernel call chain number,
1706 * 1 is PERF_CONTEXT_USER,
1707 * lbr_nr + 1 is the user call chain number.
1708 * For details, please refer to the comments
1709 * in callchain__printf
1711 int mix_chain_nr = i + 1 + lbr_nr + 1;
1713 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1714 pr_warning("corrupted callchain. skipping...\n");
1718 for (j = 0; j < mix_chain_nr; j++) {
1719 if (callchain_param.order == ORDER_CALLEE) {
1723 ip = lbr_stack->entries[j - i - 2].from;
1725 ip = lbr_stack->entries[0].to;
1728 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1729 else if (j > lbr_nr)
1730 ip = chain->ips[i + 1 - (j - lbr_nr)];
1732 ip = lbr_stack->entries[0].to;
1735 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1737 return (err < 0) ? err : 0;
1745 static int thread__resolve_callchain_sample(struct thread *thread,
1746 struct perf_evsel *evsel,
1747 struct perf_sample *sample,
1748 struct symbol **parent,
1749 struct addr_location *root_al,
1752 struct branch_stack *branch = sample->branch_stack;
1753 struct ip_callchain *chain = sample->callchain;
1754 int chain_nr = min(max_stack, (int)chain->nr);
1755 u8 cpumode = PERF_RECORD_MISC_USER;
1760 callchain_cursor_reset(&callchain_cursor);
1762 if (has_branch_callstack(evsel)) {
1763 err = resolve_lbr_callchain_sample(thread, sample, parent,
1764 root_al, max_stack);
1766 return (err < 0) ? err : 0;
1770 * Based on DWARF debug information, some architectures skip
1771 * a callchain entry saved by the kernel.
1773 if (chain->nr < PERF_MAX_STACK_DEPTH)
1774 skip_idx = arch_skip_callchain_idx(thread, chain);
1777 * Add branches to call stack for easier browsing. This gives
1778 * more context for a sample than just the callers.
1780 * This uses individual histograms of paths compared to the
1781 * aggregated histograms the normal LBR mode uses.
1783 * Limitations for now:
1784 * - No extra filters
1785 * - No annotations (should annotate somehow)
1788 if (branch && callchain_param.branch_callstack) {
1789 int nr = min(max_stack, (int)branch->nr);
1790 struct branch_entry be[nr];
1792 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1793 pr_warning("corrupted branch chain. skipping...\n");
1797 for (i = 0; i < nr; i++) {
1798 if (callchain_param.order == ORDER_CALLEE) {
1799 be[i] = branch->entries[i];
1801 * Check for overlap into the callchain.
1802 * The return address is one off compared to
1803 * the branch entry. To adjust for this
1804 * assume the calling instruction is not longer
1807 if (i == skip_idx ||
1808 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1810 else if (be[i].from < chain->ips[first_call] &&
1811 be[i].from >= chain->ips[first_call] - 8)
1814 be[i] = branch->entries[branch->nr - i - 1];
1817 nr = remove_loops(be, nr);
1819 for (i = 0; i < nr; i++) {
1820 err = add_callchain_ip(thread, parent, root_al,
1823 err = add_callchain_ip(thread, parent, root_al,
1834 if (chain->nr > PERF_MAX_STACK_DEPTH && (int)chain->nr > max_stack) {
1835 pr_warning("corrupted callchain. skipping...\n");
1839 for (i = first_call; i < chain_nr; i++) {
1842 if (callchain_param.order == ORDER_CALLEE)
1845 j = chain->nr - i - 1;
1847 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1853 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1856 return (err < 0) ? err : 0;
1862 static int unwind_entry(struct unwind_entry *entry, void *arg)
1864 struct callchain_cursor *cursor = arg;
1865 return callchain_cursor_append(cursor, entry->ip,
1866 entry->map, entry->sym);
1869 int thread__resolve_callchain(struct thread *thread,
1870 struct perf_evsel *evsel,
1871 struct perf_sample *sample,
1872 struct symbol **parent,
1873 struct addr_location *root_al,
1876 int ret = thread__resolve_callchain_sample(thread, evsel,
1878 root_al, max_stack);
1882 /* Can we do dwarf post unwind? */
1883 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1884 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1887 /* Bail out if nothing was captured. */
1888 if ((!sample->user_regs.regs) ||
1889 (!sample->user_stack.size))
1892 return unwind__get_entries(unwind_entry, &callchain_cursor,
1893 thread, sample, max_stack);
1897 int machine__for_each_thread(struct machine *machine,
1898 int (*fn)(struct thread *thread, void *p),
1902 struct thread *thread;
1905 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1906 thread = rb_entry(nd, struct thread, rb_node);
1907 rc = fn(thread, priv);
1912 list_for_each_entry(thread, &machine->dead_threads, node) {
1913 rc = fn(thread, priv);
1920 int machines__for_each_thread(struct machines *machines,
1921 int (*fn)(struct thread *thread, void *p),
1927 rc = machine__for_each_thread(&machines->host, fn, priv);
1931 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
1932 struct machine *machine = rb_entry(nd, struct machine, rb_node);
1934 rc = machine__for_each_thread(machine, fn, priv);
1941 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1942 struct target *target, struct thread_map *threads,
1943 perf_event__handler_t process, bool data_mmap,
1944 unsigned int proc_map_timeout)
1946 if (target__has_task(target))
1947 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
1948 else if (target__has_cpu(target))
1949 return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
1950 /* command specified */
1954 pid_t machine__get_current_tid(struct machine *machine, int cpu)
1956 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
1959 return machine->current_tid[cpu];
1962 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
1965 struct thread *thread;
1970 if (!machine->current_tid) {
1973 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
1974 if (!machine->current_tid)
1976 for (i = 0; i < MAX_NR_CPUS; i++)
1977 machine->current_tid[i] = -1;
1980 if (cpu >= MAX_NR_CPUS) {
1981 pr_err("Requested CPU %d too large. ", cpu);
1982 pr_err("Consider raising MAX_NR_CPUS\n");
1986 machine->current_tid[cpu] = tid;
1988 thread = machine__findnew_thread(machine, pid, tid);
1993 thread__put(thread);
1998 int machine__get_kernel_start(struct machine *machine)
2000 struct map *map = machine__kernel_map(machine);
2004 * The only addresses above 2^63 are kernel addresses of a 64-bit
2005 * kernel. Note that addresses are unsigned so that on a 32-bit system
2006 * all addresses including kernel addresses are less than 2^32. In
2007 * that case (32-bit system), if the kernel mapping is unknown, all
2008 * addresses will be assumed to be in user space - see
2009 * machine__kernel_ip().
2011 machine->kernel_start = 1ULL << 63;
2013 err = map__load(map, machine->symbol_filter);
2015 machine->kernel_start = map->start;
2020 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2022 return dsos__findnew(&machine->dsos, filename);
2025 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2027 struct machine *machine = vmachine;
2029 struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map, NULL);
2034 *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2035 *addrp = map->unmap_ip(map, sym->start);