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);
95 list_del_init(&pos->node);
99 pthread_rwlock_unlock(&dsos->lock);
102 static void dsos__exit(struct dsos *dsos)
105 pthread_rwlock_destroy(&dsos->lock);
108 void machine__delete_threads(struct machine *machine)
112 pthread_rwlock_wrlock(&machine->threads_lock);
113 nd = rb_first(&machine->threads);
115 struct thread *t = rb_entry(nd, struct thread, rb_node);
118 __machine__remove_thread(machine, t, false);
120 pthread_rwlock_unlock(&machine->threads_lock);
123 void machine__exit(struct machine *machine)
125 map_groups__exit(&machine->kmaps);
126 dsos__exit(&machine->dsos);
127 machine__exit_vdso(machine);
128 zfree(&machine->root_dir);
129 zfree(&machine->current_tid);
130 pthread_rwlock_destroy(&machine->threads_lock);
133 void machine__delete(struct machine *machine)
135 machine__exit(machine);
139 void machines__init(struct machines *machines)
141 machine__init(&machines->host, "", HOST_KERNEL_ID);
142 machines->guests = RB_ROOT;
143 machines->symbol_filter = NULL;
146 void machines__exit(struct machines *machines)
148 machine__exit(&machines->host);
152 struct machine *machines__add(struct machines *machines, pid_t pid,
153 const char *root_dir)
155 struct rb_node **p = &machines->guests.rb_node;
156 struct rb_node *parent = NULL;
157 struct machine *pos, *machine = malloc(sizeof(*machine));
162 if (machine__init(machine, root_dir, pid) != 0) {
167 machine->symbol_filter = machines->symbol_filter;
171 pos = rb_entry(parent, struct machine, rb_node);
178 rb_link_node(&machine->rb_node, parent, p);
179 rb_insert_color(&machine->rb_node, &machines->guests);
184 void machines__set_symbol_filter(struct machines *machines,
185 symbol_filter_t symbol_filter)
189 machines->symbol_filter = symbol_filter;
190 machines->host.symbol_filter = symbol_filter;
192 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
193 struct machine *machine = rb_entry(nd, struct machine, rb_node);
195 machine->symbol_filter = symbol_filter;
199 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
203 machines->host.comm_exec = comm_exec;
205 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
206 struct machine *machine = rb_entry(nd, struct machine, rb_node);
208 machine->comm_exec = comm_exec;
212 struct machine *machines__find(struct machines *machines, pid_t pid)
214 struct rb_node **p = &machines->guests.rb_node;
215 struct rb_node *parent = NULL;
216 struct machine *machine;
217 struct machine *default_machine = NULL;
219 if (pid == HOST_KERNEL_ID)
220 return &machines->host;
224 machine = rb_entry(parent, struct machine, rb_node);
225 if (pid < machine->pid)
227 else if (pid > machine->pid)
232 default_machine = machine;
235 return default_machine;
238 struct machine *machines__findnew(struct machines *machines, pid_t pid)
241 const char *root_dir = "";
242 struct machine *machine = machines__find(machines, pid);
244 if (machine && (machine->pid == pid))
247 if ((pid != HOST_KERNEL_ID) &&
248 (pid != DEFAULT_GUEST_KERNEL_ID) &&
249 (symbol_conf.guestmount)) {
250 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
251 if (access(path, R_OK)) {
252 static struct strlist *seen;
255 seen = strlist__new(NULL, NULL);
257 if (!strlist__has_entry(seen, path)) {
258 pr_err("Can't access file %s\n", path);
259 strlist__add(seen, path);
267 machine = machines__add(machines, pid, root_dir);
272 void machines__process_guests(struct machines *machines,
273 machine__process_t process, void *data)
277 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
278 struct machine *pos = rb_entry(nd, struct machine, rb_node);
283 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
285 if (machine__is_host(machine))
286 snprintf(bf, size, "[%s]", "kernel.kallsyms");
287 else if (machine__is_default_guest(machine))
288 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
290 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
297 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
299 struct rb_node *node;
300 struct machine *machine;
302 machines->host.id_hdr_size = id_hdr_size;
304 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
305 machine = rb_entry(node, struct machine, rb_node);
306 machine->id_hdr_size = id_hdr_size;
312 static void machine__update_thread_pid(struct machine *machine,
313 struct thread *th, pid_t pid)
315 struct thread *leader;
317 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
322 if (th->pid_ == th->tid)
325 leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
330 leader->mg = map_groups__new(machine);
335 if (th->mg == leader->mg)
340 * Maps are created from MMAP events which provide the pid and
341 * tid. Consequently there never should be any maps on a thread
342 * with an unknown pid. Just print an error if there are.
344 if (!map_groups__empty(th->mg))
345 pr_err("Discarding thread maps for %d:%d\n",
347 map_groups__put(th->mg);
350 th->mg = map_groups__get(leader->mg);
355 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
358 static struct thread *____machine__findnew_thread(struct machine *machine,
359 pid_t pid, pid_t tid,
362 struct rb_node **p = &machine->threads.rb_node;
363 struct rb_node *parent = NULL;
367 * Front-end cache - TID lookups come in blocks,
368 * so most of the time we dont have to look up
371 th = machine->last_match;
373 if (th->tid == tid) {
374 machine__update_thread_pid(machine, th, pid);
378 machine->last_match = NULL;
383 th = rb_entry(parent, struct thread, rb_node);
385 if (th->tid == tid) {
386 machine->last_match = th;
387 machine__update_thread_pid(machine, th, pid);
400 th = thread__new(pid, tid);
402 rb_link_node(&th->rb_node, parent, p);
403 rb_insert_color(&th->rb_node, &machine->threads);
406 * We have to initialize map_groups separately
407 * after rb tree is updated.
409 * The reason is that we call machine__findnew_thread
410 * within thread__init_map_groups to find the thread
411 * leader and that would screwed the rb tree.
413 if (thread__init_map_groups(th, machine)) {
414 rb_erase_init(&th->rb_node, &machine->threads);
415 RB_CLEAR_NODE(&th->rb_node);
420 * It is now in the rbtree, get a ref
423 machine->last_match = th;
429 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
431 return ____machine__findnew_thread(machine, pid, tid, true);
434 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
439 pthread_rwlock_wrlock(&machine->threads_lock);
440 th = thread__get(__machine__findnew_thread(machine, pid, tid));
441 pthread_rwlock_unlock(&machine->threads_lock);
445 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
449 pthread_rwlock_rdlock(&machine->threads_lock);
450 th = thread__get(____machine__findnew_thread(machine, pid, tid, false));
451 pthread_rwlock_unlock(&machine->threads_lock);
455 struct comm *machine__thread_exec_comm(struct machine *machine,
456 struct thread *thread)
458 if (machine->comm_exec)
459 return thread__exec_comm(thread);
461 return thread__comm(thread);
464 int machine__process_comm_event(struct machine *machine, union perf_event *event,
465 struct perf_sample *sample)
467 struct thread *thread = machine__findnew_thread(machine,
470 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
474 machine->comm_exec = true;
477 perf_event__fprintf_comm(event, stdout);
479 if (thread == NULL ||
480 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
481 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
490 int machine__process_lost_event(struct machine *machine __maybe_unused,
491 union perf_event *event, struct perf_sample *sample __maybe_unused)
493 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
494 event->lost.id, event->lost.lost);
498 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
499 union perf_event *event, struct perf_sample *sample)
501 dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
502 sample->id, event->lost_samples.lost);
506 static struct dso *machine__findnew_module_dso(struct machine *machine,
508 const char *filename)
512 pthread_rwlock_wrlock(&machine->dsos.lock);
514 dso = __dsos__find(&machine->dsos, m->name, true);
516 dso = __dsos__addnew(&machine->dsos, m->name);
520 if (machine__is_host(machine))
521 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
523 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
525 /* _KMODULE_COMP should be next to _KMODULE */
526 if (m->kmod && m->comp)
529 dso__set_short_name(dso, strdup(m->name), true);
530 dso__set_long_name(dso, strdup(filename), true);
535 pthread_rwlock_unlock(&machine->dsos.lock);
539 int machine__process_aux_event(struct machine *machine __maybe_unused,
540 union perf_event *event)
543 perf_event__fprintf_aux(event, stdout);
547 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
548 union perf_event *event)
551 perf_event__fprintf_itrace_start(event, stdout);
555 int machine__process_switch_event(struct machine *machine __maybe_unused,
556 union perf_event *event)
559 perf_event__fprintf_switch(event, stdout);
563 struct map *machine__findnew_module_map(struct machine *machine, u64 start,
564 const char *filename)
566 struct map *map = NULL;
570 if (kmod_path__parse_name(&m, filename))
573 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
578 dso = machine__findnew_module_dso(machine, &m, filename);
582 map = map__new2(start, dso, MAP__FUNCTION);
586 map_groups__insert(&machine->kmaps, map);
593 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
596 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
598 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
599 struct machine *pos = rb_entry(nd, struct machine, rb_node);
600 ret += __dsos__fprintf(&pos->dsos.head, fp);
606 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
607 bool (skip)(struct dso *dso, int parm), int parm)
609 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
612 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
613 bool (skip)(struct dso *dso, int parm), int parm)
616 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
618 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
619 struct machine *pos = rb_entry(nd, struct machine, rb_node);
620 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
625 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
629 struct dso *kdso = machine__kernel_map(machine)->dso;
631 if (kdso->has_build_id) {
632 char filename[PATH_MAX];
633 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
634 printed += fprintf(fp, "[0] %s\n", filename);
637 for (i = 0; i < vmlinux_path__nr_entries; ++i)
638 printed += fprintf(fp, "[%d] %s\n",
639 i + kdso->has_build_id, vmlinux_path[i]);
644 size_t machine__fprintf(struct machine *machine, FILE *fp)
649 pthread_rwlock_rdlock(&machine->threads_lock);
651 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
652 struct thread *pos = rb_entry(nd, struct thread, rb_node);
654 ret += thread__fprintf(pos, fp);
657 pthread_rwlock_unlock(&machine->threads_lock);
662 static struct dso *machine__get_kernel(struct machine *machine)
664 const char *vmlinux_name = NULL;
667 if (machine__is_host(machine)) {
668 vmlinux_name = symbol_conf.vmlinux_name;
670 vmlinux_name = "[kernel.kallsyms]";
672 kernel = machine__findnew_kernel(machine, vmlinux_name,
673 "[kernel]", DSO_TYPE_KERNEL);
677 if (machine__is_default_guest(machine))
678 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
680 vmlinux_name = machine__mmap_name(machine, bf,
683 kernel = machine__findnew_kernel(machine, vmlinux_name,
685 DSO_TYPE_GUEST_KERNEL);
688 if (kernel != NULL && (!kernel->has_build_id))
689 dso__read_running_kernel_build_id(kernel, machine);
694 struct process_args {
698 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
701 if (machine__is_default_guest(machine))
702 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
704 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
707 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
709 /* Figure out the start address of kernel map from /proc/kallsyms.
710 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
711 * symbol_name if it's not that important.
713 static u64 machine__get_running_kernel_start(struct machine *machine,
714 const char **symbol_name)
716 char filename[PATH_MAX];
721 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
723 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
726 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
727 addr = kallsyms__get_function_start(filename, name);
738 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
741 u64 start = machine__get_running_kernel_start(machine, NULL);
743 for (type = 0; type < MAP__NR_TYPES; ++type) {
747 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
748 if (machine->vmlinux_maps[type] == NULL)
751 machine->vmlinux_maps[type]->map_ip =
752 machine->vmlinux_maps[type]->unmap_ip =
754 map = __machine__kernel_map(machine, type);
755 kmap = map__kmap(map);
759 kmap->kmaps = &machine->kmaps;
760 map_groups__insert(&machine->kmaps, map);
766 void machine__destroy_kernel_maps(struct machine *machine)
770 for (type = 0; type < MAP__NR_TYPES; ++type) {
772 struct map *map = __machine__kernel_map(machine, type);
777 kmap = map__kmap(map);
778 map_groups__remove(&machine->kmaps, map);
779 if (kmap && kmap->ref_reloc_sym) {
781 * ref_reloc_sym is shared among all maps, so free just
784 if (type == MAP__FUNCTION) {
785 zfree((char **)&kmap->ref_reloc_sym->name);
786 zfree(&kmap->ref_reloc_sym);
788 kmap->ref_reloc_sym = NULL;
791 machine->vmlinux_maps[type] = NULL;
795 int machines__create_guest_kernel_maps(struct machines *machines)
798 struct dirent **namelist = NULL;
804 if (symbol_conf.default_guest_vmlinux_name ||
805 symbol_conf.default_guest_modules ||
806 symbol_conf.default_guest_kallsyms) {
807 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
810 if (symbol_conf.guestmount) {
811 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
814 for (i = 0; i < items; i++) {
815 if (!isdigit(namelist[i]->d_name[0])) {
816 /* Filter out . and .. */
819 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
820 if ((*endp != '\0') ||
821 (endp == namelist[i]->d_name) ||
823 pr_debug("invalid directory (%s). Skipping.\n",
824 namelist[i]->d_name);
827 sprintf(path, "%s/%s/proc/kallsyms",
828 symbol_conf.guestmount,
829 namelist[i]->d_name);
830 ret = access(path, R_OK);
832 pr_debug("Can't access file %s\n", path);
835 machines__create_kernel_maps(machines, pid);
844 void machines__destroy_kernel_maps(struct machines *machines)
846 struct rb_node *next = rb_first(&machines->guests);
848 machine__destroy_kernel_maps(&machines->host);
851 struct machine *pos = rb_entry(next, struct machine, rb_node);
853 next = rb_next(&pos->rb_node);
854 rb_erase(&pos->rb_node, &machines->guests);
855 machine__delete(pos);
859 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
861 struct machine *machine = machines__findnew(machines, pid);
866 return machine__create_kernel_maps(machine);
869 int machine__load_kallsyms(struct machine *machine, const char *filename,
870 enum map_type type, symbol_filter_t filter)
872 struct map *map = machine__kernel_map(machine);
873 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
876 dso__set_loaded(map->dso, type);
878 * Since /proc/kallsyms will have multiple sessions for the
879 * kernel, with modules between them, fixup the end of all
882 __map_groups__fixup_end(&machine->kmaps, type);
888 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
889 symbol_filter_t filter)
891 struct map *map = machine__kernel_map(machine);
892 int ret = dso__load_vmlinux_path(map->dso, map, filter);
895 dso__set_loaded(map->dso, type);
900 static void map_groups__fixup_end(struct map_groups *mg)
903 for (i = 0; i < MAP__NR_TYPES; ++i)
904 __map_groups__fixup_end(mg, i);
907 static char *get_kernel_version(const char *root_dir)
909 char version[PATH_MAX];
912 const char *prefix = "Linux version ";
914 sprintf(version, "%s/proc/version", root_dir);
915 file = fopen(version, "r");
920 tmp = fgets(version, sizeof(version), file);
923 name = strstr(version, prefix);
926 name += strlen(prefix);
927 tmp = strchr(name, ' ');
934 static bool is_kmod_dso(struct dso *dso)
936 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
937 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
940 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
946 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
950 long_name = strdup(path);
951 if (long_name == NULL)
954 dso__set_long_name(map->dso, long_name, true);
955 dso__kernel_module_get_build_id(map->dso, "");
958 * Full name could reveal us kmod compression, so
959 * we need to update the symtab_type if needed.
961 if (m->comp && is_kmod_dso(map->dso))
962 map->dso->symtab_type++;
967 static int map_groups__set_modules_path_dir(struct map_groups *mg,
968 const char *dir_name, int depth)
971 DIR *dir = opendir(dir_name);
975 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
979 while ((dent = readdir(dir)) != NULL) {
983 /*sshfs might return bad dent->d_type, so we have to stat*/
984 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
988 if (S_ISDIR(st.st_mode)) {
989 if (!strcmp(dent->d_name, ".") ||
990 !strcmp(dent->d_name, ".."))
993 /* Do not follow top-level source and build symlinks */
995 if (!strcmp(dent->d_name, "source") ||
996 !strcmp(dent->d_name, "build"))
1000 ret = map_groups__set_modules_path_dir(mg, path,
1007 ret = kmod_path__parse_name(&m, dent->d_name);
1012 ret = map_groups__set_module_path(mg, path, &m);
1026 static int machine__set_modules_path(struct machine *machine)
1029 char modules_path[PATH_MAX];
1031 version = get_kernel_version(machine->root_dir);
1035 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1036 machine->root_dir, version);
1039 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1042 static int machine__create_module(void *arg, const char *name, u64 start)
1044 struct machine *machine = arg;
1047 map = machine__findnew_module_map(machine, start, name);
1051 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1056 static int machine__create_modules(struct machine *machine)
1058 const char *modules;
1059 char path[PATH_MAX];
1061 if (machine__is_default_guest(machine)) {
1062 modules = symbol_conf.default_guest_modules;
1064 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1068 if (symbol__restricted_filename(modules, "/proc/modules"))
1071 if (modules__parse(modules, machine, machine__create_module))
1074 if (!machine__set_modules_path(machine))
1077 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1082 int machine__create_kernel_maps(struct machine *machine)
1084 struct dso *kernel = machine__get_kernel(machine);
1086 u64 addr = machine__get_running_kernel_start(machine, &name);
1090 if (kernel == NULL ||
1091 __machine__create_kernel_maps(machine, kernel) < 0)
1094 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1095 if (machine__is_host(machine))
1096 pr_debug("Problems creating module maps, "
1097 "continuing anyway...\n");
1099 pr_debug("Problems creating module maps for guest %d, "
1100 "continuing anyway...\n", machine->pid);
1104 * Now that we have all the maps created, just set the ->end of them:
1106 map_groups__fixup_end(&machine->kmaps);
1108 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
1110 machine__destroy_kernel_maps(machine);
1117 static void machine__set_kernel_mmap_len(struct machine *machine,
1118 union perf_event *event)
1122 for (i = 0; i < MAP__NR_TYPES; i++) {
1123 machine->vmlinux_maps[i]->start = event->mmap.start;
1124 machine->vmlinux_maps[i]->end = (event->mmap.start +
1127 * Be a bit paranoid here, some perf.data file came with
1128 * a zero sized synthesized MMAP event for the kernel.
1130 if (machine->vmlinux_maps[i]->end == 0)
1131 machine->vmlinux_maps[i]->end = ~0ULL;
1135 static bool machine__uses_kcore(struct machine *machine)
1139 list_for_each_entry(dso, &machine->dsos.head, node) {
1140 if (dso__is_kcore(dso))
1147 static int machine__process_kernel_mmap_event(struct machine *machine,
1148 union perf_event *event)
1151 char kmmap_prefix[PATH_MAX];
1152 enum dso_kernel_type kernel_type;
1153 bool is_kernel_mmap;
1155 /* If we have maps from kcore then we do not need or want any others */
1156 if (machine__uses_kcore(machine))
1159 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1160 if (machine__is_host(machine))
1161 kernel_type = DSO_TYPE_KERNEL;
1163 kernel_type = DSO_TYPE_GUEST_KERNEL;
1165 is_kernel_mmap = memcmp(event->mmap.filename,
1167 strlen(kmmap_prefix) - 1) == 0;
1168 if (event->mmap.filename[0] == '/' ||
1169 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1170 map = machine__findnew_module_map(machine, event->mmap.start,
1171 event->mmap.filename);
1175 map->end = map->start + event->mmap.len;
1176 } else if (is_kernel_mmap) {
1177 const char *symbol_name = (event->mmap.filename +
1178 strlen(kmmap_prefix));
1180 * Should be there already, from the build-id table in
1183 struct dso *kernel = NULL;
1186 pthread_rwlock_rdlock(&machine->dsos.lock);
1188 list_for_each_entry(dso, &machine->dsos.head, node) {
1191 * The cpumode passed to is_kernel_module is not the
1192 * cpumode of *this* event. If we insist on passing
1193 * correct cpumode to is_kernel_module, we should
1194 * record the cpumode when we adding this dso to the
1197 * However we don't really need passing correct
1198 * cpumode. We know the correct cpumode must be kernel
1199 * mode (if not, we should not link it onto kernel_dsos
1202 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1203 * is_kernel_module() treats it as a kernel cpumode.
1207 is_kernel_module(dso->long_name,
1208 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1216 pthread_rwlock_unlock(&machine->dsos.lock);
1219 kernel = machine__findnew_dso(machine, kmmap_prefix);
1223 kernel->kernel = kernel_type;
1224 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1229 if (strstr(kernel->long_name, "vmlinux"))
1230 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1232 machine__set_kernel_mmap_len(machine, event);
1235 * Avoid using a zero address (kptr_restrict) for the ref reloc
1236 * symbol. Effectively having zero here means that at record
1237 * time /proc/sys/kernel/kptr_restrict was non zero.
1239 if (event->mmap.pgoff != 0) {
1240 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1245 if (machine__is_default_guest(machine)) {
1247 * preload dso of guest kernel and modules
1249 dso__load(kernel, machine__kernel_map(machine), NULL);
1257 int machine__process_mmap2_event(struct machine *machine,
1258 union perf_event *event,
1259 struct perf_sample *sample __maybe_unused)
1261 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1262 struct thread *thread;
1268 perf_event__fprintf_mmap2(event, stdout);
1270 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1271 cpumode == PERF_RECORD_MISC_KERNEL) {
1272 ret = machine__process_kernel_mmap_event(machine, event);
1278 thread = machine__findnew_thread(machine, event->mmap2.pid,
1283 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1284 type = MAP__VARIABLE;
1286 type = MAP__FUNCTION;
1288 map = map__new(machine, event->mmap2.start,
1289 event->mmap2.len, event->mmap2.pgoff,
1290 event->mmap2.pid, event->mmap2.maj,
1291 event->mmap2.min, event->mmap2.ino,
1292 event->mmap2.ino_generation,
1295 event->mmap2.filename, type, thread);
1298 goto out_problem_map;
1300 thread__insert_map(thread, map);
1301 thread__put(thread);
1306 thread__put(thread);
1308 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1312 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1313 struct perf_sample *sample __maybe_unused)
1315 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1316 struct thread *thread;
1322 perf_event__fprintf_mmap(event, stdout);
1324 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1325 cpumode == PERF_RECORD_MISC_KERNEL) {
1326 ret = machine__process_kernel_mmap_event(machine, event);
1332 thread = machine__findnew_thread(machine, event->mmap.pid,
1337 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1338 type = MAP__VARIABLE;
1340 type = MAP__FUNCTION;
1342 map = map__new(machine, event->mmap.start,
1343 event->mmap.len, event->mmap.pgoff,
1344 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1345 event->mmap.filename,
1349 goto out_problem_map;
1351 thread__insert_map(thread, map);
1352 thread__put(thread);
1357 thread__put(thread);
1359 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1363 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1365 if (machine->last_match == th)
1366 machine->last_match = NULL;
1368 BUG_ON(atomic_read(&th->refcnt) == 0);
1370 pthread_rwlock_wrlock(&machine->threads_lock);
1371 rb_erase_init(&th->rb_node, &machine->threads);
1372 RB_CLEAR_NODE(&th->rb_node);
1374 * Move it first to the dead_threads list, then drop the reference,
1375 * if this is the last reference, then the thread__delete destructor
1376 * will be called and we will remove it from the dead_threads list.
1378 list_add_tail(&th->node, &machine->dead_threads);
1380 pthread_rwlock_unlock(&machine->threads_lock);
1384 void machine__remove_thread(struct machine *machine, struct thread *th)
1386 return __machine__remove_thread(machine, th, true);
1389 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1390 struct perf_sample *sample)
1392 struct thread *thread = machine__find_thread(machine,
1395 struct thread *parent = machine__findnew_thread(machine,
1401 perf_event__fprintf_task(event, stdout);
1404 * There may be an existing thread that is not actually the parent,
1405 * either because we are processing events out of order, or because the
1406 * (fork) event that would have removed the thread was lost. Assume the
1407 * latter case and continue on as best we can.
1409 if (parent->pid_ != (pid_t)event->fork.ppid) {
1410 dump_printf("removing erroneous parent thread %d/%d\n",
1411 parent->pid_, parent->tid);
1412 machine__remove_thread(machine, parent);
1413 thread__put(parent);
1414 parent = machine__findnew_thread(machine, event->fork.ppid,
1418 /* if a thread currently exists for the thread id remove it */
1419 if (thread != NULL) {
1420 machine__remove_thread(machine, thread);
1421 thread__put(thread);
1424 thread = machine__findnew_thread(machine, event->fork.pid,
1427 if (thread == NULL || parent == NULL ||
1428 thread__fork(thread, parent, sample->time) < 0) {
1429 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1432 thread__put(thread);
1433 thread__put(parent);
1438 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1439 struct perf_sample *sample __maybe_unused)
1441 struct thread *thread = machine__find_thread(machine,
1446 perf_event__fprintf_task(event, stdout);
1448 if (thread != NULL) {
1449 thread__exited(thread);
1450 thread__put(thread);
1456 int machine__process_event(struct machine *machine, union perf_event *event,
1457 struct perf_sample *sample)
1461 switch (event->header.type) {
1462 case PERF_RECORD_COMM:
1463 ret = machine__process_comm_event(machine, event, sample); break;
1464 case PERF_RECORD_MMAP:
1465 ret = machine__process_mmap_event(machine, event, sample); break;
1466 case PERF_RECORD_MMAP2:
1467 ret = machine__process_mmap2_event(machine, event, sample); break;
1468 case PERF_RECORD_FORK:
1469 ret = machine__process_fork_event(machine, event, sample); break;
1470 case PERF_RECORD_EXIT:
1471 ret = machine__process_exit_event(machine, event, sample); break;
1472 case PERF_RECORD_LOST:
1473 ret = machine__process_lost_event(machine, event, sample); break;
1474 case PERF_RECORD_AUX:
1475 ret = machine__process_aux_event(machine, event); break;
1476 case PERF_RECORD_ITRACE_START:
1477 ret = machine__process_itrace_start_event(machine, event); break;
1478 case PERF_RECORD_LOST_SAMPLES:
1479 ret = machine__process_lost_samples_event(machine, event, sample); break;
1480 case PERF_RECORD_SWITCH:
1481 case PERF_RECORD_SWITCH_CPU_WIDE:
1482 ret = machine__process_switch_event(machine, event); break;
1491 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1493 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1498 static void ip__resolve_ams(struct thread *thread,
1499 struct addr_map_symbol *ams,
1502 struct addr_location al;
1504 memset(&al, 0, sizeof(al));
1506 * We cannot use the header.misc hint to determine whether a
1507 * branch stack address is user, kernel, guest, hypervisor.
1508 * Branches may straddle the kernel/user/hypervisor boundaries.
1509 * Thus, we have to try consecutively until we find a match
1510 * or else, the symbol is unknown
1512 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1515 ams->al_addr = al.addr;
1520 static void ip__resolve_data(struct thread *thread,
1521 u8 m, struct addr_map_symbol *ams, u64 addr)
1523 struct addr_location al;
1525 memset(&al, 0, sizeof(al));
1527 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1528 if (al.map == NULL) {
1530 * some shared data regions have execute bit set which puts
1531 * their mapping in the MAP__FUNCTION type array.
1532 * Check there as a fallback option before dropping the sample.
1534 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1538 ams->al_addr = al.addr;
1543 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1544 struct addr_location *al)
1546 struct mem_info *mi = zalloc(sizeof(*mi));
1551 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1552 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1553 mi->data_src.val = sample->data_src;
1558 static int add_callchain_ip(struct thread *thread,
1559 struct symbol **parent,
1560 struct addr_location *root_al,
1564 struct addr_location al;
1569 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1572 if (ip >= PERF_CONTEXT_MAX) {
1574 case PERF_CONTEXT_HV:
1575 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1577 case PERF_CONTEXT_KERNEL:
1578 *cpumode = PERF_RECORD_MISC_KERNEL;
1580 case PERF_CONTEXT_USER:
1581 *cpumode = PERF_RECORD_MISC_USER;
1584 pr_debug("invalid callchain context: "
1585 "%"PRId64"\n", (s64) ip);
1587 * It seems the callchain is corrupted.
1590 callchain_cursor_reset(&callchain_cursor);
1595 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1599 if (al.sym != NULL) {
1600 if (sort__has_parent && !*parent &&
1601 symbol__match_regex(al.sym, &parent_regex))
1603 else if (have_ignore_callees && root_al &&
1604 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1605 /* Treat this symbol as the root,
1606 forgetting its callees. */
1608 callchain_cursor_reset(&callchain_cursor);
1612 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1615 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1616 struct addr_location *al)
1619 const struct branch_stack *bs = sample->branch_stack;
1620 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1625 for (i = 0; i < bs->nr; i++) {
1626 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1627 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1628 bi[i].flags = bs->entries[i].flags;
1635 #define NO_ENTRY 0xff
1637 #define PERF_MAX_BRANCH_DEPTH 127
1640 static int remove_loops(struct branch_entry *l, int nr)
1643 unsigned char chash[CHASHSZ];
1645 memset(chash, NO_ENTRY, sizeof(chash));
1647 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1649 for (i = 0; i < nr; i++) {
1650 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1652 /* no collision handling for now */
1653 if (chash[h] == NO_ENTRY) {
1655 } else if (l[chash[h]].from == l[i].from) {
1656 bool is_loop = true;
1657 /* check if it is a real loop */
1659 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1660 if (l[j].from != l[i + off].from) {
1665 memmove(l + i, l + i + off,
1666 (nr - (i + off)) * sizeof(*l));
1675 * Recolve LBR callstack chain sample
1677 * 1 on success get LBR callchain information
1678 * 0 no available LBR callchain information, should try fp
1679 * negative error code on other errors.
1681 static int resolve_lbr_callchain_sample(struct thread *thread,
1682 struct perf_sample *sample,
1683 struct symbol **parent,
1684 struct addr_location *root_al,
1687 struct ip_callchain *chain = sample->callchain;
1688 int chain_nr = min(max_stack, (int)chain->nr);
1689 u8 cpumode = PERF_RECORD_MISC_USER;
1693 for (i = 0; i < chain_nr; i++) {
1694 if (chain->ips[i] == PERF_CONTEXT_USER)
1698 /* LBR only affects the user callchain */
1699 if (i != chain_nr) {
1700 struct branch_stack *lbr_stack = sample->branch_stack;
1701 int lbr_nr = lbr_stack->nr;
1703 * LBR callstack can only get user call chain.
1704 * The mix_chain_nr is kernel call chain
1705 * number plus LBR user call chain number.
1706 * i is kernel call chain number,
1707 * 1 is PERF_CONTEXT_USER,
1708 * lbr_nr + 1 is the user call chain number.
1709 * For details, please refer to the comments
1710 * in callchain__printf
1712 int mix_chain_nr = i + 1 + lbr_nr + 1;
1714 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1715 pr_warning("corrupted callchain. skipping...\n");
1719 for (j = 0; j < mix_chain_nr; j++) {
1720 if (callchain_param.order == ORDER_CALLEE) {
1724 ip = lbr_stack->entries[j - i - 2].from;
1726 ip = lbr_stack->entries[0].to;
1729 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1730 else if (j > lbr_nr)
1731 ip = chain->ips[i + 1 - (j - lbr_nr)];
1733 ip = lbr_stack->entries[0].to;
1736 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1738 return (err < 0) ? err : 0;
1746 static int thread__resolve_callchain_sample(struct thread *thread,
1747 struct perf_evsel *evsel,
1748 struct perf_sample *sample,
1749 struct symbol **parent,
1750 struct addr_location *root_al,
1753 struct branch_stack *branch = sample->branch_stack;
1754 struct ip_callchain *chain = sample->callchain;
1755 int chain_nr = min(max_stack, (int)chain->nr);
1756 u8 cpumode = PERF_RECORD_MISC_USER;
1761 callchain_cursor_reset(&callchain_cursor);
1763 if (has_branch_callstack(evsel)) {
1764 err = resolve_lbr_callchain_sample(thread, sample, parent,
1765 root_al, max_stack);
1767 return (err < 0) ? err : 0;
1771 * Based on DWARF debug information, some architectures skip
1772 * a callchain entry saved by the kernel.
1774 if (chain->nr < PERF_MAX_STACK_DEPTH)
1775 skip_idx = arch_skip_callchain_idx(thread, chain);
1778 * Add branches to call stack for easier browsing. This gives
1779 * more context for a sample than just the callers.
1781 * This uses individual histograms of paths compared to the
1782 * aggregated histograms the normal LBR mode uses.
1784 * Limitations for now:
1785 * - No extra filters
1786 * - No annotations (should annotate somehow)
1789 if (branch && callchain_param.branch_callstack) {
1790 int nr = min(max_stack, (int)branch->nr);
1791 struct branch_entry be[nr];
1793 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1794 pr_warning("corrupted branch chain. skipping...\n");
1798 for (i = 0; i < nr; i++) {
1799 if (callchain_param.order == ORDER_CALLEE) {
1800 be[i] = branch->entries[i];
1802 * Check for overlap into the callchain.
1803 * The return address is one off compared to
1804 * the branch entry. To adjust for this
1805 * assume the calling instruction is not longer
1808 if (i == skip_idx ||
1809 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1811 else if (be[i].from < chain->ips[first_call] &&
1812 be[i].from >= chain->ips[first_call] - 8)
1815 be[i] = branch->entries[branch->nr - i - 1];
1818 nr = remove_loops(be, nr);
1820 for (i = 0; i < nr; i++) {
1821 err = add_callchain_ip(thread, parent, root_al,
1824 err = add_callchain_ip(thread, parent, root_al,
1835 if (chain->nr > PERF_MAX_STACK_DEPTH && (int)chain->nr > max_stack) {
1836 pr_warning("corrupted callchain. skipping...\n");
1840 for (i = first_call; i < chain_nr; i++) {
1843 if (callchain_param.order == ORDER_CALLEE)
1846 j = chain->nr - i - 1;
1848 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1854 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1857 return (err < 0) ? err : 0;
1863 static int unwind_entry(struct unwind_entry *entry, void *arg)
1865 struct callchain_cursor *cursor = arg;
1866 return callchain_cursor_append(cursor, entry->ip,
1867 entry->map, entry->sym);
1870 int thread__resolve_callchain(struct thread *thread,
1871 struct perf_evsel *evsel,
1872 struct perf_sample *sample,
1873 struct symbol **parent,
1874 struct addr_location *root_al,
1877 int ret = thread__resolve_callchain_sample(thread, evsel,
1879 root_al, max_stack);
1883 /* Can we do dwarf post unwind? */
1884 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1885 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1888 /* Bail out if nothing was captured. */
1889 if ((!sample->user_regs.regs) ||
1890 (!sample->user_stack.size))
1893 return unwind__get_entries(unwind_entry, &callchain_cursor,
1894 thread, sample, max_stack);
1898 int machine__for_each_thread(struct machine *machine,
1899 int (*fn)(struct thread *thread, void *p),
1903 struct thread *thread;
1906 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1907 thread = rb_entry(nd, struct thread, rb_node);
1908 rc = fn(thread, priv);
1913 list_for_each_entry(thread, &machine->dead_threads, node) {
1914 rc = fn(thread, priv);
1921 int machines__for_each_thread(struct machines *machines,
1922 int (*fn)(struct thread *thread, void *p),
1928 rc = machine__for_each_thread(&machines->host, fn, priv);
1932 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
1933 struct machine *machine = rb_entry(nd, struct machine, rb_node);
1935 rc = machine__for_each_thread(machine, fn, priv);
1942 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1943 struct target *target, struct thread_map *threads,
1944 perf_event__handler_t process, bool data_mmap,
1945 unsigned int proc_map_timeout)
1947 if (target__has_task(target))
1948 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
1949 else if (target__has_cpu(target))
1950 return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
1951 /* command specified */
1955 pid_t machine__get_current_tid(struct machine *machine, int cpu)
1957 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
1960 return machine->current_tid[cpu];
1963 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
1966 struct thread *thread;
1971 if (!machine->current_tid) {
1974 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
1975 if (!machine->current_tid)
1977 for (i = 0; i < MAX_NR_CPUS; i++)
1978 machine->current_tid[i] = -1;
1981 if (cpu >= MAX_NR_CPUS) {
1982 pr_err("Requested CPU %d too large. ", cpu);
1983 pr_err("Consider raising MAX_NR_CPUS\n");
1987 machine->current_tid[cpu] = tid;
1989 thread = machine__findnew_thread(machine, pid, tid);
1994 thread__put(thread);
1999 int machine__get_kernel_start(struct machine *machine)
2001 struct map *map = machine__kernel_map(machine);
2005 * The only addresses above 2^63 are kernel addresses of a 64-bit
2006 * kernel. Note that addresses are unsigned so that on a 32-bit system
2007 * all addresses including kernel addresses are less than 2^32. In
2008 * that case (32-bit system), if the kernel mapping is unknown, all
2009 * addresses will be assumed to be in user space - see
2010 * machine__kernel_ip().
2012 machine->kernel_start = 1ULL << 63;
2014 err = map__load(map, machine->symbol_filter);
2016 machine->kernel_start = map->start;
2021 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2023 return dsos__findnew(&machine->dsos, filename);
2026 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2028 struct machine *machine = vmachine;
2030 struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map, NULL);
2035 *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2036 *addrp = map->unmap_ip(map, sym->start);