13 #include <symbol/kallsyms.h>
15 #include "linux/hash.h"
17 static void dsos__init(struct dsos *dsos)
19 INIT_LIST_HEAD(&dsos->head);
23 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
25 map_groups__init(&machine->kmaps, machine);
26 RB_CLEAR_NODE(&machine->rb_node);
27 dsos__init(&machine->user_dsos);
28 dsos__init(&machine->kernel_dsos);
30 machine->threads = RB_ROOT;
31 INIT_LIST_HEAD(&machine->dead_threads);
32 machine->last_match = NULL;
34 machine->vdso_info = NULL;
38 machine->symbol_filter = NULL;
39 machine->id_hdr_size = 0;
40 machine->comm_exec = false;
41 machine->kernel_start = 0;
43 machine->root_dir = strdup(root_dir);
44 if (machine->root_dir == NULL)
47 if (pid != HOST_KERNEL_ID) {
48 struct thread *thread = machine__findnew_thread(machine, -1,
55 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
56 thread__set_comm(thread, comm, 0);
59 machine->current_tid = NULL;
64 struct machine *machine__new_host(void)
66 struct machine *machine = malloc(sizeof(*machine));
68 if (machine != NULL) {
69 machine__init(machine, "", HOST_KERNEL_ID);
71 if (machine__create_kernel_maps(machine) < 0)
81 static void dsos__delete(struct dsos *dsos)
85 list_for_each_entry_safe(pos, n, &dsos->head, node) {
86 RB_CLEAR_NODE(&pos->rb_node);
92 void machine__delete_threads(struct machine *machine)
94 struct rb_node *nd = rb_first(&machine->threads);
97 struct thread *t = rb_entry(nd, struct thread, rb_node);
100 machine__remove_thread(machine, t);
104 void machine__exit(struct machine *machine)
106 map_groups__exit(&machine->kmaps);
107 dsos__delete(&machine->user_dsos);
108 dsos__delete(&machine->kernel_dsos);
110 zfree(&machine->root_dir);
111 zfree(&machine->current_tid);
114 void machine__delete(struct machine *machine)
116 machine__exit(machine);
120 void machines__init(struct machines *machines)
122 machine__init(&machines->host, "", HOST_KERNEL_ID);
123 machines->guests = RB_ROOT;
124 machines->symbol_filter = NULL;
127 void machines__exit(struct machines *machines)
129 machine__exit(&machines->host);
133 struct machine *machines__add(struct machines *machines, pid_t pid,
134 const char *root_dir)
136 struct rb_node **p = &machines->guests.rb_node;
137 struct rb_node *parent = NULL;
138 struct machine *pos, *machine = malloc(sizeof(*machine));
143 if (machine__init(machine, root_dir, pid) != 0) {
148 machine->symbol_filter = machines->symbol_filter;
152 pos = rb_entry(parent, struct machine, rb_node);
159 rb_link_node(&machine->rb_node, parent, p);
160 rb_insert_color(&machine->rb_node, &machines->guests);
165 void machines__set_symbol_filter(struct machines *machines,
166 symbol_filter_t symbol_filter)
170 machines->symbol_filter = symbol_filter;
171 machines->host.symbol_filter = symbol_filter;
173 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
174 struct machine *machine = rb_entry(nd, struct machine, rb_node);
176 machine->symbol_filter = symbol_filter;
180 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
184 machines->host.comm_exec = comm_exec;
186 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
187 struct machine *machine = rb_entry(nd, struct machine, rb_node);
189 machine->comm_exec = comm_exec;
193 struct machine *machines__find(struct machines *machines, pid_t pid)
195 struct rb_node **p = &machines->guests.rb_node;
196 struct rb_node *parent = NULL;
197 struct machine *machine;
198 struct machine *default_machine = NULL;
200 if (pid == HOST_KERNEL_ID)
201 return &machines->host;
205 machine = rb_entry(parent, struct machine, rb_node);
206 if (pid < machine->pid)
208 else if (pid > machine->pid)
213 default_machine = machine;
216 return default_machine;
219 struct machine *machines__findnew(struct machines *machines, pid_t pid)
222 const char *root_dir = "";
223 struct machine *machine = machines__find(machines, pid);
225 if (machine && (machine->pid == pid))
228 if ((pid != HOST_KERNEL_ID) &&
229 (pid != DEFAULT_GUEST_KERNEL_ID) &&
230 (symbol_conf.guestmount)) {
231 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
232 if (access(path, R_OK)) {
233 static struct strlist *seen;
236 seen = strlist__new(true, NULL);
238 if (!strlist__has_entry(seen, path)) {
239 pr_err("Can't access file %s\n", path);
240 strlist__add(seen, path);
248 machine = machines__add(machines, pid, root_dir);
253 void machines__process_guests(struct machines *machines,
254 machine__process_t process, void *data)
258 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
259 struct machine *pos = rb_entry(nd, struct machine, rb_node);
264 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
266 if (machine__is_host(machine))
267 snprintf(bf, size, "[%s]", "kernel.kallsyms");
268 else if (machine__is_default_guest(machine))
269 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
271 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
278 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
280 struct rb_node *node;
281 struct machine *machine;
283 machines->host.id_hdr_size = id_hdr_size;
285 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
286 machine = rb_entry(node, struct machine, rb_node);
287 machine->id_hdr_size = id_hdr_size;
293 static void machine__update_thread_pid(struct machine *machine,
294 struct thread *th, pid_t pid)
296 struct thread *leader;
298 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
303 if (th->pid_ == th->tid)
306 leader = machine__findnew_thread(machine, th->pid_, th->pid_);
311 leader->mg = map_groups__new(machine);
316 if (th->mg == leader->mg)
321 * Maps are created from MMAP events which provide the pid and
322 * tid. Consequently there never should be any maps on a thread
323 * with an unknown pid. Just print an error if there are.
325 if (!map_groups__empty(th->mg))
326 pr_err("Discarding thread maps for %d:%d\n",
328 map_groups__delete(th->mg);
331 th->mg = map_groups__get(leader->mg);
336 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
339 static struct thread *__machine__findnew_thread(struct machine *machine,
340 pid_t pid, pid_t tid,
343 struct rb_node **p = &machine->threads.rb_node;
344 struct rb_node *parent = NULL;
348 * Front-end cache - TID lookups come in blocks,
349 * so most of the time we dont have to look up
352 th = machine->last_match;
354 if (th->tid == tid) {
355 machine__update_thread_pid(machine, th, pid);
359 thread__zput(machine->last_match);
364 th = rb_entry(parent, struct thread, rb_node);
366 if (th->tid == tid) {
367 machine->last_match = thread__get(th);
368 machine__update_thread_pid(machine, th, pid);
381 th = thread__new(pid, tid);
383 rb_link_node(&th->rb_node, parent, p);
384 rb_insert_color(&th->rb_node, &machine->threads);
387 * We have to initialize map_groups separately
388 * after rb tree is updated.
390 * The reason is that we call machine__findnew_thread
391 * within thread__init_map_groups to find the thread
392 * leader and that would screwed the rb tree.
394 if (thread__init_map_groups(th, machine)) {
395 rb_erase(&th->rb_node, &machine->threads);
400 * It is now in the rbtree, get a ref
403 machine->last_match = thread__get(th);
409 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
412 return __machine__findnew_thread(machine, pid, tid, true);
415 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
418 return __machine__findnew_thread(machine, pid, tid, false);
421 struct comm *machine__thread_exec_comm(struct machine *machine,
422 struct thread *thread)
424 if (machine->comm_exec)
425 return thread__exec_comm(thread);
427 return thread__comm(thread);
430 int machine__process_comm_event(struct machine *machine, union perf_event *event,
431 struct perf_sample *sample)
433 struct thread *thread = machine__findnew_thread(machine,
436 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
439 machine->comm_exec = true;
442 perf_event__fprintf_comm(event, stdout);
444 if (thread == NULL ||
445 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
446 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
453 int machine__process_lost_event(struct machine *machine __maybe_unused,
454 union perf_event *event, struct perf_sample *sample __maybe_unused)
456 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
457 event->lost.id, event->lost.lost);
462 machine__module_dso(struct machine *machine, struct kmod_path *m,
463 const char *filename)
467 dso = dsos__find(&machine->kernel_dsos, m->name, true);
469 dso = dsos__addnew(&machine->kernel_dsos, m->name);
473 if (machine__is_host(machine))
474 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
476 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
478 /* _KMODULE_COMP should be next to _KMODULE */
479 if (m->kmod && m->comp)
482 dso__set_short_name(dso, strdup(m->name), true);
483 dso__set_long_name(dso, strdup(filename), true);
489 struct map *machine__new_module(struct machine *machine, u64 start,
490 const char *filename)
492 struct map *map = NULL;
496 if (kmod_path__parse_name(&m, filename))
499 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
504 dso = machine__module_dso(machine, &m, filename);
508 map = map__new2(start, dso, MAP__FUNCTION);
512 map_groups__insert(&machine->kmaps, map);
519 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
522 size_t ret = __dsos__fprintf(&machines->host.kernel_dsos.head, fp) +
523 __dsos__fprintf(&machines->host.user_dsos.head, fp);
525 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
526 struct machine *pos = rb_entry(nd, struct machine, rb_node);
527 ret += __dsos__fprintf(&pos->kernel_dsos.head, fp);
528 ret += __dsos__fprintf(&pos->user_dsos.head, fp);
534 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
535 bool (skip)(struct dso *dso, int parm), int parm)
537 return __dsos__fprintf_buildid(&m->kernel_dsos.head, fp, skip, parm) +
538 __dsos__fprintf_buildid(&m->user_dsos.head, fp, skip, parm);
541 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
542 bool (skip)(struct dso *dso, int parm), int parm)
545 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
547 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
548 struct machine *pos = rb_entry(nd, struct machine, rb_node);
549 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
554 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
558 struct dso *kdso = machine->vmlinux_maps[MAP__FUNCTION]->dso;
560 if (kdso->has_build_id) {
561 char filename[PATH_MAX];
562 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
563 printed += fprintf(fp, "[0] %s\n", filename);
566 for (i = 0; i < vmlinux_path__nr_entries; ++i)
567 printed += fprintf(fp, "[%d] %s\n",
568 i + kdso->has_build_id, vmlinux_path[i]);
573 size_t machine__fprintf(struct machine *machine, FILE *fp)
578 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
579 struct thread *pos = rb_entry(nd, struct thread, rb_node);
581 ret += thread__fprintf(pos, fp);
587 static struct dso *machine__get_kernel(struct machine *machine)
589 const char *vmlinux_name = NULL;
592 if (machine__is_host(machine)) {
593 vmlinux_name = symbol_conf.vmlinux_name;
595 vmlinux_name = "[kernel.kallsyms]";
597 kernel = dso__kernel_findnew(machine, vmlinux_name,
603 if (machine__is_default_guest(machine))
604 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
606 vmlinux_name = machine__mmap_name(machine, bf,
609 kernel = dso__kernel_findnew(machine, vmlinux_name,
611 DSO_TYPE_GUEST_KERNEL);
614 if (kernel != NULL && (!kernel->has_build_id))
615 dso__read_running_kernel_build_id(kernel, machine);
620 struct process_args {
624 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
627 if (machine__is_default_guest(machine))
628 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
630 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
633 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
635 /* Figure out the start address of kernel map from /proc/kallsyms.
636 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
637 * symbol_name if it's not that important.
639 static u64 machine__get_running_kernel_start(struct machine *machine,
640 const char **symbol_name)
642 char filename[PATH_MAX];
647 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
649 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
652 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
653 addr = kallsyms__get_function_start(filename, name);
664 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
667 u64 start = machine__get_running_kernel_start(machine, NULL);
669 for (type = 0; type < MAP__NR_TYPES; ++type) {
672 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
673 if (machine->vmlinux_maps[type] == NULL)
676 machine->vmlinux_maps[type]->map_ip =
677 machine->vmlinux_maps[type]->unmap_ip =
679 kmap = map__kmap(machine->vmlinux_maps[type]);
683 kmap->kmaps = &machine->kmaps;
684 map_groups__insert(&machine->kmaps,
685 machine->vmlinux_maps[type]);
691 void machine__destroy_kernel_maps(struct machine *machine)
695 for (type = 0; type < MAP__NR_TYPES; ++type) {
698 if (machine->vmlinux_maps[type] == NULL)
701 kmap = map__kmap(machine->vmlinux_maps[type]);
702 map_groups__remove(&machine->kmaps,
703 machine->vmlinux_maps[type]);
704 if (kmap && kmap->ref_reloc_sym) {
706 * ref_reloc_sym is shared among all maps, so free just
709 if (type == MAP__FUNCTION) {
710 zfree((char **)&kmap->ref_reloc_sym->name);
711 zfree(&kmap->ref_reloc_sym);
713 kmap->ref_reloc_sym = NULL;
716 map__delete(machine->vmlinux_maps[type]);
717 machine->vmlinux_maps[type] = NULL;
721 int machines__create_guest_kernel_maps(struct machines *machines)
724 struct dirent **namelist = NULL;
730 if (symbol_conf.default_guest_vmlinux_name ||
731 symbol_conf.default_guest_modules ||
732 symbol_conf.default_guest_kallsyms) {
733 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
736 if (symbol_conf.guestmount) {
737 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
740 for (i = 0; i < items; i++) {
741 if (!isdigit(namelist[i]->d_name[0])) {
742 /* Filter out . and .. */
745 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
746 if ((*endp != '\0') ||
747 (endp == namelist[i]->d_name) ||
749 pr_debug("invalid directory (%s). Skipping.\n",
750 namelist[i]->d_name);
753 sprintf(path, "%s/%s/proc/kallsyms",
754 symbol_conf.guestmount,
755 namelist[i]->d_name);
756 ret = access(path, R_OK);
758 pr_debug("Can't access file %s\n", path);
761 machines__create_kernel_maps(machines, pid);
770 void machines__destroy_kernel_maps(struct machines *machines)
772 struct rb_node *next = rb_first(&machines->guests);
774 machine__destroy_kernel_maps(&machines->host);
777 struct machine *pos = rb_entry(next, struct machine, rb_node);
779 next = rb_next(&pos->rb_node);
780 rb_erase(&pos->rb_node, &machines->guests);
781 machine__delete(pos);
785 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
787 struct machine *machine = machines__findnew(machines, pid);
792 return machine__create_kernel_maps(machine);
795 int machine__load_kallsyms(struct machine *machine, const char *filename,
796 enum map_type type, symbol_filter_t filter)
798 struct map *map = machine->vmlinux_maps[type];
799 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
802 dso__set_loaded(map->dso, type);
804 * Since /proc/kallsyms will have multiple sessions for the
805 * kernel, with modules between them, fixup the end of all
808 __map_groups__fixup_end(&machine->kmaps, type);
814 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
815 symbol_filter_t filter)
817 struct map *map = machine->vmlinux_maps[type];
818 int ret = dso__load_vmlinux_path(map->dso, map, filter);
821 dso__set_loaded(map->dso, type);
826 static void map_groups__fixup_end(struct map_groups *mg)
829 for (i = 0; i < MAP__NR_TYPES; ++i)
830 __map_groups__fixup_end(mg, i);
833 static char *get_kernel_version(const char *root_dir)
835 char version[PATH_MAX];
838 const char *prefix = "Linux version ";
840 sprintf(version, "%s/proc/version", root_dir);
841 file = fopen(version, "r");
846 tmp = fgets(version, sizeof(version), file);
849 name = strstr(version, prefix);
852 name += strlen(prefix);
853 tmp = strchr(name, ' ');
860 static bool is_kmod_dso(struct dso *dso)
862 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
863 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
866 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
872 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
876 long_name = strdup(path);
877 if (long_name == NULL)
880 dso__set_long_name(map->dso, long_name, true);
881 dso__kernel_module_get_build_id(map->dso, "");
884 * Full name could reveal us kmod compression, so
885 * we need to update the symtab_type if needed.
887 if (m->comp && is_kmod_dso(map->dso))
888 map->dso->symtab_type++;
893 static int map_groups__set_modules_path_dir(struct map_groups *mg,
894 const char *dir_name, int depth)
897 DIR *dir = opendir(dir_name);
901 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
905 while ((dent = readdir(dir)) != NULL) {
909 /*sshfs might return bad dent->d_type, so we have to stat*/
910 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
914 if (S_ISDIR(st.st_mode)) {
915 if (!strcmp(dent->d_name, ".") ||
916 !strcmp(dent->d_name, ".."))
919 /* Do not follow top-level source and build symlinks */
921 if (!strcmp(dent->d_name, "source") ||
922 !strcmp(dent->d_name, "build"))
926 ret = map_groups__set_modules_path_dir(mg, path,
933 ret = kmod_path__parse_name(&m, dent->d_name);
938 ret = map_groups__set_module_path(mg, path, &m);
952 static int machine__set_modules_path(struct machine *machine)
955 char modules_path[PATH_MAX];
957 version = get_kernel_version(machine->root_dir);
961 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
962 machine->root_dir, version);
965 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
968 static int machine__create_module(void *arg, const char *name, u64 start)
970 struct machine *machine = arg;
973 map = machine__new_module(machine, start, name);
977 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
982 static int machine__create_modules(struct machine *machine)
987 if (machine__is_default_guest(machine)) {
988 modules = symbol_conf.default_guest_modules;
990 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
994 if (symbol__restricted_filename(modules, "/proc/modules"))
997 if (modules__parse(modules, machine, machine__create_module))
1000 if (!machine__set_modules_path(machine))
1003 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1008 int machine__create_kernel_maps(struct machine *machine)
1010 struct dso *kernel = machine__get_kernel(machine);
1012 u64 addr = machine__get_running_kernel_start(machine, &name);
1016 if (kernel == NULL ||
1017 __machine__create_kernel_maps(machine, kernel) < 0)
1020 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1021 if (machine__is_host(machine))
1022 pr_debug("Problems creating module maps, "
1023 "continuing anyway...\n");
1025 pr_debug("Problems creating module maps for guest %d, "
1026 "continuing anyway...\n", machine->pid);
1030 * Now that we have all the maps created, just set the ->end of them:
1032 map_groups__fixup_end(&machine->kmaps);
1034 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
1036 machine__destroy_kernel_maps(machine);
1043 static void machine__set_kernel_mmap_len(struct machine *machine,
1044 union perf_event *event)
1048 for (i = 0; i < MAP__NR_TYPES; i++) {
1049 machine->vmlinux_maps[i]->start = event->mmap.start;
1050 machine->vmlinux_maps[i]->end = (event->mmap.start +
1053 * Be a bit paranoid here, some perf.data file came with
1054 * a zero sized synthesized MMAP event for the kernel.
1056 if (machine->vmlinux_maps[i]->end == 0)
1057 machine->vmlinux_maps[i]->end = ~0ULL;
1061 static bool machine__uses_kcore(struct machine *machine)
1065 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1066 if (dso__is_kcore(dso))
1073 static int machine__process_kernel_mmap_event(struct machine *machine,
1074 union perf_event *event)
1077 char kmmap_prefix[PATH_MAX];
1078 enum dso_kernel_type kernel_type;
1079 bool is_kernel_mmap;
1081 /* If we have maps from kcore then we do not need or want any others */
1082 if (machine__uses_kcore(machine))
1085 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1086 if (machine__is_host(machine))
1087 kernel_type = DSO_TYPE_KERNEL;
1089 kernel_type = DSO_TYPE_GUEST_KERNEL;
1091 is_kernel_mmap = memcmp(event->mmap.filename,
1093 strlen(kmmap_prefix) - 1) == 0;
1094 if (event->mmap.filename[0] == '/' ||
1095 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1096 map = machine__new_module(machine, event->mmap.start,
1097 event->mmap.filename);
1101 map->end = map->start + event->mmap.len;
1102 } else if (is_kernel_mmap) {
1103 const char *symbol_name = (event->mmap.filename +
1104 strlen(kmmap_prefix));
1106 * Should be there already, from the build-id table in
1109 struct dso *kernel = NULL;
1112 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1113 if (is_kernel_module(dso->long_name))
1121 kernel = __dsos__findnew(&machine->kernel_dsos,
1126 kernel->kernel = kernel_type;
1127 if (__machine__create_kernel_maps(machine, kernel) < 0)
1130 if (strstr(kernel->long_name, "vmlinux"))
1131 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1133 machine__set_kernel_mmap_len(machine, event);
1136 * Avoid using a zero address (kptr_restrict) for the ref reloc
1137 * symbol. Effectively having zero here means that at record
1138 * time /proc/sys/kernel/kptr_restrict was non zero.
1140 if (event->mmap.pgoff != 0) {
1141 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1146 if (machine__is_default_guest(machine)) {
1148 * preload dso of guest kernel and modules
1150 dso__load(kernel, machine->vmlinux_maps[MAP__FUNCTION],
1159 int machine__process_mmap2_event(struct machine *machine,
1160 union perf_event *event,
1161 struct perf_sample *sample __maybe_unused)
1163 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1164 struct thread *thread;
1170 perf_event__fprintf_mmap2(event, stdout);
1172 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1173 cpumode == PERF_RECORD_MISC_KERNEL) {
1174 ret = machine__process_kernel_mmap_event(machine, event);
1180 thread = machine__findnew_thread(machine, event->mmap2.pid,
1185 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1186 type = MAP__VARIABLE;
1188 type = MAP__FUNCTION;
1190 map = map__new(machine, event->mmap2.start,
1191 event->mmap2.len, event->mmap2.pgoff,
1192 event->mmap2.pid, event->mmap2.maj,
1193 event->mmap2.min, event->mmap2.ino,
1194 event->mmap2.ino_generation,
1197 event->mmap2.filename, type, thread);
1202 thread__insert_map(thread, map);
1206 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1210 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1211 struct perf_sample *sample __maybe_unused)
1213 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1214 struct thread *thread;
1220 perf_event__fprintf_mmap(event, stdout);
1222 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1223 cpumode == PERF_RECORD_MISC_KERNEL) {
1224 ret = machine__process_kernel_mmap_event(machine, event);
1230 thread = machine__findnew_thread(machine, event->mmap.pid,
1235 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1236 type = MAP__VARIABLE;
1238 type = MAP__FUNCTION;
1240 map = map__new(machine, event->mmap.start,
1241 event->mmap.len, event->mmap.pgoff,
1242 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1243 event->mmap.filename,
1249 thread__insert_map(thread, map);
1253 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1257 void machine__remove_thread(struct machine *machine, struct thread *th)
1259 if (machine->last_match == th)
1260 thread__zput(machine->last_match);
1262 rb_erase(&th->rb_node, &machine->threads);
1264 * Move it first to the dead_threads list, then drop the reference,
1265 * if this is the last reference, then the thread__delete destructor
1266 * will be called and we will remove it from the dead_threads list.
1268 list_add_tail(&th->node, &machine->dead_threads);
1272 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1273 struct perf_sample *sample)
1275 struct thread *thread = machine__find_thread(machine,
1278 struct thread *parent = machine__findnew_thread(machine,
1282 /* if a thread currently exists for the thread id remove it */
1284 machine__remove_thread(machine, thread);
1286 thread = machine__findnew_thread(machine, event->fork.pid,
1289 perf_event__fprintf_task(event, stdout);
1291 if (thread == NULL || parent == NULL ||
1292 thread__fork(thread, parent, sample->time) < 0) {
1293 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1300 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1301 struct perf_sample *sample __maybe_unused)
1303 struct thread *thread = machine__find_thread(machine,
1308 perf_event__fprintf_task(event, stdout);
1311 thread__exited(thread);
1316 int machine__process_event(struct machine *machine, union perf_event *event,
1317 struct perf_sample *sample)
1321 switch (event->header.type) {
1322 case PERF_RECORD_COMM:
1323 ret = machine__process_comm_event(machine, event, sample); break;
1324 case PERF_RECORD_MMAP:
1325 ret = machine__process_mmap_event(machine, event, sample); break;
1326 case PERF_RECORD_MMAP2:
1327 ret = machine__process_mmap2_event(machine, event, sample); break;
1328 case PERF_RECORD_FORK:
1329 ret = machine__process_fork_event(machine, event, sample); break;
1330 case PERF_RECORD_EXIT:
1331 ret = machine__process_exit_event(machine, event, sample); break;
1332 case PERF_RECORD_LOST:
1333 ret = machine__process_lost_event(machine, event, sample); break;
1342 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1344 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1349 static void ip__resolve_ams(struct thread *thread,
1350 struct addr_map_symbol *ams,
1353 struct addr_location al;
1355 memset(&al, 0, sizeof(al));
1357 * We cannot use the header.misc hint to determine whether a
1358 * branch stack address is user, kernel, guest, hypervisor.
1359 * Branches may straddle the kernel/user/hypervisor boundaries.
1360 * Thus, we have to try consecutively until we find a match
1361 * or else, the symbol is unknown
1363 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1366 ams->al_addr = al.addr;
1371 static void ip__resolve_data(struct thread *thread,
1372 u8 m, struct addr_map_symbol *ams, u64 addr)
1374 struct addr_location al;
1376 memset(&al, 0, sizeof(al));
1378 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1379 if (al.map == NULL) {
1381 * some shared data regions have execute bit set which puts
1382 * their mapping in the MAP__FUNCTION type array.
1383 * Check there as a fallback option before dropping the sample.
1385 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1389 ams->al_addr = al.addr;
1394 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1395 struct addr_location *al)
1397 struct mem_info *mi = zalloc(sizeof(*mi));
1402 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1403 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1404 mi->data_src.val = sample->data_src;
1409 static int add_callchain_ip(struct thread *thread,
1410 struct symbol **parent,
1411 struct addr_location *root_al,
1415 struct addr_location al;
1420 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1423 if (ip >= PERF_CONTEXT_MAX) {
1425 case PERF_CONTEXT_HV:
1426 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1428 case PERF_CONTEXT_KERNEL:
1429 *cpumode = PERF_RECORD_MISC_KERNEL;
1431 case PERF_CONTEXT_USER:
1432 *cpumode = PERF_RECORD_MISC_USER;
1435 pr_debug("invalid callchain context: "
1436 "%"PRId64"\n", (s64) ip);
1438 * It seems the callchain is corrupted.
1441 callchain_cursor_reset(&callchain_cursor);
1446 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1450 if (al.sym != NULL) {
1451 if (sort__has_parent && !*parent &&
1452 symbol__match_regex(al.sym, &parent_regex))
1454 else if (have_ignore_callees && root_al &&
1455 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1456 /* Treat this symbol as the root,
1457 forgetting its callees. */
1459 callchain_cursor_reset(&callchain_cursor);
1463 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1466 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1467 struct addr_location *al)
1470 const struct branch_stack *bs = sample->branch_stack;
1471 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1476 for (i = 0; i < bs->nr; i++) {
1477 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1478 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1479 bi[i].flags = bs->entries[i].flags;
1486 #define NO_ENTRY 0xff
1488 #define PERF_MAX_BRANCH_DEPTH 127
1491 static int remove_loops(struct branch_entry *l, int nr)
1494 unsigned char chash[CHASHSZ];
1496 memset(chash, NO_ENTRY, sizeof(chash));
1498 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1500 for (i = 0; i < nr; i++) {
1501 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1503 /* no collision handling for now */
1504 if (chash[h] == NO_ENTRY) {
1506 } else if (l[chash[h]].from == l[i].from) {
1507 bool is_loop = true;
1508 /* check if it is a real loop */
1510 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1511 if (l[j].from != l[i + off].from) {
1516 memmove(l + i, l + i + off,
1517 (nr - (i + off)) * sizeof(*l));
1526 * Recolve LBR callstack chain sample
1528 * 1 on success get LBR callchain information
1529 * 0 no available LBR callchain information, should try fp
1530 * negative error code on other errors.
1532 static int resolve_lbr_callchain_sample(struct thread *thread,
1533 struct perf_sample *sample,
1534 struct symbol **parent,
1535 struct addr_location *root_al,
1538 struct ip_callchain *chain = sample->callchain;
1539 int chain_nr = min(max_stack, (int)chain->nr);
1540 u8 cpumode = PERF_RECORD_MISC_USER;
1544 for (i = 0; i < chain_nr; i++) {
1545 if (chain->ips[i] == PERF_CONTEXT_USER)
1549 /* LBR only affects the user callchain */
1550 if (i != chain_nr) {
1551 struct branch_stack *lbr_stack = sample->branch_stack;
1552 int lbr_nr = lbr_stack->nr;
1554 * LBR callstack can only get user call chain.
1555 * The mix_chain_nr is kernel call chain
1556 * number plus LBR user call chain number.
1557 * i is kernel call chain number,
1558 * 1 is PERF_CONTEXT_USER,
1559 * lbr_nr + 1 is the user call chain number.
1560 * For details, please refer to the comments
1561 * in callchain__printf
1563 int mix_chain_nr = i + 1 + lbr_nr + 1;
1565 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1566 pr_warning("corrupted callchain. skipping...\n");
1570 for (j = 0; j < mix_chain_nr; j++) {
1571 if (callchain_param.order == ORDER_CALLEE) {
1575 ip = lbr_stack->entries[j - i - 2].from;
1577 ip = lbr_stack->entries[0].to;
1580 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1581 else if (j > lbr_nr)
1582 ip = chain->ips[i + 1 - (j - lbr_nr)];
1584 ip = lbr_stack->entries[0].to;
1587 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1589 return (err < 0) ? err : 0;
1597 static int thread__resolve_callchain_sample(struct thread *thread,
1598 struct perf_evsel *evsel,
1599 struct perf_sample *sample,
1600 struct symbol **parent,
1601 struct addr_location *root_al,
1604 struct branch_stack *branch = sample->branch_stack;
1605 struct ip_callchain *chain = sample->callchain;
1606 int chain_nr = min(max_stack, (int)chain->nr);
1607 u8 cpumode = PERF_RECORD_MISC_USER;
1612 callchain_cursor_reset(&callchain_cursor);
1614 if (has_branch_callstack(evsel)) {
1615 err = resolve_lbr_callchain_sample(thread, sample, parent,
1616 root_al, max_stack);
1618 return (err < 0) ? err : 0;
1622 * Based on DWARF debug information, some architectures skip
1623 * a callchain entry saved by the kernel.
1625 if (chain->nr < PERF_MAX_STACK_DEPTH)
1626 skip_idx = arch_skip_callchain_idx(thread, chain);
1629 * Add branches to call stack for easier browsing. This gives
1630 * more context for a sample than just the callers.
1632 * This uses individual histograms of paths compared to the
1633 * aggregated histograms the normal LBR mode uses.
1635 * Limitations for now:
1636 * - No extra filters
1637 * - No annotations (should annotate somehow)
1640 if (branch && callchain_param.branch_callstack) {
1641 int nr = min(max_stack, (int)branch->nr);
1642 struct branch_entry be[nr];
1644 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1645 pr_warning("corrupted branch chain. skipping...\n");
1649 for (i = 0; i < nr; i++) {
1650 if (callchain_param.order == ORDER_CALLEE) {
1651 be[i] = branch->entries[i];
1653 * Check for overlap into the callchain.
1654 * The return address is one off compared to
1655 * the branch entry. To adjust for this
1656 * assume the calling instruction is not longer
1659 if (i == skip_idx ||
1660 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1662 else if (be[i].from < chain->ips[first_call] &&
1663 be[i].from >= chain->ips[first_call] - 8)
1666 be[i] = branch->entries[branch->nr - i - 1];
1669 nr = remove_loops(be, nr);
1671 for (i = 0; i < nr; i++) {
1672 err = add_callchain_ip(thread, parent, root_al,
1675 err = add_callchain_ip(thread, parent, root_al,
1686 if (chain->nr > PERF_MAX_STACK_DEPTH) {
1687 pr_warning("corrupted callchain. skipping...\n");
1691 for (i = first_call; i < chain_nr; i++) {
1694 if (callchain_param.order == ORDER_CALLEE)
1697 j = chain->nr - i - 1;
1699 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1705 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1708 return (err < 0) ? err : 0;
1714 static int unwind_entry(struct unwind_entry *entry, void *arg)
1716 struct callchain_cursor *cursor = arg;
1717 return callchain_cursor_append(cursor, entry->ip,
1718 entry->map, entry->sym);
1721 int thread__resolve_callchain(struct thread *thread,
1722 struct perf_evsel *evsel,
1723 struct perf_sample *sample,
1724 struct symbol **parent,
1725 struct addr_location *root_al,
1728 int ret = thread__resolve_callchain_sample(thread, evsel,
1730 root_al, max_stack);
1734 /* Can we do dwarf post unwind? */
1735 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1736 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1739 /* Bail out if nothing was captured. */
1740 if ((!sample->user_regs.regs) ||
1741 (!sample->user_stack.size))
1744 return unwind__get_entries(unwind_entry, &callchain_cursor,
1745 thread, sample, max_stack);
1749 int machine__for_each_thread(struct machine *machine,
1750 int (*fn)(struct thread *thread, void *p),
1754 struct thread *thread;
1757 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1758 thread = rb_entry(nd, struct thread, rb_node);
1759 rc = fn(thread, priv);
1764 list_for_each_entry(thread, &machine->dead_threads, node) {
1765 rc = fn(thread, priv);
1772 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1773 struct target *target, struct thread_map *threads,
1774 perf_event__handler_t process, bool data_mmap)
1776 if (target__has_task(target))
1777 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap);
1778 else if (target__has_cpu(target))
1779 return perf_event__synthesize_threads(tool, process, machine, data_mmap);
1780 /* command specified */
1784 pid_t machine__get_current_tid(struct machine *machine, int cpu)
1786 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
1789 return machine->current_tid[cpu];
1792 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
1795 struct thread *thread;
1800 if (!machine->current_tid) {
1803 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
1804 if (!machine->current_tid)
1806 for (i = 0; i < MAX_NR_CPUS; i++)
1807 machine->current_tid[i] = -1;
1810 if (cpu >= MAX_NR_CPUS) {
1811 pr_err("Requested CPU %d too large. ", cpu);
1812 pr_err("Consider raising MAX_NR_CPUS\n");
1816 machine->current_tid[cpu] = tid;
1818 thread = machine__findnew_thread(machine, pid, tid);
1827 int machine__get_kernel_start(struct machine *machine)
1829 struct map *map = machine__kernel_map(machine, MAP__FUNCTION);
1833 * The only addresses above 2^63 are kernel addresses of a 64-bit
1834 * kernel. Note that addresses are unsigned so that on a 32-bit system
1835 * all addresses including kernel addresses are less than 2^32. In
1836 * that case (32-bit system), if the kernel mapping is unknown, all
1837 * addresses will be assumed to be in user space - see
1838 * machine__kernel_ip().
1840 machine->kernel_start = 1ULL << 63;
1842 err = map__load(map, machine->symbol_filter);
1844 machine->kernel_start = map->start;