13 #include <symbol/kallsyms.h>
15 #include "linux/hash.h"
17 static void machine__remove_thread(struct machine *machine, struct thread *th);
19 static void dsos__init(struct dsos *dsos)
21 INIT_LIST_HEAD(&dsos->head);
25 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
27 map_groups__init(&machine->kmaps, machine);
28 RB_CLEAR_NODE(&machine->rb_node);
29 dsos__init(&machine->user_dsos);
30 dsos__init(&machine->kernel_dsos);
32 machine->threads = RB_ROOT;
33 INIT_LIST_HEAD(&machine->dead_threads);
34 machine->last_match = NULL;
36 machine->vdso_info = NULL;
40 machine->symbol_filter = NULL;
41 machine->id_hdr_size = 0;
42 machine->comm_exec = false;
43 machine->kernel_start = 0;
45 machine->root_dir = strdup(root_dir);
46 if (machine->root_dir == NULL)
49 if (pid != HOST_KERNEL_ID) {
50 struct thread *thread = machine__findnew_thread(machine, -1,
57 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
58 thread__set_comm(thread, comm, 0);
61 machine->current_tid = NULL;
66 struct machine *machine__new_host(void)
68 struct machine *machine = malloc(sizeof(*machine));
70 if (machine != NULL) {
71 machine__init(machine, "", HOST_KERNEL_ID);
73 if (machine__create_kernel_maps(machine) < 0)
83 static void dsos__delete(struct dsos *dsos)
87 list_for_each_entry_safe(pos, n, &dsos->head, node) {
88 RB_CLEAR_NODE(&pos->rb_node);
94 void machine__delete_threads(struct machine *machine)
96 struct rb_node *nd = rb_first(&machine->threads);
99 struct thread *t = rb_entry(nd, struct thread, rb_node);
102 machine__remove_thread(machine, t);
106 void machine__exit(struct machine *machine)
108 map_groups__exit(&machine->kmaps);
109 dsos__delete(&machine->user_dsos);
110 dsos__delete(&machine->kernel_dsos);
112 zfree(&machine->root_dir);
113 zfree(&machine->current_tid);
116 void machine__delete(struct machine *machine)
118 machine__exit(machine);
122 void machines__init(struct machines *machines)
124 machine__init(&machines->host, "", HOST_KERNEL_ID);
125 machines->guests = RB_ROOT;
126 machines->symbol_filter = NULL;
129 void machines__exit(struct machines *machines)
131 machine__exit(&machines->host);
135 struct machine *machines__add(struct machines *machines, pid_t pid,
136 const char *root_dir)
138 struct rb_node **p = &machines->guests.rb_node;
139 struct rb_node *parent = NULL;
140 struct machine *pos, *machine = malloc(sizeof(*machine));
145 if (machine__init(machine, root_dir, pid) != 0) {
150 machine->symbol_filter = machines->symbol_filter;
154 pos = rb_entry(parent, struct machine, rb_node);
161 rb_link_node(&machine->rb_node, parent, p);
162 rb_insert_color(&machine->rb_node, &machines->guests);
167 void machines__set_symbol_filter(struct machines *machines,
168 symbol_filter_t symbol_filter)
172 machines->symbol_filter = symbol_filter;
173 machines->host.symbol_filter = symbol_filter;
175 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
176 struct machine *machine = rb_entry(nd, struct machine, rb_node);
178 machine->symbol_filter = symbol_filter;
182 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
186 machines->host.comm_exec = comm_exec;
188 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
189 struct machine *machine = rb_entry(nd, struct machine, rb_node);
191 machine->comm_exec = comm_exec;
195 struct machine *machines__find(struct machines *machines, pid_t pid)
197 struct rb_node **p = &machines->guests.rb_node;
198 struct rb_node *parent = NULL;
199 struct machine *machine;
200 struct machine *default_machine = NULL;
202 if (pid == HOST_KERNEL_ID)
203 return &machines->host;
207 machine = rb_entry(parent, struct machine, rb_node);
208 if (pid < machine->pid)
210 else if (pid > machine->pid)
215 default_machine = machine;
218 return default_machine;
221 struct machine *machines__findnew(struct machines *machines, pid_t pid)
224 const char *root_dir = "";
225 struct machine *machine = machines__find(machines, pid);
227 if (machine && (machine->pid == pid))
230 if ((pid != HOST_KERNEL_ID) &&
231 (pid != DEFAULT_GUEST_KERNEL_ID) &&
232 (symbol_conf.guestmount)) {
233 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
234 if (access(path, R_OK)) {
235 static struct strlist *seen;
238 seen = strlist__new(true, NULL);
240 if (!strlist__has_entry(seen, path)) {
241 pr_err("Can't access file %s\n", path);
242 strlist__add(seen, path);
250 machine = machines__add(machines, pid, root_dir);
255 void machines__process_guests(struct machines *machines,
256 machine__process_t process, void *data)
260 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
261 struct machine *pos = rb_entry(nd, struct machine, rb_node);
266 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
268 if (machine__is_host(machine))
269 snprintf(bf, size, "[%s]", "kernel.kallsyms");
270 else if (machine__is_default_guest(machine))
271 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
273 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
280 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
282 struct rb_node *node;
283 struct machine *machine;
285 machines->host.id_hdr_size = id_hdr_size;
287 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
288 machine = rb_entry(node, struct machine, rb_node);
289 machine->id_hdr_size = id_hdr_size;
295 static void machine__update_thread_pid(struct machine *machine,
296 struct thread *th, pid_t pid)
298 struct thread *leader;
300 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
305 if (th->pid_ == th->tid)
308 leader = machine__findnew_thread(machine, th->pid_, th->pid_);
313 leader->mg = map_groups__new(machine);
318 if (th->mg == leader->mg)
323 * Maps are created from MMAP events which provide the pid and
324 * tid. Consequently there never should be any maps on a thread
325 * with an unknown pid. Just print an error if there are.
327 if (!map_groups__empty(th->mg))
328 pr_err("Discarding thread maps for %d:%d\n",
330 map_groups__delete(th->mg);
333 th->mg = map_groups__get(leader->mg);
338 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
341 static struct thread *__machine__findnew_thread(struct machine *machine,
342 pid_t pid, pid_t tid,
345 struct rb_node **p = &machine->threads.rb_node;
346 struct rb_node *parent = NULL;
350 * Front-end cache - TID lookups come in blocks,
351 * so most of the time we dont have to look up
354 th = machine->last_match;
356 if (th->tid == tid) {
357 machine__update_thread_pid(machine, th, pid);
361 thread__zput(machine->last_match);
366 th = rb_entry(parent, struct thread, rb_node);
368 if (th->tid == tid) {
369 machine->last_match = thread__get(th);
370 machine__update_thread_pid(machine, th, pid);
383 th = thread__new(pid, tid);
385 rb_link_node(&th->rb_node, parent, p);
386 rb_insert_color(&th->rb_node, &machine->threads);
389 * We have to initialize map_groups separately
390 * after rb tree is updated.
392 * The reason is that we call machine__findnew_thread
393 * within thread__init_map_groups to find the thread
394 * leader and that would screwed the rb tree.
396 if (thread__init_map_groups(th, machine)) {
397 rb_erase(&th->rb_node, &machine->threads);
402 * It is now in the rbtree, get a ref
405 machine->last_match = thread__get(th);
411 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
414 return __machine__findnew_thread(machine, pid, tid, true);
417 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
420 return __machine__findnew_thread(machine, pid, tid, false);
423 struct comm *machine__thread_exec_comm(struct machine *machine,
424 struct thread *thread)
426 if (machine->comm_exec)
427 return thread__exec_comm(thread);
429 return thread__comm(thread);
432 int machine__process_comm_event(struct machine *machine, union perf_event *event,
433 struct perf_sample *sample)
435 struct thread *thread = machine__findnew_thread(machine,
438 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
441 machine->comm_exec = true;
444 perf_event__fprintf_comm(event, stdout);
446 if (thread == NULL ||
447 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
448 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
455 int machine__process_lost_event(struct machine *machine __maybe_unused,
456 union perf_event *event, struct perf_sample *sample __maybe_unused)
458 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
459 event->lost.id, event->lost.lost);
463 struct map *machine__new_module(struct machine *machine, u64 start,
464 const char *filename)
467 struct dso *dso = __dsos__findnew(&machine->kernel_dsos, filename);
473 map = map__new2(start, dso, MAP__FUNCTION);
477 if (machine__is_host(machine))
478 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
480 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
482 /* _KMODULE_COMP should be next to _KMODULE */
483 if (is_kernel_module(filename, &compressed) && compressed)
486 map_groups__insert(&machine->kmaps, map);
490 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
493 size_t ret = __dsos__fprintf(&machines->host.kernel_dsos.head, fp) +
494 __dsos__fprintf(&machines->host.user_dsos.head, fp);
496 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
497 struct machine *pos = rb_entry(nd, struct machine, rb_node);
498 ret += __dsos__fprintf(&pos->kernel_dsos.head, fp);
499 ret += __dsos__fprintf(&pos->user_dsos.head, fp);
505 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
506 bool (skip)(struct dso *dso, int parm), int parm)
508 return __dsos__fprintf_buildid(&m->kernel_dsos.head, fp, skip, parm) +
509 __dsos__fprintf_buildid(&m->user_dsos.head, fp, skip, parm);
512 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
513 bool (skip)(struct dso *dso, int parm), int parm)
516 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
518 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
519 struct machine *pos = rb_entry(nd, struct machine, rb_node);
520 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
525 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
529 struct dso *kdso = machine->vmlinux_maps[MAP__FUNCTION]->dso;
531 if (kdso->has_build_id) {
532 char filename[PATH_MAX];
533 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
534 printed += fprintf(fp, "[0] %s\n", filename);
537 for (i = 0; i < vmlinux_path__nr_entries; ++i)
538 printed += fprintf(fp, "[%d] %s\n",
539 i + kdso->has_build_id, vmlinux_path[i]);
544 size_t machine__fprintf(struct machine *machine, FILE *fp)
549 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
550 struct thread *pos = rb_entry(nd, struct thread, rb_node);
552 ret += thread__fprintf(pos, fp);
558 static struct dso *machine__get_kernel(struct machine *machine)
560 const char *vmlinux_name = NULL;
563 if (machine__is_host(machine)) {
564 vmlinux_name = symbol_conf.vmlinux_name;
566 vmlinux_name = "[kernel.kallsyms]";
568 kernel = dso__kernel_findnew(machine, vmlinux_name,
574 if (machine__is_default_guest(machine))
575 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
577 vmlinux_name = machine__mmap_name(machine, bf,
580 kernel = dso__kernel_findnew(machine, vmlinux_name,
582 DSO_TYPE_GUEST_KERNEL);
585 if (kernel != NULL && (!kernel->has_build_id))
586 dso__read_running_kernel_build_id(kernel, machine);
591 struct process_args {
595 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
598 if (machine__is_default_guest(machine))
599 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
601 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
604 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
606 /* Figure out the start address of kernel map from /proc/kallsyms.
607 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
608 * symbol_name if it's not that important.
610 static u64 machine__get_running_kernel_start(struct machine *machine,
611 const char **symbol_name)
613 char filename[PATH_MAX];
618 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
620 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
623 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
624 addr = kallsyms__get_function_start(filename, name);
635 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
638 u64 start = machine__get_running_kernel_start(machine, NULL);
640 for (type = 0; type < MAP__NR_TYPES; ++type) {
643 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
644 if (machine->vmlinux_maps[type] == NULL)
647 machine->vmlinux_maps[type]->map_ip =
648 machine->vmlinux_maps[type]->unmap_ip =
650 kmap = map__kmap(machine->vmlinux_maps[type]);
651 kmap->kmaps = &machine->kmaps;
652 map_groups__insert(&machine->kmaps,
653 machine->vmlinux_maps[type]);
659 void machine__destroy_kernel_maps(struct machine *machine)
663 for (type = 0; type < MAP__NR_TYPES; ++type) {
666 if (machine->vmlinux_maps[type] == NULL)
669 kmap = map__kmap(machine->vmlinux_maps[type]);
670 map_groups__remove(&machine->kmaps,
671 machine->vmlinux_maps[type]);
672 if (kmap->ref_reloc_sym) {
674 * ref_reloc_sym is shared among all maps, so free just
677 if (type == MAP__FUNCTION) {
678 zfree((char **)&kmap->ref_reloc_sym->name);
679 zfree(&kmap->ref_reloc_sym);
681 kmap->ref_reloc_sym = NULL;
684 map__delete(machine->vmlinux_maps[type]);
685 machine->vmlinux_maps[type] = NULL;
689 int machines__create_guest_kernel_maps(struct machines *machines)
692 struct dirent **namelist = NULL;
698 if (symbol_conf.default_guest_vmlinux_name ||
699 symbol_conf.default_guest_modules ||
700 symbol_conf.default_guest_kallsyms) {
701 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
704 if (symbol_conf.guestmount) {
705 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
708 for (i = 0; i < items; i++) {
709 if (!isdigit(namelist[i]->d_name[0])) {
710 /* Filter out . and .. */
713 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
714 if ((*endp != '\0') ||
715 (endp == namelist[i]->d_name) ||
717 pr_debug("invalid directory (%s). Skipping.\n",
718 namelist[i]->d_name);
721 sprintf(path, "%s/%s/proc/kallsyms",
722 symbol_conf.guestmount,
723 namelist[i]->d_name);
724 ret = access(path, R_OK);
726 pr_debug("Can't access file %s\n", path);
729 machines__create_kernel_maps(machines, pid);
738 void machines__destroy_kernel_maps(struct machines *machines)
740 struct rb_node *next = rb_first(&machines->guests);
742 machine__destroy_kernel_maps(&machines->host);
745 struct machine *pos = rb_entry(next, struct machine, rb_node);
747 next = rb_next(&pos->rb_node);
748 rb_erase(&pos->rb_node, &machines->guests);
749 machine__delete(pos);
753 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
755 struct machine *machine = machines__findnew(machines, pid);
760 return machine__create_kernel_maps(machine);
763 int machine__load_kallsyms(struct machine *machine, const char *filename,
764 enum map_type type, symbol_filter_t filter)
766 struct map *map = machine->vmlinux_maps[type];
767 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
770 dso__set_loaded(map->dso, type);
772 * Since /proc/kallsyms will have multiple sessions for the
773 * kernel, with modules between them, fixup the end of all
776 __map_groups__fixup_end(&machine->kmaps, type);
782 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
783 symbol_filter_t filter)
785 struct map *map = machine->vmlinux_maps[type];
786 int ret = dso__load_vmlinux_path(map->dso, map, filter);
789 dso__set_loaded(map->dso, type);
794 static void map_groups__fixup_end(struct map_groups *mg)
797 for (i = 0; i < MAP__NR_TYPES; ++i)
798 __map_groups__fixup_end(mg, i);
801 static char *get_kernel_version(const char *root_dir)
803 char version[PATH_MAX];
806 const char *prefix = "Linux version ";
808 sprintf(version, "%s/proc/version", root_dir);
809 file = fopen(version, "r");
814 tmp = fgets(version, sizeof(version), file);
817 name = strstr(version, prefix);
820 name += strlen(prefix);
821 tmp = strchr(name, ' ');
828 static int map_groups__set_modules_path_dir(struct map_groups *mg,
829 const char *dir_name, int depth)
832 DIR *dir = opendir(dir_name);
836 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
840 while ((dent = readdir(dir)) != NULL) {
844 /*sshfs might return bad dent->d_type, so we have to stat*/
845 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
849 if (S_ISDIR(st.st_mode)) {
850 if (!strcmp(dent->d_name, ".") ||
851 !strcmp(dent->d_name, ".."))
854 /* Do not follow top-level source and build symlinks */
856 if (!strcmp(dent->d_name, "source") ||
857 !strcmp(dent->d_name, "build"))
861 ret = map_groups__set_modules_path_dir(mg, path,
866 char *dot = strrchr(dent->d_name, '.'),
874 /* On some system, modules are compressed like .ko.gz */
875 if (is_supported_compression(dot + 1) &&
876 is_kmodule_extension(dot - 2))
879 snprintf(dso_name, sizeof(dso_name), "[%.*s]",
880 (int)(dot - dent->d_name), dent->d_name);
882 strxfrchar(dso_name, '-', '_');
883 map = map_groups__find_by_name(mg, MAP__FUNCTION,
888 long_name = strdup(path);
889 if (long_name == NULL) {
893 dso__set_long_name(map->dso, long_name, true);
894 dso__kernel_module_get_build_id(map->dso, "");
903 static int machine__set_modules_path(struct machine *machine)
906 char modules_path[PATH_MAX];
908 version = get_kernel_version(machine->root_dir);
912 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
913 machine->root_dir, version);
916 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
919 static int machine__create_module(void *arg, const char *name, u64 start)
921 struct machine *machine = arg;
924 map = machine__new_module(machine, start, name);
928 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
933 static int machine__create_modules(struct machine *machine)
938 if (machine__is_default_guest(machine)) {
939 modules = symbol_conf.default_guest_modules;
941 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
945 if (symbol__restricted_filename(modules, "/proc/modules"))
948 if (modules__parse(modules, machine, machine__create_module))
951 if (!machine__set_modules_path(machine))
954 pr_debug("Problems setting modules path maps, continuing anyway...\n");
959 int machine__create_kernel_maps(struct machine *machine)
961 struct dso *kernel = machine__get_kernel(machine);
963 u64 addr = machine__get_running_kernel_start(machine, &name);
967 if (kernel == NULL ||
968 __machine__create_kernel_maps(machine, kernel) < 0)
971 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
972 if (machine__is_host(machine))
973 pr_debug("Problems creating module maps, "
974 "continuing anyway...\n");
976 pr_debug("Problems creating module maps for guest %d, "
977 "continuing anyway...\n", machine->pid);
981 * Now that we have all the maps created, just set the ->end of them:
983 map_groups__fixup_end(&machine->kmaps);
985 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
987 machine__destroy_kernel_maps(machine);
994 static void machine__set_kernel_mmap_len(struct machine *machine,
995 union perf_event *event)
999 for (i = 0; i < MAP__NR_TYPES; i++) {
1000 machine->vmlinux_maps[i]->start = event->mmap.start;
1001 machine->vmlinux_maps[i]->end = (event->mmap.start +
1004 * Be a bit paranoid here, some perf.data file came with
1005 * a zero sized synthesized MMAP event for the kernel.
1007 if (machine->vmlinux_maps[i]->end == 0)
1008 machine->vmlinux_maps[i]->end = ~0ULL;
1012 static bool machine__uses_kcore(struct machine *machine)
1016 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1017 if (dso__is_kcore(dso))
1024 static int machine__process_kernel_mmap_event(struct machine *machine,
1025 union perf_event *event)
1028 char kmmap_prefix[PATH_MAX];
1029 enum dso_kernel_type kernel_type;
1030 bool is_kernel_mmap;
1032 /* If we have maps from kcore then we do not need or want any others */
1033 if (machine__uses_kcore(machine))
1036 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1037 if (machine__is_host(machine))
1038 kernel_type = DSO_TYPE_KERNEL;
1040 kernel_type = DSO_TYPE_GUEST_KERNEL;
1042 is_kernel_mmap = memcmp(event->mmap.filename,
1044 strlen(kmmap_prefix) - 1) == 0;
1045 if (event->mmap.filename[0] == '/' ||
1046 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1048 char short_module_name[1024];
1051 if (event->mmap.filename[0] == '/') {
1052 name = strrchr(event->mmap.filename, '/');
1056 ++name; /* skip / */
1057 dot = strrchr(name, '.');
1060 /* On some system, modules are compressed like .ko.gz */
1061 if (is_supported_compression(dot + 1))
1063 if (!is_kmodule_extension(dot + 1))
1065 snprintf(short_module_name, sizeof(short_module_name),
1066 "[%.*s]", (int)(dot - name), name);
1067 strxfrchar(short_module_name, '-', '_');
1069 strcpy(short_module_name, event->mmap.filename);
1071 map = machine__new_module(machine, event->mmap.start,
1072 event->mmap.filename);
1076 name = strdup(short_module_name);
1080 dso__set_short_name(map->dso, name, true);
1081 map->end = map->start + event->mmap.len;
1082 } else if (is_kernel_mmap) {
1083 const char *symbol_name = (event->mmap.filename +
1084 strlen(kmmap_prefix));
1086 * Should be there already, from the build-id table in
1089 struct dso *kernel = NULL;
1092 list_for_each_entry(dso, &machine->kernel_dsos.head, node) {
1093 if (is_kernel_module(dso->long_name, NULL))
1101 kernel = __dsos__findnew(&machine->kernel_dsos,
1106 kernel->kernel = kernel_type;
1107 if (__machine__create_kernel_maps(machine, kernel) < 0)
1110 if (strstr(kernel->long_name, "vmlinux"))
1111 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1113 machine__set_kernel_mmap_len(machine, event);
1116 * Avoid using a zero address (kptr_restrict) for the ref reloc
1117 * symbol. Effectively having zero here means that at record
1118 * time /proc/sys/kernel/kptr_restrict was non zero.
1120 if (event->mmap.pgoff != 0) {
1121 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1126 if (machine__is_default_guest(machine)) {
1128 * preload dso of guest kernel and modules
1130 dso__load(kernel, machine->vmlinux_maps[MAP__FUNCTION],
1139 int machine__process_mmap2_event(struct machine *machine,
1140 union perf_event *event,
1141 struct perf_sample *sample __maybe_unused)
1143 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1144 struct thread *thread;
1150 perf_event__fprintf_mmap2(event, stdout);
1152 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1153 cpumode == PERF_RECORD_MISC_KERNEL) {
1154 ret = machine__process_kernel_mmap_event(machine, event);
1160 thread = machine__findnew_thread(machine, event->mmap2.pid,
1165 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1166 type = MAP__VARIABLE;
1168 type = MAP__FUNCTION;
1170 map = map__new(machine, event->mmap2.start,
1171 event->mmap2.len, event->mmap2.pgoff,
1172 event->mmap2.pid, event->mmap2.maj,
1173 event->mmap2.min, event->mmap2.ino,
1174 event->mmap2.ino_generation,
1177 event->mmap2.filename, type, thread);
1182 thread__insert_map(thread, map);
1186 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1190 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1191 struct perf_sample *sample __maybe_unused)
1193 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1194 struct thread *thread;
1200 perf_event__fprintf_mmap(event, stdout);
1202 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1203 cpumode == PERF_RECORD_MISC_KERNEL) {
1204 ret = machine__process_kernel_mmap_event(machine, event);
1210 thread = machine__findnew_thread(machine, event->mmap.pid,
1215 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1216 type = MAP__VARIABLE;
1218 type = MAP__FUNCTION;
1220 map = map__new(machine, event->mmap.start,
1221 event->mmap.len, event->mmap.pgoff,
1222 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1223 event->mmap.filename,
1229 thread__insert_map(thread, map);
1233 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1237 static void machine__remove_thread(struct machine *machine, struct thread *th)
1239 if (machine->last_match == th)
1240 thread__zput(machine->last_match);
1242 rb_erase(&th->rb_node, &machine->threads);
1244 * Move it first to the dead_threads list, then drop the reference,
1245 * if this is the last reference, then the thread__delete destructor
1246 * will be called and we will remove it from the dead_threads list.
1248 list_add_tail(&th->node, &machine->dead_threads);
1252 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1253 struct perf_sample *sample)
1255 struct thread *thread = machine__find_thread(machine,
1258 struct thread *parent = machine__findnew_thread(machine,
1262 /* if a thread currently exists for the thread id remove it */
1264 machine__remove_thread(machine, thread);
1266 thread = machine__findnew_thread(machine, event->fork.pid,
1269 perf_event__fprintf_task(event, stdout);
1271 if (thread == NULL || parent == NULL ||
1272 thread__fork(thread, parent, sample->time) < 0) {
1273 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1280 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1281 struct perf_sample *sample __maybe_unused)
1283 struct thread *thread = machine__find_thread(machine,
1288 perf_event__fprintf_task(event, stdout);
1291 thread__exited(thread);
1296 int machine__process_event(struct machine *machine, union perf_event *event,
1297 struct perf_sample *sample)
1301 switch (event->header.type) {
1302 case PERF_RECORD_COMM:
1303 ret = machine__process_comm_event(machine, event, sample); break;
1304 case PERF_RECORD_MMAP:
1305 ret = machine__process_mmap_event(machine, event, sample); break;
1306 case PERF_RECORD_MMAP2:
1307 ret = machine__process_mmap2_event(machine, event, sample); break;
1308 case PERF_RECORD_FORK:
1309 ret = machine__process_fork_event(machine, event, sample); break;
1310 case PERF_RECORD_EXIT:
1311 ret = machine__process_exit_event(machine, event, sample); break;
1312 case PERF_RECORD_LOST:
1313 ret = machine__process_lost_event(machine, event, sample); break;
1322 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1324 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1329 static void ip__resolve_ams(struct thread *thread,
1330 struct addr_map_symbol *ams,
1333 struct addr_location al;
1335 memset(&al, 0, sizeof(al));
1337 * We cannot use the header.misc hint to determine whether a
1338 * branch stack address is user, kernel, guest, hypervisor.
1339 * Branches may straddle the kernel/user/hypervisor boundaries.
1340 * Thus, we have to try consecutively until we find a match
1341 * or else, the symbol is unknown
1343 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1346 ams->al_addr = al.addr;
1351 static void ip__resolve_data(struct thread *thread,
1352 u8 m, struct addr_map_symbol *ams, u64 addr)
1354 struct addr_location al;
1356 memset(&al, 0, sizeof(al));
1358 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1359 if (al.map == NULL) {
1361 * some shared data regions have execute bit set which puts
1362 * their mapping in the MAP__FUNCTION type array.
1363 * Check there as a fallback option before dropping the sample.
1365 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1369 ams->al_addr = al.addr;
1374 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1375 struct addr_location *al)
1377 struct mem_info *mi = zalloc(sizeof(*mi));
1382 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1383 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1384 mi->data_src.val = sample->data_src;
1389 static int add_callchain_ip(struct thread *thread,
1390 struct symbol **parent,
1391 struct addr_location *root_al,
1392 bool branch_history,
1395 struct addr_location al;
1400 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1403 u8 cpumode = PERF_RECORD_MISC_USER;
1405 if (ip >= PERF_CONTEXT_MAX) {
1407 case PERF_CONTEXT_HV:
1408 cpumode = PERF_RECORD_MISC_HYPERVISOR;
1410 case PERF_CONTEXT_KERNEL:
1411 cpumode = PERF_RECORD_MISC_KERNEL;
1413 case PERF_CONTEXT_USER:
1414 cpumode = PERF_RECORD_MISC_USER;
1417 pr_debug("invalid callchain context: "
1418 "%"PRId64"\n", (s64) ip);
1420 * It seems the callchain is corrupted.
1423 callchain_cursor_reset(&callchain_cursor);
1428 thread__find_addr_location(thread, cpumode, MAP__FUNCTION,
1432 if (al.sym != NULL) {
1433 if (sort__has_parent && !*parent &&
1434 symbol__match_regex(al.sym, &parent_regex))
1436 else if (have_ignore_callees && root_al &&
1437 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1438 /* Treat this symbol as the root,
1439 forgetting its callees. */
1441 callchain_cursor_reset(&callchain_cursor);
1445 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1448 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1449 struct addr_location *al)
1452 const struct branch_stack *bs = sample->branch_stack;
1453 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1458 for (i = 0; i < bs->nr; i++) {
1459 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1460 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1461 bi[i].flags = bs->entries[i].flags;
1468 #define NO_ENTRY 0xff
1470 #define PERF_MAX_BRANCH_DEPTH 127
1473 static int remove_loops(struct branch_entry *l, int nr)
1476 unsigned char chash[CHASHSZ];
1478 memset(chash, NO_ENTRY, sizeof(chash));
1480 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1482 for (i = 0; i < nr; i++) {
1483 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1485 /* no collision handling for now */
1486 if (chash[h] == NO_ENTRY) {
1488 } else if (l[chash[h]].from == l[i].from) {
1489 bool is_loop = true;
1490 /* check if it is a real loop */
1492 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1493 if (l[j].from != l[i + off].from) {
1498 memmove(l + i, l + i + off,
1499 (nr - (i + off)) * sizeof(*l));
1508 * Recolve LBR callstack chain sample
1510 * 1 on success get LBR callchain information
1511 * 0 no available LBR callchain information, should try fp
1512 * negative error code on other errors.
1514 static int resolve_lbr_callchain_sample(struct thread *thread,
1515 struct perf_sample *sample,
1516 struct symbol **parent,
1517 struct addr_location *root_al,
1520 struct ip_callchain *chain = sample->callchain;
1521 int chain_nr = min(max_stack, (int)chain->nr);
1525 for (i = 0; i < chain_nr; i++) {
1526 if (chain->ips[i] == PERF_CONTEXT_USER)
1530 /* LBR only affects the user callchain */
1531 if (i != chain_nr) {
1532 struct branch_stack *lbr_stack = sample->branch_stack;
1533 int lbr_nr = lbr_stack->nr;
1535 * LBR callstack can only get user call chain.
1536 * The mix_chain_nr is kernel call chain
1537 * number plus LBR user call chain number.
1538 * i is kernel call chain number,
1539 * 1 is PERF_CONTEXT_USER,
1540 * lbr_nr + 1 is the user call chain number.
1541 * For details, please refer to the comments
1542 * in callchain__printf
1544 int mix_chain_nr = i + 1 + lbr_nr + 1;
1546 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1547 pr_warning("corrupted callchain. skipping...\n");
1551 for (j = 0; j < mix_chain_nr; j++) {
1552 if (callchain_param.order == ORDER_CALLEE) {
1556 ip = lbr_stack->entries[j - i - 2].from;
1558 ip = lbr_stack->entries[0].to;
1561 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1562 else if (j > lbr_nr)
1563 ip = chain->ips[i + 1 - (j - lbr_nr)];
1565 ip = lbr_stack->entries[0].to;
1568 err = add_callchain_ip(thread, parent, root_al, false, ip);
1570 return (err < 0) ? err : 0;
1578 static int thread__resolve_callchain_sample(struct thread *thread,
1579 struct perf_evsel *evsel,
1580 struct perf_sample *sample,
1581 struct symbol **parent,
1582 struct addr_location *root_al,
1585 struct branch_stack *branch = sample->branch_stack;
1586 struct ip_callchain *chain = sample->callchain;
1587 int chain_nr = min(max_stack, (int)chain->nr);
1592 callchain_cursor_reset(&callchain_cursor);
1594 if (has_branch_callstack(evsel)) {
1595 err = resolve_lbr_callchain_sample(thread, sample, parent,
1596 root_al, max_stack);
1598 return (err < 0) ? err : 0;
1602 * Based on DWARF debug information, some architectures skip
1603 * a callchain entry saved by the kernel.
1605 if (chain->nr < PERF_MAX_STACK_DEPTH)
1606 skip_idx = arch_skip_callchain_idx(thread, chain);
1609 * Add branches to call stack for easier browsing. This gives
1610 * more context for a sample than just the callers.
1612 * This uses individual histograms of paths compared to the
1613 * aggregated histograms the normal LBR mode uses.
1615 * Limitations for now:
1616 * - No extra filters
1617 * - No annotations (should annotate somehow)
1620 if (branch && callchain_param.branch_callstack) {
1621 int nr = min(max_stack, (int)branch->nr);
1622 struct branch_entry be[nr];
1624 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1625 pr_warning("corrupted branch chain. skipping...\n");
1629 for (i = 0; i < nr; i++) {
1630 if (callchain_param.order == ORDER_CALLEE) {
1631 be[i] = branch->entries[i];
1633 * Check for overlap into the callchain.
1634 * The return address is one off compared to
1635 * the branch entry. To adjust for this
1636 * assume the calling instruction is not longer
1639 if (i == skip_idx ||
1640 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1642 else if (be[i].from < chain->ips[first_call] &&
1643 be[i].from >= chain->ips[first_call] - 8)
1646 be[i] = branch->entries[branch->nr - i - 1];
1649 nr = remove_loops(be, nr);
1651 for (i = 0; i < nr; i++) {
1652 err = add_callchain_ip(thread, parent, root_al,
1655 err = add_callchain_ip(thread, parent, root_al,
1666 if (chain->nr > PERF_MAX_STACK_DEPTH) {
1667 pr_warning("corrupted callchain. skipping...\n");
1671 for (i = first_call; i < chain_nr; i++) {
1674 if (callchain_param.order == ORDER_CALLEE)
1677 j = chain->nr - i - 1;
1679 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1685 err = add_callchain_ip(thread, parent, root_al, false, ip);
1688 return (err < 0) ? err : 0;
1694 static int unwind_entry(struct unwind_entry *entry, void *arg)
1696 struct callchain_cursor *cursor = arg;
1697 return callchain_cursor_append(cursor, entry->ip,
1698 entry->map, entry->sym);
1701 int thread__resolve_callchain(struct thread *thread,
1702 struct perf_evsel *evsel,
1703 struct perf_sample *sample,
1704 struct symbol **parent,
1705 struct addr_location *root_al,
1708 int ret = thread__resolve_callchain_sample(thread, evsel,
1710 root_al, max_stack);
1714 /* Can we do dwarf post unwind? */
1715 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1716 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1719 /* Bail out if nothing was captured. */
1720 if ((!sample->user_regs.regs) ||
1721 (!sample->user_stack.size))
1724 return unwind__get_entries(unwind_entry, &callchain_cursor,
1725 thread, sample, max_stack);
1729 int machine__for_each_thread(struct machine *machine,
1730 int (*fn)(struct thread *thread, void *p),
1734 struct thread *thread;
1737 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1738 thread = rb_entry(nd, struct thread, rb_node);
1739 rc = fn(thread, priv);
1744 list_for_each_entry(thread, &machine->dead_threads, node) {
1745 rc = fn(thread, priv);
1752 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1753 struct target *target, struct thread_map *threads,
1754 perf_event__handler_t process, bool data_mmap)
1756 if (target__has_task(target))
1757 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap);
1758 else if (target__has_cpu(target))
1759 return perf_event__synthesize_threads(tool, process, machine, data_mmap);
1760 /* command specified */
1764 pid_t machine__get_current_tid(struct machine *machine, int cpu)
1766 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
1769 return machine->current_tid[cpu];
1772 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
1775 struct thread *thread;
1780 if (!machine->current_tid) {
1783 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
1784 if (!machine->current_tid)
1786 for (i = 0; i < MAX_NR_CPUS; i++)
1787 machine->current_tid[i] = -1;
1790 if (cpu >= MAX_NR_CPUS) {
1791 pr_err("Requested CPU %d too large. ", cpu);
1792 pr_err("Consider raising MAX_NR_CPUS\n");
1796 machine->current_tid[cpu] = tid;
1798 thread = machine__findnew_thread(machine, pid, tid);
1807 int machine__get_kernel_start(struct machine *machine)
1809 struct map *map = machine__kernel_map(machine, MAP__FUNCTION);
1813 * The only addresses above 2^63 are kernel addresses of a 64-bit
1814 * kernel. Note that addresses are unsigned so that on a 32-bit system
1815 * all addresses including kernel addresses are less than 2^32. In
1816 * that case (32-bit system), if the kernel mapping is unknown, all
1817 * addresses will be assumed to be in user space - see
1818 * machine__kernel_ip().
1820 machine->kernel_start = 1ULL << 63;
1822 err = map__load(map, machine->symbol_filter);
1824 machine->kernel_start = map->start;