Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux/fpc-iii.git] / tools / perf / util / machine.c
blob80b9b6a87990b29e4722a956e62814130c1290de
1 #include "callchain.h"
2 #include "debug.h"
3 #include "event.h"
4 #include "evsel.h"
5 #include "hist.h"
6 #include "machine.h"
7 #include "map.h"
8 #include "sort.h"
9 #include "strlist.h"
10 #include "thread.h"
11 #include "vdso.h"
12 #include <stdbool.h>
13 #include <symbol/kallsyms.h>
14 #include "unwind.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);
22 dsos->root = RB_ROOT;
23 pthread_rwlock_init(&dsos->lock, NULL);
26 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
28 memset(machine, 0, sizeof(*machine));
29 map_groups__init(&machine->kmaps, machine);
30 RB_CLEAR_NODE(&machine->rb_node);
31 dsos__init(&machine->dsos);
33 machine->threads = RB_ROOT;
34 pthread_rwlock_init(&machine->threads_lock, NULL);
35 INIT_LIST_HEAD(&machine->dead_threads);
36 machine->last_match = NULL;
38 machine->vdso_info = NULL;
39 machine->env = NULL;
41 machine->pid = pid;
43 machine->symbol_filter = NULL;
44 machine->id_hdr_size = 0;
45 machine->comm_exec = false;
46 machine->kernel_start = 0;
48 memset(machine->vmlinux_maps, 0, sizeof(machine->vmlinux_maps));
50 machine->root_dir = strdup(root_dir);
51 if (machine->root_dir == NULL)
52 return -ENOMEM;
54 if (pid != HOST_KERNEL_ID) {
55 struct thread *thread = machine__findnew_thread(machine, -1,
56 pid);
57 char comm[64];
59 if (thread == NULL)
60 return -ENOMEM;
62 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
63 thread__set_comm(thread, comm, 0);
64 thread__put(thread);
67 machine->current_tid = NULL;
69 return 0;
72 struct machine *machine__new_host(void)
74 struct machine *machine = malloc(sizeof(*machine));
76 if (machine != NULL) {
77 machine__init(machine, "", HOST_KERNEL_ID);
79 if (machine__create_kernel_maps(machine) < 0)
80 goto out_delete;
83 return machine;
84 out_delete:
85 free(machine);
86 return NULL;
89 static void dsos__purge(struct dsos *dsos)
91 struct dso *pos, *n;
93 pthread_rwlock_wrlock(&dsos->lock);
95 list_for_each_entry_safe(pos, n, &dsos->head, node) {
96 RB_CLEAR_NODE(&pos->rb_node);
97 pos->root = NULL;
98 list_del_init(&pos->node);
99 dso__put(pos);
102 pthread_rwlock_unlock(&dsos->lock);
105 static void dsos__exit(struct dsos *dsos)
107 dsos__purge(dsos);
108 pthread_rwlock_destroy(&dsos->lock);
111 void machine__delete_threads(struct machine *machine)
113 struct rb_node *nd;
115 pthread_rwlock_wrlock(&machine->threads_lock);
116 nd = rb_first(&machine->threads);
117 while (nd) {
118 struct thread *t = rb_entry(nd, struct thread, rb_node);
120 nd = rb_next(nd);
121 __machine__remove_thread(machine, t, false);
123 pthread_rwlock_unlock(&machine->threads_lock);
126 void machine__exit(struct machine *machine)
128 machine__destroy_kernel_maps(machine);
129 map_groups__exit(&machine->kmaps);
130 dsos__exit(&machine->dsos);
131 machine__exit_vdso(machine);
132 zfree(&machine->root_dir);
133 zfree(&machine->current_tid);
134 pthread_rwlock_destroy(&machine->threads_lock);
137 void machine__delete(struct machine *machine)
139 machine__exit(machine);
140 free(machine);
143 void machines__init(struct machines *machines)
145 machine__init(&machines->host, "", HOST_KERNEL_ID);
146 machines->guests = RB_ROOT;
147 machines->symbol_filter = NULL;
150 void machines__exit(struct machines *machines)
152 machine__exit(&machines->host);
153 /* XXX exit guest */
156 struct machine *machines__add(struct machines *machines, pid_t pid,
157 const char *root_dir)
159 struct rb_node **p = &machines->guests.rb_node;
160 struct rb_node *parent = NULL;
161 struct machine *pos, *machine = malloc(sizeof(*machine));
163 if (machine == NULL)
164 return NULL;
166 if (machine__init(machine, root_dir, pid) != 0) {
167 free(machine);
168 return NULL;
171 machine->symbol_filter = machines->symbol_filter;
173 while (*p != NULL) {
174 parent = *p;
175 pos = rb_entry(parent, struct machine, rb_node);
176 if (pid < pos->pid)
177 p = &(*p)->rb_left;
178 else
179 p = &(*p)->rb_right;
182 rb_link_node(&machine->rb_node, parent, p);
183 rb_insert_color(&machine->rb_node, &machines->guests);
185 return machine;
188 void machines__set_symbol_filter(struct machines *machines,
189 symbol_filter_t symbol_filter)
191 struct rb_node *nd;
193 machines->symbol_filter = symbol_filter;
194 machines->host.symbol_filter = symbol_filter;
196 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
197 struct machine *machine = rb_entry(nd, struct machine, rb_node);
199 machine->symbol_filter = symbol_filter;
203 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
205 struct rb_node *nd;
207 machines->host.comm_exec = comm_exec;
209 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
210 struct machine *machine = rb_entry(nd, struct machine, rb_node);
212 machine->comm_exec = comm_exec;
216 struct machine *machines__find(struct machines *machines, pid_t pid)
218 struct rb_node **p = &machines->guests.rb_node;
219 struct rb_node *parent = NULL;
220 struct machine *machine;
221 struct machine *default_machine = NULL;
223 if (pid == HOST_KERNEL_ID)
224 return &machines->host;
226 while (*p != NULL) {
227 parent = *p;
228 machine = rb_entry(parent, struct machine, rb_node);
229 if (pid < machine->pid)
230 p = &(*p)->rb_left;
231 else if (pid > machine->pid)
232 p = &(*p)->rb_right;
233 else
234 return machine;
235 if (!machine->pid)
236 default_machine = machine;
239 return default_machine;
242 struct machine *machines__findnew(struct machines *machines, pid_t pid)
244 char path[PATH_MAX];
245 const char *root_dir = "";
246 struct machine *machine = machines__find(machines, pid);
248 if (machine && (machine->pid == pid))
249 goto out;
251 if ((pid != HOST_KERNEL_ID) &&
252 (pid != DEFAULT_GUEST_KERNEL_ID) &&
253 (symbol_conf.guestmount)) {
254 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
255 if (access(path, R_OK)) {
256 static struct strlist *seen;
258 if (!seen)
259 seen = strlist__new(NULL, NULL);
261 if (!strlist__has_entry(seen, path)) {
262 pr_err("Can't access file %s\n", path);
263 strlist__add(seen, path);
265 machine = NULL;
266 goto out;
268 root_dir = path;
271 machine = machines__add(machines, pid, root_dir);
272 out:
273 return machine;
276 void machines__process_guests(struct machines *machines,
277 machine__process_t process, void *data)
279 struct rb_node *nd;
281 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
282 struct machine *pos = rb_entry(nd, struct machine, rb_node);
283 process(pos, data);
287 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
289 if (machine__is_host(machine))
290 snprintf(bf, size, "[%s]", "kernel.kallsyms");
291 else if (machine__is_default_guest(machine))
292 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
293 else {
294 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
295 machine->pid);
298 return bf;
301 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
303 struct rb_node *node;
304 struct machine *machine;
306 machines->host.id_hdr_size = id_hdr_size;
308 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
309 machine = rb_entry(node, struct machine, rb_node);
310 machine->id_hdr_size = id_hdr_size;
313 return;
316 static void machine__update_thread_pid(struct machine *machine,
317 struct thread *th, pid_t pid)
319 struct thread *leader;
321 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
322 return;
324 th->pid_ = pid;
326 if (th->pid_ == th->tid)
327 return;
329 leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
330 if (!leader)
331 goto out_err;
333 if (!leader->mg)
334 leader->mg = map_groups__new(machine);
336 if (!leader->mg)
337 goto out_err;
339 if (th->mg == leader->mg)
340 return;
342 if (th->mg) {
344 * Maps are created from MMAP events which provide the pid and
345 * tid. Consequently there never should be any maps on a thread
346 * with an unknown pid. Just print an error if there are.
348 if (!map_groups__empty(th->mg))
349 pr_err("Discarding thread maps for %d:%d\n",
350 th->pid_, th->tid);
351 map_groups__put(th->mg);
354 th->mg = map_groups__get(leader->mg);
355 out_put:
356 thread__put(leader);
357 return;
358 out_err:
359 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
360 goto out_put;
364 * Caller must eventually drop thread->refcnt returned with a successfull
365 * lookup/new thread inserted.
367 static struct thread *____machine__findnew_thread(struct machine *machine,
368 pid_t pid, pid_t tid,
369 bool create)
371 struct rb_node **p = &machine->threads.rb_node;
372 struct rb_node *parent = NULL;
373 struct thread *th;
376 * Front-end cache - TID lookups come in blocks,
377 * so most of the time we dont have to look up
378 * the full rbtree:
380 th = machine->last_match;
381 if (th != NULL) {
382 if (th->tid == tid) {
383 machine__update_thread_pid(machine, th, pid);
384 return thread__get(th);
387 machine->last_match = NULL;
390 while (*p != NULL) {
391 parent = *p;
392 th = rb_entry(parent, struct thread, rb_node);
394 if (th->tid == tid) {
395 machine->last_match = th;
396 machine__update_thread_pid(machine, th, pid);
397 return thread__get(th);
400 if (tid < th->tid)
401 p = &(*p)->rb_left;
402 else
403 p = &(*p)->rb_right;
406 if (!create)
407 return NULL;
409 th = thread__new(pid, tid);
410 if (th != NULL) {
411 rb_link_node(&th->rb_node, parent, p);
412 rb_insert_color(&th->rb_node, &machine->threads);
415 * We have to initialize map_groups separately
416 * after rb tree is updated.
418 * The reason is that we call machine__findnew_thread
419 * within thread__init_map_groups to find the thread
420 * leader and that would screwed the rb tree.
422 if (thread__init_map_groups(th, machine)) {
423 rb_erase_init(&th->rb_node, &machine->threads);
424 RB_CLEAR_NODE(&th->rb_node);
425 thread__put(th);
426 return NULL;
429 * It is now in the rbtree, get a ref
431 thread__get(th);
432 machine->last_match = th;
435 return th;
438 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
440 return ____machine__findnew_thread(machine, pid, tid, true);
443 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
444 pid_t tid)
446 struct thread *th;
448 pthread_rwlock_wrlock(&machine->threads_lock);
449 th = __machine__findnew_thread(machine, pid, tid);
450 pthread_rwlock_unlock(&machine->threads_lock);
451 return th;
454 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
455 pid_t tid)
457 struct thread *th;
458 pthread_rwlock_rdlock(&machine->threads_lock);
459 th = ____machine__findnew_thread(machine, pid, tid, false);
460 pthread_rwlock_unlock(&machine->threads_lock);
461 return th;
464 struct comm *machine__thread_exec_comm(struct machine *machine,
465 struct thread *thread)
467 if (machine->comm_exec)
468 return thread__exec_comm(thread);
469 else
470 return thread__comm(thread);
473 int machine__process_comm_event(struct machine *machine, union perf_event *event,
474 struct perf_sample *sample)
476 struct thread *thread = machine__findnew_thread(machine,
477 event->comm.pid,
478 event->comm.tid);
479 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
480 int err = 0;
482 if (exec)
483 machine->comm_exec = true;
485 if (dump_trace)
486 perf_event__fprintf_comm(event, stdout);
488 if (thread == NULL ||
489 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
490 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
491 err = -1;
494 thread__put(thread);
496 return err;
499 int machine__process_lost_event(struct machine *machine __maybe_unused,
500 union perf_event *event, struct perf_sample *sample __maybe_unused)
502 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
503 event->lost.id, event->lost.lost);
504 return 0;
507 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
508 union perf_event *event, struct perf_sample *sample)
510 dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
511 sample->id, event->lost_samples.lost);
512 return 0;
515 static struct dso *machine__findnew_module_dso(struct machine *machine,
516 struct kmod_path *m,
517 const char *filename)
519 struct dso *dso;
521 pthread_rwlock_wrlock(&machine->dsos.lock);
523 dso = __dsos__find(&machine->dsos, m->name, true);
524 if (!dso) {
525 dso = __dsos__addnew(&machine->dsos, m->name);
526 if (dso == NULL)
527 goto out_unlock;
529 if (machine__is_host(machine))
530 dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
531 else
532 dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
534 /* _KMODULE_COMP should be next to _KMODULE */
535 if (m->kmod && m->comp)
536 dso->symtab_type++;
538 dso__set_short_name(dso, strdup(m->name), true);
539 dso__set_long_name(dso, strdup(filename), true);
542 dso__get(dso);
543 out_unlock:
544 pthread_rwlock_unlock(&machine->dsos.lock);
545 return dso;
548 int machine__process_aux_event(struct machine *machine __maybe_unused,
549 union perf_event *event)
551 if (dump_trace)
552 perf_event__fprintf_aux(event, stdout);
553 return 0;
556 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
557 union perf_event *event)
559 if (dump_trace)
560 perf_event__fprintf_itrace_start(event, stdout);
561 return 0;
564 int machine__process_switch_event(struct machine *machine __maybe_unused,
565 union perf_event *event)
567 if (dump_trace)
568 perf_event__fprintf_switch(event, stdout);
569 return 0;
572 static void dso__adjust_kmod_long_name(struct dso *dso, const char *filename)
574 const char *dup_filename;
576 if (!filename || !dso || !dso->long_name)
577 return;
578 if (dso->long_name[0] != '[')
579 return;
580 if (!strchr(filename, '/'))
581 return;
583 dup_filename = strdup(filename);
584 if (!dup_filename)
585 return;
587 dso__set_long_name(dso, dup_filename, true);
590 struct map *machine__findnew_module_map(struct machine *machine, u64 start,
591 const char *filename)
593 struct map *map = NULL;
594 struct dso *dso = NULL;
595 struct kmod_path m;
597 if (kmod_path__parse_name(&m, filename))
598 return NULL;
600 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
601 m.name);
602 if (map) {
604 * If the map's dso is an offline module, give dso__load()
605 * a chance to find the file path of that module by fixing
606 * long_name.
608 dso__adjust_kmod_long_name(map->dso, filename);
609 goto out;
612 dso = machine__findnew_module_dso(machine, &m, filename);
613 if (dso == NULL)
614 goto out;
616 map = map__new2(start, dso, MAP__FUNCTION);
617 if (map == NULL)
618 goto out;
620 map_groups__insert(&machine->kmaps, map);
622 /* Put the map here because map_groups__insert alread got it */
623 map__put(map);
624 out:
625 /* put the dso here, corresponding to machine__findnew_module_dso */
626 dso__put(dso);
627 free(m.name);
628 return map;
631 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
633 struct rb_node *nd;
634 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
636 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
637 struct machine *pos = rb_entry(nd, struct machine, rb_node);
638 ret += __dsos__fprintf(&pos->dsos.head, fp);
641 return ret;
644 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
645 bool (skip)(struct dso *dso, int parm), int parm)
647 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
650 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
651 bool (skip)(struct dso *dso, int parm), int parm)
653 struct rb_node *nd;
654 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
656 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
657 struct machine *pos = rb_entry(nd, struct machine, rb_node);
658 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
660 return ret;
663 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
665 int i;
666 size_t printed = 0;
667 struct dso *kdso = machine__kernel_map(machine)->dso;
669 if (kdso->has_build_id) {
670 char filename[PATH_MAX];
671 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
672 printed += fprintf(fp, "[0] %s\n", filename);
675 for (i = 0; i < vmlinux_path__nr_entries; ++i)
676 printed += fprintf(fp, "[%d] %s\n",
677 i + kdso->has_build_id, vmlinux_path[i]);
679 return printed;
682 size_t machine__fprintf(struct machine *machine, FILE *fp)
684 size_t ret = 0;
685 struct rb_node *nd;
687 pthread_rwlock_rdlock(&machine->threads_lock);
689 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
690 struct thread *pos = rb_entry(nd, struct thread, rb_node);
692 ret += thread__fprintf(pos, fp);
695 pthread_rwlock_unlock(&machine->threads_lock);
697 return ret;
700 static struct dso *machine__get_kernel(struct machine *machine)
702 const char *vmlinux_name = NULL;
703 struct dso *kernel;
705 if (machine__is_host(machine)) {
706 vmlinux_name = symbol_conf.vmlinux_name;
707 if (!vmlinux_name)
708 vmlinux_name = "[kernel.kallsyms]";
710 kernel = machine__findnew_kernel(machine, vmlinux_name,
711 "[kernel]", DSO_TYPE_KERNEL);
712 } else {
713 char bf[PATH_MAX];
715 if (machine__is_default_guest(machine))
716 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
717 if (!vmlinux_name)
718 vmlinux_name = machine__mmap_name(machine, bf,
719 sizeof(bf));
721 kernel = machine__findnew_kernel(machine, vmlinux_name,
722 "[guest.kernel]",
723 DSO_TYPE_GUEST_KERNEL);
726 if (kernel != NULL && (!kernel->has_build_id))
727 dso__read_running_kernel_build_id(kernel, machine);
729 return kernel;
732 struct process_args {
733 u64 start;
736 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
737 size_t bufsz)
739 if (machine__is_default_guest(machine))
740 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
741 else
742 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
745 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
747 /* Figure out the start address of kernel map from /proc/kallsyms.
748 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
749 * symbol_name if it's not that important.
751 static u64 machine__get_running_kernel_start(struct machine *machine,
752 const char **symbol_name)
754 char filename[PATH_MAX];
755 int i;
756 const char *name;
757 u64 addr = 0;
759 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
761 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
762 return 0;
764 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
765 addr = kallsyms__get_function_start(filename, name);
766 if (addr)
767 break;
770 if (symbol_name)
771 *symbol_name = name;
773 return addr;
776 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
778 enum map_type type;
779 u64 start = machine__get_running_kernel_start(machine, NULL);
781 /* In case of renewal the kernel map, destroy previous one */
782 machine__destroy_kernel_maps(machine);
784 for (type = 0; type < MAP__NR_TYPES; ++type) {
785 struct kmap *kmap;
786 struct map *map;
788 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
789 if (machine->vmlinux_maps[type] == NULL)
790 return -1;
792 machine->vmlinux_maps[type]->map_ip =
793 machine->vmlinux_maps[type]->unmap_ip =
794 identity__map_ip;
795 map = __machine__kernel_map(machine, type);
796 kmap = map__kmap(map);
797 if (!kmap)
798 return -1;
800 kmap->kmaps = &machine->kmaps;
801 map_groups__insert(&machine->kmaps, map);
804 return 0;
807 void machine__destroy_kernel_maps(struct machine *machine)
809 enum map_type type;
811 for (type = 0; type < MAP__NR_TYPES; ++type) {
812 struct kmap *kmap;
813 struct map *map = __machine__kernel_map(machine, type);
815 if (map == NULL)
816 continue;
818 kmap = map__kmap(map);
819 map_groups__remove(&machine->kmaps, map);
820 if (kmap && kmap->ref_reloc_sym) {
822 * ref_reloc_sym is shared among all maps, so free just
823 * on one of them.
825 if (type == MAP__FUNCTION) {
826 zfree((char **)&kmap->ref_reloc_sym->name);
827 zfree(&kmap->ref_reloc_sym);
828 } else
829 kmap->ref_reloc_sym = NULL;
832 map__put(machine->vmlinux_maps[type]);
833 machine->vmlinux_maps[type] = NULL;
837 int machines__create_guest_kernel_maps(struct machines *machines)
839 int ret = 0;
840 struct dirent **namelist = NULL;
841 int i, items = 0;
842 char path[PATH_MAX];
843 pid_t pid;
844 char *endp;
846 if (symbol_conf.default_guest_vmlinux_name ||
847 symbol_conf.default_guest_modules ||
848 symbol_conf.default_guest_kallsyms) {
849 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
852 if (symbol_conf.guestmount) {
853 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
854 if (items <= 0)
855 return -ENOENT;
856 for (i = 0; i < items; i++) {
857 if (!isdigit(namelist[i]->d_name[0])) {
858 /* Filter out . and .. */
859 continue;
861 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
862 if ((*endp != '\0') ||
863 (endp == namelist[i]->d_name) ||
864 (errno == ERANGE)) {
865 pr_debug("invalid directory (%s). Skipping.\n",
866 namelist[i]->d_name);
867 continue;
869 sprintf(path, "%s/%s/proc/kallsyms",
870 symbol_conf.guestmount,
871 namelist[i]->d_name);
872 ret = access(path, R_OK);
873 if (ret) {
874 pr_debug("Can't access file %s\n", path);
875 goto failure;
877 machines__create_kernel_maps(machines, pid);
879 failure:
880 free(namelist);
883 return ret;
886 void machines__destroy_kernel_maps(struct machines *machines)
888 struct rb_node *next = rb_first(&machines->guests);
890 machine__destroy_kernel_maps(&machines->host);
892 while (next) {
893 struct machine *pos = rb_entry(next, struct machine, rb_node);
895 next = rb_next(&pos->rb_node);
896 rb_erase(&pos->rb_node, &machines->guests);
897 machine__delete(pos);
901 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
903 struct machine *machine = machines__findnew(machines, pid);
905 if (machine == NULL)
906 return -1;
908 return machine__create_kernel_maps(machine);
911 int machine__load_kallsyms(struct machine *machine, const char *filename,
912 enum map_type type, symbol_filter_t filter)
914 struct map *map = machine__kernel_map(machine);
915 int ret = dso__load_kallsyms(map->dso, filename, map, filter);
917 if (ret > 0) {
918 dso__set_loaded(map->dso, type);
920 * Since /proc/kallsyms will have multiple sessions for the
921 * kernel, with modules between them, fixup the end of all
922 * sections.
924 __map_groups__fixup_end(&machine->kmaps, type);
927 return ret;
930 int machine__load_vmlinux_path(struct machine *machine, enum map_type type,
931 symbol_filter_t filter)
933 struct map *map = machine__kernel_map(machine);
934 int ret = dso__load_vmlinux_path(map->dso, map, filter);
936 if (ret > 0)
937 dso__set_loaded(map->dso, type);
939 return ret;
942 static void map_groups__fixup_end(struct map_groups *mg)
944 int i;
945 for (i = 0; i < MAP__NR_TYPES; ++i)
946 __map_groups__fixup_end(mg, i);
949 static char *get_kernel_version(const char *root_dir)
951 char version[PATH_MAX];
952 FILE *file;
953 char *name, *tmp;
954 const char *prefix = "Linux version ";
956 sprintf(version, "%s/proc/version", root_dir);
957 file = fopen(version, "r");
958 if (!file)
959 return NULL;
961 version[0] = '\0';
962 tmp = fgets(version, sizeof(version), file);
963 fclose(file);
965 name = strstr(version, prefix);
966 if (!name)
967 return NULL;
968 name += strlen(prefix);
969 tmp = strchr(name, ' ');
970 if (tmp)
971 *tmp = '\0';
973 return strdup(name);
976 static bool is_kmod_dso(struct dso *dso)
978 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
979 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
982 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
983 struct kmod_path *m)
985 struct map *map;
986 char *long_name;
988 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
989 if (map == NULL)
990 return 0;
992 long_name = strdup(path);
993 if (long_name == NULL)
994 return -ENOMEM;
996 dso__set_long_name(map->dso, long_name, true);
997 dso__kernel_module_get_build_id(map->dso, "");
1000 * Full name could reveal us kmod compression, so
1001 * we need to update the symtab_type if needed.
1003 if (m->comp && is_kmod_dso(map->dso))
1004 map->dso->symtab_type++;
1006 return 0;
1009 static int map_groups__set_modules_path_dir(struct map_groups *mg,
1010 const char *dir_name, int depth)
1012 struct dirent *dent;
1013 DIR *dir = opendir(dir_name);
1014 int ret = 0;
1016 if (!dir) {
1017 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1018 return -1;
1021 while ((dent = readdir(dir)) != NULL) {
1022 char path[PATH_MAX];
1023 struct stat st;
1025 /*sshfs might return bad dent->d_type, so we have to stat*/
1026 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1027 if (stat(path, &st))
1028 continue;
1030 if (S_ISDIR(st.st_mode)) {
1031 if (!strcmp(dent->d_name, ".") ||
1032 !strcmp(dent->d_name, ".."))
1033 continue;
1035 /* Do not follow top-level source and build symlinks */
1036 if (depth == 0) {
1037 if (!strcmp(dent->d_name, "source") ||
1038 !strcmp(dent->d_name, "build"))
1039 continue;
1042 ret = map_groups__set_modules_path_dir(mg, path,
1043 depth + 1);
1044 if (ret < 0)
1045 goto out;
1046 } else {
1047 struct kmod_path m;
1049 ret = kmod_path__parse_name(&m, dent->d_name);
1050 if (ret)
1051 goto out;
1053 if (m.kmod)
1054 ret = map_groups__set_module_path(mg, path, &m);
1056 free(m.name);
1058 if (ret)
1059 goto out;
1063 out:
1064 closedir(dir);
1065 return ret;
1068 static int machine__set_modules_path(struct machine *machine)
1070 char *version;
1071 char modules_path[PATH_MAX];
1073 version = get_kernel_version(machine->root_dir);
1074 if (!version)
1075 return -1;
1077 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1078 machine->root_dir, version);
1079 free(version);
1081 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1084 static int machine__create_module(void *arg, const char *name, u64 start)
1086 struct machine *machine = arg;
1087 struct map *map;
1089 map = machine__findnew_module_map(machine, start, name);
1090 if (map == NULL)
1091 return -1;
1093 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1095 return 0;
1098 static int machine__create_modules(struct machine *machine)
1100 const char *modules;
1101 char path[PATH_MAX];
1103 if (machine__is_default_guest(machine)) {
1104 modules = symbol_conf.default_guest_modules;
1105 } else {
1106 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1107 modules = path;
1110 if (symbol__restricted_filename(modules, "/proc/modules"))
1111 return -1;
1113 if (modules__parse(modules, machine, machine__create_module))
1114 return -1;
1116 if (!machine__set_modules_path(machine))
1117 return 0;
1119 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1121 return 0;
1124 int machine__create_kernel_maps(struct machine *machine)
1126 struct dso *kernel = machine__get_kernel(machine);
1127 const char *name;
1128 u64 addr = machine__get_running_kernel_start(machine, &name);
1129 int ret;
1131 if (!addr || kernel == NULL)
1132 return -1;
1134 ret = __machine__create_kernel_maps(machine, kernel);
1135 dso__put(kernel);
1136 if (ret < 0)
1137 return -1;
1139 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1140 if (machine__is_host(machine))
1141 pr_debug("Problems creating module maps, "
1142 "continuing anyway...\n");
1143 else
1144 pr_debug("Problems creating module maps for guest %d, "
1145 "continuing anyway...\n", machine->pid);
1149 * Now that we have all the maps created, just set the ->end of them:
1151 map_groups__fixup_end(&machine->kmaps);
1153 if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name,
1154 addr)) {
1155 machine__destroy_kernel_maps(machine);
1156 return -1;
1159 return 0;
1162 static void machine__set_kernel_mmap_len(struct machine *machine,
1163 union perf_event *event)
1165 int i;
1167 for (i = 0; i < MAP__NR_TYPES; i++) {
1168 machine->vmlinux_maps[i]->start = event->mmap.start;
1169 machine->vmlinux_maps[i]->end = (event->mmap.start +
1170 event->mmap.len);
1172 * Be a bit paranoid here, some perf.data file came with
1173 * a zero sized synthesized MMAP event for the kernel.
1175 if (machine->vmlinux_maps[i]->end == 0)
1176 machine->vmlinux_maps[i]->end = ~0ULL;
1180 static bool machine__uses_kcore(struct machine *machine)
1182 struct dso *dso;
1184 list_for_each_entry(dso, &machine->dsos.head, node) {
1185 if (dso__is_kcore(dso))
1186 return true;
1189 return false;
1192 static int machine__process_kernel_mmap_event(struct machine *machine,
1193 union perf_event *event)
1195 struct map *map;
1196 char kmmap_prefix[PATH_MAX];
1197 enum dso_kernel_type kernel_type;
1198 bool is_kernel_mmap;
1200 /* If we have maps from kcore then we do not need or want any others */
1201 if (machine__uses_kcore(machine))
1202 return 0;
1204 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1205 if (machine__is_host(machine))
1206 kernel_type = DSO_TYPE_KERNEL;
1207 else
1208 kernel_type = DSO_TYPE_GUEST_KERNEL;
1210 is_kernel_mmap = memcmp(event->mmap.filename,
1211 kmmap_prefix,
1212 strlen(kmmap_prefix) - 1) == 0;
1213 if (event->mmap.filename[0] == '/' ||
1214 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1215 map = machine__findnew_module_map(machine, event->mmap.start,
1216 event->mmap.filename);
1217 if (map == NULL)
1218 goto out_problem;
1220 map->end = map->start + event->mmap.len;
1221 } else if (is_kernel_mmap) {
1222 const char *symbol_name = (event->mmap.filename +
1223 strlen(kmmap_prefix));
1225 * Should be there already, from the build-id table in
1226 * the header.
1228 struct dso *kernel = NULL;
1229 struct dso *dso;
1231 pthread_rwlock_rdlock(&machine->dsos.lock);
1233 list_for_each_entry(dso, &machine->dsos.head, node) {
1236 * The cpumode passed to is_kernel_module is not the
1237 * cpumode of *this* event. If we insist on passing
1238 * correct cpumode to is_kernel_module, we should
1239 * record the cpumode when we adding this dso to the
1240 * linked list.
1242 * However we don't really need passing correct
1243 * cpumode. We know the correct cpumode must be kernel
1244 * mode (if not, we should not link it onto kernel_dsos
1245 * list).
1247 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1248 * is_kernel_module() treats it as a kernel cpumode.
1251 if (!dso->kernel ||
1252 is_kernel_module(dso->long_name,
1253 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1254 continue;
1257 kernel = dso;
1258 break;
1261 pthread_rwlock_unlock(&machine->dsos.lock);
1263 if (kernel == NULL)
1264 kernel = machine__findnew_dso(machine, kmmap_prefix);
1265 if (kernel == NULL)
1266 goto out_problem;
1268 kernel->kernel = kernel_type;
1269 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1270 dso__put(kernel);
1271 goto out_problem;
1274 if (strstr(kernel->long_name, "vmlinux"))
1275 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1277 machine__set_kernel_mmap_len(machine, event);
1280 * Avoid using a zero address (kptr_restrict) for the ref reloc
1281 * symbol. Effectively having zero here means that at record
1282 * time /proc/sys/kernel/kptr_restrict was non zero.
1284 if (event->mmap.pgoff != 0) {
1285 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1286 symbol_name,
1287 event->mmap.pgoff);
1290 if (machine__is_default_guest(machine)) {
1292 * preload dso of guest kernel and modules
1294 dso__load(kernel, machine__kernel_map(machine), NULL);
1297 return 0;
1298 out_problem:
1299 return -1;
1302 int machine__process_mmap2_event(struct machine *machine,
1303 union perf_event *event,
1304 struct perf_sample *sample)
1306 struct thread *thread;
1307 struct map *map;
1308 enum map_type type;
1309 int ret = 0;
1311 if (dump_trace)
1312 perf_event__fprintf_mmap2(event, stdout);
1314 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1315 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1316 ret = machine__process_kernel_mmap_event(machine, event);
1317 if (ret < 0)
1318 goto out_problem;
1319 return 0;
1322 thread = machine__findnew_thread(machine, event->mmap2.pid,
1323 event->mmap2.tid);
1324 if (thread == NULL)
1325 goto out_problem;
1327 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1328 type = MAP__VARIABLE;
1329 else
1330 type = MAP__FUNCTION;
1332 map = map__new(machine, event->mmap2.start,
1333 event->mmap2.len, event->mmap2.pgoff,
1334 event->mmap2.pid, event->mmap2.maj,
1335 event->mmap2.min, event->mmap2.ino,
1336 event->mmap2.ino_generation,
1337 event->mmap2.prot,
1338 event->mmap2.flags,
1339 event->mmap2.filename, type, thread);
1341 if (map == NULL)
1342 goto out_problem_map;
1344 thread__insert_map(thread, map);
1345 thread__put(thread);
1346 map__put(map);
1347 return 0;
1349 out_problem_map:
1350 thread__put(thread);
1351 out_problem:
1352 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1353 return 0;
1356 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1357 struct perf_sample *sample)
1359 struct thread *thread;
1360 struct map *map;
1361 enum map_type type;
1362 int ret = 0;
1364 if (dump_trace)
1365 perf_event__fprintf_mmap(event, stdout);
1367 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1368 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1369 ret = machine__process_kernel_mmap_event(machine, event);
1370 if (ret < 0)
1371 goto out_problem;
1372 return 0;
1375 thread = machine__findnew_thread(machine, event->mmap.pid,
1376 event->mmap.tid);
1377 if (thread == NULL)
1378 goto out_problem;
1380 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1381 type = MAP__VARIABLE;
1382 else
1383 type = MAP__FUNCTION;
1385 map = map__new(machine, event->mmap.start,
1386 event->mmap.len, event->mmap.pgoff,
1387 event->mmap.pid, 0, 0, 0, 0, 0, 0,
1388 event->mmap.filename,
1389 type, thread);
1391 if (map == NULL)
1392 goto out_problem_map;
1394 thread__insert_map(thread, map);
1395 thread__put(thread);
1396 map__put(map);
1397 return 0;
1399 out_problem_map:
1400 thread__put(thread);
1401 out_problem:
1402 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1403 return 0;
1406 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1408 if (machine->last_match == th)
1409 machine->last_match = NULL;
1411 BUG_ON(atomic_read(&th->refcnt) == 0);
1412 if (lock)
1413 pthread_rwlock_wrlock(&machine->threads_lock);
1414 rb_erase_init(&th->rb_node, &machine->threads);
1415 RB_CLEAR_NODE(&th->rb_node);
1417 * Move it first to the dead_threads list, then drop the reference,
1418 * if this is the last reference, then the thread__delete destructor
1419 * will be called and we will remove it from the dead_threads list.
1421 list_add_tail(&th->node, &machine->dead_threads);
1422 if (lock)
1423 pthread_rwlock_unlock(&machine->threads_lock);
1424 thread__put(th);
1427 void machine__remove_thread(struct machine *machine, struct thread *th)
1429 return __machine__remove_thread(machine, th, true);
1432 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1433 struct perf_sample *sample)
1435 struct thread *thread = machine__find_thread(machine,
1436 event->fork.pid,
1437 event->fork.tid);
1438 struct thread *parent = machine__findnew_thread(machine,
1439 event->fork.ppid,
1440 event->fork.ptid);
1441 int err = 0;
1443 if (dump_trace)
1444 perf_event__fprintf_task(event, stdout);
1447 * There may be an existing thread that is not actually the parent,
1448 * either because we are processing events out of order, or because the
1449 * (fork) event that would have removed the thread was lost. Assume the
1450 * latter case and continue on as best we can.
1452 if (parent->pid_ != (pid_t)event->fork.ppid) {
1453 dump_printf("removing erroneous parent thread %d/%d\n",
1454 parent->pid_, parent->tid);
1455 machine__remove_thread(machine, parent);
1456 thread__put(parent);
1457 parent = machine__findnew_thread(machine, event->fork.ppid,
1458 event->fork.ptid);
1461 /* if a thread currently exists for the thread id remove it */
1462 if (thread != NULL) {
1463 machine__remove_thread(machine, thread);
1464 thread__put(thread);
1467 thread = machine__findnew_thread(machine, event->fork.pid,
1468 event->fork.tid);
1470 if (thread == NULL || parent == NULL ||
1471 thread__fork(thread, parent, sample->time) < 0) {
1472 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1473 err = -1;
1475 thread__put(thread);
1476 thread__put(parent);
1478 return err;
1481 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1482 struct perf_sample *sample __maybe_unused)
1484 struct thread *thread = machine__find_thread(machine,
1485 event->fork.pid,
1486 event->fork.tid);
1488 if (dump_trace)
1489 perf_event__fprintf_task(event, stdout);
1491 if (thread != NULL) {
1492 thread__exited(thread);
1493 thread__put(thread);
1496 return 0;
1499 int machine__process_event(struct machine *machine, union perf_event *event,
1500 struct perf_sample *sample)
1502 int ret;
1504 switch (event->header.type) {
1505 case PERF_RECORD_COMM:
1506 ret = machine__process_comm_event(machine, event, sample); break;
1507 case PERF_RECORD_MMAP:
1508 ret = machine__process_mmap_event(machine, event, sample); break;
1509 case PERF_RECORD_MMAP2:
1510 ret = machine__process_mmap2_event(machine, event, sample); break;
1511 case PERF_RECORD_FORK:
1512 ret = machine__process_fork_event(machine, event, sample); break;
1513 case PERF_RECORD_EXIT:
1514 ret = machine__process_exit_event(machine, event, sample); break;
1515 case PERF_RECORD_LOST:
1516 ret = machine__process_lost_event(machine, event, sample); break;
1517 case PERF_RECORD_AUX:
1518 ret = machine__process_aux_event(machine, event); break;
1519 case PERF_RECORD_ITRACE_START:
1520 ret = machine__process_itrace_start_event(machine, event); break;
1521 case PERF_RECORD_LOST_SAMPLES:
1522 ret = machine__process_lost_samples_event(machine, event, sample); break;
1523 case PERF_RECORD_SWITCH:
1524 case PERF_RECORD_SWITCH_CPU_WIDE:
1525 ret = machine__process_switch_event(machine, event); break;
1526 default:
1527 ret = -1;
1528 break;
1531 return ret;
1534 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1536 if (sym->name && !regexec(regex, sym->name, 0, NULL, 0))
1537 return 1;
1538 return 0;
1541 static void ip__resolve_ams(struct thread *thread,
1542 struct addr_map_symbol *ams,
1543 u64 ip)
1545 struct addr_location al;
1547 memset(&al, 0, sizeof(al));
1549 * We cannot use the header.misc hint to determine whether a
1550 * branch stack address is user, kernel, guest, hypervisor.
1551 * Branches may straddle the kernel/user/hypervisor boundaries.
1552 * Thus, we have to try consecutively until we find a match
1553 * or else, the symbol is unknown
1555 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1557 ams->addr = ip;
1558 ams->al_addr = al.addr;
1559 ams->sym = al.sym;
1560 ams->map = al.map;
1563 static void ip__resolve_data(struct thread *thread,
1564 u8 m, struct addr_map_symbol *ams, u64 addr)
1566 struct addr_location al;
1568 memset(&al, 0, sizeof(al));
1570 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1571 if (al.map == NULL) {
1573 * some shared data regions have execute bit set which puts
1574 * their mapping in the MAP__FUNCTION type array.
1575 * Check there as a fallback option before dropping the sample.
1577 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1580 ams->addr = addr;
1581 ams->al_addr = al.addr;
1582 ams->sym = al.sym;
1583 ams->map = al.map;
1586 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1587 struct addr_location *al)
1589 struct mem_info *mi = zalloc(sizeof(*mi));
1591 if (!mi)
1592 return NULL;
1594 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1595 ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1596 mi->data_src.val = sample->data_src;
1598 return mi;
1601 static int add_callchain_ip(struct thread *thread,
1602 struct symbol **parent,
1603 struct addr_location *root_al,
1604 u8 *cpumode,
1605 u64 ip)
1607 struct addr_location al;
1609 al.filtered = 0;
1610 al.sym = NULL;
1611 if (!cpumode) {
1612 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1613 ip, &al);
1614 } else {
1615 if (ip >= PERF_CONTEXT_MAX) {
1616 switch (ip) {
1617 case PERF_CONTEXT_HV:
1618 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1619 break;
1620 case PERF_CONTEXT_KERNEL:
1621 *cpumode = PERF_RECORD_MISC_KERNEL;
1622 break;
1623 case PERF_CONTEXT_USER:
1624 *cpumode = PERF_RECORD_MISC_USER;
1625 break;
1626 default:
1627 pr_debug("invalid callchain context: "
1628 "%"PRId64"\n", (s64) ip);
1630 * It seems the callchain is corrupted.
1631 * Discard all.
1633 callchain_cursor_reset(&callchain_cursor);
1634 return 1;
1636 return 0;
1638 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1639 ip, &al);
1642 if (al.sym != NULL) {
1643 if (sort__has_parent && !*parent &&
1644 symbol__match_regex(al.sym, &parent_regex))
1645 *parent = al.sym;
1646 else if (have_ignore_callees && root_al &&
1647 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1648 /* Treat this symbol as the root,
1649 forgetting its callees. */
1650 *root_al = al;
1651 callchain_cursor_reset(&callchain_cursor);
1655 if (symbol_conf.hide_unresolved && al.sym == NULL)
1656 return 0;
1657 return callchain_cursor_append(&callchain_cursor, al.addr, al.map, al.sym);
1660 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1661 struct addr_location *al)
1663 unsigned int i;
1664 const struct branch_stack *bs = sample->branch_stack;
1665 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1667 if (!bi)
1668 return NULL;
1670 for (i = 0; i < bs->nr; i++) {
1671 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1672 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1673 bi[i].flags = bs->entries[i].flags;
1675 return bi;
1678 #define CHASHSZ 127
1679 #define CHASHBITS 7
1680 #define NO_ENTRY 0xff
1682 #define PERF_MAX_BRANCH_DEPTH 127
1684 /* Remove loops. */
1685 static int remove_loops(struct branch_entry *l, int nr)
1687 int i, j, off;
1688 unsigned char chash[CHASHSZ];
1690 memset(chash, NO_ENTRY, sizeof(chash));
1692 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1694 for (i = 0; i < nr; i++) {
1695 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1697 /* no collision handling for now */
1698 if (chash[h] == NO_ENTRY) {
1699 chash[h] = i;
1700 } else if (l[chash[h]].from == l[i].from) {
1701 bool is_loop = true;
1702 /* check if it is a real loop */
1703 off = 0;
1704 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1705 if (l[j].from != l[i + off].from) {
1706 is_loop = false;
1707 break;
1709 if (is_loop) {
1710 memmove(l + i, l + i + off,
1711 (nr - (i + off)) * sizeof(*l));
1712 nr -= off;
1716 return nr;
1720 * Recolve LBR callstack chain sample
1721 * Return:
1722 * 1 on success get LBR callchain information
1723 * 0 no available LBR callchain information, should try fp
1724 * negative error code on other errors.
1726 static int resolve_lbr_callchain_sample(struct thread *thread,
1727 struct perf_sample *sample,
1728 struct symbol **parent,
1729 struct addr_location *root_al,
1730 int max_stack)
1732 struct ip_callchain *chain = sample->callchain;
1733 int chain_nr = min(max_stack, (int)chain->nr);
1734 u8 cpumode = PERF_RECORD_MISC_USER;
1735 int i, j, err;
1736 u64 ip;
1738 for (i = 0; i < chain_nr; i++) {
1739 if (chain->ips[i] == PERF_CONTEXT_USER)
1740 break;
1743 /* LBR only affects the user callchain */
1744 if (i != chain_nr) {
1745 struct branch_stack *lbr_stack = sample->branch_stack;
1746 int lbr_nr = lbr_stack->nr;
1748 * LBR callstack can only get user call chain.
1749 * The mix_chain_nr is kernel call chain
1750 * number plus LBR user call chain number.
1751 * i is kernel call chain number,
1752 * 1 is PERF_CONTEXT_USER,
1753 * lbr_nr + 1 is the user call chain number.
1754 * For details, please refer to the comments
1755 * in callchain__printf
1757 int mix_chain_nr = i + 1 + lbr_nr + 1;
1759 if (mix_chain_nr > PERF_MAX_STACK_DEPTH + PERF_MAX_BRANCH_DEPTH) {
1760 pr_warning("corrupted callchain. skipping...\n");
1761 return 0;
1764 for (j = 0; j < mix_chain_nr; j++) {
1765 if (callchain_param.order == ORDER_CALLEE) {
1766 if (j < i + 1)
1767 ip = chain->ips[j];
1768 else if (j > i + 1)
1769 ip = lbr_stack->entries[j - i - 2].from;
1770 else
1771 ip = lbr_stack->entries[0].to;
1772 } else {
1773 if (j < lbr_nr)
1774 ip = lbr_stack->entries[lbr_nr - j - 1].from;
1775 else if (j > lbr_nr)
1776 ip = chain->ips[i + 1 - (j - lbr_nr)];
1777 else
1778 ip = lbr_stack->entries[0].to;
1781 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1782 if (err)
1783 return (err < 0) ? err : 0;
1785 return 1;
1788 return 0;
1791 static int thread__resolve_callchain_sample(struct thread *thread,
1792 struct perf_evsel *evsel,
1793 struct perf_sample *sample,
1794 struct symbol **parent,
1795 struct addr_location *root_al,
1796 int max_stack)
1798 struct branch_stack *branch = sample->branch_stack;
1799 struct ip_callchain *chain = sample->callchain;
1800 int chain_nr = min(max_stack, (int)chain->nr);
1801 u8 cpumode = PERF_RECORD_MISC_USER;
1802 int i, j, err;
1803 int skip_idx = -1;
1804 int first_call = 0;
1806 callchain_cursor_reset(&callchain_cursor);
1808 if (has_branch_callstack(evsel)) {
1809 err = resolve_lbr_callchain_sample(thread, sample, parent,
1810 root_al, max_stack);
1811 if (err)
1812 return (err < 0) ? err : 0;
1816 * Based on DWARF debug information, some architectures skip
1817 * a callchain entry saved by the kernel.
1819 if (chain->nr < PERF_MAX_STACK_DEPTH)
1820 skip_idx = arch_skip_callchain_idx(thread, chain);
1823 * Add branches to call stack for easier browsing. This gives
1824 * more context for a sample than just the callers.
1826 * This uses individual histograms of paths compared to the
1827 * aggregated histograms the normal LBR mode uses.
1829 * Limitations for now:
1830 * - No extra filters
1831 * - No annotations (should annotate somehow)
1834 if (branch && callchain_param.branch_callstack) {
1835 int nr = min(max_stack, (int)branch->nr);
1836 struct branch_entry be[nr];
1838 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1839 pr_warning("corrupted branch chain. skipping...\n");
1840 goto check_calls;
1843 for (i = 0; i < nr; i++) {
1844 if (callchain_param.order == ORDER_CALLEE) {
1845 be[i] = branch->entries[i];
1847 * Check for overlap into the callchain.
1848 * The return address is one off compared to
1849 * the branch entry. To adjust for this
1850 * assume the calling instruction is not longer
1851 * than 8 bytes.
1853 if (i == skip_idx ||
1854 chain->ips[first_call] >= PERF_CONTEXT_MAX)
1855 first_call++;
1856 else if (be[i].from < chain->ips[first_call] &&
1857 be[i].from >= chain->ips[first_call] - 8)
1858 first_call++;
1859 } else
1860 be[i] = branch->entries[branch->nr - i - 1];
1863 nr = remove_loops(be, nr);
1865 for (i = 0; i < nr; i++) {
1866 err = add_callchain_ip(thread, parent, root_al,
1867 NULL, be[i].to);
1868 if (!err)
1869 err = add_callchain_ip(thread, parent, root_al,
1870 NULL, be[i].from);
1871 if (err == -EINVAL)
1872 break;
1873 if (err)
1874 return err;
1876 chain_nr -= nr;
1879 check_calls:
1880 if (chain->nr > PERF_MAX_STACK_DEPTH && (int)chain->nr > max_stack) {
1881 pr_warning("corrupted callchain. skipping...\n");
1882 return 0;
1885 for (i = first_call; i < chain_nr; i++) {
1886 u64 ip;
1888 if (callchain_param.order == ORDER_CALLEE)
1889 j = i;
1890 else
1891 j = chain->nr - i - 1;
1893 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1894 if (j == skip_idx)
1895 continue;
1896 #endif
1897 ip = chain->ips[j];
1899 err = add_callchain_ip(thread, parent, root_al, &cpumode, ip);
1901 if (err)
1902 return (err < 0) ? err : 0;
1905 return 0;
1908 static int unwind_entry(struct unwind_entry *entry, void *arg)
1910 struct callchain_cursor *cursor = arg;
1912 if (symbol_conf.hide_unresolved && entry->sym == NULL)
1913 return 0;
1914 return callchain_cursor_append(cursor, entry->ip,
1915 entry->map, entry->sym);
1918 int thread__resolve_callchain(struct thread *thread,
1919 struct perf_evsel *evsel,
1920 struct perf_sample *sample,
1921 struct symbol **parent,
1922 struct addr_location *root_al,
1923 int max_stack)
1925 int ret = thread__resolve_callchain_sample(thread, evsel,
1926 sample, parent,
1927 root_al, max_stack);
1928 if (ret)
1929 return ret;
1931 /* Can we do dwarf post unwind? */
1932 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
1933 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
1934 return 0;
1936 /* Bail out if nothing was captured. */
1937 if ((!sample->user_regs.regs) ||
1938 (!sample->user_stack.size))
1939 return 0;
1941 return unwind__get_entries(unwind_entry, &callchain_cursor,
1942 thread, sample, max_stack);
1946 int machine__for_each_thread(struct machine *machine,
1947 int (*fn)(struct thread *thread, void *p),
1948 void *priv)
1950 struct rb_node *nd;
1951 struct thread *thread;
1952 int rc = 0;
1954 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
1955 thread = rb_entry(nd, struct thread, rb_node);
1956 rc = fn(thread, priv);
1957 if (rc != 0)
1958 return rc;
1961 list_for_each_entry(thread, &machine->dead_threads, node) {
1962 rc = fn(thread, priv);
1963 if (rc != 0)
1964 return rc;
1966 return rc;
1969 int machines__for_each_thread(struct machines *machines,
1970 int (*fn)(struct thread *thread, void *p),
1971 void *priv)
1973 struct rb_node *nd;
1974 int rc = 0;
1976 rc = machine__for_each_thread(&machines->host, fn, priv);
1977 if (rc != 0)
1978 return rc;
1980 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
1981 struct machine *machine = rb_entry(nd, struct machine, rb_node);
1983 rc = machine__for_each_thread(machine, fn, priv);
1984 if (rc != 0)
1985 return rc;
1987 return rc;
1990 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
1991 struct target *target, struct thread_map *threads,
1992 perf_event__handler_t process, bool data_mmap,
1993 unsigned int proc_map_timeout)
1995 if (target__has_task(target))
1996 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
1997 else if (target__has_cpu(target))
1998 return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
1999 /* command specified */
2000 return 0;
2003 pid_t machine__get_current_tid(struct machine *machine, int cpu)
2005 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
2006 return -1;
2008 return machine->current_tid[cpu];
2011 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
2012 pid_t tid)
2014 struct thread *thread;
2016 if (cpu < 0)
2017 return -EINVAL;
2019 if (!machine->current_tid) {
2020 int i;
2022 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
2023 if (!machine->current_tid)
2024 return -ENOMEM;
2025 for (i = 0; i < MAX_NR_CPUS; i++)
2026 machine->current_tid[i] = -1;
2029 if (cpu >= MAX_NR_CPUS) {
2030 pr_err("Requested CPU %d too large. ", cpu);
2031 pr_err("Consider raising MAX_NR_CPUS\n");
2032 return -EINVAL;
2035 machine->current_tid[cpu] = tid;
2037 thread = machine__findnew_thread(machine, pid, tid);
2038 if (!thread)
2039 return -ENOMEM;
2041 thread->cpu = cpu;
2042 thread__put(thread);
2044 return 0;
2047 int machine__get_kernel_start(struct machine *machine)
2049 struct map *map = machine__kernel_map(machine);
2050 int err = 0;
2053 * The only addresses above 2^63 are kernel addresses of a 64-bit
2054 * kernel. Note that addresses are unsigned so that on a 32-bit system
2055 * all addresses including kernel addresses are less than 2^32. In
2056 * that case (32-bit system), if the kernel mapping is unknown, all
2057 * addresses will be assumed to be in user space - see
2058 * machine__kernel_ip().
2060 machine->kernel_start = 1ULL << 63;
2061 if (map) {
2062 err = map__load(map, machine->symbol_filter);
2063 if (map->start)
2064 machine->kernel_start = map->start;
2066 return err;
2069 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2071 return dsos__findnew(&machine->dsos, filename);
2074 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2076 struct machine *machine = vmachine;
2077 struct map *map;
2078 struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map, NULL);
2080 if (sym == NULL)
2081 return NULL;
2083 *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2084 *addrp = map->unmap_ip(map, sym->start);
2085 return sym->name;