]> git.proxmox.com Git - mirror_ubuntu-focal-kernel.git/blob - tools/perf/util/machine.c
projects / mirror_ubuntu-focal-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'trace-v4.14-rc3' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[mirror_ubuntu-focal-kernel.git] / tools / perf / util / machine.c
1 #include <dirent.h>
2 #include <errno.h>
3 #include <inttypes.h>
4 #include <regex.h>
5 #include "callchain.h"
6 #include "debug.h"
7 #include "event.h"
8 #include "evsel.h"
9 #include "hist.h"
10 #include "machine.h"
11 #include "map.h"
12 #include "sort.h"
13 #include "strlist.h"
14 #include "thread.h"
15 #include "vdso.h"
16 #include <stdbool.h>
17 #include <sys/types.h>
18 #include <sys/stat.h>
19 #include <unistd.h>
20 #include "unwind.h"
21 #include "linux/hash.h"
22 #include "asm/bug.h"
23
24 #include "sane_ctype.h"
25 #include <symbol/kallsyms.h>
26
27 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
28
29 static void dsos__init(struct dsos *dsos)
30 {
31 INIT_LIST_HEAD(&dsos->head);
32 dsos->root = RB_ROOT;
33 pthread_rwlock_init(&dsos->lock, NULL);
34 }
35
36 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
37 {
38 memset(machine, 0, sizeof(*machine));
39 map_groups__init(&machine->kmaps, machine);
40 RB_CLEAR_NODE(&machine->rb_node);
41 dsos__init(&machine->dsos);
42
43 machine->threads = RB_ROOT;
44 pthread_rwlock_init(&machine->threads_lock, NULL);
45 machine->nr_threads = 0;
46 INIT_LIST_HEAD(&machine->dead_threads);
47 machine->last_match = NULL;
48
49 machine->vdso_info = NULL;
50 machine->env = NULL;
51
52 machine->pid = pid;
53
54 machine->id_hdr_size = 0;
55 machine->kptr_restrict_warned = false;
56 machine->comm_exec = false;
57 machine->kernel_start = 0;
58
59 memset(machine->vmlinux_maps, 0, sizeof(machine->vmlinux_maps));
60
61 machine->root_dir = strdup(root_dir);
62 if (machine->root_dir == NULL)
63 return -ENOMEM;
64
65 if (pid != HOST_KERNEL_ID) {
66 struct thread *thread = machine__findnew_thread(machine, -1,
67 pid);
68 char comm[64];
69
70 if (thread == NULL)
71 return -ENOMEM;
72
73 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
74 thread__set_comm(thread, comm, 0);
75 thread__put(thread);
76 }
77
78 machine->current_tid = NULL;
79
80 return 0;
81 }
82
83 struct machine *machine__new_host(void)
84 {
85 struct machine *machine = malloc(sizeof(*machine));
86
87 if (machine != NULL) {
88 machine__init(machine, "", HOST_KERNEL_ID);
89
90 if (machine__create_kernel_maps(machine) < 0)
91 goto out_delete;
92 }
93
94 return machine;
95 out_delete:
96 free(machine);
97 return NULL;
98 }
99
100 struct machine *machine__new_kallsyms(void)
101 {
102 struct machine *machine = machine__new_host();
103 /*
104 * FIXME:
105 * 1) MAP__FUNCTION will go away when we stop loading separate maps for
106 * functions and data objects.
107 * 2) We should switch to machine__load_kallsyms(), i.e. not explicitely
108 * ask for not using the kcore parsing code, once this one is fixed
109 * to create a map per module.
110 */
111 if (machine && __machine__load_kallsyms(machine, "/proc/kallsyms", MAP__FUNCTION, true) <= 0) {
112 machine__delete(machine);
113 machine = NULL;
114 }
115
116 return machine;
117 }
118
119 static void dsos__purge(struct dsos *dsos)
120 {
121 struct dso *pos, *n;
122
123 pthread_rwlock_wrlock(&dsos->lock);
124
125 list_for_each_entry_safe(pos, n, &dsos->head, node) {
126 RB_CLEAR_NODE(&pos->rb_node);
127 pos->root = NULL;
128 list_del_init(&pos->node);
129 dso__put(pos);
130 }
131
132 pthread_rwlock_unlock(&dsos->lock);
133 }
134
135 static void dsos__exit(struct dsos *dsos)
136 {
137 dsos__purge(dsos);
138 pthread_rwlock_destroy(&dsos->lock);
139 }
140
141 void machine__delete_threads(struct machine *machine)
142 {
143 struct rb_node *nd;
144
145 pthread_rwlock_wrlock(&machine->threads_lock);
146 nd = rb_first(&machine->threads);
147 while (nd) {
148 struct thread *t = rb_entry(nd, struct thread, rb_node);
149
150 nd = rb_next(nd);
151 __machine__remove_thread(machine, t, false);
152 }
153 pthread_rwlock_unlock(&machine->threads_lock);
154 }
155
156 void machine__exit(struct machine *machine)
157 {
158 machine__destroy_kernel_maps(machine);
159 map_groups__exit(&machine->kmaps);
160 dsos__exit(&machine->dsos);
161 machine__exit_vdso(machine);
162 zfree(&machine->root_dir);
163 zfree(&machine->current_tid);
164 pthread_rwlock_destroy(&machine->threads_lock);
165 }
166
167 void machine__delete(struct machine *machine)
168 {
169 if (machine) {
170 machine__exit(machine);
171 free(machine);
172 }
173 }
174
175 void machines__init(struct machines *machines)
176 {
177 machine__init(&machines->host, "", HOST_KERNEL_ID);
178 machines->guests = RB_ROOT;
179 }
180
181 void machines__exit(struct machines *machines)
182 {
183 machine__exit(&machines->host);
184 /* XXX exit guest */
185 }
186
187 struct machine *machines__add(struct machines *machines, pid_t pid,
188 const char *root_dir)
189 {
190 struct rb_node **p = &machines->guests.rb_node;
191 struct rb_node *parent = NULL;
192 struct machine *pos, *machine = malloc(sizeof(*machine));
193
194 if (machine == NULL)
195 return NULL;
196
197 if (machine__init(machine, root_dir, pid) != 0) {
198 free(machine);
199 return NULL;
200 }
201
202 while (*p != NULL) {
203 parent = *p;
204 pos = rb_entry(parent, struct machine, rb_node);
205 if (pid < pos->pid)
206 p = &(*p)->rb_left;
207 else
208 p = &(*p)->rb_right;
209 }
210
211 rb_link_node(&machine->rb_node, parent, p);
212 rb_insert_color(&machine->rb_node, &machines->guests);
213
214 return machine;
215 }
216
217 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
218 {
219 struct rb_node *nd;
220
221 machines->host.comm_exec = comm_exec;
222
223 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
224 struct machine *machine = rb_entry(nd, struct machine, rb_node);
225
226 machine->comm_exec = comm_exec;
227 }
228 }
229
230 struct machine *machines__find(struct machines *machines, pid_t pid)
231 {
232 struct rb_node **p = &machines->guests.rb_node;
233 struct rb_node *parent = NULL;
234 struct machine *machine;
235 struct machine *default_machine = NULL;
236
237 if (pid == HOST_KERNEL_ID)
238 return &machines->host;
239
240 while (*p != NULL) {
241 parent = *p;
242 machine = rb_entry(parent, struct machine, rb_node);
243 if (pid < machine->pid)
244 p = &(*p)->rb_left;
245 else if (pid > machine->pid)
246 p = &(*p)->rb_right;
247 else
248 return machine;
249 if (!machine->pid)
250 default_machine = machine;
251 }
252
253 return default_machine;
254 }
255
256 struct machine *machines__findnew(struct machines *machines, pid_t pid)
257 {
258 char path[PATH_MAX];
259 const char *root_dir = "";
260 struct machine *machine = machines__find(machines, pid);
261
262 if (machine && (machine->pid == pid))
263 goto out;
264
265 if ((pid != HOST_KERNEL_ID) &&
266 (pid != DEFAULT_GUEST_KERNEL_ID) &&
267 (symbol_conf.guestmount)) {
268 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
269 if (access(path, R_OK)) {
270 static struct strlist *seen;
271
272 if (!seen)
273 seen = strlist__new(NULL, NULL);
274
275 if (!strlist__has_entry(seen, path)) {
276 pr_err("Can't access file %s\n", path);
277 strlist__add(seen, path);
278 }
279 machine = NULL;
280 goto out;
281 }
282 root_dir = path;
283 }
284
285 machine = machines__add(machines, pid, root_dir);
286 out:
287 return machine;
288 }
289
290 void machines__process_guests(struct machines *machines,
291 machine__process_t process, void *data)
292 {
293 struct rb_node *nd;
294
295 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
296 struct machine *pos = rb_entry(nd, struct machine, rb_node);
297 process(pos, data);
298 }
299 }
300
301 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
302 {
303 if (machine__is_host(machine))
304 snprintf(bf, size, "[%s]", "kernel.kallsyms");
305 else if (machine__is_default_guest(machine))
306 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
307 else {
308 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
309 machine->pid);
310 }
311
312 return bf;
313 }
314
315 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
316 {
317 struct rb_node *node;
318 struct machine *machine;
319
320 machines->host.id_hdr_size = id_hdr_size;
321
322 for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
323 machine = rb_entry(node, struct machine, rb_node);
324 machine->id_hdr_size = id_hdr_size;
325 }
326
327 return;
328 }
329
330 static void machine__update_thread_pid(struct machine *machine,
331 struct thread *th, pid_t pid)
332 {
333 struct thread *leader;
334
335 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
336 return;
337
338 th->pid_ = pid;
339
340 if (th->pid_ == th->tid)
341 return;
342
343 leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
344 if (!leader)
345 goto out_err;
346
347 if (!leader->mg)
348 leader->mg = map_groups__new(machine);
349
350 if (!leader->mg)
351 goto out_err;
352
353 if (th->mg == leader->mg)
354 return;
355
356 if (th->mg) {
357 /*
358 * Maps are created from MMAP events which provide the pid and
359 * tid. Consequently there never should be any maps on a thread
360 * with an unknown pid. Just print an error if there are.
361 */
362 if (!map_groups__empty(th->mg))
363 pr_err("Discarding thread maps for %d:%d\n",
364 th->pid_, th->tid);
365 map_groups__put(th->mg);
366 }
367
368 th->mg = map_groups__get(leader->mg);
369 out_put:
370 thread__put(leader);
371 return;
372 out_err:
373 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
374 goto out_put;
375 }
376
377 /*
378 * Caller must eventually drop thread->refcnt returned with a successful
379 * lookup/new thread inserted.
380 */
381 static struct thread *____machine__findnew_thread(struct machine *machine,
382 pid_t pid, pid_t tid,
383 bool create)
384 {
385 struct rb_node **p = &machine->threads.rb_node;
386 struct rb_node *parent = NULL;
387 struct thread *th;
388
389 /*
390 * Front-end cache - TID lookups come in blocks,
391 * so most of the time we dont have to look up
392 * the full rbtree:
393 */
394 th = machine->last_match;
395 if (th != NULL) {
396 if (th->tid == tid) {
397 machine__update_thread_pid(machine, th, pid);
398 return thread__get(th);
399 }
400
401 machine->last_match = NULL;
402 }
403
404 while (*p != NULL) {
405 parent = *p;
406 th = rb_entry(parent, struct thread, rb_node);
407
408 if (th->tid == tid) {
409 machine->last_match = th;
410 machine__update_thread_pid(machine, th, pid);
411 return thread__get(th);
412 }
413
414 if (tid < th->tid)
415 p = &(*p)->rb_left;
416 else
417 p = &(*p)->rb_right;
418 }
419
420 if (!create)
421 return NULL;
422
423 th = thread__new(pid, tid);
424 if (th != NULL) {
425 rb_link_node(&th->rb_node, parent, p);
426 rb_insert_color(&th->rb_node, &machine->threads);
427
428 /*
429 * We have to initialize map_groups separately
430 * after rb tree is updated.
431 *
432 * The reason is that we call machine__findnew_thread
433 * within thread__init_map_groups to find the thread
434 * leader and that would screwed the rb tree.
435 */
436 if (thread__init_map_groups(th, machine)) {
437 rb_erase_init(&th->rb_node, &machine->threads);
438 RB_CLEAR_NODE(&th->rb_node);
439 thread__put(th);
440 return NULL;
441 }
442 /*
443 * It is now in the rbtree, get a ref
444 */
445 thread__get(th);
446 machine->last_match = th;
447 ++machine->nr_threads;
448 }
449
450 return th;
451 }
452
453 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
454 {
455 return ____machine__findnew_thread(machine, pid, tid, true);
456 }
457
458 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
459 pid_t tid)
460 {
461 struct thread *th;
462
463 pthread_rwlock_wrlock(&machine->threads_lock);
464 th = __machine__findnew_thread(machine, pid, tid);
465 pthread_rwlock_unlock(&machine->threads_lock);
466 return th;
467 }
468
469 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
470 pid_t tid)
471 {
472 struct thread *th;
473 pthread_rwlock_rdlock(&machine->threads_lock);
474 th = ____machine__findnew_thread(machine, pid, tid, false);
475 pthread_rwlock_unlock(&machine->threads_lock);
476 return th;
477 }
478
479 struct comm *machine__thread_exec_comm(struct machine *machine,
480 struct thread *thread)
481 {
482 if (machine->comm_exec)
483 return thread__exec_comm(thread);
484 else
485 return thread__comm(thread);
486 }
487
488 int machine__process_comm_event(struct machine *machine, union perf_event *event,
489 struct perf_sample *sample)
490 {
491 struct thread *thread = machine__findnew_thread(machine,
492 event->comm.pid,
493 event->comm.tid);
494 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
495 int err = 0;
496
497 if (exec)
498 machine->comm_exec = true;
499
500 if (dump_trace)
501 perf_event__fprintf_comm(event, stdout);
502
503 if (thread == NULL ||
504 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
505 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
506 err = -1;
507 }
508
509 thread__put(thread);
510
511 return err;
512 }
513
514 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
515 union perf_event *event,
516 struct perf_sample *sample __maybe_unused)
517 {
518 struct thread *thread = machine__findnew_thread(machine,
519 event->namespaces.pid,
520 event->namespaces.tid);
521 int err = 0;
522
523 WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
524 "\nWARNING: kernel seems to support more namespaces than perf"
525 " tool.\nTry updating the perf tool..\n\n");
526
527 WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
528 "\nWARNING: perf tool seems to support more namespaces than"
529 " the kernel.\nTry updating the kernel..\n\n");
530
531 if (dump_trace)
532 perf_event__fprintf_namespaces(event, stdout);
533
534 if (thread == NULL ||
535 thread__set_namespaces(thread, sample->time, &event->namespaces)) {
536 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
537 err = -1;
538 }
539
540 thread__put(thread);
541
542 return err;
543 }
544
545 int machine__process_lost_event(struct machine *machine __maybe_unused,
546 union perf_event *event, struct perf_sample *sample __maybe_unused)
547 {
548 dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
549 event->lost.id, event->lost.lost);
550 return 0;
551 }
552
553 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
554 union perf_event *event, struct perf_sample *sample)
555 {
556 dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
557 sample->id, event->lost_samples.lost);
558 return 0;
559 }
560
561 static struct dso *machine__findnew_module_dso(struct machine *machine,
562 struct kmod_path *m,
563 const char *filename)
564 {
565 struct dso *dso;
566
567 pthread_rwlock_wrlock(&machine->dsos.lock);
568
569 dso = __dsos__find(&machine->dsos, m->name, true);
570 if (!dso) {
571 dso = __dsos__addnew(&machine->dsos, m->name);
572 if (dso == NULL)
573 goto out_unlock;
574
575 dso__set_module_info(dso, m, machine);
576 dso__set_long_name(dso, strdup(filename), true);
577 }
578
579 dso__get(dso);
580 out_unlock:
581 pthread_rwlock_unlock(&machine->dsos.lock);
582 return dso;
583 }
584
585 int machine__process_aux_event(struct machine *machine __maybe_unused,
586 union perf_event *event)
587 {
588 if (dump_trace)
589 perf_event__fprintf_aux(event, stdout);
590 return 0;
591 }
592
593 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
594 union perf_event *event)
595 {
596 if (dump_trace)
597 perf_event__fprintf_itrace_start(event, stdout);
598 return 0;
599 }
600
601 int machine__process_switch_event(struct machine *machine __maybe_unused,
602 union perf_event *event)
603 {
604 if (dump_trace)
605 perf_event__fprintf_switch(event, stdout);
606 return 0;
607 }
608
609 static void dso__adjust_kmod_long_name(struct dso *dso, const char *filename)
610 {
611 const char *dup_filename;
612
613 if (!filename || !dso || !dso->long_name)
614 return;
615 if (dso->long_name[0] != '[')
616 return;
617 if (!strchr(filename, '/'))
618 return;
619
620 dup_filename = strdup(filename);
621 if (!dup_filename)
622 return;
623
624 dso__set_long_name(dso, dup_filename, true);
625 }
626
627 struct map *machine__findnew_module_map(struct machine *machine, u64 start,
628 const char *filename)
629 {
630 struct map *map = NULL;
631 struct dso *dso = NULL;
632 struct kmod_path m;
633
634 if (kmod_path__parse_name(&m, filename))
635 return NULL;
636
637 map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
638 m.name);
639 if (map) {
640 /*
641 * If the map's dso is an offline module, give dso__load()
642 * a chance to find the file path of that module by fixing
643 * long_name.
644 */
645 dso__adjust_kmod_long_name(map->dso, filename);
646 goto out;
647 }
648
649 dso = machine__findnew_module_dso(machine, &m, filename);
650 if (dso == NULL)
651 goto out;
652
653 map = map__new2(start, dso, MAP__FUNCTION);
654 if (map == NULL)
655 goto out;
656
657 map_groups__insert(&machine->kmaps, map);
658
659 /* Put the map here because map_groups__insert alread got it */
660 map__put(map);
661 out:
662 /* put the dso here, corresponding to machine__findnew_module_dso */
663 dso__put(dso);
664 free(m.name);
665 return map;
666 }
667
668 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
669 {
670 struct rb_node *nd;
671 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
672
673 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
674 struct machine *pos = rb_entry(nd, struct machine, rb_node);
675 ret += __dsos__fprintf(&pos->dsos.head, fp);
676 }
677
678 return ret;
679 }
680
681 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
682 bool (skip)(struct dso *dso, int parm), int parm)
683 {
684 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
685 }
686
687 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
688 bool (skip)(struct dso *dso, int parm), int parm)
689 {
690 struct rb_node *nd;
691 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
692
693 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
694 struct machine *pos = rb_entry(nd, struct machine, rb_node);
695 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
696 }
697 return ret;
698 }
699
700 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
701 {
702 int i;
703 size_t printed = 0;
704 struct dso *kdso = machine__kernel_map(machine)->dso;
705
706 if (kdso->has_build_id) {
707 char filename[PATH_MAX];
708 if (dso__build_id_filename(kdso, filename, sizeof(filename),
709 false))
710 printed += fprintf(fp, "[0] %s\n", filename);
711 }
712
713 for (i = 0; i < vmlinux_path__nr_entries; ++i)
714 printed += fprintf(fp, "[%d] %s\n",
715 i + kdso->has_build_id, vmlinux_path[i]);
716
717 return printed;
718 }
719
720 size_t machine__fprintf(struct machine *machine, FILE *fp)
721 {
722 size_t ret;
723 struct rb_node *nd;
724
725 pthread_rwlock_rdlock(&machine->threads_lock);
726
727 ret = fprintf(fp, "Threads: %u\n", machine->nr_threads);
728
729 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
730 struct thread *pos = rb_entry(nd, struct thread, rb_node);
731
732 ret += thread__fprintf(pos, fp);
733 }
734
735 pthread_rwlock_unlock(&machine->threads_lock);
736
737 return ret;
738 }
739
740 static struct dso *machine__get_kernel(struct machine *machine)
741 {
742 const char *vmlinux_name = NULL;
743 struct dso *kernel;
744
745 if (machine__is_host(machine)) {
746 vmlinux_name = symbol_conf.vmlinux_name;
747 if (!vmlinux_name)
748 vmlinux_name = DSO__NAME_KALLSYMS;
749
750 kernel = machine__findnew_kernel(machine, vmlinux_name,
751 "[kernel]", DSO_TYPE_KERNEL);
752 } else {
753 char bf[PATH_MAX];
754
755 if (machine__is_default_guest(machine))
756 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
757 if (!vmlinux_name)
758 vmlinux_name = machine__mmap_name(machine, bf,
759 sizeof(bf));
760
761 kernel = machine__findnew_kernel(machine, vmlinux_name,
762 "[guest.kernel]",
763 DSO_TYPE_GUEST_KERNEL);
764 }
765
766 if (kernel != NULL && (!kernel->has_build_id))
767 dso__read_running_kernel_build_id(kernel, machine);
768
769 return kernel;
770 }
771
772 struct process_args {
773 u64 start;
774 };
775
776 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
777 size_t bufsz)
778 {
779 if (machine__is_default_guest(machine))
780 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
781 else
782 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
783 }
784
785 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
786
787 /* Figure out the start address of kernel map from /proc/kallsyms.
788 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
789 * symbol_name if it's not that important.
790 */
791 static int machine__get_running_kernel_start(struct machine *machine,
792 const char **symbol_name, u64 *start)
793 {
794 char filename[PATH_MAX];
795 int i, err = -1;
796 const char *name;
797 u64 addr = 0;
798
799 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
800
801 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
802 return 0;
803
804 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
805 err = kallsyms__get_function_start(filename, name, &addr);
806 if (!err)
807 break;
808 }
809
810 if (err)
811 return -1;
812
813 if (symbol_name)
814 *symbol_name = name;
815
816 *start = addr;
817 return 0;
818 }
819
820 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
821 {
822 int type;
823 u64 start = 0;
824
825 if (machine__get_running_kernel_start(machine, NULL, &start))
826 return -1;
827
828 /* In case of renewal the kernel map, destroy previous one */
829 machine__destroy_kernel_maps(machine);
830
831 for (type = 0; type < MAP__NR_TYPES; ++type) {
832 struct kmap *kmap;
833 struct map *map;
834
835 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
836 if (machine->vmlinux_maps[type] == NULL)
837 return -1;
838
839 machine->vmlinux_maps[type]->map_ip =
840 machine->vmlinux_maps[type]->unmap_ip =
841 identity__map_ip;
842 map = __machine__kernel_map(machine, type);
843 kmap = map__kmap(map);
844 if (!kmap)
845 return -1;
846
847 kmap->kmaps = &machine->kmaps;
848 map_groups__insert(&machine->kmaps, map);
849 }
850
851 return 0;
852 }
853
854 void machine__destroy_kernel_maps(struct machine *machine)
855 {
856 int type;
857
858 for (type = 0; type < MAP__NR_TYPES; ++type) {
859 struct kmap *kmap;
860 struct map *map = __machine__kernel_map(machine, type);
861
862 if (map == NULL)
863 continue;
864
865 kmap = map__kmap(map);
866 map_groups__remove(&machine->kmaps, map);
867 if (kmap && kmap->ref_reloc_sym) {
868 /*
869 * ref_reloc_sym is shared among all maps, so free just
870 * on one of them.
871 */
872 if (type == MAP__FUNCTION) {
873 zfree((char **)&kmap->ref_reloc_sym->name);
874 zfree(&kmap->ref_reloc_sym);
875 } else
876 kmap->ref_reloc_sym = NULL;
877 }
878
879 map__put(machine->vmlinux_maps[type]);
880 machine->vmlinux_maps[type] = NULL;
881 }
882 }
883
884 int machines__create_guest_kernel_maps(struct machines *machines)
885 {
886 int ret = 0;
887 struct dirent **namelist = NULL;
888 int i, items = 0;
889 char path[PATH_MAX];
890 pid_t pid;
891 char *endp;
892
893 if (symbol_conf.default_guest_vmlinux_name ||
894 symbol_conf.default_guest_modules ||
895 symbol_conf.default_guest_kallsyms) {
896 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
897 }
898
899 if (symbol_conf.guestmount) {
900 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
901 if (items <= 0)
902 return -ENOENT;
903 for (i = 0; i < items; i++) {
904 if (!isdigit(namelist[i]->d_name[0])) {
905 /* Filter out . and .. */
906 continue;
907 }
908 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
909 if ((*endp != '\0') ||
910 (endp == namelist[i]->d_name) ||
911 (errno == ERANGE)) {
912 pr_debug("invalid directory (%s). Skipping.\n",
913 namelist[i]->d_name);
914 continue;
915 }
916 sprintf(path, "%s/%s/proc/kallsyms",
917 symbol_conf.guestmount,
918 namelist[i]->d_name);
919 ret = access(path, R_OK);
920 if (ret) {
921 pr_debug("Can't access file %s\n", path);
922 goto failure;
923 }
924 machines__create_kernel_maps(machines, pid);
925 }
926 failure:
927 free(namelist);
928 }
929
930 return ret;
931 }
932
933 void machines__destroy_kernel_maps(struct machines *machines)
934 {
935 struct rb_node *next = rb_first(&machines->guests);
936
937 machine__destroy_kernel_maps(&machines->host);
938
939 while (next) {
940 struct machine *pos = rb_entry(next, struct machine, rb_node);
941
942 next = rb_next(&pos->rb_node);
943 rb_erase(&pos->rb_node, &machines->guests);
944 machine__delete(pos);
945 }
946 }
947
948 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
949 {
950 struct machine *machine = machines__findnew(machines, pid);
951
952 if (machine == NULL)
953 return -1;
954
955 return machine__create_kernel_maps(machine);
956 }
957
958 int __machine__load_kallsyms(struct machine *machine, const char *filename,
959 enum map_type type, bool no_kcore)
960 {
961 struct map *map = machine__kernel_map(machine);
962 int ret = __dso__load_kallsyms(map->dso, filename, map, no_kcore);
963
964 if (ret > 0) {
965 dso__set_loaded(map->dso, type);
966 /*
967 * Since /proc/kallsyms will have multiple sessions for the
968 * kernel, with modules between them, fixup the end of all
969 * sections.
970 */
971 __map_groups__fixup_end(&machine->kmaps, type);
972 }
973
974 return ret;
975 }
976
977 int machine__load_kallsyms(struct machine *machine, const char *filename,
978 enum map_type type)
979 {
980 return __machine__load_kallsyms(machine, filename, type, false);
981 }
982
983 int machine__load_vmlinux_path(struct machine *machine, enum map_type type)
984 {
985 struct map *map = machine__kernel_map(machine);
986 int ret = dso__load_vmlinux_path(map->dso, map);
987
988 if (ret > 0)
989 dso__set_loaded(map->dso, type);
990
991 return ret;
992 }
993
994 static void map_groups__fixup_end(struct map_groups *mg)
995 {
996 int i;
997 for (i = 0; i < MAP__NR_TYPES; ++i)
998 __map_groups__fixup_end(mg, i);
999 }
1000
1001 static char *get_kernel_version(const char *root_dir)
1002 {
1003 char version[PATH_MAX];
1004 FILE *file;
1005 char *name, *tmp;
1006 const char *prefix = "Linux version ";
1007
1008 sprintf(version, "%s/proc/version", root_dir);
1009 file = fopen(version, "r");
1010 if (!file)
1011 return NULL;
1012
1013 version[0] = '\0';
1014 tmp = fgets(version, sizeof(version), file);
1015 fclose(file);
1016
1017 name = strstr(version, prefix);
1018 if (!name)
1019 return NULL;
1020 name += strlen(prefix);
1021 tmp = strchr(name, ' ');
1022 if (tmp)
1023 *tmp = '\0';
1024
1025 return strdup(name);
1026 }
1027
1028 static bool is_kmod_dso(struct dso *dso)
1029 {
1030 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1031 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1032 }
1033
1034 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
1035 struct kmod_path *m)
1036 {
1037 struct map *map;
1038 char *long_name;
1039
1040 map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
1041 if (map == NULL)
1042 return 0;
1043
1044 long_name = strdup(path);
1045 if (long_name == NULL)
1046 return -ENOMEM;
1047
1048 dso__set_long_name(map->dso, long_name, true);
1049 dso__kernel_module_get_build_id(map->dso, "");
1050
1051 /*
1052 * Full name could reveal us kmod compression, so
1053 * we need to update the symtab_type if needed.
1054 */
1055 if (m->comp && is_kmod_dso(map->dso))
1056 map->dso->symtab_type++;
1057
1058 return 0;
1059 }
1060
1061 static int map_groups__set_modules_path_dir(struct map_groups *mg,
1062 const char *dir_name, int depth)
1063 {
1064 struct dirent *dent;
1065 DIR *dir = opendir(dir_name);
1066 int ret = 0;
1067
1068 if (!dir) {
1069 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1070 return -1;
1071 }
1072
1073 while ((dent = readdir(dir)) != NULL) {
1074 char path[PATH_MAX];
1075 struct stat st;
1076
1077 /*sshfs might return bad dent->d_type, so we have to stat*/
1078 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1079 if (stat(path, &st))
1080 continue;
1081
1082 if (S_ISDIR(st.st_mode)) {
1083 if (!strcmp(dent->d_name, ".") ||
1084 !strcmp(dent->d_name, ".."))
1085 continue;
1086
1087 /* Do not follow top-level source and build symlinks */
1088 if (depth == 0) {
1089 if (!strcmp(dent->d_name, "source") ||
1090 !strcmp(dent->d_name, "build"))
1091 continue;
1092 }
1093
1094 ret = map_groups__set_modules_path_dir(mg, path,
1095 depth + 1);
1096 if (ret < 0)
1097 goto out;
1098 } else {
1099 struct kmod_path m;
1100
1101 ret = kmod_path__parse_name(&m, dent->d_name);
1102 if (ret)
1103 goto out;
1104
1105 if (m.kmod)
1106 ret = map_groups__set_module_path(mg, path, &m);
1107
1108 free(m.name);
1109
1110 if (ret)
1111 goto out;
1112 }
1113 }
1114
1115 out:
1116 closedir(dir);
1117 return ret;
1118 }
1119
1120 static int machine__set_modules_path(struct machine *machine)
1121 {
1122 char *version;
1123 char modules_path[PATH_MAX];
1124
1125 version = get_kernel_version(machine->root_dir);
1126 if (!version)
1127 return -1;
1128
1129 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1130 machine->root_dir, version);
1131 free(version);
1132
1133 return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1134 }
1135 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1136 const char *name __maybe_unused)
1137 {
1138 return 0;
1139 }
1140
1141 static int machine__create_module(void *arg, const char *name, u64 start,
1142 u64 size)
1143 {
1144 struct machine *machine = arg;
1145 struct map *map;
1146
1147 if (arch__fix_module_text_start(&start, name) < 0)
1148 return -1;
1149
1150 map = machine__findnew_module_map(machine, start, name);
1151 if (map == NULL)
1152 return -1;
1153 map->end = start + size;
1154
1155 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1156
1157 return 0;
1158 }
1159
1160 static int machine__create_modules(struct machine *machine)
1161 {
1162 const char *modules;
1163 char path[PATH_MAX];
1164
1165 if (machine__is_default_guest(machine)) {
1166 modules = symbol_conf.default_guest_modules;
1167 } else {
1168 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1169 modules = path;
1170 }
1171
1172 if (symbol__restricted_filename(modules, "/proc/modules"))
1173 return -1;
1174
1175 if (modules__parse(modules, machine, machine__create_module))
1176 return -1;
1177
1178 if (!machine__set_modules_path(machine))
1179 return 0;
1180
1181 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1182
1183 return 0;
1184 }
1185
1186 int machine__create_kernel_maps(struct machine *machine)
1187 {
1188 struct dso *kernel = machine__get_kernel(machine);
1189 const char *name = NULL;
1190 u64 addr = 0;
1191 int ret;
1192
1193 if (kernel == NULL)
1194 return -1;
1195
1196 ret = __machine__create_kernel_maps(machine, kernel);
1197 dso__put(kernel);
1198 if (ret < 0)
1199 return -1;
1200
1201 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1202 if (machine__is_host(machine))
1203 pr_debug("Problems creating module maps, "
1204 "continuing anyway...\n");
1205 else
1206 pr_debug("Problems creating module maps for guest %d, "
1207 "continuing anyway...\n", machine->pid);
1208 }
1209
1210 /*
1211 * Now that we have all the maps created, just set the ->end of them:
1212 */
1213 map_groups__fixup_end(&machine->kmaps);
1214
1215 if (!machine__get_running_kernel_start(machine, &name, &addr)) {
1216 if (name &&
1217 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name, addr)) {
1218 machine__destroy_kernel_maps(machine);
1219 return -1;
1220 }
1221 }
1222
1223 return 0;
1224 }
1225
1226 static void machine__set_kernel_mmap_len(struct machine *machine,
1227 union perf_event *event)
1228 {
1229 int i;
1230
1231 for (i = 0; i < MAP__NR_TYPES; i++) {
1232 machine->vmlinux_maps[i]->start = event->mmap.start;
1233 machine->vmlinux_maps[i]->end = (event->mmap.start +
1234 event->mmap.len);
1235 /*
1236 * Be a bit paranoid here, some perf.data file came with
1237 * a zero sized synthesized MMAP event for the kernel.
1238 */
1239 if (machine->vmlinux_maps[i]->end == 0)
1240 machine->vmlinux_maps[i]->end = ~0ULL;
1241 }
1242 }
1243
1244 static bool machine__uses_kcore(struct machine *machine)
1245 {
1246 struct dso *dso;
1247
1248 list_for_each_entry(dso, &machine->dsos.head, node) {
1249 if (dso__is_kcore(dso))
1250 return true;
1251 }
1252
1253 return false;
1254 }
1255
1256 static int machine__process_kernel_mmap_event(struct machine *machine,
1257 union perf_event *event)
1258 {
1259 struct map *map;
1260 char kmmap_prefix[PATH_MAX];
1261 enum dso_kernel_type kernel_type;
1262 bool is_kernel_mmap;
1263
1264 /* If we have maps from kcore then we do not need or want any others */
1265 if (machine__uses_kcore(machine))
1266 return 0;
1267
1268 machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1269 if (machine__is_host(machine))
1270 kernel_type = DSO_TYPE_KERNEL;
1271 else
1272 kernel_type = DSO_TYPE_GUEST_KERNEL;
1273
1274 is_kernel_mmap = memcmp(event->mmap.filename,
1275 kmmap_prefix,
1276 strlen(kmmap_prefix) - 1) == 0;
1277 if (event->mmap.filename[0] == '/' ||
1278 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1279 map = machine__findnew_module_map(machine, event->mmap.start,
1280 event->mmap.filename);
1281 if (map == NULL)
1282 goto out_problem;
1283
1284 map->end = map->start + event->mmap.len;
1285 } else if (is_kernel_mmap) {
1286 const char *symbol_name = (event->mmap.filename +
1287 strlen(kmmap_prefix));
1288 /*
1289 * Should be there already, from the build-id table in
1290 * the header.
1291 */
1292 struct dso *kernel = NULL;
1293 struct dso *dso;
1294
1295 pthread_rwlock_rdlock(&machine->dsos.lock);
1296
1297 list_for_each_entry(dso, &machine->dsos.head, node) {
1298
1299 /*
1300 * The cpumode passed to is_kernel_module is not the
1301 * cpumode of *this* event. If we insist on passing
1302 * correct cpumode to is_kernel_module, we should
1303 * record the cpumode when we adding this dso to the
1304 * linked list.
1305 *
1306 * However we don't really need passing correct
1307 * cpumode. We know the correct cpumode must be kernel
1308 * mode (if not, we should not link it onto kernel_dsos
1309 * list).
1310 *
1311 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1312 * is_kernel_module() treats it as a kernel cpumode.
1313 */
1314
1315 if (!dso->kernel ||
1316 is_kernel_module(dso->long_name,
1317 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1318 continue;
1319
1320
1321 kernel = dso;
1322 break;
1323 }
1324
1325 pthread_rwlock_unlock(&machine->dsos.lock);
1326
1327 if (kernel == NULL)
1328 kernel = machine__findnew_dso(machine, kmmap_prefix);
1329 if (kernel == NULL)
1330 goto out_problem;
1331
1332 kernel->kernel = kernel_type;
1333 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1334 dso__put(kernel);
1335 goto out_problem;
1336 }
1337
1338 if (strstr(kernel->long_name, "vmlinux"))
1339 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1340
1341 machine__set_kernel_mmap_len(machine, event);
1342
1343 /*
1344 * Avoid using a zero address (kptr_restrict) for the ref reloc
1345 * symbol. Effectively having zero here means that at record
1346 * time /proc/sys/kernel/kptr_restrict was non zero.
1347 */
1348 if (event->mmap.pgoff != 0) {
1349 maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1350 symbol_name,
1351 event->mmap.pgoff);
1352 }
1353
1354 if (machine__is_default_guest(machine)) {
1355 /*
1356 * preload dso of guest kernel and modules
1357 */
1358 dso__load(kernel, machine__kernel_map(machine));
1359 }
1360 }
1361 return 0;
1362 out_problem:
1363 return -1;
1364 }
1365
1366 int machine__process_mmap2_event(struct machine *machine,
1367 union perf_event *event,
1368 struct perf_sample *sample)
1369 {
1370 struct thread *thread;
1371 struct map *map;
1372 enum map_type type;
1373 int ret = 0;
1374
1375 if (dump_trace)
1376 perf_event__fprintf_mmap2(event, stdout);
1377
1378 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1379 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1380 ret = machine__process_kernel_mmap_event(machine, event);
1381 if (ret < 0)
1382 goto out_problem;
1383 return 0;
1384 }
1385
1386 thread = machine__findnew_thread(machine, event->mmap2.pid,
1387 event->mmap2.tid);
1388 if (thread == NULL)
1389 goto out_problem;
1390
1391 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1392 type = MAP__VARIABLE;
1393 else
1394 type = MAP__FUNCTION;
1395
1396 map = map__new(machine, event->mmap2.start,
1397 event->mmap2.len, event->mmap2.pgoff,
1398 event->mmap2.maj,
1399 event->mmap2.min, event->mmap2.ino,
1400 event->mmap2.ino_generation,
1401 event->mmap2.prot,
1402 event->mmap2.flags,
1403 event->mmap2.filename, type, thread);
1404
1405 if (map == NULL)
1406 goto out_problem_map;
1407
1408 ret = thread__insert_map(thread, map);
1409 if (ret)
1410 goto out_problem_insert;
1411
1412 thread__put(thread);
1413 map__put(map);
1414 return 0;
1415
1416 out_problem_insert:
1417 map__put(map);
1418 out_problem_map:
1419 thread__put(thread);
1420 out_problem:
1421 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1422 return 0;
1423 }
1424
1425 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1426 struct perf_sample *sample)
1427 {
1428 struct thread *thread;
1429 struct map *map;
1430 enum map_type type;
1431 int ret = 0;
1432
1433 if (dump_trace)
1434 perf_event__fprintf_mmap(event, stdout);
1435
1436 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1437 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1438 ret = machine__process_kernel_mmap_event(machine, event);
1439 if (ret < 0)
1440 goto out_problem;
1441 return 0;
1442 }
1443
1444 thread = machine__findnew_thread(machine, event->mmap.pid,
1445 event->mmap.tid);
1446 if (thread == NULL)
1447 goto out_problem;
1448
1449 if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1450 type = MAP__VARIABLE;
1451 else
1452 type = MAP__FUNCTION;
1453
1454 map = map__new(machine, event->mmap.start,
1455 event->mmap.len, event->mmap.pgoff,
1456 0, 0, 0, 0, 0, 0,
1457 event->mmap.filename,
1458 type, thread);
1459
1460 if (map == NULL)
1461 goto out_problem_map;
1462
1463 ret = thread__insert_map(thread, map);
1464 if (ret)
1465 goto out_problem_insert;
1466
1467 thread__put(thread);
1468 map__put(map);
1469 return 0;
1470
1471 out_problem_insert:
1472 map__put(map);
1473 out_problem_map:
1474 thread__put(thread);
1475 out_problem:
1476 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1477 return 0;
1478 }
1479
1480 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1481 {
1482 if (machine->last_match == th)
1483 machine->last_match = NULL;
1484
1485 BUG_ON(refcount_read(&th->refcnt) == 0);
1486 if (lock)
1487 pthread_rwlock_wrlock(&machine->threads_lock);
1488 rb_erase_init(&th->rb_node, &machine->threads);
1489 RB_CLEAR_NODE(&th->rb_node);
1490 --machine->nr_threads;
1491 /*
1492 * Move it first to the dead_threads list, then drop the reference,
1493 * if this is the last reference, then the thread__delete destructor
1494 * will be called and we will remove it from the dead_threads list.
1495 */
1496 list_add_tail(&th->node, &machine->dead_threads);
1497 if (lock)
1498 pthread_rwlock_unlock(&machine->threads_lock);
1499 thread__put(th);
1500 }
1501
1502 void machine__remove_thread(struct machine *machine, struct thread *th)
1503 {
1504 return __machine__remove_thread(machine, th, true);
1505 }
1506
1507 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1508 struct perf_sample *sample)
1509 {
1510 struct thread *thread = machine__find_thread(machine,
1511 event->fork.pid,
1512 event->fork.tid);
1513 struct thread *parent = machine__findnew_thread(machine,
1514 event->fork.ppid,
1515 event->fork.ptid);
1516 int err = 0;
1517
1518 if (dump_trace)
1519 perf_event__fprintf_task(event, stdout);
1520
1521 /*
1522 * There may be an existing thread that is not actually the parent,
1523 * either because we are processing events out of order, or because the
1524 * (fork) event that would have removed the thread was lost. Assume the
1525 * latter case and continue on as best we can.
1526 */
1527 if (parent->pid_ != (pid_t)event->fork.ppid) {
1528 dump_printf("removing erroneous parent thread %d/%d\n",
1529 parent->pid_, parent->tid);
1530 machine__remove_thread(machine, parent);
1531 thread__put(parent);
1532 parent = machine__findnew_thread(machine, event->fork.ppid,
1533 event->fork.ptid);
1534 }
1535
1536 /* if a thread currently exists for the thread id remove it */
1537 if (thread != NULL) {
1538 machine__remove_thread(machine, thread);
1539 thread__put(thread);
1540 }
1541
1542 thread = machine__findnew_thread(machine, event->fork.pid,
1543 event->fork.tid);
1544
1545 if (thread == NULL || parent == NULL ||
1546 thread__fork(thread, parent, sample->time) < 0) {
1547 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1548 err = -1;
1549 }
1550 thread__put(thread);
1551 thread__put(parent);
1552
1553 return err;
1554 }
1555
1556 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1557 struct perf_sample *sample __maybe_unused)
1558 {
1559 struct thread *thread = machine__find_thread(machine,
1560 event->fork.pid,
1561 event->fork.tid);
1562
1563 if (dump_trace)
1564 perf_event__fprintf_task(event, stdout);
1565
1566 if (thread != NULL) {
1567 thread__exited(thread);
1568 thread__put(thread);
1569 }
1570
1571 return 0;
1572 }
1573
1574 int machine__process_event(struct machine *machine, union perf_event *event,
1575 struct perf_sample *sample)
1576 {
1577 int ret;
1578
1579 switch (event->header.type) {
1580 case PERF_RECORD_COMM:
1581 ret = machine__process_comm_event(machine, event, sample); break;
1582 case PERF_RECORD_MMAP:
1583 ret = machine__process_mmap_event(machine, event, sample); break;
1584 case PERF_RECORD_NAMESPACES:
1585 ret = machine__process_namespaces_event(machine, event, sample); break;
1586 case PERF_RECORD_MMAP2:
1587 ret = machine__process_mmap2_event(machine, event, sample); break;
1588 case PERF_RECORD_FORK:
1589 ret = machine__process_fork_event(machine, event, sample); break;
1590 case PERF_RECORD_EXIT:
1591 ret = machine__process_exit_event(machine, event, sample); break;
1592 case PERF_RECORD_LOST:
1593 ret = machine__process_lost_event(machine, event, sample); break;
1594 case PERF_RECORD_AUX:
1595 ret = machine__process_aux_event(machine, event); break;
1596 case PERF_RECORD_ITRACE_START:
1597 ret = machine__process_itrace_start_event(machine, event); break;
1598 case PERF_RECORD_LOST_SAMPLES:
1599 ret = machine__process_lost_samples_event(machine, event, sample); break;
1600 case PERF_RECORD_SWITCH:
1601 case PERF_RECORD_SWITCH_CPU_WIDE:
1602 ret = machine__process_switch_event(machine, event); break;
1603 default:
1604 ret = -1;
1605 break;
1606 }
1607
1608 return ret;
1609 }
1610
1611 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1612 {
1613 if (!regexec(regex, sym->name, 0, NULL, 0))
1614 return 1;
1615 return 0;
1616 }
1617
1618 static void ip__resolve_ams(struct thread *thread,
1619 struct addr_map_symbol *ams,
1620 u64 ip)
1621 {
1622 struct addr_location al;
1623
1624 memset(&al, 0, sizeof(al));
1625 /*
1626 * We cannot use the header.misc hint to determine whether a
1627 * branch stack address is user, kernel, guest, hypervisor.
1628 * Branches may straddle the kernel/user/hypervisor boundaries.
1629 * Thus, we have to try consecutively until we find a match
1630 * or else, the symbol is unknown
1631 */
1632 thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1633
1634 ams->addr = ip;
1635 ams->al_addr = al.addr;
1636 ams->sym = al.sym;
1637 ams->map = al.map;
1638 ams->phys_addr = 0;
1639 }
1640
1641 static void ip__resolve_data(struct thread *thread,
1642 u8 m, struct addr_map_symbol *ams,
1643 u64 addr, u64 phys_addr)
1644 {
1645 struct addr_location al;
1646
1647 memset(&al, 0, sizeof(al));
1648
1649 thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1650 if (al.map == NULL) {
1651 /*
1652 * some shared data regions have execute bit set which puts
1653 * their mapping in the MAP__FUNCTION type array.
1654 * Check there as a fallback option before dropping the sample.
1655 */
1656 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1657 }
1658
1659 ams->addr = addr;
1660 ams->al_addr = al.addr;
1661 ams->sym = al.sym;
1662 ams->map = al.map;
1663 ams->phys_addr = phys_addr;
1664 }
1665
1666 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1667 struct addr_location *al)
1668 {
1669 struct mem_info *mi = zalloc(sizeof(*mi));
1670
1671 if (!mi)
1672 return NULL;
1673
1674 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1675 ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
1676 sample->addr, sample->phys_addr);
1677 mi->data_src.val = sample->data_src;
1678
1679 return mi;
1680 }
1681
1682 struct iterations {
1683 int nr_loop_iter;
1684 u64 cycles;
1685 };
1686
1687 static int add_callchain_ip(struct thread *thread,
1688 struct callchain_cursor *cursor,
1689 struct symbol **parent,
1690 struct addr_location *root_al,
1691 u8 *cpumode,
1692 u64 ip,
1693 bool branch,
1694 struct branch_flags *flags,
1695 struct iterations *iter,
1696 u64 branch_from)
1697 {
1698 struct addr_location al;
1699 int nr_loop_iter = 0;
1700 u64 iter_cycles = 0;
1701
1702 al.filtered = 0;
1703 al.sym = NULL;
1704 if (!cpumode) {
1705 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1706 ip, &al);
1707 } else {
1708 if (ip >= PERF_CONTEXT_MAX) {
1709 switch (ip) {
1710 case PERF_CONTEXT_HV:
1711 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1712 break;
1713 case PERF_CONTEXT_KERNEL:
1714 *cpumode = PERF_RECORD_MISC_KERNEL;
1715 break;
1716 case PERF_CONTEXT_USER:
1717 *cpumode = PERF_RECORD_MISC_USER;
1718 break;
1719 default:
1720 pr_debug("invalid callchain context: "
1721 "%"PRId64"\n", (s64) ip);
1722 /*
1723 * It seems the callchain is corrupted.
1724 * Discard all.
1725 */
1726 callchain_cursor_reset(cursor);
1727 return 1;
1728 }
1729 return 0;
1730 }
1731 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1732 ip, &al);
1733 }
1734
1735 if (al.sym != NULL) {
1736 if (perf_hpp_list.parent && !*parent &&
1737 symbol__match_regex(al.sym, &parent_regex))
1738 *parent = al.sym;
1739 else if (have_ignore_callees && root_al &&
1740 symbol__match_regex(al.sym, &ignore_callees_regex)) {
1741 /* Treat this symbol as the root,
1742 forgetting its callees. */
1743 *root_al = al;
1744 callchain_cursor_reset(cursor);
1745 }
1746 }
1747
1748 if (symbol_conf.hide_unresolved && al.sym == NULL)
1749 return 0;
1750
1751 if (iter) {
1752 nr_loop_iter = iter->nr_loop_iter;
1753 iter_cycles = iter->cycles;
1754 }
1755
1756 return callchain_cursor_append(cursor, al.addr, al.map, al.sym,
1757 branch, flags, nr_loop_iter,
1758 iter_cycles, branch_from);
1759 }
1760
1761 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1762 struct addr_location *al)
1763 {
1764 unsigned int i;
1765 const struct branch_stack *bs = sample->branch_stack;
1766 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1767
1768 if (!bi)
1769 return NULL;
1770
1771 for (i = 0; i < bs->nr; i++) {
1772 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1773 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1774 bi[i].flags = bs->entries[i].flags;
1775 }
1776 return bi;
1777 }
1778
1779 static void save_iterations(struct iterations *iter,
1780 struct branch_entry *be, int nr)
1781 {
1782 int i;
1783
1784 iter->nr_loop_iter = nr;
1785 iter->cycles = 0;
1786
1787 for (i = 0; i < nr; i++)
1788 iter->cycles += be[i].flags.cycles;
1789 }
1790
1791 #define CHASHSZ 127
1792 #define CHASHBITS 7
1793 #define NO_ENTRY 0xff
1794
1795 #define PERF_MAX_BRANCH_DEPTH 127
1796
1797 /* Remove loops. */
1798 static int remove_loops(struct branch_entry *l, int nr,
1799 struct iterations *iter)
1800 {
1801 int i, j, off;
1802 unsigned char chash[CHASHSZ];
1803
1804 memset(chash, NO_ENTRY, sizeof(chash));
1805
1806 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1807
1808 for (i = 0; i < nr; i++) {
1809 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1810
1811 /* no collision handling for now */
1812 if (chash[h] == NO_ENTRY) {
1813 chash[h] = i;
1814 } else if (l[chash[h]].from == l[i].from) {
1815 bool is_loop = true;
1816 /* check if it is a real loop */
1817 off = 0;
1818 for (j = chash[h]; j < i && i + off < nr; j++, off++)
1819 if (l[j].from != l[i + off].from) {
1820 is_loop = false;
1821 break;
1822 }
1823 if (is_loop) {
1824 j = nr - (i + off);
1825 if (j > 0) {
1826 save_iterations(iter + i + off,
1827 l + i, off);
1828
1829 memmove(iter + i, iter + i + off,
1830 j * sizeof(*iter));
1831
1832 memmove(l + i, l + i + off,
1833 j * sizeof(*l));
1834 }
1835
1836 nr -= off;
1837 }
1838 }
1839 }
1840 return nr;
1841 }
1842
1843 /*
1844 * Recolve LBR callstack chain sample
1845 * Return:
1846 * 1 on success get LBR callchain information
1847 * 0 no available LBR callchain information, should try fp
1848 * negative error code on other errors.
1849 */
1850 static int resolve_lbr_callchain_sample(struct thread *thread,
1851 struct callchain_cursor *cursor,
1852 struct perf_sample *sample,
1853 struct symbol **parent,
1854 struct addr_location *root_al,
1855 int max_stack)
1856 {
1857 struct ip_callchain *chain = sample->callchain;
1858 int chain_nr = min(max_stack, (int)chain->nr), i;
1859 u8 cpumode = PERF_RECORD_MISC_USER;
1860 u64 ip, branch_from = 0;
1861
1862 for (i = 0; i < chain_nr; i++) {
1863 if (chain->ips[i] == PERF_CONTEXT_USER)
1864 break;
1865 }
1866
1867 /* LBR only affects the user callchain */
1868 if (i != chain_nr) {
1869 struct branch_stack *lbr_stack = sample->branch_stack;
1870 int lbr_nr = lbr_stack->nr, j, k;
1871 bool branch;
1872 struct branch_flags *flags;
1873 /*
1874 * LBR callstack can only get user call chain.
1875 * The mix_chain_nr is kernel call chain
1876 * number plus LBR user call chain number.
1877 * i is kernel call chain number,
1878 * 1 is PERF_CONTEXT_USER,
1879 * lbr_nr + 1 is the user call chain number.
1880 * For details, please refer to the comments
1881 * in callchain__printf
1882 */
1883 int mix_chain_nr = i + 1 + lbr_nr + 1;
1884
1885 for (j = 0; j < mix_chain_nr; j++) {
1886 int err;
1887 branch = false;
1888 flags = NULL;
1889
1890 if (callchain_param.order == ORDER_CALLEE) {
1891 if (j < i + 1)
1892 ip = chain->ips[j];
1893 else if (j > i + 1) {
1894 k = j - i - 2;
1895 ip = lbr_stack->entries[k].from;
1896 branch = true;
1897 flags = &lbr_stack->entries[k].flags;
1898 } else {
1899 ip = lbr_stack->entries[0].to;
1900 branch = true;
1901 flags = &lbr_stack->entries[0].flags;
1902 branch_from =
1903 lbr_stack->entries[0].from;
1904 }
1905 } else {
1906 if (j < lbr_nr) {
1907 k = lbr_nr - j - 1;
1908 ip = lbr_stack->entries[k].from;
1909 branch = true;
1910 flags = &lbr_stack->entries[k].flags;
1911 }
1912 else if (j > lbr_nr)
1913 ip = chain->ips[i + 1 - (j - lbr_nr)];
1914 else {
1915 ip = lbr_stack->entries[0].to;
1916 branch = true;
1917 flags = &lbr_stack->entries[0].flags;
1918 branch_from =
1919 lbr_stack->entries[0].from;
1920 }
1921 }
1922
1923 err = add_callchain_ip(thread, cursor, parent,
1924 root_al, &cpumode, ip,
1925 branch, flags, NULL,
1926 branch_from);
1927 if (err)
1928 return (err < 0) ? err : 0;
1929 }
1930 return 1;
1931 }
1932
1933 return 0;
1934 }
1935
1936 static int thread__resolve_callchain_sample(struct thread *thread,
1937 struct callchain_cursor *cursor,
1938 struct perf_evsel *evsel,
1939 struct perf_sample *sample,
1940 struct symbol **parent,
1941 struct addr_location *root_al,
1942 int max_stack)
1943 {
1944 struct branch_stack *branch = sample->branch_stack;
1945 struct ip_callchain *chain = sample->callchain;
1946 int chain_nr = 0;
1947 u8 cpumode = PERF_RECORD_MISC_USER;
1948 int i, j, err, nr_entries;
1949 int skip_idx = -1;
1950 int first_call = 0;
1951
1952 if (chain)
1953 chain_nr = chain->nr;
1954
1955 if (perf_evsel__has_branch_callstack(evsel)) {
1956 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
1957 root_al, max_stack);
1958 if (err)
1959 return (err < 0) ? err : 0;
1960 }
1961
1962 /*
1963 * Based on DWARF debug information, some architectures skip
1964 * a callchain entry saved by the kernel.
1965 */
1966 skip_idx = arch_skip_callchain_idx(thread, chain);
1967
1968 /*
1969 * Add branches to call stack for easier browsing. This gives
1970 * more context for a sample than just the callers.
1971 *
1972 * This uses individual histograms of paths compared to the
1973 * aggregated histograms the normal LBR mode uses.
1974 *
1975 * Limitations for now:
1976 * - No extra filters
1977 * - No annotations (should annotate somehow)
1978 */
1979
1980 if (branch && callchain_param.branch_callstack) {
1981 int nr = min(max_stack, (int)branch->nr);
1982 struct branch_entry be[nr];
1983 struct iterations iter[nr];
1984
1985 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1986 pr_warning("corrupted branch chain. skipping...\n");
1987 goto check_calls;
1988 }
1989
1990 for (i = 0; i < nr; i++) {
1991 if (callchain_param.order == ORDER_CALLEE) {
1992 be[i] = branch->entries[i];
1993
1994 if (chain == NULL)
1995 continue;
1996
1997 /*
1998 * Check for overlap into the callchain.
1999 * The return address is one off compared to
2000 * the branch entry. To adjust for this
2001 * assume the calling instruction is not longer
2002 * than 8 bytes.
2003 */
2004 if (i == skip_idx ||
2005 chain->ips[first_call] >= PERF_CONTEXT_MAX)
2006 first_call++;
2007 else if (be[i].from < chain->ips[first_call] &&
2008 be[i].from >= chain->ips[first_call] - 8)
2009 first_call++;
2010 } else
2011 be[i] = branch->entries[branch->nr - i - 1];
2012 }
2013
2014 memset(iter, 0, sizeof(struct iterations) * nr);
2015 nr = remove_loops(be, nr, iter);
2016
2017 for (i = 0; i < nr; i++) {
2018 err = add_callchain_ip(thread, cursor, parent,
2019 root_al,
2020 NULL, be[i].to,
2021 true, &be[i].flags,
2022 NULL, be[i].from);
2023
2024 if (!err)
2025 err = add_callchain_ip(thread, cursor, parent, root_al,
2026 NULL, be[i].from,
2027 true, &be[i].flags,
2028 &iter[i], 0);
2029 if (err == -EINVAL)
2030 break;
2031 if (err)
2032 return err;
2033 }
2034
2035 if (chain_nr == 0)
2036 return 0;
2037
2038 chain_nr -= nr;
2039 }
2040
2041 check_calls:
2042 for (i = first_call, nr_entries = 0;
2043 i < chain_nr && nr_entries < max_stack; i++) {
2044 u64 ip;
2045
2046 if (callchain_param.order == ORDER_CALLEE)
2047 j = i;
2048 else
2049 j = chain->nr - i - 1;
2050
2051 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2052 if (j == skip_idx)
2053 continue;
2054 #endif
2055 ip = chain->ips[j];
2056
2057 if (ip < PERF_CONTEXT_MAX)
2058 ++nr_entries;
2059
2060 err = add_callchain_ip(thread, cursor, parent,
2061 root_al, &cpumode, ip,
2062 false, NULL, NULL, 0);
2063
2064 if (err)
2065 return (err < 0) ? err : 0;
2066 }
2067
2068 return 0;
2069 }
2070
2071 static int unwind_entry(struct unwind_entry *entry, void *arg)
2072 {
2073 struct callchain_cursor *cursor = arg;
2074
2075 if (symbol_conf.hide_unresolved && entry->sym == NULL)
2076 return 0;
2077 return callchain_cursor_append(cursor, entry->ip,
2078 entry->map, entry->sym,
2079 false, NULL, 0, 0, 0);
2080 }
2081
2082 static int thread__resolve_callchain_unwind(struct thread *thread,
2083 struct callchain_cursor *cursor,
2084 struct perf_evsel *evsel,
2085 struct perf_sample *sample,
2086 int max_stack)
2087 {
2088 /* Can we do dwarf post unwind? */
2089 if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2090 (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
2091 return 0;
2092
2093 /* Bail out if nothing was captured. */
2094 if ((!sample->user_regs.regs) ||
2095 (!sample->user_stack.size))
2096 return 0;
2097
2098 return unwind__get_entries(unwind_entry, cursor,
2099 thread, sample, max_stack);
2100 }
2101
2102 int thread__resolve_callchain(struct thread *thread,
2103 struct callchain_cursor *cursor,
2104 struct perf_evsel *evsel,
2105 struct perf_sample *sample,
2106 struct symbol **parent,
2107 struct addr_location *root_al,
2108 int max_stack)
2109 {
2110 int ret = 0;
2111
2112 callchain_cursor_reset(&callchain_cursor);
2113
2114 if (callchain_param.order == ORDER_CALLEE) {
2115 ret = thread__resolve_callchain_sample(thread, cursor,
2116 evsel, sample,
2117 parent, root_al,
2118 max_stack);
2119 if (ret)
2120 return ret;
2121 ret = thread__resolve_callchain_unwind(thread, cursor,
2122 evsel, sample,
2123 max_stack);
2124 } else {
2125 ret = thread__resolve_callchain_unwind(thread, cursor,
2126 evsel, sample,
2127 max_stack);
2128 if (ret)
2129 return ret;
2130 ret = thread__resolve_callchain_sample(thread, cursor,
2131 evsel, sample,
2132 parent, root_al,
2133 max_stack);
2134 }
2135
2136 return ret;
2137 }
2138
2139 int machine__for_each_thread(struct machine *machine,
2140 int (*fn)(struct thread *thread, void *p),
2141 void *priv)
2142 {
2143 struct rb_node *nd;
2144 struct thread *thread;
2145 int rc = 0;
2146
2147 for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
2148 thread = rb_entry(nd, struct thread, rb_node);
2149 rc = fn(thread, priv);
2150 if (rc != 0)
2151 return rc;
2152 }
2153
2154 list_for_each_entry(thread, &machine->dead_threads, node) {
2155 rc = fn(thread, priv);
2156 if (rc != 0)
2157 return rc;
2158 }
2159 return rc;
2160 }
2161
2162 int machines__for_each_thread(struct machines *machines,
2163 int (*fn)(struct thread *thread, void *p),
2164 void *priv)
2165 {
2166 struct rb_node *nd;
2167 int rc = 0;
2168
2169 rc = machine__for_each_thread(&machines->host, fn, priv);
2170 if (rc != 0)
2171 return rc;
2172
2173 for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
2174 struct machine *machine = rb_entry(nd, struct machine, rb_node);
2175
2176 rc = machine__for_each_thread(machine, fn, priv);
2177 if (rc != 0)
2178 return rc;
2179 }
2180 return rc;
2181 }
2182
2183 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
2184 struct target *target, struct thread_map *threads,
2185 perf_event__handler_t process, bool data_mmap,
2186 unsigned int proc_map_timeout)
2187 {
2188 if (target__has_task(target))
2189 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
2190 else if (target__has_cpu(target))
2191 return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
2192 /* command specified */
2193 return 0;
2194 }
2195
2196 pid_t machine__get_current_tid(struct machine *machine, int cpu)
2197 {
2198 if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
2199 return -1;
2200
2201 return machine->current_tid[cpu];
2202 }
2203
2204 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
2205 pid_t tid)
2206 {
2207 struct thread *thread;
2208
2209 if (cpu < 0)
2210 return -EINVAL;
2211
2212 if (!machine->current_tid) {
2213 int i;
2214
2215 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
2216 if (!machine->current_tid)
2217 return -ENOMEM;
2218 for (i = 0; i < MAX_NR_CPUS; i++)
2219 machine->current_tid[i] = -1;
2220 }
2221
2222 if (cpu >= MAX_NR_CPUS) {
2223 pr_err("Requested CPU %d too large. ", cpu);
2224 pr_err("Consider raising MAX_NR_CPUS\n");
2225 return -EINVAL;
2226 }
2227
2228 machine->current_tid[cpu] = tid;
2229
2230 thread = machine__findnew_thread(machine, pid, tid);
2231 if (!thread)
2232 return -ENOMEM;
2233
2234 thread->cpu = cpu;
2235 thread__put(thread);
2236
2237 return 0;
2238 }
2239
2240 int machine__get_kernel_start(struct machine *machine)
2241 {
2242 struct map *map = machine__kernel_map(machine);
2243 int err = 0;
2244
2245 /*
2246 * The only addresses above 2^63 are kernel addresses of a 64-bit
2247 * kernel. Note that addresses are unsigned so that on a 32-bit system
2248 * all addresses including kernel addresses are less than 2^32. In
2249 * that case (32-bit system), if the kernel mapping is unknown, all
2250 * addresses will be assumed to be in user space - see
2251 * machine__kernel_ip().
2252 */
2253 machine->kernel_start = 1ULL << 63;
2254 if (map) {
2255 err = map__load(map);
2256 if (!err)
2257 machine->kernel_start = map->start;
2258 }
2259 return err;
2260 }
2261
2262 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2263 {
2264 return dsos__findnew(&machine->dsos, filename);
2265 }
2266
2267 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2268 {
2269 struct machine *machine = vmachine;
2270 struct map *map;
2271 struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map);
2272
2273 if (sym == NULL)
2274 return NULL;
2275
2276 *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2277 *addrp = map->unmap_ip(map, sym->start);
2278 return sym->name;
2279 }
Ubuntu Focal Linux kernel mirror