xref: /openbmc/linux/tools/perf/util/machine.c (revision 165f2d28)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <dirent.h>
3 #include <errno.h>
4 #include <inttypes.h>
5 #include <regex.h>
6 #include <stdlib.h>
7 #include "callchain.h"
8 #include "debug.h"
9 #include "dso.h"
10 #include "env.h"
11 #include "event.h"
12 #include "evsel.h"
13 #include "hist.h"
14 #include "machine.h"
15 #include "map.h"
16 #include "map_symbol.h"
17 #include "branch.h"
18 #include "mem-events.h"
19 #include "srcline.h"
20 #include "symbol.h"
21 #include "sort.h"
22 #include "strlist.h"
23 #include "target.h"
24 #include "thread.h"
25 #include "util.h"
26 #include "vdso.h"
27 #include <stdbool.h>
28 #include <sys/types.h>
29 #include <sys/stat.h>
30 #include <unistd.h>
31 #include "unwind.h"
32 #include "linux/hash.h"
33 #include "asm/bug.h"
34 #include "bpf-event.h"
35 #include <internal/lib.h> // page_size
36 #include "cgroup.h"
37 
38 #include <linux/ctype.h>
39 #include <symbol/kallsyms.h>
40 #include <linux/mman.h>
41 #include <linux/string.h>
42 #include <linux/zalloc.h>
43 
44 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
45 
46 static struct dso *machine__kernel_dso(struct machine *machine)
47 {
48 	return machine->vmlinux_map->dso;
49 }
50 
51 static void dsos__init(struct dsos *dsos)
52 {
53 	INIT_LIST_HEAD(&dsos->head);
54 	dsos->root = RB_ROOT;
55 	init_rwsem(&dsos->lock);
56 }
57 
58 static void machine__threads_init(struct machine *machine)
59 {
60 	int i;
61 
62 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
63 		struct threads *threads = &machine->threads[i];
64 		threads->entries = RB_ROOT_CACHED;
65 		init_rwsem(&threads->lock);
66 		threads->nr = 0;
67 		INIT_LIST_HEAD(&threads->dead);
68 		threads->last_match = NULL;
69 	}
70 }
71 
72 static int machine__set_mmap_name(struct machine *machine)
73 {
74 	if (machine__is_host(machine))
75 		machine->mmap_name = strdup("[kernel.kallsyms]");
76 	else if (machine__is_default_guest(machine))
77 		machine->mmap_name = strdup("[guest.kernel.kallsyms]");
78 	else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
79 			  machine->pid) < 0)
80 		machine->mmap_name = NULL;
81 
82 	return machine->mmap_name ? 0 : -ENOMEM;
83 }
84 
85 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
86 {
87 	int err = -ENOMEM;
88 
89 	memset(machine, 0, sizeof(*machine));
90 	maps__init(&machine->kmaps, machine);
91 	RB_CLEAR_NODE(&machine->rb_node);
92 	dsos__init(&machine->dsos);
93 
94 	machine__threads_init(machine);
95 
96 	machine->vdso_info = NULL;
97 	machine->env = NULL;
98 
99 	machine->pid = pid;
100 
101 	machine->id_hdr_size = 0;
102 	machine->kptr_restrict_warned = false;
103 	machine->comm_exec = false;
104 	machine->kernel_start = 0;
105 	machine->vmlinux_map = NULL;
106 
107 	machine->root_dir = strdup(root_dir);
108 	if (machine->root_dir == NULL)
109 		return -ENOMEM;
110 
111 	if (machine__set_mmap_name(machine))
112 		goto out;
113 
114 	if (pid != HOST_KERNEL_ID) {
115 		struct thread *thread = machine__findnew_thread(machine, -1,
116 								pid);
117 		char comm[64];
118 
119 		if (thread == NULL)
120 			goto out;
121 
122 		snprintf(comm, sizeof(comm), "[guest/%d]", pid);
123 		thread__set_comm(thread, comm, 0);
124 		thread__put(thread);
125 	}
126 
127 	machine->current_tid = NULL;
128 	err = 0;
129 
130 out:
131 	if (err) {
132 		zfree(&machine->root_dir);
133 		zfree(&machine->mmap_name);
134 	}
135 	return 0;
136 }
137 
138 struct machine *machine__new_host(void)
139 {
140 	struct machine *machine = malloc(sizeof(*machine));
141 
142 	if (machine != NULL) {
143 		machine__init(machine, "", HOST_KERNEL_ID);
144 
145 		if (machine__create_kernel_maps(machine) < 0)
146 			goto out_delete;
147 	}
148 
149 	return machine;
150 out_delete:
151 	free(machine);
152 	return NULL;
153 }
154 
155 struct machine *machine__new_kallsyms(void)
156 {
157 	struct machine *machine = machine__new_host();
158 	/*
159 	 * FIXME:
160 	 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
161 	 *    ask for not using the kcore parsing code, once this one is fixed
162 	 *    to create a map per module.
163 	 */
164 	if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
165 		machine__delete(machine);
166 		machine = NULL;
167 	}
168 
169 	return machine;
170 }
171 
172 static void dsos__purge(struct dsos *dsos)
173 {
174 	struct dso *pos, *n;
175 
176 	down_write(&dsos->lock);
177 
178 	list_for_each_entry_safe(pos, n, &dsos->head, node) {
179 		RB_CLEAR_NODE(&pos->rb_node);
180 		pos->root = NULL;
181 		list_del_init(&pos->node);
182 		dso__put(pos);
183 	}
184 
185 	up_write(&dsos->lock);
186 }
187 
188 static void dsos__exit(struct dsos *dsos)
189 {
190 	dsos__purge(dsos);
191 	exit_rwsem(&dsos->lock);
192 }
193 
194 void machine__delete_threads(struct machine *machine)
195 {
196 	struct rb_node *nd;
197 	int i;
198 
199 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
200 		struct threads *threads = &machine->threads[i];
201 		down_write(&threads->lock);
202 		nd = rb_first_cached(&threads->entries);
203 		while (nd) {
204 			struct thread *t = rb_entry(nd, struct thread, rb_node);
205 
206 			nd = rb_next(nd);
207 			__machine__remove_thread(machine, t, false);
208 		}
209 		up_write(&threads->lock);
210 	}
211 }
212 
213 void machine__exit(struct machine *machine)
214 {
215 	int i;
216 
217 	if (machine == NULL)
218 		return;
219 
220 	machine__destroy_kernel_maps(machine);
221 	maps__exit(&machine->kmaps);
222 	dsos__exit(&machine->dsos);
223 	machine__exit_vdso(machine);
224 	zfree(&machine->root_dir);
225 	zfree(&machine->mmap_name);
226 	zfree(&machine->current_tid);
227 
228 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
229 		struct threads *threads = &machine->threads[i];
230 		struct thread *thread, *n;
231 		/*
232 		 * Forget about the dead, at this point whatever threads were
233 		 * left in the dead lists better have a reference count taken
234 		 * by who is using them, and then, when they drop those references
235 		 * and it finally hits zero, thread__put() will check and see that
236 		 * its not in the dead threads list and will not try to remove it
237 		 * from there, just calling thread__delete() straight away.
238 		 */
239 		list_for_each_entry_safe(thread, n, &threads->dead, node)
240 			list_del_init(&thread->node);
241 
242 		exit_rwsem(&threads->lock);
243 	}
244 }
245 
246 void machine__delete(struct machine *machine)
247 {
248 	if (machine) {
249 		machine__exit(machine);
250 		free(machine);
251 	}
252 }
253 
254 void machines__init(struct machines *machines)
255 {
256 	machine__init(&machines->host, "", HOST_KERNEL_ID);
257 	machines->guests = RB_ROOT_CACHED;
258 }
259 
260 void machines__exit(struct machines *machines)
261 {
262 	machine__exit(&machines->host);
263 	/* XXX exit guest */
264 }
265 
266 struct machine *machines__add(struct machines *machines, pid_t pid,
267 			      const char *root_dir)
268 {
269 	struct rb_node **p = &machines->guests.rb_root.rb_node;
270 	struct rb_node *parent = NULL;
271 	struct machine *pos, *machine = malloc(sizeof(*machine));
272 	bool leftmost = true;
273 
274 	if (machine == NULL)
275 		return NULL;
276 
277 	if (machine__init(machine, root_dir, pid) != 0) {
278 		free(machine);
279 		return NULL;
280 	}
281 
282 	while (*p != NULL) {
283 		parent = *p;
284 		pos = rb_entry(parent, struct machine, rb_node);
285 		if (pid < pos->pid)
286 			p = &(*p)->rb_left;
287 		else {
288 			p = &(*p)->rb_right;
289 			leftmost = false;
290 		}
291 	}
292 
293 	rb_link_node(&machine->rb_node, parent, p);
294 	rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
295 
296 	return machine;
297 }
298 
299 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
300 {
301 	struct rb_node *nd;
302 
303 	machines->host.comm_exec = comm_exec;
304 
305 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
306 		struct machine *machine = rb_entry(nd, struct machine, rb_node);
307 
308 		machine->comm_exec = comm_exec;
309 	}
310 }
311 
312 struct machine *machines__find(struct machines *machines, pid_t pid)
313 {
314 	struct rb_node **p = &machines->guests.rb_root.rb_node;
315 	struct rb_node *parent = NULL;
316 	struct machine *machine;
317 	struct machine *default_machine = NULL;
318 
319 	if (pid == HOST_KERNEL_ID)
320 		return &machines->host;
321 
322 	while (*p != NULL) {
323 		parent = *p;
324 		machine = rb_entry(parent, struct machine, rb_node);
325 		if (pid < machine->pid)
326 			p = &(*p)->rb_left;
327 		else if (pid > machine->pid)
328 			p = &(*p)->rb_right;
329 		else
330 			return machine;
331 		if (!machine->pid)
332 			default_machine = machine;
333 	}
334 
335 	return default_machine;
336 }
337 
338 struct machine *machines__findnew(struct machines *machines, pid_t pid)
339 {
340 	char path[PATH_MAX];
341 	const char *root_dir = "";
342 	struct machine *machine = machines__find(machines, pid);
343 
344 	if (machine && (machine->pid == pid))
345 		goto out;
346 
347 	if ((pid != HOST_KERNEL_ID) &&
348 	    (pid != DEFAULT_GUEST_KERNEL_ID) &&
349 	    (symbol_conf.guestmount)) {
350 		sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
351 		if (access(path, R_OK)) {
352 			static struct strlist *seen;
353 
354 			if (!seen)
355 				seen = strlist__new(NULL, NULL);
356 
357 			if (!strlist__has_entry(seen, path)) {
358 				pr_err("Can't access file %s\n", path);
359 				strlist__add(seen, path);
360 			}
361 			machine = NULL;
362 			goto out;
363 		}
364 		root_dir = path;
365 	}
366 
367 	machine = machines__add(machines, pid, root_dir);
368 out:
369 	return machine;
370 }
371 
372 void machines__process_guests(struct machines *machines,
373 			      machine__process_t process, void *data)
374 {
375 	struct rb_node *nd;
376 
377 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
378 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
379 		process(pos, data);
380 	}
381 }
382 
383 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
384 {
385 	struct rb_node *node;
386 	struct machine *machine;
387 
388 	machines->host.id_hdr_size = id_hdr_size;
389 
390 	for (node = rb_first_cached(&machines->guests); node;
391 	     node = rb_next(node)) {
392 		machine = rb_entry(node, struct machine, rb_node);
393 		machine->id_hdr_size = id_hdr_size;
394 	}
395 
396 	return;
397 }
398 
399 static void machine__update_thread_pid(struct machine *machine,
400 				       struct thread *th, pid_t pid)
401 {
402 	struct thread *leader;
403 
404 	if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
405 		return;
406 
407 	th->pid_ = pid;
408 
409 	if (th->pid_ == th->tid)
410 		return;
411 
412 	leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
413 	if (!leader)
414 		goto out_err;
415 
416 	if (!leader->maps)
417 		leader->maps = maps__new(machine);
418 
419 	if (!leader->maps)
420 		goto out_err;
421 
422 	if (th->maps == leader->maps)
423 		return;
424 
425 	if (th->maps) {
426 		/*
427 		 * Maps are created from MMAP events which provide the pid and
428 		 * tid.  Consequently there never should be any maps on a thread
429 		 * with an unknown pid.  Just print an error if there are.
430 		 */
431 		if (!maps__empty(th->maps))
432 			pr_err("Discarding thread maps for %d:%d\n",
433 			       th->pid_, th->tid);
434 		maps__put(th->maps);
435 	}
436 
437 	th->maps = maps__get(leader->maps);
438 out_put:
439 	thread__put(leader);
440 	return;
441 out_err:
442 	pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
443 	goto out_put;
444 }
445 
446 /*
447  * Front-end cache - TID lookups come in blocks,
448  * so most of the time we dont have to look up
449  * the full rbtree:
450  */
451 static struct thread*
452 __threads__get_last_match(struct threads *threads, struct machine *machine,
453 			  int pid, int tid)
454 {
455 	struct thread *th;
456 
457 	th = threads->last_match;
458 	if (th != NULL) {
459 		if (th->tid == tid) {
460 			machine__update_thread_pid(machine, th, pid);
461 			return thread__get(th);
462 		}
463 
464 		threads->last_match = NULL;
465 	}
466 
467 	return NULL;
468 }
469 
470 static struct thread*
471 threads__get_last_match(struct threads *threads, struct machine *machine,
472 			int pid, int tid)
473 {
474 	struct thread *th = NULL;
475 
476 	if (perf_singlethreaded)
477 		th = __threads__get_last_match(threads, machine, pid, tid);
478 
479 	return th;
480 }
481 
482 static void
483 __threads__set_last_match(struct threads *threads, struct thread *th)
484 {
485 	threads->last_match = th;
486 }
487 
488 static void
489 threads__set_last_match(struct threads *threads, struct thread *th)
490 {
491 	if (perf_singlethreaded)
492 		__threads__set_last_match(threads, th);
493 }
494 
495 /*
496  * Caller must eventually drop thread->refcnt returned with a successful
497  * lookup/new thread inserted.
498  */
499 static struct thread *____machine__findnew_thread(struct machine *machine,
500 						  struct threads *threads,
501 						  pid_t pid, pid_t tid,
502 						  bool create)
503 {
504 	struct rb_node **p = &threads->entries.rb_root.rb_node;
505 	struct rb_node *parent = NULL;
506 	struct thread *th;
507 	bool leftmost = true;
508 
509 	th = threads__get_last_match(threads, machine, pid, tid);
510 	if (th)
511 		return th;
512 
513 	while (*p != NULL) {
514 		parent = *p;
515 		th = rb_entry(parent, struct thread, rb_node);
516 
517 		if (th->tid == tid) {
518 			threads__set_last_match(threads, th);
519 			machine__update_thread_pid(machine, th, pid);
520 			return thread__get(th);
521 		}
522 
523 		if (tid < th->tid)
524 			p = &(*p)->rb_left;
525 		else {
526 			p = &(*p)->rb_right;
527 			leftmost = false;
528 		}
529 	}
530 
531 	if (!create)
532 		return NULL;
533 
534 	th = thread__new(pid, tid);
535 	if (th != NULL) {
536 		rb_link_node(&th->rb_node, parent, p);
537 		rb_insert_color_cached(&th->rb_node, &threads->entries, leftmost);
538 
539 		/*
540 		 * We have to initialize maps separately after rb tree is updated.
541 		 *
542 		 * The reason is that we call machine__findnew_thread
543 		 * within thread__init_maps to find the thread
544 		 * leader and that would screwed the rb tree.
545 		 */
546 		if (thread__init_maps(th, machine)) {
547 			rb_erase_cached(&th->rb_node, &threads->entries);
548 			RB_CLEAR_NODE(&th->rb_node);
549 			thread__put(th);
550 			return NULL;
551 		}
552 		/*
553 		 * It is now in the rbtree, get a ref
554 		 */
555 		thread__get(th);
556 		threads__set_last_match(threads, th);
557 		++threads->nr;
558 	}
559 
560 	return th;
561 }
562 
563 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
564 {
565 	return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
566 }
567 
568 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
569 				       pid_t tid)
570 {
571 	struct threads *threads = machine__threads(machine, tid);
572 	struct thread *th;
573 
574 	down_write(&threads->lock);
575 	th = __machine__findnew_thread(machine, pid, tid);
576 	up_write(&threads->lock);
577 	return th;
578 }
579 
580 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
581 				    pid_t tid)
582 {
583 	struct threads *threads = machine__threads(machine, tid);
584 	struct thread *th;
585 
586 	down_read(&threads->lock);
587 	th =  ____machine__findnew_thread(machine, threads, pid, tid, false);
588 	up_read(&threads->lock);
589 	return th;
590 }
591 
592 struct comm *machine__thread_exec_comm(struct machine *machine,
593 				       struct thread *thread)
594 {
595 	if (machine->comm_exec)
596 		return thread__exec_comm(thread);
597 	else
598 		return thread__comm(thread);
599 }
600 
601 int machine__process_comm_event(struct machine *machine, union perf_event *event,
602 				struct perf_sample *sample)
603 {
604 	struct thread *thread = machine__findnew_thread(machine,
605 							event->comm.pid,
606 							event->comm.tid);
607 	bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
608 	int err = 0;
609 
610 	if (exec)
611 		machine->comm_exec = true;
612 
613 	if (dump_trace)
614 		perf_event__fprintf_comm(event, stdout);
615 
616 	if (thread == NULL ||
617 	    __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
618 		dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
619 		err = -1;
620 	}
621 
622 	thread__put(thread);
623 
624 	return err;
625 }
626 
627 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
628 				      union perf_event *event,
629 				      struct perf_sample *sample __maybe_unused)
630 {
631 	struct thread *thread = machine__findnew_thread(machine,
632 							event->namespaces.pid,
633 							event->namespaces.tid);
634 	int err = 0;
635 
636 	WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
637 		  "\nWARNING: kernel seems to support more namespaces than perf"
638 		  " tool.\nTry updating the perf tool..\n\n");
639 
640 	WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
641 		  "\nWARNING: perf tool seems to support more namespaces than"
642 		  " the kernel.\nTry updating the kernel..\n\n");
643 
644 	if (dump_trace)
645 		perf_event__fprintf_namespaces(event, stdout);
646 
647 	if (thread == NULL ||
648 	    thread__set_namespaces(thread, sample->time, &event->namespaces)) {
649 		dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
650 		err = -1;
651 	}
652 
653 	thread__put(thread);
654 
655 	return err;
656 }
657 
658 int machine__process_cgroup_event(struct machine *machine,
659 				  union perf_event *event,
660 				  struct perf_sample *sample __maybe_unused)
661 {
662 	struct cgroup *cgrp;
663 
664 	if (dump_trace)
665 		perf_event__fprintf_cgroup(event, stdout);
666 
667 	cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
668 	if (cgrp == NULL)
669 		return -ENOMEM;
670 
671 	return 0;
672 }
673 
674 int machine__process_lost_event(struct machine *machine __maybe_unused,
675 				union perf_event *event, struct perf_sample *sample __maybe_unused)
676 {
677 	dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
678 		    event->lost.id, event->lost.lost);
679 	return 0;
680 }
681 
682 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
683 					union perf_event *event, struct perf_sample *sample)
684 {
685 	dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
686 		    sample->id, event->lost_samples.lost);
687 	return 0;
688 }
689 
690 static struct dso *machine__findnew_module_dso(struct machine *machine,
691 					       struct kmod_path *m,
692 					       const char *filename)
693 {
694 	struct dso *dso;
695 
696 	down_write(&machine->dsos.lock);
697 
698 	dso = __dsos__find(&machine->dsos, m->name, true);
699 	if (!dso) {
700 		dso = __dsos__addnew(&machine->dsos, m->name);
701 		if (dso == NULL)
702 			goto out_unlock;
703 
704 		dso__set_module_info(dso, m, machine);
705 		dso__set_long_name(dso, strdup(filename), true);
706 		dso->kernel = DSO_TYPE_KERNEL;
707 	}
708 
709 	dso__get(dso);
710 out_unlock:
711 	up_write(&machine->dsos.lock);
712 	return dso;
713 }
714 
715 int machine__process_aux_event(struct machine *machine __maybe_unused,
716 			       union perf_event *event)
717 {
718 	if (dump_trace)
719 		perf_event__fprintf_aux(event, stdout);
720 	return 0;
721 }
722 
723 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
724 					union perf_event *event)
725 {
726 	if (dump_trace)
727 		perf_event__fprintf_itrace_start(event, stdout);
728 	return 0;
729 }
730 
731 int machine__process_switch_event(struct machine *machine __maybe_unused,
732 				  union perf_event *event)
733 {
734 	if (dump_trace)
735 		perf_event__fprintf_switch(event, stdout);
736 	return 0;
737 }
738 
739 static int machine__process_ksymbol_register(struct machine *machine,
740 					     union perf_event *event,
741 					     struct perf_sample *sample __maybe_unused)
742 {
743 	struct symbol *sym;
744 	struct map *map = maps__find(&machine->kmaps, event->ksymbol.addr);
745 
746 	if (!map) {
747 		struct dso *dso = dso__new(event->ksymbol.name);
748 
749 		if (dso) {
750 			dso->kernel = DSO_TYPE_KERNEL;
751 			map = map__new2(0, dso);
752 		}
753 
754 		if (!dso || !map) {
755 			dso__put(dso);
756 			return -ENOMEM;
757 		}
758 
759 		map->start = event->ksymbol.addr;
760 		map->end = map->start + event->ksymbol.len;
761 		maps__insert(&machine->kmaps, map);
762 	}
763 
764 	sym = symbol__new(map->map_ip(map, map->start),
765 			  event->ksymbol.len,
766 			  0, 0, event->ksymbol.name);
767 	if (!sym)
768 		return -ENOMEM;
769 	dso__insert_symbol(map->dso, sym);
770 	return 0;
771 }
772 
773 static int machine__process_ksymbol_unregister(struct machine *machine,
774 					       union perf_event *event,
775 					       struct perf_sample *sample __maybe_unused)
776 {
777 	struct map *map;
778 
779 	map = maps__find(&machine->kmaps, event->ksymbol.addr);
780 	if (map)
781 		maps__remove(&machine->kmaps, map);
782 
783 	return 0;
784 }
785 
786 int machine__process_ksymbol(struct machine *machine __maybe_unused,
787 			     union perf_event *event,
788 			     struct perf_sample *sample)
789 {
790 	if (dump_trace)
791 		perf_event__fprintf_ksymbol(event, stdout);
792 
793 	if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
794 		return machine__process_ksymbol_unregister(machine, event,
795 							   sample);
796 	return machine__process_ksymbol_register(machine, event, sample);
797 }
798 
799 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
800 					      const char *filename)
801 {
802 	struct map *map = NULL;
803 	struct kmod_path m;
804 	struct dso *dso;
805 
806 	if (kmod_path__parse_name(&m, filename))
807 		return NULL;
808 
809 	dso = machine__findnew_module_dso(machine, &m, filename);
810 	if (dso == NULL)
811 		goto out;
812 
813 	map = map__new2(start, dso);
814 	if (map == NULL)
815 		goto out;
816 
817 	maps__insert(&machine->kmaps, map);
818 
819 	/* Put the map here because maps__insert alread got it */
820 	map__put(map);
821 out:
822 	/* put the dso here, corresponding to  machine__findnew_module_dso */
823 	dso__put(dso);
824 	zfree(&m.name);
825 	return map;
826 }
827 
828 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
829 {
830 	struct rb_node *nd;
831 	size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
832 
833 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
834 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
835 		ret += __dsos__fprintf(&pos->dsos.head, fp);
836 	}
837 
838 	return ret;
839 }
840 
841 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
842 				     bool (skip)(struct dso *dso, int parm), int parm)
843 {
844 	return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
845 }
846 
847 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
848 				     bool (skip)(struct dso *dso, int parm), int parm)
849 {
850 	struct rb_node *nd;
851 	size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
852 
853 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
854 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
855 		ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
856 	}
857 	return ret;
858 }
859 
860 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
861 {
862 	int i;
863 	size_t printed = 0;
864 	struct dso *kdso = machine__kernel_dso(machine);
865 
866 	if (kdso->has_build_id) {
867 		char filename[PATH_MAX];
868 		if (dso__build_id_filename(kdso, filename, sizeof(filename),
869 					   false))
870 			printed += fprintf(fp, "[0] %s\n", filename);
871 	}
872 
873 	for (i = 0; i < vmlinux_path__nr_entries; ++i)
874 		printed += fprintf(fp, "[%d] %s\n",
875 				   i + kdso->has_build_id, vmlinux_path[i]);
876 
877 	return printed;
878 }
879 
880 size_t machine__fprintf(struct machine *machine, FILE *fp)
881 {
882 	struct rb_node *nd;
883 	size_t ret;
884 	int i;
885 
886 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
887 		struct threads *threads = &machine->threads[i];
888 
889 		down_read(&threads->lock);
890 
891 		ret = fprintf(fp, "Threads: %u\n", threads->nr);
892 
893 		for (nd = rb_first_cached(&threads->entries); nd;
894 		     nd = rb_next(nd)) {
895 			struct thread *pos = rb_entry(nd, struct thread, rb_node);
896 
897 			ret += thread__fprintf(pos, fp);
898 		}
899 
900 		up_read(&threads->lock);
901 	}
902 	return ret;
903 }
904 
905 static struct dso *machine__get_kernel(struct machine *machine)
906 {
907 	const char *vmlinux_name = machine->mmap_name;
908 	struct dso *kernel;
909 
910 	if (machine__is_host(machine)) {
911 		if (symbol_conf.vmlinux_name)
912 			vmlinux_name = symbol_conf.vmlinux_name;
913 
914 		kernel = machine__findnew_kernel(machine, vmlinux_name,
915 						 "[kernel]", DSO_TYPE_KERNEL);
916 	} else {
917 		if (symbol_conf.default_guest_vmlinux_name)
918 			vmlinux_name = symbol_conf.default_guest_vmlinux_name;
919 
920 		kernel = machine__findnew_kernel(machine, vmlinux_name,
921 						 "[guest.kernel]",
922 						 DSO_TYPE_GUEST_KERNEL);
923 	}
924 
925 	if (kernel != NULL && (!kernel->has_build_id))
926 		dso__read_running_kernel_build_id(kernel, machine);
927 
928 	return kernel;
929 }
930 
931 struct process_args {
932 	u64 start;
933 };
934 
935 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
936 				    size_t bufsz)
937 {
938 	if (machine__is_default_guest(machine))
939 		scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
940 	else
941 		scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
942 }
943 
944 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
945 
946 /* Figure out the start address of kernel map from /proc/kallsyms.
947  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
948  * symbol_name if it's not that important.
949  */
950 static int machine__get_running_kernel_start(struct machine *machine,
951 					     const char **symbol_name,
952 					     u64 *start, u64 *end)
953 {
954 	char filename[PATH_MAX];
955 	int i, err = -1;
956 	const char *name;
957 	u64 addr = 0;
958 
959 	machine__get_kallsyms_filename(machine, filename, PATH_MAX);
960 
961 	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
962 		return 0;
963 
964 	for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
965 		err = kallsyms__get_function_start(filename, name, &addr);
966 		if (!err)
967 			break;
968 	}
969 
970 	if (err)
971 		return -1;
972 
973 	if (symbol_name)
974 		*symbol_name = name;
975 
976 	*start = addr;
977 
978 	err = kallsyms__get_function_start(filename, "_etext", &addr);
979 	if (!err)
980 		*end = addr;
981 
982 	return 0;
983 }
984 
985 int machine__create_extra_kernel_map(struct machine *machine,
986 				     struct dso *kernel,
987 				     struct extra_kernel_map *xm)
988 {
989 	struct kmap *kmap;
990 	struct map *map;
991 
992 	map = map__new2(xm->start, kernel);
993 	if (!map)
994 		return -1;
995 
996 	map->end   = xm->end;
997 	map->pgoff = xm->pgoff;
998 
999 	kmap = map__kmap(map);
1000 
1001 	strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1002 
1003 	maps__insert(&machine->kmaps, map);
1004 
1005 	pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1006 		  kmap->name, map->start, map->end);
1007 
1008 	map__put(map);
1009 
1010 	return 0;
1011 }
1012 
1013 static u64 find_entry_trampoline(struct dso *dso)
1014 {
1015 	/* Duplicates are removed so lookup all aliases */
1016 	const char *syms[] = {
1017 		"_entry_trampoline",
1018 		"__entry_trampoline_start",
1019 		"entry_SYSCALL_64_trampoline",
1020 	};
1021 	struct symbol *sym = dso__first_symbol(dso);
1022 	unsigned int i;
1023 
1024 	for (; sym; sym = dso__next_symbol(sym)) {
1025 		if (sym->binding != STB_GLOBAL)
1026 			continue;
1027 		for (i = 0; i < ARRAY_SIZE(syms); i++) {
1028 			if (!strcmp(sym->name, syms[i]))
1029 				return sym->start;
1030 		}
1031 	}
1032 
1033 	return 0;
1034 }
1035 
1036 /*
1037  * These values can be used for kernels that do not have symbols for the entry
1038  * trampolines in kallsyms.
1039  */
1040 #define X86_64_CPU_ENTRY_AREA_PER_CPU	0xfffffe0000000000ULL
1041 #define X86_64_CPU_ENTRY_AREA_SIZE	0x2c000
1042 #define X86_64_ENTRY_TRAMPOLINE		0x6000
1043 
1044 /* Map x86_64 PTI entry trampolines */
1045 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1046 					  struct dso *kernel)
1047 {
1048 	struct maps *kmaps = &machine->kmaps;
1049 	int nr_cpus_avail, cpu;
1050 	bool found = false;
1051 	struct map *map;
1052 	u64 pgoff;
1053 
1054 	/*
1055 	 * In the vmlinux case, pgoff is a virtual address which must now be
1056 	 * mapped to a vmlinux offset.
1057 	 */
1058 	maps__for_each_entry(kmaps, map) {
1059 		struct kmap *kmap = __map__kmap(map);
1060 		struct map *dest_map;
1061 
1062 		if (!kmap || !is_entry_trampoline(kmap->name))
1063 			continue;
1064 
1065 		dest_map = maps__find(kmaps, map->pgoff);
1066 		if (dest_map != map)
1067 			map->pgoff = dest_map->map_ip(dest_map, map->pgoff);
1068 		found = true;
1069 	}
1070 	if (found || machine->trampolines_mapped)
1071 		return 0;
1072 
1073 	pgoff = find_entry_trampoline(kernel);
1074 	if (!pgoff)
1075 		return 0;
1076 
1077 	nr_cpus_avail = machine__nr_cpus_avail(machine);
1078 
1079 	/* Add a 1 page map for each CPU's entry trampoline */
1080 	for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1081 		u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1082 			 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1083 			 X86_64_ENTRY_TRAMPOLINE;
1084 		struct extra_kernel_map xm = {
1085 			.start = va,
1086 			.end   = va + page_size,
1087 			.pgoff = pgoff,
1088 		};
1089 
1090 		strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1091 
1092 		if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1093 			return -1;
1094 	}
1095 
1096 	machine->trampolines_mapped = nr_cpus_avail;
1097 
1098 	return 0;
1099 }
1100 
1101 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1102 					     struct dso *kernel __maybe_unused)
1103 {
1104 	return 0;
1105 }
1106 
1107 static int
1108 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1109 {
1110 	/* In case of renewal the kernel map, destroy previous one */
1111 	machine__destroy_kernel_maps(machine);
1112 
1113 	machine->vmlinux_map = map__new2(0, kernel);
1114 	if (machine->vmlinux_map == NULL)
1115 		return -1;
1116 
1117 	machine->vmlinux_map->map_ip = machine->vmlinux_map->unmap_ip = identity__map_ip;
1118 	maps__insert(&machine->kmaps, machine->vmlinux_map);
1119 	return 0;
1120 }
1121 
1122 void machine__destroy_kernel_maps(struct machine *machine)
1123 {
1124 	struct kmap *kmap;
1125 	struct map *map = machine__kernel_map(machine);
1126 
1127 	if (map == NULL)
1128 		return;
1129 
1130 	kmap = map__kmap(map);
1131 	maps__remove(&machine->kmaps, map);
1132 	if (kmap && kmap->ref_reloc_sym) {
1133 		zfree((char **)&kmap->ref_reloc_sym->name);
1134 		zfree(&kmap->ref_reloc_sym);
1135 	}
1136 
1137 	map__zput(machine->vmlinux_map);
1138 }
1139 
1140 int machines__create_guest_kernel_maps(struct machines *machines)
1141 {
1142 	int ret = 0;
1143 	struct dirent **namelist = NULL;
1144 	int i, items = 0;
1145 	char path[PATH_MAX];
1146 	pid_t pid;
1147 	char *endp;
1148 
1149 	if (symbol_conf.default_guest_vmlinux_name ||
1150 	    symbol_conf.default_guest_modules ||
1151 	    symbol_conf.default_guest_kallsyms) {
1152 		machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1153 	}
1154 
1155 	if (symbol_conf.guestmount) {
1156 		items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1157 		if (items <= 0)
1158 			return -ENOENT;
1159 		for (i = 0; i < items; i++) {
1160 			if (!isdigit(namelist[i]->d_name[0])) {
1161 				/* Filter out . and .. */
1162 				continue;
1163 			}
1164 			pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1165 			if ((*endp != '\0') ||
1166 			    (endp == namelist[i]->d_name) ||
1167 			    (errno == ERANGE)) {
1168 				pr_debug("invalid directory (%s). Skipping.\n",
1169 					 namelist[i]->d_name);
1170 				continue;
1171 			}
1172 			sprintf(path, "%s/%s/proc/kallsyms",
1173 				symbol_conf.guestmount,
1174 				namelist[i]->d_name);
1175 			ret = access(path, R_OK);
1176 			if (ret) {
1177 				pr_debug("Can't access file %s\n", path);
1178 				goto failure;
1179 			}
1180 			machines__create_kernel_maps(machines, pid);
1181 		}
1182 failure:
1183 		free(namelist);
1184 	}
1185 
1186 	return ret;
1187 }
1188 
1189 void machines__destroy_kernel_maps(struct machines *machines)
1190 {
1191 	struct rb_node *next = rb_first_cached(&machines->guests);
1192 
1193 	machine__destroy_kernel_maps(&machines->host);
1194 
1195 	while (next) {
1196 		struct machine *pos = rb_entry(next, struct machine, rb_node);
1197 
1198 		next = rb_next(&pos->rb_node);
1199 		rb_erase_cached(&pos->rb_node, &machines->guests);
1200 		machine__delete(pos);
1201 	}
1202 }
1203 
1204 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1205 {
1206 	struct machine *machine = machines__findnew(machines, pid);
1207 
1208 	if (machine == NULL)
1209 		return -1;
1210 
1211 	return machine__create_kernel_maps(machine);
1212 }
1213 
1214 int machine__load_kallsyms(struct machine *machine, const char *filename)
1215 {
1216 	struct map *map = machine__kernel_map(machine);
1217 	int ret = __dso__load_kallsyms(map->dso, filename, map, true);
1218 
1219 	if (ret > 0) {
1220 		dso__set_loaded(map->dso);
1221 		/*
1222 		 * Since /proc/kallsyms will have multiple sessions for the
1223 		 * kernel, with modules between them, fixup the end of all
1224 		 * sections.
1225 		 */
1226 		maps__fixup_end(&machine->kmaps);
1227 	}
1228 
1229 	return ret;
1230 }
1231 
1232 int machine__load_vmlinux_path(struct machine *machine)
1233 {
1234 	struct map *map = machine__kernel_map(machine);
1235 	int ret = dso__load_vmlinux_path(map->dso, map);
1236 
1237 	if (ret > 0)
1238 		dso__set_loaded(map->dso);
1239 
1240 	return ret;
1241 }
1242 
1243 static char *get_kernel_version(const char *root_dir)
1244 {
1245 	char version[PATH_MAX];
1246 	FILE *file;
1247 	char *name, *tmp;
1248 	const char *prefix = "Linux version ";
1249 
1250 	sprintf(version, "%s/proc/version", root_dir);
1251 	file = fopen(version, "r");
1252 	if (!file)
1253 		return NULL;
1254 
1255 	tmp = fgets(version, sizeof(version), file);
1256 	fclose(file);
1257 	if (!tmp)
1258 		return NULL;
1259 
1260 	name = strstr(version, prefix);
1261 	if (!name)
1262 		return NULL;
1263 	name += strlen(prefix);
1264 	tmp = strchr(name, ' ');
1265 	if (tmp)
1266 		*tmp = '\0';
1267 
1268 	return strdup(name);
1269 }
1270 
1271 static bool is_kmod_dso(struct dso *dso)
1272 {
1273 	return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1274 	       dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1275 }
1276 
1277 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1278 {
1279 	char *long_name;
1280 	struct map *map = maps__find_by_name(maps, m->name);
1281 
1282 	if (map == NULL)
1283 		return 0;
1284 
1285 	long_name = strdup(path);
1286 	if (long_name == NULL)
1287 		return -ENOMEM;
1288 
1289 	dso__set_long_name(map->dso, long_name, true);
1290 	dso__kernel_module_get_build_id(map->dso, "");
1291 
1292 	/*
1293 	 * Full name could reveal us kmod compression, so
1294 	 * we need to update the symtab_type if needed.
1295 	 */
1296 	if (m->comp && is_kmod_dso(map->dso)) {
1297 		map->dso->symtab_type++;
1298 		map->dso->comp = m->comp;
1299 	}
1300 
1301 	return 0;
1302 }
1303 
1304 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1305 {
1306 	struct dirent *dent;
1307 	DIR *dir = opendir(dir_name);
1308 	int ret = 0;
1309 
1310 	if (!dir) {
1311 		pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1312 		return -1;
1313 	}
1314 
1315 	while ((dent = readdir(dir)) != NULL) {
1316 		char path[PATH_MAX];
1317 		struct stat st;
1318 
1319 		/*sshfs might return bad dent->d_type, so we have to stat*/
1320 		snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1321 		if (stat(path, &st))
1322 			continue;
1323 
1324 		if (S_ISDIR(st.st_mode)) {
1325 			if (!strcmp(dent->d_name, ".") ||
1326 			    !strcmp(dent->d_name, ".."))
1327 				continue;
1328 
1329 			/* Do not follow top-level source and build symlinks */
1330 			if (depth == 0) {
1331 				if (!strcmp(dent->d_name, "source") ||
1332 				    !strcmp(dent->d_name, "build"))
1333 					continue;
1334 			}
1335 
1336 			ret = maps__set_modules_path_dir(maps, path, depth + 1);
1337 			if (ret < 0)
1338 				goto out;
1339 		} else {
1340 			struct kmod_path m;
1341 
1342 			ret = kmod_path__parse_name(&m, dent->d_name);
1343 			if (ret)
1344 				goto out;
1345 
1346 			if (m.kmod)
1347 				ret = maps__set_module_path(maps, path, &m);
1348 
1349 			zfree(&m.name);
1350 
1351 			if (ret)
1352 				goto out;
1353 		}
1354 	}
1355 
1356 out:
1357 	closedir(dir);
1358 	return ret;
1359 }
1360 
1361 static int machine__set_modules_path(struct machine *machine)
1362 {
1363 	char *version;
1364 	char modules_path[PATH_MAX];
1365 
1366 	version = get_kernel_version(machine->root_dir);
1367 	if (!version)
1368 		return -1;
1369 
1370 	snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1371 		 machine->root_dir, version);
1372 	free(version);
1373 
1374 	return maps__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1375 }
1376 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1377 				u64 *size __maybe_unused,
1378 				const char *name __maybe_unused)
1379 {
1380 	return 0;
1381 }
1382 
1383 static int machine__create_module(void *arg, const char *name, u64 start,
1384 				  u64 size)
1385 {
1386 	struct machine *machine = arg;
1387 	struct map *map;
1388 
1389 	if (arch__fix_module_text_start(&start, &size, name) < 0)
1390 		return -1;
1391 
1392 	map = machine__addnew_module_map(machine, start, name);
1393 	if (map == NULL)
1394 		return -1;
1395 	map->end = start + size;
1396 
1397 	dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1398 
1399 	return 0;
1400 }
1401 
1402 static int machine__create_modules(struct machine *machine)
1403 {
1404 	const char *modules;
1405 	char path[PATH_MAX];
1406 
1407 	if (machine__is_default_guest(machine)) {
1408 		modules = symbol_conf.default_guest_modules;
1409 	} else {
1410 		snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1411 		modules = path;
1412 	}
1413 
1414 	if (symbol__restricted_filename(modules, "/proc/modules"))
1415 		return -1;
1416 
1417 	if (modules__parse(modules, machine, machine__create_module))
1418 		return -1;
1419 
1420 	if (!machine__set_modules_path(machine))
1421 		return 0;
1422 
1423 	pr_debug("Problems setting modules path maps, continuing anyway...\n");
1424 
1425 	return 0;
1426 }
1427 
1428 static void machine__set_kernel_mmap(struct machine *machine,
1429 				     u64 start, u64 end)
1430 {
1431 	machine->vmlinux_map->start = start;
1432 	machine->vmlinux_map->end   = end;
1433 	/*
1434 	 * Be a bit paranoid here, some perf.data file came with
1435 	 * a zero sized synthesized MMAP event for the kernel.
1436 	 */
1437 	if (start == 0 && end == 0)
1438 		machine->vmlinux_map->end = ~0ULL;
1439 }
1440 
1441 static void machine__update_kernel_mmap(struct machine *machine,
1442 				     u64 start, u64 end)
1443 {
1444 	struct map *map = machine__kernel_map(machine);
1445 
1446 	map__get(map);
1447 	maps__remove(&machine->kmaps, map);
1448 
1449 	machine__set_kernel_mmap(machine, start, end);
1450 
1451 	maps__insert(&machine->kmaps, map);
1452 	map__put(map);
1453 }
1454 
1455 int machine__create_kernel_maps(struct machine *machine)
1456 {
1457 	struct dso *kernel = machine__get_kernel(machine);
1458 	const char *name = NULL;
1459 	struct map *map;
1460 	u64 start = 0, end = ~0ULL;
1461 	int ret;
1462 
1463 	if (kernel == NULL)
1464 		return -1;
1465 
1466 	ret = __machine__create_kernel_maps(machine, kernel);
1467 	if (ret < 0)
1468 		goto out_put;
1469 
1470 	if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1471 		if (machine__is_host(machine))
1472 			pr_debug("Problems creating module maps, "
1473 				 "continuing anyway...\n");
1474 		else
1475 			pr_debug("Problems creating module maps for guest %d, "
1476 				 "continuing anyway...\n", machine->pid);
1477 	}
1478 
1479 	if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1480 		if (name &&
1481 		    map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1482 			machine__destroy_kernel_maps(machine);
1483 			ret = -1;
1484 			goto out_put;
1485 		}
1486 
1487 		/*
1488 		 * we have a real start address now, so re-order the kmaps
1489 		 * assume it's the last in the kmaps
1490 		 */
1491 		machine__update_kernel_mmap(machine, start, end);
1492 	}
1493 
1494 	if (machine__create_extra_kernel_maps(machine, kernel))
1495 		pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1496 
1497 	if (end == ~0ULL) {
1498 		/* update end address of the kernel map using adjacent module address */
1499 		map = map__next(machine__kernel_map(machine));
1500 		if (map)
1501 			machine__set_kernel_mmap(machine, start, map->start);
1502 	}
1503 
1504 out_put:
1505 	dso__put(kernel);
1506 	return ret;
1507 }
1508 
1509 static bool machine__uses_kcore(struct machine *machine)
1510 {
1511 	struct dso *dso;
1512 
1513 	list_for_each_entry(dso, &machine->dsos.head, node) {
1514 		if (dso__is_kcore(dso))
1515 			return true;
1516 	}
1517 
1518 	return false;
1519 }
1520 
1521 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1522 					     union perf_event *event)
1523 {
1524 	return machine__is(machine, "x86_64") &&
1525 	       is_entry_trampoline(event->mmap.filename);
1526 }
1527 
1528 static int machine__process_extra_kernel_map(struct machine *machine,
1529 					     union perf_event *event)
1530 {
1531 	struct dso *kernel = machine__kernel_dso(machine);
1532 	struct extra_kernel_map xm = {
1533 		.start = event->mmap.start,
1534 		.end   = event->mmap.start + event->mmap.len,
1535 		.pgoff = event->mmap.pgoff,
1536 	};
1537 
1538 	if (kernel == NULL)
1539 		return -1;
1540 
1541 	strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1542 
1543 	return machine__create_extra_kernel_map(machine, kernel, &xm);
1544 }
1545 
1546 static int machine__process_kernel_mmap_event(struct machine *machine,
1547 					      union perf_event *event)
1548 {
1549 	struct map *map;
1550 	enum dso_kernel_type kernel_type;
1551 	bool is_kernel_mmap;
1552 
1553 	/* If we have maps from kcore then we do not need or want any others */
1554 	if (machine__uses_kcore(machine))
1555 		return 0;
1556 
1557 	if (machine__is_host(machine))
1558 		kernel_type = DSO_TYPE_KERNEL;
1559 	else
1560 		kernel_type = DSO_TYPE_GUEST_KERNEL;
1561 
1562 	is_kernel_mmap = memcmp(event->mmap.filename,
1563 				machine->mmap_name,
1564 				strlen(machine->mmap_name) - 1) == 0;
1565 	if (event->mmap.filename[0] == '/' ||
1566 	    (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1567 		map = machine__addnew_module_map(machine, event->mmap.start,
1568 						 event->mmap.filename);
1569 		if (map == NULL)
1570 			goto out_problem;
1571 
1572 		map->end = map->start + event->mmap.len;
1573 	} else if (is_kernel_mmap) {
1574 		const char *symbol_name = (event->mmap.filename +
1575 				strlen(machine->mmap_name));
1576 		/*
1577 		 * Should be there already, from the build-id table in
1578 		 * the header.
1579 		 */
1580 		struct dso *kernel = NULL;
1581 		struct dso *dso;
1582 
1583 		down_read(&machine->dsos.lock);
1584 
1585 		list_for_each_entry(dso, &machine->dsos.head, node) {
1586 
1587 			/*
1588 			 * The cpumode passed to is_kernel_module is not the
1589 			 * cpumode of *this* event. If we insist on passing
1590 			 * correct cpumode to is_kernel_module, we should
1591 			 * record the cpumode when we adding this dso to the
1592 			 * linked list.
1593 			 *
1594 			 * However we don't really need passing correct
1595 			 * cpumode.  We know the correct cpumode must be kernel
1596 			 * mode (if not, we should not link it onto kernel_dsos
1597 			 * list).
1598 			 *
1599 			 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1600 			 * is_kernel_module() treats it as a kernel cpumode.
1601 			 */
1602 
1603 			if (!dso->kernel ||
1604 			    is_kernel_module(dso->long_name,
1605 					     PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1606 				continue;
1607 
1608 
1609 			kernel = dso;
1610 			break;
1611 		}
1612 
1613 		up_read(&machine->dsos.lock);
1614 
1615 		if (kernel == NULL)
1616 			kernel = machine__findnew_dso(machine, machine->mmap_name);
1617 		if (kernel == NULL)
1618 			goto out_problem;
1619 
1620 		kernel->kernel = kernel_type;
1621 		if (__machine__create_kernel_maps(machine, kernel) < 0) {
1622 			dso__put(kernel);
1623 			goto out_problem;
1624 		}
1625 
1626 		if (strstr(kernel->long_name, "vmlinux"))
1627 			dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1628 
1629 		machine__update_kernel_mmap(machine, event->mmap.start,
1630 					 event->mmap.start + event->mmap.len);
1631 
1632 		/*
1633 		 * Avoid using a zero address (kptr_restrict) for the ref reloc
1634 		 * symbol. Effectively having zero here means that at record
1635 		 * time /proc/sys/kernel/kptr_restrict was non zero.
1636 		 */
1637 		if (event->mmap.pgoff != 0) {
1638 			map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1639 							symbol_name,
1640 							event->mmap.pgoff);
1641 		}
1642 
1643 		if (machine__is_default_guest(machine)) {
1644 			/*
1645 			 * preload dso of guest kernel and modules
1646 			 */
1647 			dso__load(kernel, machine__kernel_map(machine));
1648 		}
1649 	} else if (perf_event__is_extra_kernel_mmap(machine, event)) {
1650 		return machine__process_extra_kernel_map(machine, event);
1651 	}
1652 	return 0;
1653 out_problem:
1654 	return -1;
1655 }
1656 
1657 int machine__process_mmap2_event(struct machine *machine,
1658 				 union perf_event *event,
1659 				 struct perf_sample *sample)
1660 {
1661 	struct thread *thread;
1662 	struct map *map;
1663 	struct dso_id dso_id = {
1664 		.maj = event->mmap2.maj,
1665 		.min = event->mmap2.min,
1666 		.ino = event->mmap2.ino,
1667 		.ino_generation = event->mmap2.ino_generation,
1668 	};
1669 	int ret = 0;
1670 
1671 	if (dump_trace)
1672 		perf_event__fprintf_mmap2(event, stdout);
1673 
1674 	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1675 	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1676 		ret = machine__process_kernel_mmap_event(machine, event);
1677 		if (ret < 0)
1678 			goto out_problem;
1679 		return 0;
1680 	}
1681 
1682 	thread = machine__findnew_thread(machine, event->mmap2.pid,
1683 					event->mmap2.tid);
1684 	if (thread == NULL)
1685 		goto out_problem;
1686 
1687 	map = map__new(machine, event->mmap2.start,
1688 			event->mmap2.len, event->mmap2.pgoff,
1689 			&dso_id, event->mmap2.prot,
1690 			event->mmap2.flags,
1691 			event->mmap2.filename, thread);
1692 
1693 	if (map == NULL)
1694 		goto out_problem_map;
1695 
1696 	ret = thread__insert_map(thread, map);
1697 	if (ret)
1698 		goto out_problem_insert;
1699 
1700 	thread__put(thread);
1701 	map__put(map);
1702 	return 0;
1703 
1704 out_problem_insert:
1705 	map__put(map);
1706 out_problem_map:
1707 	thread__put(thread);
1708 out_problem:
1709 	dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1710 	return 0;
1711 }
1712 
1713 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1714 				struct perf_sample *sample)
1715 {
1716 	struct thread *thread;
1717 	struct map *map;
1718 	u32 prot = 0;
1719 	int ret = 0;
1720 
1721 	if (dump_trace)
1722 		perf_event__fprintf_mmap(event, stdout);
1723 
1724 	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1725 	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1726 		ret = machine__process_kernel_mmap_event(machine, event);
1727 		if (ret < 0)
1728 			goto out_problem;
1729 		return 0;
1730 	}
1731 
1732 	thread = machine__findnew_thread(machine, event->mmap.pid,
1733 					 event->mmap.tid);
1734 	if (thread == NULL)
1735 		goto out_problem;
1736 
1737 	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1738 		prot = PROT_EXEC;
1739 
1740 	map = map__new(machine, event->mmap.start,
1741 			event->mmap.len, event->mmap.pgoff,
1742 			NULL, prot, 0, event->mmap.filename, thread);
1743 
1744 	if (map == NULL)
1745 		goto out_problem_map;
1746 
1747 	ret = thread__insert_map(thread, map);
1748 	if (ret)
1749 		goto out_problem_insert;
1750 
1751 	thread__put(thread);
1752 	map__put(map);
1753 	return 0;
1754 
1755 out_problem_insert:
1756 	map__put(map);
1757 out_problem_map:
1758 	thread__put(thread);
1759 out_problem:
1760 	dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1761 	return 0;
1762 }
1763 
1764 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1765 {
1766 	struct threads *threads = machine__threads(machine, th->tid);
1767 
1768 	if (threads->last_match == th)
1769 		threads__set_last_match(threads, NULL);
1770 
1771 	if (lock)
1772 		down_write(&threads->lock);
1773 
1774 	BUG_ON(refcount_read(&th->refcnt) == 0);
1775 
1776 	rb_erase_cached(&th->rb_node, &threads->entries);
1777 	RB_CLEAR_NODE(&th->rb_node);
1778 	--threads->nr;
1779 	/*
1780 	 * Move it first to the dead_threads list, then drop the reference,
1781 	 * if this is the last reference, then the thread__delete destructor
1782 	 * will be called and we will remove it from the dead_threads list.
1783 	 */
1784 	list_add_tail(&th->node, &threads->dead);
1785 
1786 	/*
1787 	 * We need to do the put here because if this is the last refcount,
1788 	 * then we will be touching the threads->dead head when removing the
1789 	 * thread.
1790 	 */
1791 	thread__put(th);
1792 
1793 	if (lock)
1794 		up_write(&threads->lock);
1795 }
1796 
1797 void machine__remove_thread(struct machine *machine, struct thread *th)
1798 {
1799 	return __machine__remove_thread(machine, th, true);
1800 }
1801 
1802 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1803 				struct perf_sample *sample)
1804 {
1805 	struct thread *thread = machine__find_thread(machine,
1806 						     event->fork.pid,
1807 						     event->fork.tid);
1808 	struct thread *parent = machine__findnew_thread(machine,
1809 							event->fork.ppid,
1810 							event->fork.ptid);
1811 	bool do_maps_clone = true;
1812 	int err = 0;
1813 
1814 	if (dump_trace)
1815 		perf_event__fprintf_task(event, stdout);
1816 
1817 	/*
1818 	 * There may be an existing thread that is not actually the parent,
1819 	 * either because we are processing events out of order, or because the
1820 	 * (fork) event that would have removed the thread was lost. Assume the
1821 	 * latter case and continue on as best we can.
1822 	 */
1823 	if (parent->pid_ != (pid_t)event->fork.ppid) {
1824 		dump_printf("removing erroneous parent thread %d/%d\n",
1825 			    parent->pid_, parent->tid);
1826 		machine__remove_thread(machine, parent);
1827 		thread__put(parent);
1828 		parent = machine__findnew_thread(machine, event->fork.ppid,
1829 						 event->fork.ptid);
1830 	}
1831 
1832 	/* if a thread currently exists for the thread id remove it */
1833 	if (thread != NULL) {
1834 		machine__remove_thread(machine, thread);
1835 		thread__put(thread);
1836 	}
1837 
1838 	thread = machine__findnew_thread(machine, event->fork.pid,
1839 					 event->fork.tid);
1840 	/*
1841 	 * When synthesizing FORK events, we are trying to create thread
1842 	 * objects for the already running tasks on the machine.
1843 	 *
1844 	 * Normally, for a kernel FORK event, we want to clone the parent's
1845 	 * maps because that is what the kernel just did.
1846 	 *
1847 	 * But when synthesizing, this should not be done.  If we do, we end up
1848 	 * with overlapping maps as we process the sythesized MMAP2 events that
1849 	 * get delivered shortly thereafter.
1850 	 *
1851 	 * Use the FORK event misc flags in an internal way to signal this
1852 	 * situation, so we can elide the map clone when appropriate.
1853 	 */
1854 	if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1855 		do_maps_clone = false;
1856 
1857 	if (thread == NULL || parent == NULL ||
1858 	    thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1859 		dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1860 		err = -1;
1861 	}
1862 	thread__put(thread);
1863 	thread__put(parent);
1864 
1865 	return err;
1866 }
1867 
1868 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1869 				struct perf_sample *sample __maybe_unused)
1870 {
1871 	struct thread *thread = machine__find_thread(machine,
1872 						     event->fork.pid,
1873 						     event->fork.tid);
1874 
1875 	if (dump_trace)
1876 		perf_event__fprintf_task(event, stdout);
1877 
1878 	if (thread != NULL) {
1879 		thread__exited(thread);
1880 		thread__put(thread);
1881 	}
1882 
1883 	return 0;
1884 }
1885 
1886 int machine__process_event(struct machine *machine, union perf_event *event,
1887 			   struct perf_sample *sample)
1888 {
1889 	int ret;
1890 
1891 	switch (event->header.type) {
1892 	case PERF_RECORD_COMM:
1893 		ret = machine__process_comm_event(machine, event, sample); break;
1894 	case PERF_RECORD_MMAP:
1895 		ret = machine__process_mmap_event(machine, event, sample); break;
1896 	case PERF_RECORD_NAMESPACES:
1897 		ret = machine__process_namespaces_event(machine, event, sample); break;
1898 	case PERF_RECORD_CGROUP:
1899 		ret = machine__process_cgroup_event(machine, event, sample); break;
1900 	case PERF_RECORD_MMAP2:
1901 		ret = machine__process_mmap2_event(machine, event, sample); break;
1902 	case PERF_RECORD_FORK:
1903 		ret = machine__process_fork_event(machine, event, sample); break;
1904 	case PERF_RECORD_EXIT:
1905 		ret = machine__process_exit_event(machine, event, sample); break;
1906 	case PERF_RECORD_LOST:
1907 		ret = machine__process_lost_event(machine, event, sample); break;
1908 	case PERF_RECORD_AUX:
1909 		ret = machine__process_aux_event(machine, event); break;
1910 	case PERF_RECORD_ITRACE_START:
1911 		ret = machine__process_itrace_start_event(machine, event); break;
1912 	case PERF_RECORD_LOST_SAMPLES:
1913 		ret = machine__process_lost_samples_event(machine, event, sample); break;
1914 	case PERF_RECORD_SWITCH:
1915 	case PERF_RECORD_SWITCH_CPU_WIDE:
1916 		ret = machine__process_switch_event(machine, event); break;
1917 	case PERF_RECORD_KSYMBOL:
1918 		ret = machine__process_ksymbol(machine, event, sample); break;
1919 	case PERF_RECORD_BPF_EVENT:
1920 		ret = machine__process_bpf(machine, event, sample); break;
1921 	default:
1922 		ret = -1;
1923 		break;
1924 	}
1925 
1926 	return ret;
1927 }
1928 
1929 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1930 {
1931 	if (!regexec(regex, sym->name, 0, NULL, 0))
1932 		return 1;
1933 	return 0;
1934 }
1935 
1936 static void ip__resolve_ams(struct thread *thread,
1937 			    struct addr_map_symbol *ams,
1938 			    u64 ip)
1939 {
1940 	struct addr_location al;
1941 
1942 	memset(&al, 0, sizeof(al));
1943 	/*
1944 	 * We cannot use the header.misc hint to determine whether a
1945 	 * branch stack address is user, kernel, guest, hypervisor.
1946 	 * Branches may straddle the kernel/user/hypervisor boundaries.
1947 	 * Thus, we have to try consecutively until we find a match
1948 	 * or else, the symbol is unknown
1949 	 */
1950 	thread__find_cpumode_addr_location(thread, ip, &al);
1951 
1952 	ams->addr = ip;
1953 	ams->al_addr = al.addr;
1954 	ams->ms.maps = al.maps;
1955 	ams->ms.sym = al.sym;
1956 	ams->ms.map = al.map;
1957 	ams->phys_addr = 0;
1958 }
1959 
1960 static void ip__resolve_data(struct thread *thread,
1961 			     u8 m, struct addr_map_symbol *ams,
1962 			     u64 addr, u64 phys_addr)
1963 {
1964 	struct addr_location al;
1965 
1966 	memset(&al, 0, sizeof(al));
1967 
1968 	thread__find_symbol(thread, m, addr, &al);
1969 
1970 	ams->addr = addr;
1971 	ams->al_addr = al.addr;
1972 	ams->ms.maps = al.maps;
1973 	ams->ms.sym = al.sym;
1974 	ams->ms.map = al.map;
1975 	ams->phys_addr = phys_addr;
1976 }
1977 
1978 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1979 				     struct addr_location *al)
1980 {
1981 	struct mem_info *mi = mem_info__new();
1982 
1983 	if (!mi)
1984 		return NULL;
1985 
1986 	ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1987 	ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
1988 			 sample->addr, sample->phys_addr);
1989 	mi->data_src.val = sample->data_src;
1990 
1991 	return mi;
1992 }
1993 
1994 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
1995 {
1996 	struct map *map = ms->map;
1997 	char *srcline = NULL;
1998 
1999 	if (!map || callchain_param.key == CCKEY_FUNCTION)
2000 		return srcline;
2001 
2002 	srcline = srcline__tree_find(&map->dso->srclines, ip);
2003 	if (!srcline) {
2004 		bool show_sym = false;
2005 		bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2006 
2007 		srcline = get_srcline(map->dso, map__rip_2objdump(map, ip),
2008 				      ms->sym, show_sym, show_addr, ip);
2009 		srcline__tree_insert(&map->dso->srclines, ip, srcline);
2010 	}
2011 
2012 	return srcline;
2013 }
2014 
2015 struct iterations {
2016 	int nr_loop_iter;
2017 	u64 cycles;
2018 };
2019 
2020 static int add_callchain_ip(struct thread *thread,
2021 			    struct callchain_cursor *cursor,
2022 			    struct symbol **parent,
2023 			    struct addr_location *root_al,
2024 			    u8 *cpumode,
2025 			    u64 ip,
2026 			    bool branch,
2027 			    struct branch_flags *flags,
2028 			    struct iterations *iter,
2029 			    u64 branch_from)
2030 {
2031 	struct map_symbol ms;
2032 	struct addr_location al;
2033 	int nr_loop_iter = 0;
2034 	u64 iter_cycles = 0;
2035 	const char *srcline = NULL;
2036 
2037 	al.filtered = 0;
2038 	al.sym = NULL;
2039 	if (!cpumode) {
2040 		thread__find_cpumode_addr_location(thread, ip, &al);
2041 	} else {
2042 		if (ip >= PERF_CONTEXT_MAX) {
2043 			switch (ip) {
2044 			case PERF_CONTEXT_HV:
2045 				*cpumode = PERF_RECORD_MISC_HYPERVISOR;
2046 				break;
2047 			case PERF_CONTEXT_KERNEL:
2048 				*cpumode = PERF_RECORD_MISC_KERNEL;
2049 				break;
2050 			case PERF_CONTEXT_USER:
2051 				*cpumode = PERF_RECORD_MISC_USER;
2052 				break;
2053 			default:
2054 				pr_debug("invalid callchain context: "
2055 					 "%"PRId64"\n", (s64) ip);
2056 				/*
2057 				 * It seems the callchain is corrupted.
2058 				 * Discard all.
2059 				 */
2060 				callchain_cursor_reset(cursor);
2061 				return 1;
2062 			}
2063 			return 0;
2064 		}
2065 		thread__find_symbol(thread, *cpumode, ip, &al);
2066 	}
2067 
2068 	if (al.sym != NULL) {
2069 		if (perf_hpp_list.parent && !*parent &&
2070 		    symbol__match_regex(al.sym, &parent_regex))
2071 			*parent = al.sym;
2072 		else if (have_ignore_callees && root_al &&
2073 		  symbol__match_regex(al.sym, &ignore_callees_regex)) {
2074 			/* Treat this symbol as the root,
2075 			   forgetting its callees. */
2076 			*root_al = al;
2077 			callchain_cursor_reset(cursor);
2078 		}
2079 	}
2080 
2081 	if (symbol_conf.hide_unresolved && al.sym == NULL)
2082 		return 0;
2083 
2084 	if (iter) {
2085 		nr_loop_iter = iter->nr_loop_iter;
2086 		iter_cycles = iter->cycles;
2087 	}
2088 
2089 	ms.maps = al.maps;
2090 	ms.map = al.map;
2091 	ms.sym = al.sym;
2092 	srcline = callchain_srcline(&ms, al.addr);
2093 	return callchain_cursor_append(cursor, ip, &ms,
2094 				       branch, flags, nr_loop_iter,
2095 				       iter_cycles, branch_from, srcline);
2096 }
2097 
2098 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2099 					   struct addr_location *al)
2100 {
2101 	unsigned int i;
2102 	const struct branch_stack *bs = sample->branch_stack;
2103 	struct branch_entry *entries = perf_sample__branch_entries(sample);
2104 	struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2105 
2106 	if (!bi)
2107 		return NULL;
2108 
2109 	for (i = 0; i < bs->nr; i++) {
2110 		ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2111 		ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2112 		bi[i].flags = entries[i].flags;
2113 	}
2114 	return bi;
2115 }
2116 
2117 static void save_iterations(struct iterations *iter,
2118 			    struct branch_entry *be, int nr)
2119 {
2120 	int i;
2121 
2122 	iter->nr_loop_iter++;
2123 	iter->cycles = 0;
2124 
2125 	for (i = 0; i < nr; i++)
2126 		iter->cycles += be[i].flags.cycles;
2127 }
2128 
2129 #define CHASHSZ 127
2130 #define CHASHBITS 7
2131 #define NO_ENTRY 0xff
2132 
2133 #define PERF_MAX_BRANCH_DEPTH 127
2134 
2135 /* Remove loops. */
2136 static int remove_loops(struct branch_entry *l, int nr,
2137 			struct iterations *iter)
2138 {
2139 	int i, j, off;
2140 	unsigned char chash[CHASHSZ];
2141 
2142 	memset(chash, NO_ENTRY, sizeof(chash));
2143 
2144 	BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2145 
2146 	for (i = 0; i < nr; i++) {
2147 		int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2148 
2149 		/* no collision handling for now */
2150 		if (chash[h] == NO_ENTRY) {
2151 			chash[h] = i;
2152 		} else if (l[chash[h]].from == l[i].from) {
2153 			bool is_loop = true;
2154 			/* check if it is a real loop */
2155 			off = 0;
2156 			for (j = chash[h]; j < i && i + off < nr; j++, off++)
2157 				if (l[j].from != l[i + off].from) {
2158 					is_loop = false;
2159 					break;
2160 				}
2161 			if (is_loop) {
2162 				j = nr - (i + off);
2163 				if (j > 0) {
2164 					save_iterations(iter + i + off,
2165 						l + i, off);
2166 
2167 					memmove(iter + i, iter + i + off,
2168 						j * sizeof(*iter));
2169 
2170 					memmove(l + i, l + i + off,
2171 						j * sizeof(*l));
2172 				}
2173 
2174 				nr -= off;
2175 			}
2176 		}
2177 	}
2178 	return nr;
2179 }
2180 
2181 /*
2182  * Recolve LBR callstack chain sample
2183  * Return:
2184  * 1 on success get LBR callchain information
2185  * 0 no available LBR callchain information, should try fp
2186  * negative error code on other errors.
2187  */
2188 static int resolve_lbr_callchain_sample(struct thread *thread,
2189 					struct callchain_cursor *cursor,
2190 					struct perf_sample *sample,
2191 					struct symbol **parent,
2192 					struct addr_location *root_al,
2193 					int max_stack)
2194 {
2195 	struct ip_callchain *chain = sample->callchain;
2196 	int chain_nr = min(max_stack, (int)chain->nr), i;
2197 	u8 cpumode = PERF_RECORD_MISC_USER;
2198 	u64 ip, branch_from = 0;
2199 
2200 	for (i = 0; i < chain_nr; i++) {
2201 		if (chain->ips[i] == PERF_CONTEXT_USER)
2202 			break;
2203 	}
2204 
2205 	/* LBR only affects the user callchain */
2206 	if (i != chain_nr) {
2207 		struct branch_stack *lbr_stack = sample->branch_stack;
2208 		struct branch_entry *entries = perf_sample__branch_entries(sample);
2209 		int lbr_nr = lbr_stack->nr, j, k;
2210 		bool branch;
2211 		struct branch_flags *flags;
2212 		/*
2213 		 * LBR callstack can only get user call chain.
2214 		 * The mix_chain_nr is kernel call chain
2215 		 * number plus LBR user call chain number.
2216 		 * i is kernel call chain number,
2217 		 * 1 is PERF_CONTEXT_USER,
2218 		 * lbr_nr + 1 is the user call chain number.
2219 		 * For details, please refer to the comments
2220 		 * in callchain__printf
2221 		 */
2222 		int mix_chain_nr = i + 1 + lbr_nr + 1;
2223 
2224 		for (j = 0; j < mix_chain_nr; j++) {
2225 			int err;
2226 			branch = false;
2227 			flags = NULL;
2228 
2229 			if (callchain_param.order == ORDER_CALLEE) {
2230 				if (j < i + 1)
2231 					ip = chain->ips[j];
2232 				else if (j > i + 1) {
2233 					k = j - i - 2;
2234 					ip = entries[k].from;
2235 					branch = true;
2236 					flags = &entries[k].flags;
2237 				} else {
2238 					ip = entries[0].to;
2239 					branch = true;
2240 					flags = &entries[0].flags;
2241 					branch_from = entries[0].from;
2242 				}
2243 			} else {
2244 				if (j < lbr_nr) {
2245 					k = lbr_nr - j - 1;
2246 					ip = entries[k].from;
2247 					branch = true;
2248 					flags = &entries[k].flags;
2249 				}
2250 				else if (j > lbr_nr)
2251 					ip = chain->ips[i + 1 - (j - lbr_nr)];
2252 				else {
2253 					ip = entries[0].to;
2254 					branch = true;
2255 					flags = &entries[0].flags;
2256 					branch_from = entries[0].from;
2257 				}
2258 			}
2259 
2260 			err = add_callchain_ip(thread, cursor, parent,
2261 					       root_al, &cpumode, ip,
2262 					       branch, flags, NULL,
2263 					       branch_from);
2264 			if (err)
2265 				return (err < 0) ? err : 0;
2266 		}
2267 		return 1;
2268 	}
2269 
2270 	return 0;
2271 }
2272 
2273 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2274 			     struct callchain_cursor *cursor,
2275 			     struct symbol **parent,
2276 			     struct addr_location *root_al,
2277 			     u8 *cpumode, int ent)
2278 {
2279 	int err = 0;
2280 
2281 	while (--ent >= 0) {
2282 		u64 ip = chain->ips[ent];
2283 
2284 		if (ip >= PERF_CONTEXT_MAX) {
2285 			err = add_callchain_ip(thread, cursor, parent,
2286 					       root_al, cpumode, ip,
2287 					       false, NULL, NULL, 0);
2288 			break;
2289 		}
2290 	}
2291 	return err;
2292 }
2293 
2294 static int thread__resolve_callchain_sample(struct thread *thread,
2295 					    struct callchain_cursor *cursor,
2296 					    struct evsel *evsel,
2297 					    struct perf_sample *sample,
2298 					    struct symbol **parent,
2299 					    struct addr_location *root_al,
2300 					    int max_stack)
2301 {
2302 	struct branch_stack *branch = sample->branch_stack;
2303 	struct branch_entry *entries = perf_sample__branch_entries(sample);
2304 	struct ip_callchain *chain = sample->callchain;
2305 	int chain_nr = 0;
2306 	u8 cpumode = PERF_RECORD_MISC_USER;
2307 	int i, j, err, nr_entries;
2308 	int skip_idx = -1;
2309 	int first_call = 0;
2310 
2311 	if (chain)
2312 		chain_nr = chain->nr;
2313 
2314 	if (perf_evsel__has_branch_callstack(evsel)) {
2315 		err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2316 						   root_al, max_stack);
2317 		if (err)
2318 			return (err < 0) ? err : 0;
2319 	}
2320 
2321 	/*
2322 	 * Based on DWARF debug information, some architectures skip
2323 	 * a callchain entry saved by the kernel.
2324 	 */
2325 	skip_idx = arch_skip_callchain_idx(thread, chain);
2326 
2327 	/*
2328 	 * Add branches to call stack for easier browsing. This gives
2329 	 * more context for a sample than just the callers.
2330 	 *
2331 	 * This uses individual histograms of paths compared to the
2332 	 * aggregated histograms the normal LBR mode uses.
2333 	 *
2334 	 * Limitations for now:
2335 	 * - No extra filters
2336 	 * - No annotations (should annotate somehow)
2337 	 */
2338 
2339 	if (branch && callchain_param.branch_callstack) {
2340 		int nr = min(max_stack, (int)branch->nr);
2341 		struct branch_entry be[nr];
2342 		struct iterations iter[nr];
2343 
2344 		if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2345 			pr_warning("corrupted branch chain. skipping...\n");
2346 			goto check_calls;
2347 		}
2348 
2349 		for (i = 0; i < nr; i++) {
2350 			if (callchain_param.order == ORDER_CALLEE) {
2351 				be[i] = entries[i];
2352 
2353 				if (chain == NULL)
2354 					continue;
2355 
2356 				/*
2357 				 * Check for overlap into the callchain.
2358 				 * The return address is one off compared to
2359 				 * the branch entry. To adjust for this
2360 				 * assume the calling instruction is not longer
2361 				 * than 8 bytes.
2362 				 */
2363 				if (i == skip_idx ||
2364 				    chain->ips[first_call] >= PERF_CONTEXT_MAX)
2365 					first_call++;
2366 				else if (be[i].from < chain->ips[first_call] &&
2367 				    be[i].from >= chain->ips[first_call] - 8)
2368 					first_call++;
2369 			} else
2370 				be[i] = entries[branch->nr - i - 1];
2371 		}
2372 
2373 		memset(iter, 0, sizeof(struct iterations) * nr);
2374 		nr = remove_loops(be, nr, iter);
2375 
2376 		for (i = 0; i < nr; i++) {
2377 			err = add_callchain_ip(thread, cursor, parent,
2378 					       root_al,
2379 					       NULL, be[i].to,
2380 					       true, &be[i].flags,
2381 					       NULL, be[i].from);
2382 
2383 			if (!err)
2384 				err = add_callchain_ip(thread, cursor, parent, root_al,
2385 						       NULL, be[i].from,
2386 						       true, &be[i].flags,
2387 						       &iter[i], 0);
2388 			if (err == -EINVAL)
2389 				break;
2390 			if (err)
2391 				return err;
2392 		}
2393 
2394 		if (chain_nr == 0)
2395 			return 0;
2396 
2397 		chain_nr -= nr;
2398 	}
2399 
2400 check_calls:
2401 	if (chain && callchain_param.order != ORDER_CALLEE) {
2402 		err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2403 					&cpumode, chain->nr - first_call);
2404 		if (err)
2405 			return (err < 0) ? err : 0;
2406 	}
2407 	for (i = first_call, nr_entries = 0;
2408 	     i < chain_nr && nr_entries < max_stack; i++) {
2409 		u64 ip;
2410 
2411 		if (callchain_param.order == ORDER_CALLEE)
2412 			j = i;
2413 		else
2414 			j = chain->nr - i - 1;
2415 
2416 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2417 		if (j == skip_idx)
2418 			continue;
2419 #endif
2420 		ip = chain->ips[j];
2421 		if (ip < PERF_CONTEXT_MAX)
2422                        ++nr_entries;
2423 		else if (callchain_param.order != ORDER_CALLEE) {
2424 			err = find_prev_cpumode(chain, thread, cursor, parent,
2425 						root_al, &cpumode, j);
2426 			if (err)
2427 				return (err < 0) ? err : 0;
2428 			continue;
2429 		}
2430 
2431 		err = add_callchain_ip(thread, cursor, parent,
2432 				       root_al, &cpumode, ip,
2433 				       false, NULL, NULL, 0);
2434 
2435 		if (err)
2436 			return (err < 0) ? err : 0;
2437 	}
2438 
2439 	return 0;
2440 }
2441 
2442 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2443 {
2444 	struct symbol *sym = ms->sym;
2445 	struct map *map = ms->map;
2446 	struct inline_node *inline_node;
2447 	struct inline_list *ilist;
2448 	u64 addr;
2449 	int ret = 1;
2450 
2451 	if (!symbol_conf.inline_name || !map || !sym)
2452 		return ret;
2453 
2454 	addr = map__map_ip(map, ip);
2455 	addr = map__rip_2objdump(map, addr);
2456 
2457 	inline_node = inlines__tree_find(&map->dso->inlined_nodes, addr);
2458 	if (!inline_node) {
2459 		inline_node = dso__parse_addr_inlines(map->dso, addr, sym);
2460 		if (!inline_node)
2461 			return ret;
2462 		inlines__tree_insert(&map->dso->inlined_nodes, inline_node);
2463 	}
2464 
2465 	list_for_each_entry(ilist, &inline_node->val, list) {
2466 		struct map_symbol ilist_ms = {
2467 			.maps = ms->maps,
2468 			.map = map,
2469 			.sym = ilist->symbol,
2470 		};
2471 		ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2472 					      NULL, 0, 0, 0, ilist->srcline);
2473 
2474 		if (ret != 0)
2475 			return ret;
2476 	}
2477 
2478 	return ret;
2479 }
2480 
2481 static int unwind_entry(struct unwind_entry *entry, void *arg)
2482 {
2483 	struct callchain_cursor *cursor = arg;
2484 	const char *srcline = NULL;
2485 	u64 addr = entry->ip;
2486 
2487 	if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2488 		return 0;
2489 
2490 	if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2491 		return 0;
2492 
2493 	/*
2494 	 * Convert entry->ip from a virtual address to an offset in
2495 	 * its corresponding binary.
2496 	 */
2497 	if (entry->ms.map)
2498 		addr = map__map_ip(entry->ms.map, entry->ip);
2499 
2500 	srcline = callchain_srcline(&entry->ms, addr);
2501 	return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2502 				       false, NULL, 0, 0, 0, srcline);
2503 }
2504 
2505 static int thread__resolve_callchain_unwind(struct thread *thread,
2506 					    struct callchain_cursor *cursor,
2507 					    struct evsel *evsel,
2508 					    struct perf_sample *sample,
2509 					    int max_stack)
2510 {
2511 	/* Can we do dwarf post unwind? */
2512 	if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2513 	      (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2514 		return 0;
2515 
2516 	/* Bail out if nothing was captured. */
2517 	if ((!sample->user_regs.regs) ||
2518 	    (!sample->user_stack.size))
2519 		return 0;
2520 
2521 	return unwind__get_entries(unwind_entry, cursor,
2522 				   thread, sample, max_stack);
2523 }
2524 
2525 int thread__resolve_callchain(struct thread *thread,
2526 			      struct callchain_cursor *cursor,
2527 			      struct evsel *evsel,
2528 			      struct perf_sample *sample,
2529 			      struct symbol **parent,
2530 			      struct addr_location *root_al,
2531 			      int max_stack)
2532 {
2533 	int ret = 0;
2534 
2535 	callchain_cursor_reset(cursor);
2536 
2537 	if (callchain_param.order == ORDER_CALLEE) {
2538 		ret = thread__resolve_callchain_sample(thread, cursor,
2539 						       evsel, sample,
2540 						       parent, root_al,
2541 						       max_stack);
2542 		if (ret)
2543 			return ret;
2544 		ret = thread__resolve_callchain_unwind(thread, cursor,
2545 						       evsel, sample,
2546 						       max_stack);
2547 	} else {
2548 		ret = thread__resolve_callchain_unwind(thread, cursor,
2549 						       evsel, sample,
2550 						       max_stack);
2551 		if (ret)
2552 			return ret;
2553 		ret = thread__resolve_callchain_sample(thread, cursor,
2554 						       evsel, sample,
2555 						       parent, root_al,
2556 						       max_stack);
2557 	}
2558 
2559 	return ret;
2560 }
2561 
2562 int machine__for_each_thread(struct machine *machine,
2563 			     int (*fn)(struct thread *thread, void *p),
2564 			     void *priv)
2565 {
2566 	struct threads *threads;
2567 	struct rb_node *nd;
2568 	struct thread *thread;
2569 	int rc = 0;
2570 	int i;
2571 
2572 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
2573 		threads = &machine->threads[i];
2574 		for (nd = rb_first_cached(&threads->entries); nd;
2575 		     nd = rb_next(nd)) {
2576 			thread = rb_entry(nd, struct thread, rb_node);
2577 			rc = fn(thread, priv);
2578 			if (rc != 0)
2579 				return rc;
2580 		}
2581 
2582 		list_for_each_entry(thread, &threads->dead, node) {
2583 			rc = fn(thread, priv);
2584 			if (rc != 0)
2585 				return rc;
2586 		}
2587 	}
2588 	return rc;
2589 }
2590 
2591 int machines__for_each_thread(struct machines *machines,
2592 			      int (*fn)(struct thread *thread, void *p),
2593 			      void *priv)
2594 {
2595 	struct rb_node *nd;
2596 	int rc = 0;
2597 
2598 	rc = machine__for_each_thread(&machines->host, fn, priv);
2599 	if (rc != 0)
2600 		return rc;
2601 
2602 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
2603 		struct machine *machine = rb_entry(nd, struct machine, rb_node);
2604 
2605 		rc = machine__for_each_thread(machine, fn, priv);
2606 		if (rc != 0)
2607 			return rc;
2608 	}
2609 	return rc;
2610 }
2611 
2612 pid_t machine__get_current_tid(struct machine *machine, int cpu)
2613 {
2614 	int nr_cpus = min(machine->env->nr_cpus_online, MAX_NR_CPUS);
2615 
2616 	if (cpu < 0 || cpu >= nr_cpus || !machine->current_tid)
2617 		return -1;
2618 
2619 	return machine->current_tid[cpu];
2620 }
2621 
2622 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
2623 			     pid_t tid)
2624 {
2625 	struct thread *thread;
2626 	int nr_cpus = min(machine->env->nr_cpus_online, MAX_NR_CPUS);
2627 
2628 	if (cpu < 0)
2629 		return -EINVAL;
2630 
2631 	if (!machine->current_tid) {
2632 		int i;
2633 
2634 		machine->current_tid = calloc(nr_cpus, sizeof(pid_t));
2635 		if (!machine->current_tid)
2636 			return -ENOMEM;
2637 		for (i = 0; i < nr_cpus; i++)
2638 			machine->current_tid[i] = -1;
2639 	}
2640 
2641 	if (cpu >= nr_cpus) {
2642 		pr_err("Requested CPU %d too large. ", cpu);
2643 		pr_err("Consider raising MAX_NR_CPUS\n");
2644 		return -EINVAL;
2645 	}
2646 
2647 	machine->current_tid[cpu] = tid;
2648 
2649 	thread = machine__findnew_thread(machine, pid, tid);
2650 	if (!thread)
2651 		return -ENOMEM;
2652 
2653 	thread->cpu = cpu;
2654 	thread__put(thread);
2655 
2656 	return 0;
2657 }
2658 
2659 /*
2660  * Compares the raw arch string. N.B. see instead perf_env__arch() if a
2661  * normalized arch is needed.
2662  */
2663 bool machine__is(struct machine *machine, const char *arch)
2664 {
2665 	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
2666 }
2667 
2668 int machine__nr_cpus_avail(struct machine *machine)
2669 {
2670 	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
2671 }
2672 
2673 int machine__get_kernel_start(struct machine *machine)
2674 {
2675 	struct map *map = machine__kernel_map(machine);
2676 	int err = 0;
2677 
2678 	/*
2679 	 * The only addresses above 2^63 are kernel addresses of a 64-bit
2680 	 * kernel.  Note that addresses are unsigned so that on a 32-bit system
2681 	 * all addresses including kernel addresses are less than 2^32.  In
2682 	 * that case (32-bit system), if the kernel mapping is unknown, all
2683 	 * addresses will be assumed to be in user space - see
2684 	 * machine__kernel_ip().
2685 	 */
2686 	machine->kernel_start = 1ULL << 63;
2687 	if (map) {
2688 		err = map__load(map);
2689 		/*
2690 		 * On x86_64, PTI entry trampolines are less than the
2691 		 * start of kernel text, but still above 2^63. So leave
2692 		 * kernel_start = 1ULL << 63 for x86_64.
2693 		 */
2694 		if (!err && !machine__is(machine, "x86_64"))
2695 			machine->kernel_start = map->start;
2696 	}
2697 	return err;
2698 }
2699 
2700 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
2701 {
2702 	u8 addr_cpumode = cpumode;
2703 	bool kernel_ip;
2704 
2705 	if (!machine->single_address_space)
2706 		goto out;
2707 
2708 	kernel_ip = machine__kernel_ip(machine, addr);
2709 	switch (cpumode) {
2710 	case PERF_RECORD_MISC_KERNEL:
2711 	case PERF_RECORD_MISC_USER:
2712 		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
2713 					   PERF_RECORD_MISC_USER;
2714 		break;
2715 	case PERF_RECORD_MISC_GUEST_KERNEL:
2716 	case PERF_RECORD_MISC_GUEST_USER:
2717 		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
2718 					   PERF_RECORD_MISC_GUEST_USER;
2719 		break;
2720 	default:
2721 		break;
2722 	}
2723 out:
2724 	return addr_cpumode;
2725 }
2726 
2727 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
2728 {
2729 	return dsos__findnew_id(&machine->dsos, filename, id);
2730 }
2731 
2732 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2733 {
2734 	return machine__findnew_dso_id(machine, filename, NULL);
2735 }
2736 
2737 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2738 {
2739 	struct machine *machine = vmachine;
2740 	struct map *map;
2741 	struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
2742 
2743 	if (sym == NULL)
2744 		return NULL;
2745 
2746 	*modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2747 	*addrp = map->unmap_ip(map, sym->start);
2748 	return sym->name;
2749 }
2750