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 "path.h"
20 #include "srcline.h"
21 #include "symbol.h"
22 #include "sort.h"
23 #include "strlist.h"
24 #include "target.h"
25 #include "thread.h"
26 #include "util.h"
27 #include "vdso.h"
28 #include <stdbool.h>
29 #include <sys/types.h>
30 #include <sys/stat.h>
31 #include <unistd.h>
32 #include "unwind.h"
33 #include "linux/hash.h"
34 #include "asm/bug.h"
35 #include "bpf-event.h"
36 #include <internal/lib.h> // page_size
37 #include "cgroup.h"
38 #include "arm64-frame-pointer-unwind-support.h"
39 
40 #include <linux/ctype.h>
41 #include <symbol/kallsyms.h>
42 #include <linux/mman.h>
43 #include <linux/string.h>
44 #include <linux/zalloc.h>
45 
46 static void __machine__remove_thread(struct machine *machine, struct thread_rb_node *nd,
47 				     struct thread *th, bool lock);
48 
machine__kernel_dso(struct machine * machine)49 static struct dso *machine__kernel_dso(struct machine *machine)
50 {
51 	return map__dso(machine->vmlinux_map);
52 }
53 
dsos__init(struct dsos * dsos)54 static void dsos__init(struct dsos *dsos)
55 {
56 	INIT_LIST_HEAD(&dsos->head);
57 	dsos->root = RB_ROOT;
58 	init_rwsem(&dsos->lock);
59 }
60 
machine__threads_init(struct machine * machine)61 static void machine__threads_init(struct machine *machine)
62 {
63 	int i;
64 
65 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
66 		struct threads *threads = &machine->threads[i];
67 		threads->entries = RB_ROOT_CACHED;
68 		init_rwsem(&threads->lock);
69 		threads->nr = 0;
70 		INIT_LIST_HEAD(&threads->dead);
71 		threads->last_match = NULL;
72 	}
73 }
74 
thread_rb_node__cmp_tid(const void * key,const struct rb_node * nd)75 static int thread_rb_node__cmp_tid(const void *key, const struct rb_node *nd)
76 {
77 	int to_find = (int) *((pid_t *)key);
78 
79 	return to_find - (int)thread__tid(rb_entry(nd, struct thread_rb_node, rb_node)->thread);
80 }
81 
thread_rb_node__find(const struct thread * th,struct rb_root * tree)82 static struct thread_rb_node *thread_rb_node__find(const struct thread *th,
83 						   struct rb_root *tree)
84 {
85 	pid_t to_find = thread__tid(th);
86 	struct rb_node *nd = rb_find(&to_find, tree, thread_rb_node__cmp_tid);
87 
88 	return rb_entry(nd, struct thread_rb_node, rb_node);
89 }
90 
machine__set_mmap_name(struct machine * machine)91 static int machine__set_mmap_name(struct machine *machine)
92 {
93 	if (machine__is_host(machine))
94 		machine->mmap_name = strdup("[kernel.kallsyms]");
95 	else if (machine__is_default_guest(machine))
96 		machine->mmap_name = strdup("[guest.kernel.kallsyms]");
97 	else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
98 			  machine->pid) < 0)
99 		machine->mmap_name = NULL;
100 
101 	return machine->mmap_name ? 0 : -ENOMEM;
102 }
103 
thread__set_guest_comm(struct thread * thread,pid_t pid)104 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
105 {
106 	char comm[64];
107 
108 	snprintf(comm, sizeof(comm), "[guest/%d]", pid);
109 	thread__set_comm(thread, comm, 0);
110 }
111 
machine__init(struct machine * machine,const char * root_dir,pid_t pid)112 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
113 {
114 	int err = -ENOMEM;
115 
116 	memset(machine, 0, sizeof(*machine));
117 	machine->kmaps = maps__new(machine);
118 	if (machine->kmaps == NULL)
119 		return -ENOMEM;
120 
121 	RB_CLEAR_NODE(&machine->rb_node);
122 	dsos__init(&machine->dsos);
123 
124 	machine__threads_init(machine);
125 
126 	machine->vdso_info = NULL;
127 	machine->env = NULL;
128 
129 	machine->pid = pid;
130 
131 	machine->id_hdr_size = 0;
132 	machine->kptr_restrict_warned = false;
133 	machine->comm_exec = false;
134 	machine->kernel_start = 0;
135 	machine->vmlinux_map = NULL;
136 
137 	machine->root_dir = strdup(root_dir);
138 	if (machine->root_dir == NULL)
139 		goto out;
140 
141 	if (machine__set_mmap_name(machine))
142 		goto out;
143 
144 	if (pid != HOST_KERNEL_ID) {
145 		struct thread *thread = machine__findnew_thread(machine, -1,
146 								pid);
147 
148 		if (thread == NULL)
149 			goto out;
150 
151 		thread__set_guest_comm(thread, pid);
152 		thread__put(thread);
153 	}
154 
155 	machine->current_tid = NULL;
156 	err = 0;
157 
158 out:
159 	if (err) {
160 		zfree(&machine->kmaps);
161 		zfree(&machine->root_dir);
162 		zfree(&machine->mmap_name);
163 	}
164 	return 0;
165 }
166 
machine__new_host(void)167 struct machine *machine__new_host(void)
168 {
169 	struct machine *machine = malloc(sizeof(*machine));
170 
171 	if (machine != NULL) {
172 		machine__init(machine, "", HOST_KERNEL_ID);
173 
174 		if (machine__create_kernel_maps(machine) < 0)
175 			goto out_delete;
176 	}
177 
178 	return machine;
179 out_delete:
180 	free(machine);
181 	return NULL;
182 }
183 
machine__new_kallsyms(void)184 struct machine *machine__new_kallsyms(void)
185 {
186 	struct machine *machine = machine__new_host();
187 	/*
188 	 * FIXME:
189 	 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
190 	 *    ask for not using the kcore parsing code, once this one is fixed
191 	 *    to create a map per module.
192 	 */
193 	if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
194 		machine__delete(machine);
195 		machine = NULL;
196 	}
197 
198 	return machine;
199 }
200 
dsos__purge(struct dsos * dsos)201 static void dsos__purge(struct dsos *dsos)
202 {
203 	struct dso *pos, *n;
204 
205 	down_write(&dsos->lock);
206 
207 	list_for_each_entry_safe(pos, n, &dsos->head, node) {
208 		RB_CLEAR_NODE(&pos->rb_node);
209 		pos->root = NULL;
210 		list_del_init(&pos->node);
211 		dso__put(pos);
212 	}
213 
214 	up_write(&dsos->lock);
215 }
216 
dsos__exit(struct dsos * dsos)217 static void dsos__exit(struct dsos *dsos)
218 {
219 	dsos__purge(dsos);
220 	exit_rwsem(&dsos->lock);
221 }
222 
machine__delete_threads(struct machine * machine)223 void machine__delete_threads(struct machine *machine)
224 {
225 	struct rb_node *nd;
226 	int i;
227 
228 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
229 		struct threads *threads = &machine->threads[i];
230 		down_write(&threads->lock);
231 		nd = rb_first_cached(&threads->entries);
232 		while (nd) {
233 			struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
234 
235 			nd = rb_next(nd);
236 			__machine__remove_thread(machine, trb, trb->thread, false);
237 		}
238 		up_write(&threads->lock);
239 	}
240 }
241 
machine__exit(struct machine * machine)242 void machine__exit(struct machine *machine)
243 {
244 	int i;
245 
246 	if (machine == NULL)
247 		return;
248 
249 	machine__destroy_kernel_maps(machine);
250 	maps__zput(machine->kmaps);
251 	dsos__exit(&machine->dsos);
252 	machine__exit_vdso(machine);
253 	zfree(&machine->root_dir);
254 	zfree(&machine->mmap_name);
255 	zfree(&machine->current_tid);
256 	zfree(&machine->kallsyms_filename);
257 
258 	machine__delete_threads(machine);
259 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
260 		struct threads *threads = &machine->threads[i];
261 
262 		exit_rwsem(&threads->lock);
263 	}
264 }
265 
machine__delete(struct machine * machine)266 void machine__delete(struct machine *machine)
267 {
268 	if (machine) {
269 		machine__exit(machine);
270 		free(machine);
271 	}
272 }
273 
machines__init(struct machines * machines)274 void machines__init(struct machines *machines)
275 {
276 	machine__init(&machines->host, "", HOST_KERNEL_ID);
277 	machines->guests = RB_ROOT_CACHED;
278 }
279 
machines__exit(struct machines * machines)280 void machines__exit(struct machines *machines)
281 {
282 	machine__exit(&machines->host);
283 	/* XXX exit guest */
284 }
285 
machines__add(struct machines * machines,pid_t pid,const char * root_dir)286 struct machine *machines__add(struct machines *machines, pid_t pid,
287 			      const char *root_dir)
288 {
289 	struct rb_node **p = &machines->guests.rb_root.rb_node;
290 	struct rb_node *parent = NULL;
291 	struct machine *pos, *machine = malloc(sizeof(*machine));
292 	bool leftmost = true;
293 
294 	if (machine == NULL)
295 		return NULL;
296 
297 	if (machine__init(machine, root_dir, pid) != 0) {
298 		free(machine);
299 		return NULL;
300 	}
301 
302 	while (*p != NULL) {
303 		parent = *p;
304 		pos = rb_entry(parent, struct machine, rb_node);
305 		if (pid < pos->pid)
306 			p = &(*p)->rb_left;
307 		else {
308 			p = &(*p)->rb_right;
309 			leftmost = false;
310 		}
311 	}
312 
313 	rb_link_node(&machine->rb_node, parent, p);
314 	rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
315 
316 	machine->machines = machines;
317 
318 	return machine;
319 }
320 
machines__set_comm_exec(struct machines * machines,bool comm_exec)321 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
322 {
323 	struct rb_node *nd;
324 
325 	machines->host.comm_exec = comm_exec;
326 
327 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
328 		struct machine *machine = rb_entry(nd, struct machine, rb_node);
329 
330 		machine->comm_exec = comm_exec;
331 	}
332 }
333 
machines__find(struct machines * machines,pid_t pid)334 struct machine *machines__find(struct machines *machines, pid_t pid)
335 {
336 	struct rb_node **p = &machines->guests.rb_root.rb_node;
337 	struct rb_node *parent = NULL;
338 	struct machine *machine;
339 	struct machine *default_machine = NULL;
340 
341 	if (pid == HOST_KERNEL_ID)
342 		return &machines->host;
343 
344 	while (*p != NULL) {
345 		parent = *p;
346 		machine = rb_entry(parent, struct machine, rb_node);
347 		if (pid < machine->pid)
348 			p = &(*p)->rb_left;
349 		else if (pid > machine->pid)
350 			p = &(*p)->rb_right;
351 		else
352 			return machine;
353 		if (!machine->pid)
354 			default_machine = machine;
355 	}
356 
357 	return default_machine;
358 }
359 
machines__findnew(struct machines * machines,pid_t pid)360 struct machine *machines__findnew(struct machines *machines, pid_t pid)
361 {
362 	char path[PATH_MAX];
363 	const char *root_dir = "";
364 	struct machine *machine = machines__find(machines, pid);
365 
366 	if (machine && (machine->pid == pid))
367 		goto out;
368 
369 	if ((pid != HOST_KERNEL_ID) &&
370 	    (pid != DEFAULT_GUEST_KERNEL_ID) &&
371 	    (symbol_conf.guestmount)) {
372 		sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
373 		if (access(path, R_OK)) {
374 			static struct strlist *seen;
375 
376 			if (!seen)
377 				seen = strlist__new(NULL, NULL);
378 
379 			if (!strlist__has_entry(seen, path)) {
380 				pr_err("Can't access file %s\n", path);
381 				strlist__add(seen, path);
382 			}
383 			machine = NULL;
384 			goto out;
385 		}
386 		root_dir = path;
387 	}
388 
389 	machine = machines__add(machines, pid, root_dir);
390 out:
391 	return machine;
392 }
393 
machines__find_guest(struct machines * machines,pid_t pid)394 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
395 {
396 	struct machine *machine = machines__find(machines, pid);
397 
398 	if (!machine)
399 		machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
400 	return machine;
401 }
402 
403 /*
404  * A common case for KVM test programs is that the test program acts as the
405  * hypervisor, creating, running and destroying the virtual machine, and
406  * providing the guest object code from its own object code. In this case,
407  * the VM is not running an OS, but only the functions loaded into it by the
408  * hypervisor test program, and conveniently, loaded at the same virtual
409  * addresses.
410  *
411  * Normally to resolve addresses, MMAP events are needed to map addresses
412  * back to the object code and debug symbols for that object code.
413  *
414  * Currently, there is no way to get such mapping information from guests
415  * but, in the scenario described above, the guest has the same mappings
416  * as the hypervisor, so support for that scenario can be achieved.
417  *
418  * To support that, copy the host thread's maps to the guest thread's maps.
419  * Note, we do not discover the guest until we encounter a guest event,
420  * which works well because it is not until then that we know that the host
421  * thread's maps have been set up.
422  *
423  * This function returns the guest thread. Apart from keeping the data
424  * structures sane, using a thread belonging to the guest machine, instead
425  * of the host thread, allows it to have its own comm (refer
426  * thread__set_guest_comm()).
427  */
findnew_guest_code(struct machine * machine,struct machine * host_machine,pid_t pid)428 static struct thread *findnew_guest_code(struct machine *machine,
429 					 struct machine *host_machine,
430 					 pid_t pid)
431 {
432 	struct thread *host_thread;
433 	struct thread *thread;
434 	int err;
435 
436 	if (!machine)
437 		return NULL;
438 
439 	thread = machine__findnew_thread(machine, -1, pid);
440 	if (!thread)
441 		return NULL;
442 
443 	/* Assume maps are set up if there are any */
444 	if (maps__nr_maps(thread__maps(thread)))
445 		return thread;
446 
447 	host_thread = machine__find_thread(host_machine, -1, pid);
448 	if (!host_thread)
449 		goto out_err;
450 
451 	thread__set_guest_comm(thread, pid);
452 
453 	/*
454 	 * Guest code can be found in hypervisor process at the same address
455 	 * so copy host maps.
456 	 */
457 	err = maps__clone(thread, thread__maps(host_thread));
458 	thread__put(host_thread);
459 	if (err)
460 		goto out_err;
461 
462 	return thread;
463 
464 out_err:
465 	thread__zput(thread);
466 	return NULL;
467 }
468 
machines__findnew_guest_code(struct machines * machines,pid_t pid)469 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
470 {
471 	struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
472 	struct machine *machine = machines__findnew(machines, pid);
473 
474 	return findnew_guest_code(machine, host_machine, pid);
475 }
476 
machine__findnew_guest_code(struct machine * machine,pid_t pid)477 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
478 {
479 	struct machines *machines = machine->machines;
480 	struct machine *host_machine;
481 
482 	if (!machines)
483 		return NULL;
484 
485 	host_machine = machines__find(machines, HOST_KERNEL_ID);
486 
487 	return findnew_guest_code(machine, host_machine, pid);
488 }
489 
machines__process_guests(struct machines * machines,machine__process_t process,void * data)490 void machines__process_guests(struct machines *machines,
491 			      machine__process_t process, void *data)
492 {
493 	struct rb_node *nd;
494 
495 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
496 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
497 		process(pos, data);
498 	}
499 }
500 
machines__set_id_hdr_size(struct machines * machines,u16 id_hdr_size)501 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
502 {
503 	struct rb_node *node;
504 	struct machine *machine;
505 
506 	machines->host.id_hdr_size = id_hdr_size;
507 
508 	for (node = rb_first_cached(&machines->guests); node;
509 	     node = rb_next(node)) {
510 		machine = rb_entry(node, struct machine, rb_node);
511 		machine->id_hdr_size = id_hdr_size;
512 	}
513 
514 	return;
515 }
516 
machine__update_thread_pid(struct machine * machine,struct thread * th,pid_t pid)517 static void machine__update_thread_pid(struct machine *machine,
518 				       struct thread *th, pid_t pid)
519 {
520 	struct thread *leader;
521 
522 	if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
523 		return;
524 
525 	thread__set_pid(th, pid);
526 
527 	if (thread__pid(th) == thread__tid(th))
528 		return;
529 
530 	leader = __machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
531 	if (!leader)
532 		goto out_err;
533 
534 	if (!thread__maps(leader))
535 		thread__set_maps(leader, maps__new(machine));
536 
537 	if (!thread__maps(leader))
538 		goto out_err;
539 
540 	if (thread__maps(th) == thread__maps(leader))
541 		goto out_put;
542 
543 	if (thread__maps(th)) {
544 		/*
545 		 * Maps are created from MMAP events which provide the pid and
546 		 * tid.  Consequently there never should be any maps on a thread
547 		 * with an unknown pid.  Just print an error if there are.
548 		 */
549 		if (!maps__empty(thread__maps(th)))
550 			pr_err("Discarding thread maps for %d:%d\n",
551 				thread__pid(th), thread__tid(th));
552 		maps__put(thread__maps(th));
553 	}
554 
555 	thread__set_maps(th, maps__get(thread__maps(leader)));
556 out_put:
557 	thread__put(leader);
558 	return;
559 out_err:
560 	pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
561 	goto out_put;
562 }
563 
564 /*
565  * Front-end cache - TID lookups come in blocks,
566  * so most of the time we dont have to look up
567  * the full rbtree:
568  */
569 static struct thread*
__threads__get_last_match(struct threads * threads,struct machine * machine,int pid,int tid)570 __threads__get_last_match(struct threads *threads, struct machine *machine,
571 			  int pid, int tid)
572 {
573 	struct thread *th;
574 
575 	th = threads->last_match;
576 	if (th != NULL) {
577 		if (thread__tid(th) == tid) {
578 			machine__update_thread_pid(machine, th, pid);
579 			return thread__get(th);
580 		}
581 		thread__put(threads->last_match);
582 		threads->last_match = NULL;
583 	}
584 
585 	return NULL;
586 }
587 
588 static struct thread*
threads__get_last_match(struct threads * threads,struct machine * machine,int pid,int tid)589 threads__get_last_match(struct threads *threads, struct machine *machine,
590 			int pid, int tid)
591 {
592 	struct thread *th = NULL;
593 
594 	if (perf_singlethreaded)
595 		th = __threads__get_last_match(threads, machine, pid, tid);
596 
597 	return th;
598 }
599 
600 static void
__threads__set_last_match(struct threads * threads,struct thread * th)601 __threads__set_last_match(struct threads *threads, struct thread *th)
602 {
603 	thread__put(threads->last_match);
604 	threads->last_match = thread__get(th);
605 }
606 
607 static void
threads__set_last_match(struct threads * threads,struct thread * th)608 threads__set_last_match(struct threads *threads, struct thread *th)
609 {
610 	if (perf_singlethreaded)
611 		__threads__set_last_match(threads, th);
612 }
613 
614 /*
615  * Caller must eventually drop thread->refcnt returned with a successful
616  * lookup/new thread inserted.
617  */
____machine__findnew_thread(struct machine * machine,struct threads * threads,pid_t pid,pid_t tid,bool create)618 static struct thread *____machine__findnew_thread(struct machine *machine,
619 						  struct threads *threads,
620 						  pid_t pid, pid_t tid,
621 						  bool create)
622 {
623 	struct rb_node **p = &threads->entries.rb_root.rb_node;
624 	struct rb_node *parent = NULL;
625 	struct thread *th;
626 	struct thread_rb_node *nd;
627 	bool leftmost = true;
628 
629 	th = threads__get_last_match(threads, machine, pid, tid);
630 	if (th)
631 		return th;
632 
633 	while (*p != NULL) {
634 		parent = *p;
635 		th = rb_entry(parent, struct thread_rb_node, rb_node)->thread;
636 
637 		if (thread__tid(th) == tid) {
638 			threads__set_last_match(threads, th);
639 			machine__update_thread_pid(machine, th, pid);
640 			return thread__get(th);
641 		}
642 
643 		if (tid < thread__tid(th))
644 			p = &(*p)->rb_left;
645 		else {
646 			p = &(*p)->rb_right;
647 			leftmost = false;
648 		}
649 	}
650 
651 	if (!create)
652 		return NULL;
653 
654 	th = thread__new(pid, tid);
655 	if (th == NULL)
656 		return NULL;
657 
658 	nd = malloc(sizeof(*nd));
659 	if (nd == NULL) {
660 		thread__put(th);
661 		return NULL;
662 	}
663 	nd->thread = th;
664 
665 	rb_link_node(&nd->rb_node, parent, p);
666 	rb_insert_color_cached(&nd->rb_node, &threads->entries, leftmost);
667 	/*
668 	 * We have to initialize maps separately after rb tree is updated.
669 	 *
670 	 * The reason is that we call machine__findnew_thread within
671 	 * thread__init_maps to find the thread leader and that would screwed
672 	 * the rb tree.
673 	 */
674 	if (thread__init_maps(th, machine)) {
675 		pr_err("Thread init failed thread %d\n", pid);
676 		rb_erase_cached(&nd->rb_node, &threads->entries);
677 		RB_CLEAR_NODE(&nd->rb_node);
678 		free(nd);
679 		thread__put(th);
680 		return NULL;
681 	}
682 	/*
683 	 * It is now in the rbtree, get a ref
684 	 */
685 	threads__set_last_match(threads, th);
686 	++threads->nr;
687 
688 	return thread__get(th);
689 }
690 
__machine__findnew_thread(struct machine * machine,pid_t pid,pid_t tid)691 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
692 {
693 	return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
694 }
695 
machine__findnew_thread(struct machine * machine,pid_t pid,pid_t tid)696 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
697 				       pid_t tid)
698 {
699 	struct threads *threads = machine__threads(machine, tid);
700 	struct thread *th;
701 
702 	down_write(&threads->lock);
703 	th = __machine__findnew_thread(machine, pid, tid);
704 	up_write(&threads->lock);
705 	return th;
706 }
707 
machine__find_thread(struct machine * machine,pid_t pid,pid_t tid)708 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
709 				    pid_t tid)
710 {
711 	struct threads *threads = machine__threads(machine, tid);
712 	struct thread *th;
713 
714 	down_read(&threads->lock);
715 	th =  ____machine__findnew_thread(machine, threads, pid, tid, false);
716 	up_read(&threads->lock);
717 	return th;
718 }
719 
720 /*
721  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
722  * So here a single thread is created for that, but actually there is a separate
723  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
724  * is only 1. That causes problems for some tools, requiring workarounds. For
725  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
726  */
machine__idle_thread(struct machine * machine)727 struct thread *machine__idle_thread(struct machine *machine)
728 {
729 	struct thread *thread = machine__findnew_thread(machine, 0, 0);
730 
731 	if (!thread || thread__set_comm(thread, "swapper", 0) ||
732 	    thread__set_namespaces(thread, 0, NULL))
733 		pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
734 
735 	return thread;
736 }
737 
machine__thread_exec_comm(struct machine * machine,struct thread * thread)738 struct comm *machine__thread_exec_comm(struct machine *machine,
739 				       struct thread *thread)
740 {
741 	if (machine->comm_exec)
742 		return thread__exec_comm(thread);
743 	else
744 		return thread__comm(thread);
745 }
746 
machine__process_comm_event(struct machine * machine,union perf_event * event,struct perf_sample * sample)747 int machine__process_comm_event(struct machine *machine, union perf_event *event,
748 				struct perf_sample *sample)
749 {
750 	struct thread *thread = machine__findnew_thread(machine,
751 							event->comm.pid,
752 							event->comm.tid);
753 	bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
754 	int err = 0;
755 
756 	if (exec)
757 		machine->comm_exec = true;
758 
759 	if (dump_trace)
760 		perf_event__fprintf_comm(event, stdout);
761 
762 	if (thread == NULL ||
763 	    __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
764 		dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
765 		err = -1;
766 	}
767 
768 	thread__put(thread);
769 
770 	return err;
771 }
772 
machine__process_namespaces_event(struct machine * machine __maybe_unused,union perf_event * event,struct perf_sample * sample __maybe_unused)773 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
774 				      union perf_event *event,
775 				      struct perf_sample *sample __maybe_unused)
776 {
777 	struct thread *thread = machine__findnew_thread(machine,
778 							event->namespaces.pid,
779 							event->namespaces.tid);
780 	int err = 0;
781 
782 	WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
783 		  "\nWARNING: kernel seems to support more namespaces than perf"
784 		  " tool.\nTry updating the perf tool..\n\n");
785 
786 	WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
787 		  "\nWARNING: perf tool seems to support more namespaces than"
788 		  " the kernel.\nTry updating the kernel..\n\n");
789 
790 	if (dump_trace)
791 		perf_event__fprintf_namespaces(event, stdout);
792 
793 	if (thread == NULL ||
794 	    thread__set_namespaces(thread, sample->time, &event->namespaces)) {
795 		dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
796 		err = -1;
797 	}
798 
799 	thread__put(thread);
800 
801 	return err;
802 }
803 
machine__process_cgroup_event(struct machine * machine,union perf_event * event,struct perf_sample * sample __maybe_unused)804 int machine__process_cgroup_event(struct machine *machine,
805 				  union perf_event *event,
806 				  struct perf_sample *sample __maybe_unused)
807 {
808 	struct cgroup *cgrp;
809 
810 	if (dump_trace)
811 		perf_event__fprintf_cgroup(event, stdout);
812 
813 	cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
814 	if (cgrp == NULL)
815 		return -ENOMEM;
816 
817 	return 0;
818 }
819 
machine__process_lost_event(struct machine * machine __maybe_unused,union perf_event * event,struct perf_sample * sample __maybe_unused)820 int machine__process_lost_event(struct machine *machine __maybe_unused,
821 				union perf_event *event, struct perf_sample *sample __maybe_unused)
822 {
823 	dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
824 		    event->lost.id, event->lost.lost);
825 	return 0;
826 }
827 
machine__process_lost_samples_event(struct machine * machine __maybe_unused,union perf_event * event,struct perf_sample * sample)828 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
829 					union perf_event *event, struct perf_sample *sample)
830 {
831 	dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
832 		    sample->id, event->lost_samples.lost);
833 	return 0;
834 }
835 
machine__findnew_module_dso(struct machine * machine,struct kmod_path * m,const char * filename)836 static struct dso *machine__findnew_module_dso(struct machine *machine,
837 					       struct kmod_path *m,
838 					       const char *filename)
839 {
840 	struct dso *dso;
841 
842 	down_write(&machine->dsos.lock);
843 
844 	dso = __dsos__find(&machine->dsos, m->name, true);
845 	if (!dso) {
846 		dso = __dsos__addnew(&machine->dsos, m->name);
847 		if (dso == NULL)
848 			goto out_unlock;
849 
850 		dso__set_module_info(dso, m, machine);
851 		dso__set_long_name(dso, strdup(filename), true);
852 		dso->kernel = DSO_SPACE__KERNEL;
853 	}
854 
855 	dso__get(dso);
856 out_unlock:
857 	up_write(&machine->dsos.lock);
858 	return dso;
859 }
860 
machine__process_aux_event(struct machine * machine __maybe_unused,union perf_event * event)861 int machine__process_aux_event(struct machine *machine __maybe_unused,
862 			       union perf_event *event)
863 {
864 	if (dump_trace)
865 		perf_event__fprintf_aux(event, stdout);
866 	return 0;
867 }
868 
machine__process_itrace_start_event(struct machine * machine __maybe_unused,union perf_event * event)869 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
870 					union perf_event *event)
871 {
872 	if (dump_trace)
873 		perf_event__fprintf_itrace_start(event, stdout);
874 	return 0;
875 }
876 
machine__process_aux_output_hw_id_event(struct machine * machine __maybe_unused,union perf_event * event)877 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
878 					    union perf_event *event)
879 {
880 	if (dump_trace)
881 		perf_event__fprintf_aux_output_hw_id(event, stdout);
882 	return 0;
883 }
884 
machine__process_switch_event(struct machine * machine __maybe_unused,union perf_event * event)885 int machine__process_switch_event(struct machine *machine __maybe_unused,
886 				  union perf_event *event)
887 {
888 	if (dump_trace)
889 		perf_event__fprintf_switch(event, stdout);
890 	return 0;
891 }
892 
machine__process_ksymbol_register(struct machine * machine,union perf_event * event,struct perf_sample * sample __maybe_unused)893 static int machine__process_ksymbol_register(struct machine *machine,
894 					     union perf_event *event,
895 					     struct perf_sample *sample __maybe_unused)
896 {
897 	struct symbol *sym;
898 	struct dso *dso;
899 	struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
900 	bool put_map = false;
901 	int err = 0;
902 
903 	if (!map) {
904 		dso = dso__new(event->ksymbol.name);
905 
906 		if (!dso) {
907 			err = -ENOMEM;
908 			goto out;
909 		}
910 		dso->kernel = DSO_SPACE__KERNEL;
911 		map = map__new2(0, dso);
912 		dso__put(dso);
913 		if (!map) {
914 			err = -ENOMEM;
915 			goto out;
916 		}
917 		/*
918 		 * The inserted map has a get on it, we need to put to release
919 		 * the reference count here, but do it after all accesses are
920 		 * done.
921 		 */
922 		put_map = true;
923 		if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
924 			dso->binary_type = DSO_BINARY_TYPE__OOL;
925 			dso->data.file_size = event->ksymbol.len;
926 			dso__set_loaded(dso);
927 		}
928 
929 		map__set_start(map, event->ksymbol.addr);
930 		map__set_end(map, map__start(map) + event->ksymbol.len);
931 		err = maps__insert(machine__kernel_maps(machine), map);
932 		if (err) {
933 			err = -ENOMEM;
934 			goto out;
935 		}
936 
937 		dso__set_loaded(dso);
938 
939 		if (is_bpf_image(event->ksymbol.name)) {
940 			dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
941 			dso__set_long_name(dso, "", false);
942 		}
943 	} else {
944 		dso = map__dso(map);
945 	}
946 
947 	sym = symbol__new(map__map_ip(map, map__start(map)),
948 			  event->ksymbol.len,
949 			  0, 0, event->ksymbol.name);
950 	if (!sym) {
951 		err = -ENOMEM;
952 		goto out;
953 	}
954 	dso__insert_symbol(dso, sym);
955 out:
956 	if (put_map)
957 		map__put(map);
958 	return err;
959 }
960 
machine__process_ksymbol_unregister(struct machine * machine,union perf_event * event,struct perf_sample * sample __maybe_unused)961 static int machine__process_ksymbol_unregister(struct machine *machine,
962 					       union perf_event *event,
963 					       struct perf_sample *sample __maybe_unused)
964 {
965 	struct symbol *sym;
966 	struct map *map;
967 
968 	map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
969 	if (!map)
970 		return 0;
971 
972 	if (RC_CHK_ACCESS(map) != RC_CHK_ACCESS(machine->vmlinux_map))
973 		maps__remove(machine__kernel_maps(machine), map);
974 	else {
975 		struct dso *dso = map__dso(map);
976 
977 		sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
978 		if (sym)
979 			dso__delete_symbol(dso, sym);
980 	}
981 
982 	return 0;
983 }
984 
machine__process_ksymbol(struct machine * machine __maybe_unused,union perf_event * event,struct perf_sample * sample)985 int machine__process_ksymbol(struct machine *machine __maybe_unused,
986 			     union perf_event *event,
987 			     struct perf_sample *sample)
988 {
989 	if (dump_trace)
990 		perf_event__fprintf_ksymbol(event, stdout);
991 
992 	if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
993 		return machine__process_ksymbol_unregister(machine, event,
994 							   sample);
995 	return machine__process_ksymbol_register(machine, event, sample);
996 }
997 
machine__process_text_poke(struct machine * machine,union perf_event * event,struct perf_sample * sample __maybe_unused)998 int machine__process_text_poke(struct machine *machine, union perf_event *event,
999 			       struct perf_sample *sample __maybe_unused)
1000 {
1001 	struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
1002 	u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1003 	struct dso *dso = map ? map__dso(map) : NULL;
1004 
1005 	if (dump_trace)
1006 		perf_event__fprintf_text_poke(event, machine, stdout);
1007 
1008 	if (!event->text_poke.new_len)
1009 		return 0;
1010 
1011 	if (cpumode != PERF_RECORD_MISC_KERNEL) {
1012 		pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
1013 		return 0;
1014 	}
1015 
1016 	if (dso) {
1017 		u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
1018 		int ret;
1019 
1020 		/*
1021 		 * Kernel maps might be changed when loading symbols so loading
1022 		 * must be done prior to using kernel maps.
1023 		 */
1024 		map__load(map);
1025 		ret = dso__data_write_cache_addr(dso, map, machine,
1026 						 event->text_poke.addr,
1027 						 new_bytes,
1028 						 event->text_poke.new_len);
1029 		if (ret != event->text_poke.new_len)
1030 			pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
1031 				 event->text_poke.addr);
1032 	} else {
1033 		pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
1034 			 event->text_poke.addr);
1035 	}
1036 
1037 	return 0;
1038 }
1039 
machine__addnew_module_map(struct machine * machine,u64 start,const char * filename)1040 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
1041 					      const char *filename)
1042 {
1043 	struct map *map = NULL;
1044 	struct kmod_path m;
1045 	struct dso *dso;
1046 	int err;
1047 
1048 	if (kmod_path__parse_name(&m, filename))
1049 		return NULL;
1050 
1051 	dso = machine__findnew_module_dso(machine, &m, filename);
1052 	if (dso == NULL)
1053 		goto out;
1054 
1055 	map = map__new2(start, dso);
1056 	if (map == NULL)
1057 		goto out;
1058 
1059 	err = maps__insert(machine__kernel_maps(machine), map);
1060 	/* If maps__insert failed, return NULL. */
1061 	if (err) {
1062 		map__put(map);
1063 		map = NULL;
1064 	}
1065 out:
1066 	/* put the dso here, corresponding to  machine__findnew_module_dso */
1067 	dso__put(dso);
1068 	zfree(&m.name);
1069 	return map;
1070 }
1071 
machines__fprintf_dsos(struct machines * machines,FILE * fp)1072 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
1073 {
1074 	struct rb_node *nd;
1075 	size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
1076 
1077 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1078 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
1079 		ret += __dsos__fprintf(&pos->dsos.head, fp);
1080 	}
1081 
1082 	return ret;
1083 }
1084 
machine__fprintf_dsos_buildid(struct machine * m,FILE * fp,bool (skip)(struct dso * dso,int parm),int parm)1085 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
1086 				     bool (skip)(struct dso *dso, int parm), int parm)
1087 {
1088 	return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
1089 }
1090 
machines__fprintf_dsos_buildid(struct machines * machines,FILE * fp,bool (skip)(struct dso * dso,int parm),int parm)1091 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
1092 				     bool (skip)(struct dso *dso, int parm), int parm)
1093 {
1094 	struct rb_node *nd;
1095 	size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
1096 
1097 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1098 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
1099 		ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
1100 	}
1101 	return ret;
1102 }
1103 
machine__fprintf_vmlinux_path(struct machine * machine,FILE * fp)1104 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
1105 {
1106 	int i;
1107 	size_t printed = 0;
1108 	struct dso *kdso = machine__kernel_dso(machine);
1109 
1110 	if (kdso->has_build_id) {
1111 		char filename[PATH_MAX];
1112 		if (dso__build_id_filename(kdso, filename, sizeof(filename),
1113 					   false))
1114 			printed += fprintf(fp, "[0] %s\n", filename);
1115 	}
1116 
1117 	for (i = 0; i < vmlinux_path__nr_entries; ++i)
1118 		printed += fprintf(fp, "[%d] %s\n",
1119 				   i + kdso->has_build_id, vmlinux_path[i]);
1120 
1121 	return printed;
1122 }
1123 
machine__fprintf(struct machine * machine,FILE * fp)1124 size_t machine__fprintf(struct machine *machine, FILE *fp)
1125 {
1126 	struct rb_node *nd;
1127 	size_t ret;
1128 	int i;
1129 
1130 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
1131 		struct threads *threads = &machine->threads[i];
1132 
1133 		down_read(&threads->lock);
1134 
1135 		ret = fprintf(fp, "Threads: %u\n", threads->nr);
1136 
1137 		for (nd = rb_first_cached(&threads->entries); nd;
1138 		     nd = rb_next(nd)) {
1139 			struct thread *pos = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
1140 
1141 			ret += thread__fprintf(pos, fp);
1142 		}
1143 
1144 		up_read(&threads->lock);
1145 	}
1146 	return ret;
1147 }
1148 
machine__get_kernel(struct machine * machine)1149 static struct dso *machine__get_kernel(struct machine *machine)
1150 {
1151 	const char *vmlinux_name = machine->mmap_name;
1152 	struct dso *kernel;
1153 
1154 	if (machine__is_host(machine)) {
1155 		if (symbol_conf.vmlinux_name)
1156 			vmlinux_name = symbol_conf.vmlinux_name;
1157 
1158 		kernel = machine__findnew_kernel(machine, vmlinux_name,
1159 						 "[kernel]", DSO_SPACE__KERNEL);
1160 	} else {
1161 		if (symbol_conf.default_guest_vmlinux_name)
1162 			vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1163 
1164 		kernel = machine__findnew_kernel(machine, vmlinux_name,
1165 						 "[guest.kernel]",
1166 						 DSO_SPACE__KERNEL_GUEST);
1167 	}
1168 
1169 	if (kernel != NULL && (!kernel->has_build_id))
1170 		dso__read_running_kernel_build_id(kernel, machine);
1171 
1172 	return kernel;
1173 }
1174 
machine__get_kallsyms_filename(struct machine * machine,char * buf,size_t bufsz)1175 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1176 				    size_t bufsz)
1177 {
1178 	if (machine__is_default_guest(machine))
1179 		scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1180 	else
1181 		scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1182 }
1183 
1184 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1185 
1186 /* Figure out the start address of kernel map from /proc/kallsyms.
1187  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1188  * symbol_name if it's not that important.
1189  */
machine__get_running_kernel_start(struct machine * machine,const char ** symbol_name,u64 * start,u64 * end)1190 static int machine__get_running_kernel_start(struct machine *machine,
1191 					     const char **symbol_name,
1192 					     u64 *start, u64 *end)
1193 {
1194 	char filename[PATH_MAX];
1195 	int i, err = -1;
1196 	const char *name;
1197 	u64 addr = 0;
1198 
1199 	machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1200 
1201 	if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1202 		return 0;
1203 
1204 	for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1205 		err = kallsyms__get_function_start(filename, name, &addr);
1206 		if (!err)
1207 			break;
1208 	}
1209 
1210 	if (err)
1211 		return -1;
1212 
1213 	if (symbol_name)
1214 		*symbol_name = name;
1215 
1216 	*start = addr;
1217 
1218 	err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1219 	if (err)
1220 		err = kallsyms__get_function_start(filename, "_etext", &addr);
1221 	if (!err)
1222 		*end = addr;
1223 
1224 	return 0;
1225 }
1226 
machine__create_extra_kernel_map(struct machine * machine,struct dso * kernel,struct extra_kernel_map * xm)1227 int machine__create_extra_kernel_map(struct machine *machine,
1228 				     struct dso *kernel,
1229 				     struct extra_kernel_map *xm)
1230 {
1231 	struct kmap *kmap;
1232 	struct map *map;
1233 	int err;
1234 
1235 	map = map__new2(xm->start, kernel);
1236 	if (!map)
1237 		return -ENOMEM;
1238 
1239 	map__set_end(map, xm->end);
1240 	map__set_pgoff(map, xm->pgoff);
1241 
1242 	kmap = map__kmap(map);
1243 
1244 	strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1245 
1246 	err = maps__insert(machine__kernel_maps(machine), map);
1247 
1248 	if (!err) {
1249 		pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1250 			kmap->name, map__start(map), map__end(map));
1251 	}
1252 
1253 	map__put(map);
1254 
1255 	return err;
1256 }
1257 
find_entry_trampoline(struct dso * dso)1258 static u64 find_entry_trampoline(struct dso *dso)
1259 {
1260 	/* Duplicates are removed so lookup all aliases */
1261 	const char *syms[] = {
1262 		"_entry_trampoline",
1263 		"__entry_trampoline_start",
1264 		"entry_SYSCALL_64_trampoline",
1265 	};
1266 	struct symbol *sym = dso__first_symbol(dso);
1267 	unsigned int i;
1268 
1269 	for (; sym; sym = dso__next_symbol(sym)) {
1270 		if (sym->binding != STB_GLOBAL)
1271 			continue;
1272 		for (i = 0; i < ARRAY_SIZE(syms); i++) {
1273 			if (!strcmp(sym->name, syms[i]))
1274 				return sym->start;
1275 		}
1276 	}
1277 
1278 	return 0;
1279 }
1280 
1281 /*
1282  * These values can be used for kernels that do not have symbols for the entry
1283  * trampolines in kallsyms.
1284  */
1285 #define X86_64_CPU_ENTRY_AREA_PER_CPU	0xfffffe0000000000ULL
1286 #define X86_64_CPU_ENTRY_AREA_SIZE	0x2c000
1287 #define X86_64_ENTRY_TRAMPOLINE		0x6000
1288 
1289 /* Map x86_64 PTI entry trampolines */
machine__map_x86_64_entry_trampolines(struct machine * machine,struct dso * kernel)1290 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1291 					  struct dso *kernel)
1292 {
1293 	struct maps *kmaps = machine__kernel_maps(machine);
1294 	int nr_cpus_avail, cpu;
1295 	bool found = false;
1296 	struct map_rb_node *rb_node;
1297 	u64 pgoff;
1298 
1299 	/*
1300 	 * In the vmlinux case, pgoff is a virtual address which must now be
1301 	 * mapped to a vmlinux offset.
1302 	 */
1303 	maps__for_each_entry(kmaps, rb_node) {
1304 		struct map *dest_map, *map = rb_node->map;
1305 		struct kmap *kmap = __map__kmap(map);
1306 
1307 		if (!kmap || !is_entry_trampoline(kmap->name))
1308 			continue;
1309 
1310 		dest_map = maps__find(kmaps, map__pgoff(map));
1311 		if (dest_map != map)
1312 			map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1313 		found = true;
1314 	}
1315 	if (found || machine->trampolines_mapped)
1316 		return 0;
1317 
1318 	pgoff = find_entry_trampoline(kernel);
1319 	if (!pgoff)
1320 		return 0;
1321 
1322 	nr_cpus_avail = machine__nr_cpus_avail(machine);
1323 
1324 	/* Add a 1 page map for each CPU's entry trampoline */
1325 	for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1326 		u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1327 			 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1328 			 X86_64_ENTRY_TRAMPOLINE;
1329 		struct extra_kernel_map xm = {
1330 			.start = va,
1331 			.end   = va + page_size,
1332 			.pgoff = pgoff,
1333 		};
1334 
1335 		strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1336 
1337 		if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1338 			return -1;
1339 	}
1340 
1341 	machine->trampolines_mapped = nr_cpus_avail;
1342 
1343 	return 0;
1344 }
1345 
machine__create_extra_kernel_maps(struct machine * machine __maybe_unused,struct dso * kernel __maybe_unused)1346 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1347 					     struct dso *kernel __maybe_unused)
1348 {
1349 	return 0;
1350 }
1351 
1352 static int
__machine__create_kernel_maps(struct machine * machine,struct dso * kernel)1353 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1354 {
1355 	/* In case of renewal the kernel map, destroy previous one */
1356 	machine__destroy_kernel_maps(machine);
1357 
1358 	map__put(machine->vmlinux_map);
1359 	machine->vmlinux_map = map__new2(0, kernel);
1360 	if (machine->vmlinux_map == NULL)
1361 		return -ENOMEM;
1362 
1363 	map__set_map_ip(machine->vmlinux_map, identity__map_ip);
1364 	map__set_unmap_ip(machine->vmlinux_map, identity__map_ip);
1365 	return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1366 }
1367 
machine__destroy_kernel_maps(struct machine * machine)1368 void machine__destroy_kernel_maps(struct machine *machine)
1369 {
1370 	struct kmap *kmap;
1371 	struct map *map = machine__kernel_map(machine);
1372 
1373 	if (map == NULL)
1374 		return;
1375 
1376 	kmap = map__kmap(map);
1377 	maps__remove(machine__kernel_maps(machine), map);
1378 	if (kmap && kmap->ref_reloc_sym) {
1379 		zfree((char **)&kmap->ref_reloc_sym->name);
1380 		zfree(&kmap->ref_reloc_sym);
1381 	}
1382 
1383 	map__zput(machine->vmlinux_map);
1384 }
1385 
machines__create_guest_kernel_maps(struct machines * machines)1386 int machines__create_guest_kernel_maps(struct machines *machines)
1387 {
1388 	int ret = 0;
1389 	struct dirent **namelist = NULL;
1390 	int i, items = 0;
1391 	char path[PATH_MAX];
1392 	pid_t pid;
1393 	char *endp;
1394 
1395 	if (symbol_conf.default_guest_vmlinux_name ||
1396 	    symbol_conf.default_guest_modules ||
1397 	    symbol_conf.default_guest_kallsyms) {
1398 		machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1399 	}
1400 
1401 	if (symbol_conf.guestmount) {
1402 		items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1403 		if (items <= 0)
1404 			return -ENOENT;
1405 		for (i = 0; i < items; i++) {
1406 			if (!isdigit(namelist[i]->d_name[0])) {
1407 				/* Filter out . and .. */
1408 				continue;
1409 			}
1410 			pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1411 			if ((*endp != '\0') ||
1412 			    (endp == namelist[i]->d_name) ||
1413 			    (errno == ERANGE)) {
1414 				pr_debug("invalid directory (%s). Skipping.\n",
1415 					 namelist[i]->d_name);
1416 				continue;
1417 			}
1418 			sprintf(path, "%s/%s/proc/kallsyms",
1419 				symbol_conf.guestmount,
1420 				namelist[i]->d_name);
1421 			ret = access(path, R_OK);
1422 			if (ret) {
1423 				pr_debug("Can't access file %s\n", path);
1424 				goto failure;
1425 			}
1426 			machines__create_kernel_maps(machines, pid);
1427 		}
1428 failure:
1429 		free(namelist);
1430 	}
1431 
1432 	return ret;
1433 }
1434 
machines__destroy_kernel_maps(struct machines * machines)1435 void machines__destroy_kernel_maps(struct machines *machines)
1436 {
1437 	struct rb_node *next = rb_first_cached(&machines->guests);
1438 
1439 	machine__destroy_kernel_maps(&machines->host);
1440 
1441 	while (next) {
1442 		struct machine *pos = rb_entry(next, struct machine, rb_node);
1443 
1444 		next = rb_next(&pos->rb_node);
1445 		rb_erase_cached(&pos->rb_node, &machines->guests);
1446 		machine__delete(pos);
1447 	}
1448 }
1449 
machines__create_kernel_maps(struct machines * machines,pid_t pid)1450 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1451 {
1452 	struct machine *machine = machines__findnew(machines, pid);
1453 
1454 	if (machine == NULL)
1455 		return -1;
1456 
1457 	return machine__create_kernel_maps(machine);
1458 }
1459 
machine__load_kallsyms(struct machine * machine,const char * filename)1460 int machine__load_kallsyms(struct machine *machine, const char *filename)
1461 {
1462 	struct map *map = machine__kernel_map(machine);
1463 	struct dso *dso = map__dso(map);
1464 	int ret = __dso__load_kallsyms(dso, filename, map, true);
1465 
1466 	if (ret > 0) {
1467 		dso__set_loaded(dso);
1468 		/*
1469 		 * Since /proc/kallsyms will have multiple sessions for the
1470 		 * kernel, with modules between them, fixup the end of all
1471 		 * sections.
1472 		 */
1473 		maps__fixup_end(machine__kernel_maps(machine));
1474 	}
1475 
1476 	return ret;
1477 }
1478 
machine__load_vmlinux_path(struct machine * machine)1479 int machine__load_vmlinux_path(struct machine *machine)
1480 {
1481 	struct map *map = machine__kernel_map(machine);
1482 	struct dso *dso = map__dso(map);
1483 	int ret = dso__load_vmlinux_path(dso, map);
1484 
1485 	if (ret > 0)
1486 		dso__set_loaded(dso);
1487 
1488 	return ret;
1489 }
1490 
get_kernel_version(const char * root_dir)1491 static char *get_kernel_version(const char *root_dir)
1492 {
1493 	char version[PATH_MAX];
1494 	FILE *file;
1495 	char *name, *tmp;
1496 	const char *prefix = "Linux version ";
1497 
1498 	sprintf(version, "%s/proc/version", root_dir);
1499 	file = fopen(version, "r");
1500 	if (!file)
1501 		return NULL;
1502 
1503 	tmp = fgets(version, sizeof(version), file);
1504 	fclose(file);
1505 	if (!tmp)
1506 		return NULL;
1507 
1508 	name = strstr(version, prefix);
1509 	if (!name)
1510 		return NULL;
1511 	name += strlen(prefix);
1512 	tmp = strchr(name, ' ');
1513 	if (tmp)
1514 		*tmp = '\0';
1515 
1516 	return strdup(name);
1517 }
1518 
is_kmod_dso(struct dso * dso)1519 static bool is_kmod_dso(struct dso *dso)
1520 {
1521 	return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1522 	       dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1523 }
1524 
maps__set_module_path(struct maps * maps,const char * path,struct kmod_path * m)1525 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1526 {
1527 	char *long_name;
1528 	struct dso *dso;
1529 	struct map *map = maps__find_by_name(maps, m->name);
1530 
1531 	if (map == NULL)
1532 		return 0;
1533 
1534 	long_name = strdup(path);
1535 	if (long_name == NULL)
1536 		return -ENOMEM;
1537 
1538 	dso = map__dso(map);
1539 	dso__set_long_name(dso, long_name, true);
1540 	dso__kernel_module_get_build_id(dso, "");
1541 
1542 	/*
1543 	 * Full name could reveal us kmod compression, so
1544 	 * we need to update the symtab_type if needed.
1545 	 */
1546 	if (m->comp && is_kmod_dso(dso)) {
1547 		dso->symtab_type++;
1548 		dso->comp = m->comp;
1549 	}
1550 
1551 	return 0;
1552 }
1553 
maps__set_modules_path_dir(struct maps * maps,const char * dir_name,int depth)1554 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1555 {
1556 	struct dirent *dent;
1557 	DIR *dir = opendir(dir_name);
1558 	int ret = 0;
1559 
1560 	if (!dir) {
1561 		pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1562 		return -1;
1563 	}
1564 
1565 	while ((dent = readdir(dir)) != NULL) {
1566 		char path[PATH_MAX];
1567 		struct stat st;
1568 
1569 		/*sshfs might return bad dent->d_type, so we have to stat*/
1570 		path__join(path, sizeof(path), dir_name, dent->d_name);
1571 		if (stat(path, &st))
1572 			continue;
1573 
1574 		if (S_ISDIR(st.st_mode)) {
1575 			if (!strcmp(dent->d_name, ".") ||
1576 			    !strcmp(dent->d_name, ".."))
1577 				continue;
1578 
1579 			/* Do not follow top-level source and build symlinks */
1580 			if (depth == 0) {
1581 				if (!strcmp(dent->d_name, "source") ||
1582 				    !strcmp(dent->d_name, "build"))
1583 					continue;
1584 			}
1585 
1586 			ret = maps__set_modules_path_dir(maps, path, depth + 1);
1587 			if (ret < 0)
1588 				goto out;
1589 		} else {
1590 			struct kmod_path m;
1591 
1592 			ret = kmod_path__parse_name(&m, dent->d_name);
1593 			if (ret)
1594 				goto out;
1595 
1596 			if (m.kmod)
1597 				ret = maps__set_module_path(maps, path, &m);
1598 
1599 			zfree(&m.name);
1600 
1601 			if (ret)
1602 				goto out;
1603 		}
1604 	}
1605 
1606 out:
1607 	closedir(dir);
1608 	return ret;
1609 }
1610 
machine__set_modules_path(struct machine * machine)1611 static int machine__set_modules_path(struct machine *machine)
1612 {
1613 	char *version;
1614 	char modules_path[PATH_MAX];
1615 
1616 	version = get_kernel_version(machine->root_dir);
1617 	if (!version)
1618 		return -1;
1619 
1620 	snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1621 		 machine->root_dir, version);
1622 	free(version);
1623 
1624 	return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1625 }
arch__fix_module_text_start(u64 * start __maybe_unused,u64 * size __maybe_unused,const char * name __maybe_unused)1626 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1627 				u64 *size __maybe_unused,
1628 				const char *name __maybe_unused)
1629 {
1630 	return 0;
1631 }
1632 
machine__create_module(void * arg,const char * name,u64 start,u64 size)1633 static int machine__create_module(void *arg, const char *name, u64 start,
1634 				  u64 size)
1635 {
1636 	struct machine *machine = arg;
1637 	struct map *map;
1638 
1639 	if (arch__fix_module_text_start(&start, &size, name) < 0)
1640 		return -1;
1641 
1642 	map = machine__addnew_module_map(machine, start, name);
1643 	if (map == NULL)
1644 		return -1;
1645 	map__set_end(map, start + size);
1646 
1647 	dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1648 	map__put(map);
1649 	return 0;
1650 }
1651 
machine__create_modules(struct machine * machine)1652 static int machine__create_modules(struct machine *machine)
1653 {
1654 	const char *modules;
1655 	char path[PATH_MAX];
1656 
1657 	if (machine__is_default_guest(machine)) {
1658 		modules = symbol_conf.default_guest_modules;
1659 	} else {
1660 		snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1661 		modules = path;
1662 	}
1663 
1664 	if (symbol__restricted_filename(modules, "/proc/modules"))
1665 		return -1;
1666 
1667 	if (modules__parse(modules, machine, machine__create_module))
1668 		return -1;
1669 
1670 	if (!machine__set_modules_path(machine))
1671 		return 0;
1672 
1673 	pr_debug("Problems setting modules path maps, continuing anyway...\n");
1674 
1675 	return 0;
1676 }
1677 
machine__set_kernel_mmap(struct machine * machine,u64 start,u64 end)1678 static void machine__set_kernel_mmap(struct machine *machine,
1679 				     u64 start, u64 end)
1680 {
1681 	map__set_start(machine->vmlinux_map, start);
1682 	map__set_end(machine->vmlinux_map, end);
1683 	/*
1684 	 * Be a bit paranoid here, some perf.data file came with
1685 	 * a zero sized synthesized MMAP event for the kernel.
1686 	 */
1687 	if (start == 0 && end == 0)
1688 		map__set_end(machine->vmlinux_map, ~0ULL);
1689 }
1690 
machine__update_kernel_mmap(struct machine * machine,u64 start,u64 end)1691 static int machine__update_kernel_mmap(struct machine *machine,
1692 				     u64 start, u64 end)
1693 {
1694 	struct map *orig, *updated;
1695 	int err;
1696 
1697 	orig = machine->vmlinux_map;
1698 	updated = map__get(orig);
1699 
1700 	machine->vmlinux_map = updated;
1701 	machine__set_kernel_mmap(machine, start, end);
1702 	maps__remove(machine__kernel_maps(machine), orig);
1703 	err = maps__insert(machine__kernel_maps(machine), updated);
1704 	map__put(orig);
1705 
1706 	return err;
1707 }
1708 
machine__create_kernel_maps(struct machine * machine)1709 int machine__create_kernel_maps(struct machine *machine)
1710 {
1711 	struct dso *kernel = machine__get_kernel(machine);
1712 	const char *name = NULL;
1713 	u64 start = 0, end = ~0ULL;
1714 	int ret;
1715 
1716 	if (kernel == NULL)
1717 		return -1;
1718 
1719 	ret = __machine__create_kernel_maps(machine, kernel);
1720 	if (ret < 0)
1721 		goto out_put;
1722 
1723 	if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1724 		if (machine__is_host(machine))
1725 			pr_debug("Problems creating module maps, "
1726 				 "continuing anyway...\n");
1727 		else
1728 			pr_debug("Problems creating module maps for guest %d, "
1729 				 "continuing anyway...\n", machine->pid);
1730 	}
1731 
1732 	if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1733 		if (name &&
1734 		    map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1735 			machine__destroy_kernel_maps(machine);
1736 			ret = -1;
1737 			goto out_put;
1738 		}
1739 
1740 		/*
1741 		 * we have a real start address now, so re-order the kmaps
1742 		 * assume it's the last in the kmaps
1743 		 */
1744 		ret = machine__update_kernel_mmap(machine, start, end);
1745 		if (ret < 0)
1746 			goto out_put;
1747 	}
1748 
1749 	if (machine__create_extra_kernel_maps(machine, kernel))
1750 		pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1751 
1752 	if (end == ~0ULL) {
1753 		/* update end address of the kernel map using adjacent module address */
1754 		struct map_rb_node *rb_node = maps__find_node(machine__kernel_maps(machine),
1755 							machine__kernel_map(machine));
1756 		struct map_rb_node *next = map_rb_node__next(rb_node);
1757 
1758 		if (next)
1759 			machine__set_kernel_mmap(machine, start, map__start(next->map));
1760 	}
1761 
1762 out_put:
1763 	dso__put(kernel);
1764 	return ret;
1765 }
1766 
machine__uses_kcore(struct machine * machine)1767 static bool machine__uses_kcore(struct machine *machine)
1768 {
1769 	struct dso *dso;
1770 
1771 	list_for_each_entry(dso, &machine->dsos.head, node) {
1772 		if (dso__is_kcore(dso))
1773 			return true;
1774 	}
1775 
1776 	return false;
1777 }
1778 
perf_event__is_extra_kernel_mmap(struct machine * machine,struct extra_kernel_map * xm)1779 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1780 					     struct extra_kernel_map *xm)
1781 {
1782 	return machine__is(machine, "x86_64") &&
1783 	       is_entry_trampoline(xm->name);
1784 }
1785 
machine__process_extra_kernel_map(struct machine * machine,struct extra_kernel_map * xm)1786 static int machine__process_extra_kernel_map(struct machine *machine,
1787 					     struct extra_kernel_map *xm)
1788 {
1789 	struct dso *kernel = machine__kernel_dso(machine);
1790 
1791 	if (kernel == NULL)
1792 		return -1;
1793 
1794 	return machine__create_extra_kernel_map(machine, kernel, xm);
1795 }
1796 
machine__process_kernel_mmap_event(struct machine * machine,struct extra_kernel_map * xm,struct build_id * bid)1797 static int machine__process_kernel_mmap_event(struct machine *machine,
1798 					      struct extra_kernel_map *xm,
1799 					      struct build_id *bid)
1800 {
1801 	enum dso_space_type dso_space;
1802 	bool is_kernel_mmap;
1803 	const char *mmap_name = machine->mmap_name;
1804 
1805 	/* If we have maps from kcore then we do not need or want any others */
1806 	if (machine__uses_kcore(machine))
1807 		return 0;
1808 
1809 	if (machine__is_host(machine))
1810 		dso_space = DSO_SPACE__KERNEL;
1811 	else
1812 		dso_space = DSO_SPACE__KERNEL_GUEST;
1813 
1814 	is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1815 	if (!is_kernel_mmap && !machine__is_host(machine)) {
1816 		/*
1817 		 * If the event was recorded inside the guest and injected into
1818 		 * the host perf.data file, then it will match a host mmap_name,
1819 		 * so try that - see machine__set_mmap_name().
1820 		 */
1821 		mmap_name = "[kernel.kallsyms]";
1822 		is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1823 	}
1824 	if (xm->name[0] == '/' ||
1825 	    (!is_kernel_mmap && xm->name[0] == '[')) {
1826 		struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1827 
1828 		if (map == NULL)
1829 			goto out_problem;
1830 
1831 		map__set_end(map, map__start(map) + xm->end - xm->start);
1832 
1833 		if (build_id__is_defined(bid))
1834 			dso__set_build_id(map__dso(map), bid);
1835 
1836 		map__put(map);
1837 	} else if (is_kernel_mmap) {
1838 		const char *symbol_name = xm->name + strlen(mmap_name);
1839 		/*
1840 		 * Should be there already, from the build-id table in
1841 		 * the header.
1842 		 */
1843 		struct dso *kernel = NULL;
1844 		struct dso *dso;
1845 
1846 		down_read(&machine->dsos.lock);
1847 
1848 		list_for_each_entry(dso, &machine->dsos.head, node) {
1849 
1850 			/*
1851 			 * The cpumode passed to is_kernel_module is not the
1852 			 * cpumode of *this* event. If we insist on passing
1853 			 * correct cpumode to is_kernel_module, we should
1854 			 * record the cpumode when we adding this dso to the
1855 			 * linked list.
1856 			 *
1857 			 * However we don't really need passing correct
1858 			 * cpumode.  We know the correct cpumode must be kernel
1859 			 * mode (if not, we should not link it onto kernel_dsos
1860 			 * list).
1861 			 *
1862 			 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1863 			 * is_kernel_module() treats it as a kernel cpumode.
1864 			 */
1865 
1866 			if (!dso->kernel ||
1867 			    is_kernel_module(dso->long_name,
1868 					     PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1869 				continue;
1870 
1871 
1872 			kernel = dso__get(dso);
1873 			break;
1874 		}
1875 
1876 		up_read(&machine->dsos.lock);
1877 
1878 		if (kernel == NULL)
1879 			kernel = machine__findnew_dso(machine, machine->mmap_name);
1880 		if (kernel == NULL)
1881 			goto out_problem;
1882 
1883 		kernel->kernel = dso_space;
1884 		if (__machine__create_kernel_maps(machine, kernel) < 0) {
1885 			dso__put(kernel);
1886 			goto out_problem;
1887 		}
1888 
1889 		if (strstr(kernel->long_name, "vmlinux"))
1890 			dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1891 
1892 		if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1893 			dso__put(kernel);
1894 			goto out_problem;
1895 		}
1896 
1897 		if (build_id__is_defined(bid))
1898 			dso__set_build_id(kernel, bid);
1899 
1900 		/*
1901 		 * Avoid using a zero address (kptr_restrict) for the ref reloc
1902 		 * symbol. Effectively having zero here means that at record
1903 		 * time /proc/sys/kernel/kptr_restrict was non zero.
1904 		 */
1905 		if (xm->pgoff != 0) {
1906 			map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1907 							symbol_name,
1908 							xm->pgoff);
1909 		}
1910 
1911 		if (machine__is_default_guest(machine)) {
1912 			/*
1913 			 * preload dso of guest kernel and modules
1914 			 */
1915 			dso__load(kernel, machine__kernel_map(machine));
1916 		}
1917 		dso__put(kernel);
1918 	} else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1919 		return machine__process_extra_kernel_map(machine, xm);
1920 	}
1921 	return 0;
1922 out_problem:
1923 	return -1;
1924 }
1925 
machine__process_mmap2_event(struct machine * machine,union perf_event * event,struct perf_sample * sample)1926 int machine__process_mmap2_event(struct machine *machine,
1927 				 union perf_event *event,
1928 				 struct perf_sample *sample)
1929 {
1930 	struct thread *thread;
1931 	struct map *map;
1932 	struct dso_id dso_id = {
1933 		.maj = event->mmap2.maj,
1934 		.min = event->mmap2.min,
1935 		.ino = event->mmap2.ino,
1936 		.ino_generation = event->mmap2.ino_generation,
1937 	};
1938 	struct build_id __bid, *bid = NULL;
1939 	int ret = 0;
1940 
1941 	if (dump_trace)
1942 		perf_event__fprintf_mmap2(event, stdout);
1943 
1944 	if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1945 		bid = &__bid;
1946 		build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1947 	}
1948 
1949 	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1950 	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1951 		struct extra_kernel_map xm = {
1952 			.start = event->mmap2.start,
1953 			.end   = event->mmap2.start + event->mmap2.len,
1954 			.pgoff = event->mmap2.pgoff,
1955 		};
1956 
1957 		strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1958 		ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1959 		if (ret < 0)
1960 			goto out_problem;
1961 		return 0;
1962 	}
1963 
1964 	thread = machine__findnew_thread(machine, event->mmap2.pid,
1965 					event->mmap2.tid);
1966 	if (thread == NULL)
1967 		goto out_problem;
1968 
1969 	map = map__new(machine, event->mmap2.start,
1970 			event->mmap2.len, event->mmap2.pgoff,
1971 			&dso_id, event->mmap2.prot,
1972 			event->mmap2.flags, bid,
1973 			event->mmap2.filename, thread);
1974 
1975 	if (map == NULL)
1976 		goto out_problem_map;
1977 
1978 	ret = thread__insert_map(thread, map);
1979 	if (ret)
1980 		goto out_problem_insert;
1981 
1982 	thread__put(thread);
1983 	map__put(map);
1984 	return 0;
1985 
1986 out_problem_insert:
1987 	map__put(map);
1988 out_problem_map:
1989 	thread__put(thread);
1990 out_problem:
1991 	dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1992 	return 0;
1993 }
1994 
machine__process_mmap_event(struct machine * machine,union perf_event * event,struct perf_sample * sample)1995 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1996 				struct perf_sample *sample)
1997 {
1998 	struct thread *thread;
1999 	struct map *map;
2000 	u32 prot = 0;
2001 	int ret = 0;
2002 
2003 	if (dump_trace)
2004 		perf_event__fprintf_mmap(event, stdout);
2005 
2006 	if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
2007 	    sample->cpumode == PERF_RECORD_MISC_KERNEL) {
2008 		struct extra_kernel_map xm = {
2009 			.start = event->mmap.start,
2010 			.end   = event->mmap.start + event->mmap.len,
2011 			.pgoff = event->mmap.pgoff,
2012 		};
2013 
2014 		strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
2015 		ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
2016 		if (ret < 0)
2017 			goto out_problem;
2018 		return 0;
2019 	}
2020 
2021 	thread = machine__findnew_thread(machine, event->mmap.pid,
2022 					 event->mmap.tid);
2023 	if (thread == NULL)
2024 		goto out_problem;
2025 
2026 	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
2027 		prot = PROT_EXEC;
2028 
2029 	map = map__new(machine, event->mmap.start,
2030 			event->mmap.len, event->mmap.pgoff,
2031 			NULL, prot, 0, NULL, event->mmap.filename, thread);
2032 
2033 	if (map == NULL)
2034 		goto out_problem_map;
2035 
2036 	ret = thread__insert_map(thread, map);
2037 	if (ret)
2038 		goto out_problem_insert;
2039 
2040 	thread__put(thread);
2041 	map__put(map);
2042 	return 0;
2043 
2044 out_problem_insert:
2045 	map__put(map);
2046 out_problem_map:
2047 	thread__put(thread);
2048 out_problem:
2049 	dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
2050 	return 0;
2051 }
2052 
__machine__remove_thread(struct machine * machine,struct thread_rb_node * nd,struct thread * th,bool lock)2053 static void __machine__remove_thread(struct machine *machine, struct thread_rb_node *nd,
2054 				     struct thread *th, bool lock)
2055 {
2056 	struct threads *threads = machine__threads(machine, thread__tid(th));
2057 
2058 	if (!nd)
2059 		nd = thread_rb_node__find(th, &threads->entries.rb_root);
2060 
2061 	if (threads->last_match && RC_CHK_ACCESS(threads->last_match) == RC_CHK_ACCESS(th))
2062 		threads__set_last_match(threads, NULL);
2063 
2064 	if (lock)
2065 		down_write(&threads->lock);
2066 
2067 	BUG_ON(refcount_read(thread__refcnt(th)) == 0);
2068 
2069 	thread__put(nd->thread);
2070 	rb_erase_cached(&nd->rb_node, &threads->entries);
2071 	RB_CLEAR_NODE(&nd->rb_node);
2072 	--threads->nr;
2073 
2074 	free(nd);
2075 
2076 	if (lock)
2077 		up_write(&threads->lock);
2078 }
2079 
machine__remove_thread(struct machine * machine,struct thread * th)2080 void machine__remove_thread(struct machine *machine, struct thread *th)
2081 {
2082 	return __machine__remove_thread(machine, NULL, th, true);
2083 }
2084 
machine__process_fork_event(struct machine * machine,union perf_event * event,struct perf_sample * sample)2085 int machine__process_fork_event(struct machine *machine, union perf_event *event,
2086 				struct perf_sample *sample)
2087 {
2088 	struct thread *thread = machine__find_thread(machine,
2089 						     event->fork.pid,
2090 						     event->fork.tid);
2091 	struct thread *parent = machine__findnew_thread(machine,
2092 							event->fork.ppid,
2093 							event->fork.ptid);
2094 	bool do_maps_clone = true;
2095 	int err = 0;
2096 
2097 	if (dump_trace)
2098 		perf_event__fprintf_task(event, stdout);
2099 
2100 	/*
2101 	 * There may be an existing thread that is not actually the parent,
2102 	 * either because we are processing events out of order, or because the
2103 	 * (fork) event that would have removed the thread was lost. Assume the
2104 	 * latter case and continue on as best we can.
2105 	 */
2106 	if (thread__pid(parent) != (pid_t)event->fork.ppid) {
2107 		dump_printf("removing erroneous parent thread %d/%d\n",
2108 			    thread__pid(parent), thread__tid(parent));
2109 		machine__remove_thread(machine, parent);
2110 		thread__put(parent);
2111 		parent = machine__findnew_thread(machine, event->fork.ppid,
2112 						 event->fork.ptid);
2113 	}
2114 
2115 	/* if a thread currently exists for the thread id remove it */
2116 	if (thread != NULL) {
2117 		machine__remove_thread(machine, thread);
2118 		thread__put(thread);
2119 	}
2120 
2121 	thread = machine__findnew_thread(machine, event->fork.pid,
2122 					 event->fork.tid);
2123 	/*
2124 	 * When synthesizing FORK events, we are trying to create thread
2125 	 * objects for the already running tasks on the machine.
2126 	 *
2127 	 * Normally, for a kernel FORK event, we want to clone the parent's
2128 	 * maps because that is what the kernel just did.
2129 	 *
2130 	 * But when synthesizing, this should not be done.  If we do, we end up
2131 	 * with overlapping maps as we process the synthesized MMAP2 events that
2132 	 * get delivered shortly thereafter.
2133 	 *
2134 	 * Use the FORK event misc flags in an internal way to signal this
2135 	 * situation, so we can elide the map clone when appropriate.
2136 	 */
2137 	if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
2138 		do_maps_clone = false;
2139 
2140 	if (thread == NULL || parent == NULL ||
2141 	    thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
2142 		dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
2143 		err = -1;
2144 	}
2145 	thread__put(thread);
2146 	thread__put(parent);
2147 
2148 	return err;
2149 }
2150 
machine__process_exit_event(struct machine * machine,union perf_event * event,struct perf_sample * sample __maybe_unused)2151 int machine__process_exit_event(struct machine *machine, union perf_event *event,
2152 				struct perf_sample *sample __maybe_unused)
2153 {
2154 	struct thread *thread = machine__find_thread(machine,
2155 						     event->fork.pid,
2156 						     event->fork.tid);
2157 
2158 	if (dump_trace)
2159 		perf_event__fprintf_task(event, stdout);
2160 
2161 	if (thread != NULL)
2162 		thread__put(thread);
2163 
2164 	return 0;
2165 }
2166 
machine__process_event(struct machine * machine,union perf_event * event,struct perf_sample * sample)2167 int machine__process_event(struct machine *machine, union perf_event *event,
2168 			   struct perf_sample *sample)
2169 {
2170 	int ret;
2171 
2172 	switch (event->header.type) {
2173 	case PERF_RECORD_COMM:
2174 		ret = machine__process_comm_event(machine, event, sample); break;
2175 	case PERF_RECORD_MMAP:
2176 		ret = machine__process_mmap_event(machine, event, sample); break;
2177 	case PERF_RECORD_NAMESPACES:
2178 		ret = machine__process_namespaces_event(machine, event, sample); break;
2179 	case PERF_RECORD_CGROUP:
2180 		ret = machine__process_cgroup_event(machine, event, sample); break;
2181 	case PERF_RECORD_MMAP2:
2182 		ret = machine__process_mmap2_event(machine, event, sample); break;
2183 	case PERF_RECORD_FORK:
2184 		ret = machine__process_fork_event(machine, event, sample); break;
2185 	case PERF_RECORD_EXIT:
2186 		ret = machine__process_exit_event(machine, event, sample); break;
2187 	case PERF_RECORD_LOST:
2188 		ret = machine__process_lost_event(machine, event, sample); break;
2189 	case PERF_RECORD_AUX:
2190 		ret = machine__process_aux_event(machine, event); break;
2191 	case PERF_RECORD_ITRACE_START:
2192 		ret = machine__process_itrace_start_event(machine, event); break;
2193 	case PERF_RECORD_LOST_SAMPLES:
2194 		ret = machine__process_lost_samples_event(machine, event, sample); break;
2195 	case PERF_RECORD_SWITCH:
2196 	case PERF_RECORD_SWITCH_CPU_WIDE:
2197 		ret = machine__process_switch_event(machine, event); break;
2198 	case PERF_RECORD_KSYMBOL:
2199 		ret = machine__process_ksymbol(machine, event, sample); break;
2200 	case PERF_RECORD_BPF_EVENT:
2201 		ret = machine__process_bpf(machine, event, sample); break;
2202 	case PERF_RECORD_TEXT_POKE:
2203 		ret = machine__process_text_poke(machine, event, sample); break;
2204 	case PERF_RECORD_AUX_OUTPUT_HW_ID:
2205 		ret = machine__process_aux_output_hw_id_event(machine, event); break;
2206 	default:
2207 		ret = -1;
2208 		break;
2209 	}
2210 
2211 	return ret;
2212 }
2213 
symbol__match_regex(struct symbol * sym,regex_t * regex)2214 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2215 {
2216 	if (!regexec(regex, sym->name, 0, NULL, 0))
2217 		return true;
2218 	return false;
2219 }
2220 
ip__resolve_ams(struct thread * thread,struct addr_map_symbol * ams,u64 ip)2221 static void ip__resolve_ams(struct thread *thread,
2222 			    struct addr_map_symbol *ams,
2223 			    u64 ip)
2224 {
2225 	struct addr_location al;
2226 
2227 	addr_location__init(&al);
2228 	/*
2229 	 * We cannot use the header.misc hint to determine whether a
2230 	 * branch stack address is user, kernel, guest, hypervisor.
2231 	 * Branches may straddle the kernel/user/hypervisor boundaries.
2232 	 * Thus, we have to try consecutively until we find a match
2233 	 * or else, the symbol is unknown
2234 	 */
2235 	thread__find_cpumode_addr_location(thread, ip, &al);
2236 
2237 	ams->addr = ip;
2238 	ams->al_addr = al.addr;
2239 	ams->al_level = al.level;
2240 	ams->ms.maps = maps__get(al.maps);
2241 	ams->ms.sym = al.sym;
2242 	ams->ms.map = map__get(al.map);
2243 	ams->phys_addr = 0;
2244 	ams->data_page_size = 0;
2245 	addr_location__exit(&al);
2246 }
2247 
ip__resolve_data(struct thread * thread,u8 m,struct addr_map_symbol * ams,u64 addr,u64 phys_addr,u64 daddr_page_size)2248 static void ip__resolve_data(struct thread *thread,
2249 			     u8 m, struct addr_map_symbol *ams,
2250 			     u64 addr, u64 phys_addr, u64 daddr_page_size)
2251 {
2252 	struct addr_location al;
2253 
2254 	addr_location__init(&al);
2255 
2256 	thread__find_symbol(thread, m, addr, &al);
2257 
2258 	ams->addr = addr;
2259 	ams->al_addr = al.addr;
2260 	ams->al_level = al.level;
2261 	ams->ms.maps = maps__get(al.maps);
2262 	ams->ms.sym = al.sym;
2263 	ams->ms.map = map__get(al.map);
2264 	ams->phys_addr = phys_addr;
2265 	ams->data_page_size = daddr_page_size;
2266 	addr_location__exit(&al);
2267 }
2268 
sample__resolve_mem(struct perf_sample * sample,struct addr_location * al)2269 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2270 				     struct addr_location *al)
2271 {
2272 	struct mem_info *mi = mem_info__new();
2273 
2274 	if (!mi)
2275 		return NULL;
2276 
2277 	ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2278 	ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2279 			 sample->addr, sample->phys_addr,
2280 			 sample->data_page_size);
2281 	mi->data_src.val = sample->data_src;
2282 
2283 	return mi;
2284 }
2285 
callchain_srcline(struct map_symbol * ms,u64 ip)2286 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2287 {
2288 	struct map *map = ms->map;
2289 	char *srcline = NULL;
2290 	struct dso *dso;
2291 
2292 	if (!map || callchain_param.key == CCKEY_FUNCTION)
2293 		return srcline;
2294 
2295 	dso = map__dso(map);
2296 	srcline = srcline__tree_find(&dso->srclines, ip);
2297 	if (!srcline) {
2298 		bool show_sym = false;
2299 		bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2300 
2301 		srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2302 				      ms->sym, show_sym, show_addr, ip);
2303 		srcline__tree_insert(&dso->srclines, ip, srcline);
2304 	}
2305 
2306 	return srcline;
2307 }
2308 
2309 struct iterations {
2310 	int nr_loop_iter;
2311 	u64 cycles;
2312 };
2313 
add_callchain_ip(struct thread * thread,struct callchain_cursor * cursor,struct symbol ** parent,struct addr_location * root_al,u8 * cpumode,u64 ip,bool branch,struct branch_flags * flags,struct iterations * iter,u64 branch_from)2314 static int add_callchain_ip(struct thread *thread,
2315 			    struct callchain_cursor *cursor,
2316 			    struct symbol **parent,
2317 			    struct addr_location *root_al,
2318 			    u8 *cpumode,
2319 			    u64 ip,
2320 			    bool branch,
2321 			    struct branch_flags *flags,
2322 			    struct iterations *iter,
2323 			    u64 branch_from)
2324 {
2325 	struct map_symbol ms = {};
2326 	struct addr_location al;
2327 	int nr_loop_iter = 0, err = 0;
2328 	u64 iter_cycles = 0;
2329 	const char *srcline = NULL;
2330 
2331 	addr_location__init(&al);
2332 	al.filtered = 0;
2333 	al.sym = NULL;
2334 	al.srcline = NULL;
2335 	if (!cpumode) {
2336 		thread__find_cpumode_addr_location(thread, ip, &al);
2337 	} else {
2338 		if (ip >= PERF_CONTEXT_MAX) {
2339 			switch (ip) {
2340 			case PERF_CONTEXT_HV:
2341 				*cpumode = PERF_RECORD_MISC_HYPERVISOR;
2342 				break;
2343 			case PERF_CONTEXT_KERNEL:
2344 				*cpumode = PERF_RECORD_MISC_KERNEL;
2345 				break;
2346 			case PERF_CONTEXT_USER:
2347 				*cpumode = PERF_RECORD_MISC_USER;
2348 				break;
2349 			default:
2350 				pr_debug("invalid callchain context: "
2351 					 "%"PRId64"\n", (s64) ip);
2352 				/*
2353 				 * It seems the callchain is corrupted.
2354 				 * Discard all.
2355 				 */
2356 				callchain_cursor_reset(cursor);
2357 				err = 1;
2358 				goto out;
2359 			}
2360 			goto out;
2361 		}
2362 		thread__find_symbol(thread, *cpumode, ip, &al);
2363 	}
2364 
2365 	if (al.sym != NULL) {
2366 		if (perf_hpp_list.parent && !*parent &&
2367 		    symbol__match_regex(al.sym, &parent_regex))
2368 			*parent = al.sym;
2369 		else if (have_ignore_callees && root_al &&
2370 		  symbol__match_regex(al.sym, &ignore_callees_regex)) {
2371 			/* Treat this symbol as the root,
2372 			   forgetting its callees. */
2373 			addr_location__copy(root_al, &al);
2374 			callchain_cursor_reset(cursor);
2375 		}
2376 	}
2377 
2378 	if (symbol_conf.hide_unresolved && al.sym == NULL)
2379 		goto out;
2380 
2381 	if (iter) {
2382 		nr_loop_iter = iter->nr_loop_iter;
2383 		iter_cycles = iter->cycles;
2384 	}
2385 
2386 	ms.maps = maps__get(al.maps);
2387 	ms.map = map__get(al.map);
2388 	ms.sym = al.sym;
2389 	srcline = callchain_srcline(&ms, al.addr);
2390 	err = callchain_cursor_append(cursor, ip, &ms,
2391 				      branch, flags, nr_loop_iter,
2392 				      iter_cycles, branch_from, srcline);
2393 out:
2394 	addr_location__exit(&al);
2395 	maps__put(ms.maps);
2396 	map__put(ms.map);
2397 	return err;
2398 }
2399 
sample__resolve_bstack(struct perf_sample * sample,struct addr_location * al)2400 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2401 					   struct addr_location *al)
2402 {
2403 	unsigned int i;
2404 	const struct branch_stack *bs = sample->branch_stack;
2405 	struct branch_entry *entries = perf_sample__branch_entries(sample);
2406 	struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2407 
2408 	if (!bi)
2409 		return NULL;
2410 
2411 	for (i = 0; i < bs->nr; i++) {
2412 		ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2413 		ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2414 		bi[i].flags = entries[i].flags;
2415 	}
2416 	return bi;
2417 }
2418 
save_iterations(struct iterations * iter,struct branch_entry * be,int nr)2419 static void save_iterations(struct iterations *iter,
2420 			    struct branch_entry *be, int nr)
2421 {
2422 	int i;
2423 
2424 	iter->nr_loop_iter++;
2425 	iter->cycles = 0;
2426 
2427 	for (i = 0; i < nr; i++)
2428 		iter->cycles += be[i].flags.cycles;
2429 }
2430 
2431 #define CHASHSZ 127
2432 #define CHASHBITS 7
2433 #define NO_ENTRY 0xff
2434 
2435 #define PERF_MAX_BRANCH_DEPTH 127
2436 
2437 /* Remove loops. */
remove_loops(struct branch_entry * l,int nr,struct iterations * iter)2438 static int remove_loops(struct branch_entry *l, int nr,
2439 			struct iterations *iter)
2440 {
2441 	int i, j, off;
2442 	unsigned char chash[CHASHSZ];
2443 
2444 	memset(chash, NO_ENTRY, sizeof(chash));
2445 
2446 	BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2447 
2448 	for (i = 0; i < nr; i++) {
2449 		int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2450 
2451 		/* no collision handling for now */
2452 		if (chash[h] == NO_ENTRY) {
2453 			chash[h] = i;
2454 		} else if (l[chash[h]].from == l[i].from) {
2455 			bool is_loop = true;
2456 			/* check if it is a real loop */
2457 			off = 0;
2458 			for (j = chash[h]; j < i && i + off < nr; j++, off++)
2459 				if (l[j].from != l[i + off].from) {
2460 					is_loop = false;
2461 					break;
2462 				}
2463 			if (is_loop) {
2464 				j = nr - (i + off);
2465 				if (j > 0) {
2466 					save_iterations(iter + i + off,
2467 						l + i, off);
2468 
2469 					memmove(iter + i, iter + i + off,
2470 						j * sizeof(*iter));
2471 
2472 					memmove(l + i, l + i + off,
2473 						j * sizeof(*l));
2474 				}
2475 
2476 				nr -= off;
2477 			}
2478 		}
2479 	}
2480 	return nr;
2481 }
2482 
lbr_callchain_add_kernel_ip(struct thread * thread,struct callchain_cursor * cursor,struct perf_sample * sample,struct symbol ** parent,struct addr_location * root_al,u64 branch_from,bool callee,int end)2483 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2484 				       struct callchain_cursor *cursor,
2485 				       struct perf_sample *sample,
2486 				       struct symbol **parent,
2487 				       struct addr_location *root_al,
2488 				       u64 branch_from,
2489 				       bool callee, int end)
2490 {
2491 	struct ip_callchain *chain = sample->callchain;
2492 	u8 cpumode = PERF_RECORD_MISC_USER;
2493 	int err, i;
2494 
2495 	if (callee) {
2496 		for (i = 0; i < end + 1; i++) {
2497 			err = add_callchain_ip(thread, cursor, parent,
2498 					       root_al, &cpumode, chain->ips[i],
2499 					       false, NULL, NULL, branch_from);
2500 			if (err)
2501 				return err;
2502 		}
2503 		return 0;
2504 	}
2505 
2506 	for (i = end; i >= 0; i--) {
2507 		err = add_callchain_ip(thread, cursor, parent,
2508 				       root_al, &cpumode, chain->ips[i],
2509 				       false, NULL, NULL, branch_from);
2510 		if (err)
2511 			return err;
2512 	}
2513 
2514 	return 0;
2515 }
2516 
save_lbr_cursor_node(struct thread * thread,struct callchain_cursor * cursor,int idx)2517 static void save_lbr_cursor_node(struct thread *thread,
2518 				 struct callchain_cursor *cursor,
2519 				 int idx)
2520 {
2521 	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2522 
2523 	if (!lbr_stitch)
2524 		return;
2525 
2526 	if (cursor->pos == cursor->nr) {
2527 		lbr_stitch->prev_lbr_cursor[idx].valid = false;
2528 		return;
2529 	}
2530 
2531 	if (!cursor->curr)
2532 		cursor->curr = cursor->first;
2533 	else
2534 		cursor->curr = cursor->curr->next;
2535 	memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2536 	       sizeof(struct callchain_cursor_node));
2537 
2538 	lbr_stitch->prev_lbr_cursor[idx].valid = true;
2539 	cursor->pos++;
2540 }
2541 
lbr_callchain_add_lbr_ip(struct thread * thread,struct callchain_cursor * cursor,struct perf_sample * sample,struct symbol ** parent,struct addr_location * root_al,u64 * branch_from,bool callee)2542 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2543 				    struct callchain_cursor *cursor,
2544 				    struct perf_sample *sample,
2545 				    struct symbol **parent,
2546 				    struct addr_location *root_al,
2547 				    u64 *branch_from,
2548 				    bool callee)
2549 {
2550 	struct branch_stack *lbr_stack = sample->branch_stack;
2551 	struct branch_entry *entries = perf_sample__branch_entries(sample);
2552 	u8 cpumode = PERF_RECORD_MISC_USER;
2553 	int lbr_nr = lbr_stack->nr;
2554 	struct branch_flags *flags;
2555 	int err, i;
2556 	u64 ip;
2557 
2558 	/*
2559 	 * The curr and pos are not used in writing session. They are cleared
2560 	 * in callchain_cursor_commit() when the writing session is closed.
2561 	 * Using curr and pos to track the current cursor node.
2562 	 */
2563 	if (thread__lbr_stitch(thread)) {
2564 		cursor->curr = NULL;
2565 		cursor->pos = cursor->nr;
2566 		if (cursor->nr) {
2567 			cursor->curr = cursor->first;
2568 			for (i = 0; i < (int)(cursor->nr - 1); i++)
2569 				cursor->curr = cursor->curr->next;
2570 		}
2571 	}
2572 
2573 	if (callee) {
2574 		/* Add LBR ip from first entries.to */
2575 		ip = entries[0].to;
2576 		flags = &entries[0].flags;
2577 		*branch_from = entries[0].from;
2578 		err = add_callchain_ip(thread, cursor, parent,
2579 				       root_al, &cpumode, ip,
2580 				       true, flags, NULL,
2581 				       *branch_from);
2582 		if (err)
2583 			return err;
2584 
2585 		/*
2586 		 * The number of cursor node increases.
2587 		 * Move the current cursor node.
2588 		 * But does not need to save current cursor node for entry 0.
2589 		 * It's impossible to stitch the whole LBRs of previous sample.
2590 		 */
2591 		if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2592 			if (!cursor->curr)
2593 				cursor->curr = cursor->first;
2594 			else
2595 				cursor->curr = cursor->curr->next;
2596 			cursor->pos++;
2597 		}
2598 
2599 		/* Add LBR ip from entries.from one by one. */
2600 		for (i = 0; i < lbr_nr; i++) {
2601 			ip = entries[i].from;
2602 			flags = &entries[i].flags;
2603 			err = add_callchain_ip(thread, cursor, parent,
2604 					       root_al, &cpumode, ip,
2605 					       true, flags, NULL,
2606 					       *branch_from);
2607 			if (err)
2608 				return err;
2609 			save_lbr_cursor_node(thread, cursor, i);
2610 		}
2611 		return 0;
2612 	}
2613 
2614 	/* Add LBR ip from entries.from one by one. */
2615 	for (i = lbr_nr - 1; i >= 0; i--) {
2616 		ip = entries[i].from;
2617 		flags = &entries[i].flags;
2618 		err = add_callchain_ip(thread, cursor, parent,
2619 				       root_al, &cpumode, ip,
2620 				       true, flags, NULL,
2621 				       *branch_from);
2622 		if (err)
2623 			return err;
2624 		save_lbr_cursor_node(thread, cursor, i);
2625 	}
2626 
2627 	if (lbr_nr > 0) {
2628 		/* Add LBR ip from first entries.to */
2629 		ip = entries[0].to;
2630 		flags = &entries[0].flags;
2631 		*branch_from = entries[0].from;
2632 		err = add_callchain_ip(thread, cursor, parent,
2633 				root_al, &cpumode, ip,
2634 				true, flags, NULL,
2635 				*branch_from);
2636 		if (err)
2637 			return err;
2638 	}
2639 
2640 	return 0;
2641 }
2642 
lbr_callchain_add_stitched_lbr_ip(struct thread * thread,struct callchain_cursor * cursor)2643 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2644 					     struct callchain_cursor *cursor)
2645 {
2646 	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2647 	struct callchain_cursor_node *cnode;
2648 	struct stitch_list *stitch_node;
2649 	int err;
2650 
2651 	list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2652 		cnode = &stitch_node->cursor;
2653 
2654 		err = callchain_cursor_append(cursor, cnode->ip,
2655 					      &cnode->ms,
2656 					      cnode->branch,
2657 					      &cnode->branch_flags,
2658 					      cnode->nr_loop_iter,
2659 					      cnode->iter_cycles,
2660 					      cnode->branch_from,
2661 					      cnode->srcline);
2662 		if (err)
2663 			return err;
2664 	}
2665 	return 0;
2666 }
2667 
get_stitch_node(struct thread * thread)2668 static struct stitch_list *get_stitch_node(struct thread *thread)
2669 {
2670 	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2671 	struct stitch_list *stitch_node;
2672 
2673 	if (!list_empty(&lbr_stitch->free_lists)) {
2674 		stitch_node = list_first_entry(&lbr_stitch->free_lists,
2675 					       struct stitch_list, node);
2676 		list_del(&stitch_node->node);
2677 
2678 		return stitch_node;
2679 	}
2680 
2681 	return malloc(sizeof(struct stitch_list));
2682 }
2683 
has_stitched_lbr(struct thread * thread,struct perf_sample * cur,struct perf_sample * prev,unsigned int max_lbr,bool callee)2684 static bool has_stitched_lbr(struct thread *thread,
2685 			     struct perf_sample *cur,
2686 			     struct perf_sample *prev,
2687 			     unsigned int max_lbr,
2688 			     bool callee)
2689 {
2690 	struct branch_stack *cur_stack = cur->branch_stack;
2691 	struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2692 	struct branch_stack *prev_stack = prev->branch_stack;
2693 	struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2694 	struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2695 	int i, j, nr_identical_branches = 0;
2696 	struct stitch_list *stitch_node;
2697 	u64 cur_base, distance;
2698 
2699 	if (!cur_stack || !prev_stack)
2700 		return false;
2701 
2702 	/* Find the physical index of the base-of-stack for current sample. */
2703 	cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2704 
2705 	distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2706 						     (max_lbr + prev_stack->hw_idx - cur_base);
2707 	/* Previous sample has shorter stack. Nothing can be stitched. */
2708 	if (distance + 1 > prev_stack->nr)
2709 		return false;
2710 
2711 	/*
2712 	 * Check if there are identical LBRs between two samples.
2713 	 * Identical LBRs must have same from, to and flags values. Also,
2714 	 * they have to be saved in the same LBR registers (same physical
2715 	 * index).
2716 	 *
2717 	 * Starts from the base-of-stack of current sample.
2718 	 */
2719 	for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2720 		if ((prev_entries[i].from != cur_entries[j].from) ||
2721 		    (prev_entries[i].to != cur_entries[j].to) ||
2722 		    (prev_entries[i].flags.value != cur_entries[j].flags.value))
2723 			break;
2724 		nr_identical_branches++;
2725 	}
2726 
2727 	if (!nr_identical_branches)
2728 		return false;
2729 
2730 	/*
2731 	 * Save the LBRs between the base-of-stack of previous sample
2732 	 * and the base-of-stack of current sample into lbr_stitch->lists.
2733 	 * These LBRs will be stitched later.
2734 	 */
2735 	for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2736 
2737 		if (!lbr_stitch->prev_lbr_cursor[i].valid)
2738 			continue;
2739 
2740 		stitch_node = get_stitch_node(thread);
2741 		if (!stitch_node)
2742 			return false;
2743 
2744 		memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2745 		       sizeof(struct callchain_cursor_node));
2746 
2747 		if (callee)
2748 			list_add(&stitch_node->node, &lbr_stitch->lists);
2749 		else
2750 			list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2751 	}
2752 
2753 	return true;
2754 }
2755 
alloc_lbr_stitch(struct thread * thread,unsigned int max_lbr)2756 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2757 {
2758 	if (thread__lbr_stitch(thread))
2759 		return true;
2760 
2761 	thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2762 	if (!thread__lbr_stitch(thread))
2763 		goto err;
2764 
2765 	thread__lbr_stitch(thread)->prev_lbr_cursor =
2766 		calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2767 	if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2768 		goto free_lbr_stitch;
2769 
2770 	INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2771 	INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2772 
2773 	return true;
2774 
2775 free_lbr_stitch:
2776 	free(thread__lbr_stitch(thread));
2777 	thread__set_lbr_stitch(thread, NULL);
2778 err:
2779 	pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2780 	thread__set_lbr_stitch_enable(thread, false);
2781 	return false;
2782 }
2783 
2784 /*
2785  * Resolve LBR callstack chain sample
2786  * Return:
2787  * 1 on success get LBR callchain information
2788  * 0 no available LBR callchain information, should try fp
2789  * negative error code on other errors.
2790  */
resolve_lbr_callchain_sample(struct thread * thread,struct callchain_cursor * cursor,struct perf_sample * sample,struct symbol ** parent,struct addr_location * root_al,int max_stack,unsigned int max_lbr)2791 static int resolve_lbr_callchain_sample(struct thread *thread,
2792 					struct callchain_cursor *cursor,
2793 					struct perf_sample *sample,
2794 					struct symbol **parent,
2795 					struct addr_location *root_al,
2796 					int max_stack,
2797 					unsigned int max_lbr)
2798 {
2799 	bool callee = (callchain_param.order == ORDER_CALLEE);
2800 	struct ip_callchain *chain = sample->callchain;
2801 	int chain_nr = min(max_stack, (int)chain->nr), i;
2802 	struct lbr_stitch *lbr_stitch;
2803 	bool stitched_lbr = false;
2804 	u64 branch_from = 0;
2805 	int err;
2806 
2807 	for (i = 0; i < chain_nr; i++) {
2808 		if (chain->ips[i] == PERF_CONTEXT_USER)
2809 			break;
2810 	}
2811 
2812 	/* LBR only affects the user callchain */
2813 	if (i == chain_nr)
2814 		return 0;
2815 
2816 	if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2817 	    (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2818 		lbr_stitch = thread__lbr_stitch(thread);
2819 
2820 		stitched_lbr = has_stitched_lbr(thread, sample,
2821 						&lbr_stitch->prev_sample,
2822 						max_lbr, callee);
2823 
2824 		if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2825 			list_replace_init(&lbr_stitch->lists,
2826 					  &lbr_stitch->free_lists);
2827 		}
2828 		memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2829 	}
2830 
2831 	if (callee) {
2832 		/* Add kernel ip */
2833 		err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2834 						  parent, root_al, branch_from,
2835 						  true, i);
2836 		if (err)
2837 			goto error;
2838 
2839 		err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2840 					       root_al, &branch_from, true);
2841 		if (err)
2842 			goto error;
2843 
2844 		if (stitched_lbr) {
2845 			err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2846 			if (err)
2847 				goto error;
2848 		}
2849 
2850 	} else {
2851 		if (stitched_lbr) {
2852 			err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2853 			if (err)
2854 				goto error;
2855 		}
2856 		err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2857 					       root_al, &branch_from, false);
2858 		if (err)
2859 			goto error;
2860 
2861 		/* Add kernel ip */
2862 		err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2863 						  parent, root_al, branch_from,
2864 						  false, i);
2865 		if (err)
2866 			goto error;
2867 	}
2868 	return 1;
2869 
2870 error:
2871 	return (err < 0) ? err : 0;
2872 }
2873 
find_prev_cpumode(struct ip_callchain * chain,struct thread * thread,struct callchain_cursor * cursor,struct symbol ** parent,struct addr_location * root_al,u8 * cpumode,int ent)2874 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2875 			     struct callchain_cursor *cursor,
2876 			     struct symbol **parent,
2877 			     struct addr_location *root_al,
2878 			     u8 *cpumode, int ent)
2879 {
2880 	int err = 0;
2881 
2882 	while (--ent >= 0) {
2883 		u64 ip = chain->ips[ent];
2884 
2885 		if (ip >= PERF_CONTEXT_MAX) {
2886 			err = add_callchain_ip(thread, cursor, parent,
2887 					       root_al, cpumode, ip,
2888 					       false, NULL, NULL, 0);
2889 			break;
2890 		}
2891 	}
2892 	return err;
2893 }
2894 
get_leaf_frame_caller(struct perf_sample * sample,struct thread * thread,int usr_idx)2895 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2896 		struct thread *thread, int usr_idx)
2897 {
2898 	if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2899 		return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2900 	else
2901 		return 0;
2902 }
2903 
thread__resolve_callchain_sample(struct thread * thread,struct callchain_cursor * cursor,struct evsel * evsel,struct perf_sample * sample,struct symbol ** parent,struct addr_location * root_al,int max_stack)2904 static int thread__resolve_callchain_sample(struct thread *thread,
2905 					    struct callchain_cursor *cursor,
2906 					    struct evsel *evsel,
2907 					    struct perf_sample *sample,
2908 					    struct symbol **parent,
2909 					    struct addr_location *root_al,
2910 					    int max_stack)
2911 {
2912 	struct branch_stack *branch = sample->branch_stack;
2913 	struct branch_entry *entries = perf_sample__branch_entries(sample);
2914 	struct ip_callchain *chain = sample->callchain;
2915 	int chain_nr = 0;
2916 	u8 cpumode = PERF_RECORD_MISC_USER;
2917 	int i, j, err, nr_entries, usr_idx;
2918 	int skip_idx = -1;
2919 	int first_call = 0;
2920 	u64 leaf_frame_caller;
2921 
2922 	if (chain)
2923 		chain_nr = chain->nr;
2924 
2925 	if (evsel__has_branch_callstack(evsel)) {
2926 		struct perf_env *env = evsel__env(evsel);
2927 
2928 		err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2929 						   root_al, max_stack,
2930 						   !env ? 0 : env->max_branches);
2931 		if (err)
2932 			return (err < 0) ? err : 0;
2933 	}
2934 
2935 	/*
2936 	 * Based on DWARF debug information, some architectures skip
2937 	 * a callchain entry saved by the kernel.
2938 	 */
2939 	skip_idx = arch_skip_callchain_idx(thread, chain);
2940 
2941 	/*
2942 	 * Add branches to call stack for easier browsing. This gives
2943 	 * more context for a sample than just the callers.
2944 	 *
2945 	 * This uses individual histograms of paths compared to the
2946 	 * aggregated histograms the normal LBR mode uses.
2947 	 *
2948 	 * Limitations for now:
2949 	 * - No extra filters
2950 	 * - No annotations (should annotate somehow)
2951 	 */
2952 
2953 	if (branch && callchain_param.branch_callstack) {
2954 		int nr = min(max_stack, (int)branch->nr);
2955 		struct branch_entry be[nr];
2956 		struct iterations iter[nr];
2957 
2958 		if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2959 			pr_warning("corrupted branch chain. skipping...\n");
2960 			goto check_calls;
2961 		}
2962 
2963 		for (i = 0; i < nr; i++) {
2964 			if (callchain_param.order == ORDER_CALLEE) {
2965 				be[i] = entries[i];
2966 
2967 				if (chain == NULL)
2968 					continue;
2969 
2970 				/*
2971 				 * Check for overlap into the callchain.
2972 				 * The return address is one off compared to
2973 				 * the branch entry. To adjust for this
2974 				 * assume the calling instruction is not longer
2975 				 * than 8 bytes.
2976 				 */
2977 				if (i == skip_idx ||
2978 				    chain->ips[first_call] >= PERF_CONTEXT_MAX)
2979 					first_call++;
2980 				else if (be[i].from < chain->ips[first_call] &&
2981 				    be[i].from >= chain->ips[first_call] - 8)
2982 					first_call++;
2983 			} else
2984 				be[i] = entries[branch->nr - i - 1];
2985 		}
2986 
2987 		memset(iter, 0, sizeof(struct iterations) * nr);
2988 		nr = remove_loops(be, nr, iter);
2989 
2990 		for (i = 0; i < nr; i++) {
2991 			err = add_callchain_ip(thread, cursor, parent,
2992 					       root_al,
2993 					       NULL, be[i].to,
2994 					       true, &be[i].flags,
2995 					       NULL, be[i].from);
2996 
2997 			if (!err)
2998 				err = add_callchain_ip(thread, cursor, parent, root_al,
2999 						       NULL, be[i].from,
3000 						       true, &be[i].flags,
3001 						       &iter[i], 0);
3002 			if (err == -EINVAL)
3003 				break;
3004 			if (err)
3005 				return err;
3006 		}
3007 
3008 		if (chain_nr == 0)
3009 			return 0;
3010 
3011 		chain_nr -= nr;
3012 	}
3013 
3014 check_calls:
3015 	if (chain && callchain_param.order != ORDER_CALLEE) {
3016 		err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
3017 					&cpumode, chain->nr - first_call);
3018 		if (err)
3019 			return (err < 0) ? err : 0;
3020 	}
3021 	for (i = first_call, nr_entries = 0;
3022 	     i < chain_nr && nr_entries < max_stack; i++) {
3023 		u64 ip;
3024 
3025 		if (callchain_param.order == ORDER_CALLEE)
3026 			j = i;
3027 		else
3028 			j = chain->nr - i - 1;
3029 
3030 #ifdef HAVE_SKIP_CALLCHAIN_IDX
3031 		if (j == skip_idx)
3032 			continue;
3033 #endif
3034 		ip = chain->ips[j];
3035 		if (ip < PERF_CONTEXT_MAX)
3036                        ++nr_entries;
3037 		else if (callchain_param.order != ORDER_CALLEE) {
3038 			err = find_prev_cpumode(chain, thread, cursor, parent,
3039 						root_al, &cpumode, j);
3040 			if (err)
3041 				return (err < 0) ? err : 0;
3042 			continue;
3043 		}
3044 
3045 		/*
3046 		 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
3047 		 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
3048 		 * the index will be different in order to add the missing frame
3049 		 * at the right place.
3050 		 */
3051 
3052 		usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
3053 
3054 		if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
3055 
3056 			leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
3057 
3058 			/*
3059 			 * check if leaf_frame_Caller != ip to not add the same
3060 			 * value twice.
3061 			 */
3062 
3063 			if (leaf_frame_caller && leaf_frame_caller != ip) {
3064 
3065 				err = add_callchain_ip(thread, cursor, parent,
3066 					       root_al, &cpumode, leaf_frame_caller,
3067 					       false, NULL, NULL, 0);
3068 				if (err)
3069 					return (err < 0) ? err : 0;
3070 			}
3071 		}
3072 
3073 		err = add_callchain_ip(thread, cursor, parent,
3074 				       root_al, &cpumode, ip,
3075 				       false, NULL, NULL, 0);
3076 
3077 		if (err)
3078 			return (err < 0) ? err : 0;
3079 	}
3080 
3081 	return 0;
3082 }
3083 
append_inlines(struct callchain_cursor * cursor,struct map_symbol * ms,u64 ip)3084 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
3085 {
3086 	struct symbol *sym = ms->sym;
3087 	struct map *map = ms->map;
3088 	struct inline_node *inline_node;
3089 	struct inline_list *ilist;
3090 	struct dso *dso;
3091 	u64 addr;
3092 	int ret = 1;
3093 	struct map_symbol ilist_ms;
3094 
3095 	if (!symbol_conf.inline_name || !map || !sym)
3096 		return ret;
3097 
3098 	addr = map__dso_map_ip(map, ip);
3099 	addr = map__rip_2objdump(map, addr);
3100 	dso = map__dso(map);
3101 
3102 	inline_node = inlines__tree_find(&dso->inlined_nodes, addr);
3103 	if (!inline_node) {
3104 		inline_node = dso__parse_addr_inlines(dso, addr, sym);
3105 		if (!inline_node)
3106 			return ret;
3107 		inlines__tree_insert(&dso->inlined_nodes, inline_node);
3108 	}
3109 
3110 	ilist_ms = (struct map_symbol) {
3111 		.maps = maps__get(ms->maps),
3112 		.map = map__get(map),
3113 	};
3114 	list_for_each_entry(ilist, &inline_node->val, list) {
3115 		ilist_ms.sym = ilist->symbol;
3116 		ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
3117 					      NULL, 0, 0, 0, ilist->srcline);
3118 
3119 		if (ret != 0)
3120 			return ret;
3121 	}
3122 	map__put(ilist_ms.map);
3123 	maps__put(ilist_ms.maps);
3124 
3125 	return ret;
3126 }
3127 
unwind_entry(struct unwind_entry * entry,void * arg)3128 static int unwind_entry(struct unwind_entry *entry, void *arg)
3129 {
3130 	struct callchain_cursor *cursor = arg;
3131 	const char *srcline = NULL;
3132 	u64 addr = entry->ip;
3133 
3134 	if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
3135 		return 0;
3136 
3137 	if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
3138 		return 0;
3139 
3140 	/*
3141 	 * Convert entry->ip from a virtual address to an offset in
3142 	 * its corresponding binary.
3143 	 */
3144 	if (entry->ms.map)
3145 		addr = map__dso_map_ip(entry->ms.map, entry->ip);
3146 
3147 	srcline = callchain_srcline(&entry->ms, addr);
3148 	return callchain_cursor_append(cursor, entry->ip, &entry->ms,
3149 				       false, NULL, 0, 0, 0, srcline);
3150 }
3151 
thread__resolve_callchain_unwind(struct thread * thread,struct callchain_cursor * cursor,struct evsel * evsel,struct perf_sample * sample,int max_stack)3152 static int thread__resolve_callchain_unwind(struct thread *thread,
3153 					    struct callchain_cursor *cursor,
3154 					    struct evsel *evsel,
3155 					    struct perf_sample *sample,
3156 					    int max_stack)
3157 {
3158 	/* Can we do dwarf post unwind? */
3159 	if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
3160 	      (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
3161 		return 0;
3162 
3163 	/* Bail out if nothing was captured. */
3164 	if ((!sample->user_regs.regs) ||
3165 	    (!sample->user_stack.size))
3166 		return 0;
3167 
3168 	return unwind__get_entries(unwind_entry, cursor,
3169 				   thread, sample, max_stack, false);
3170 }
3171 
thread__resolve_callchain(struct thread * thread,struct callchain_cursor * cursor,struct evsel * evsel,struct perf_sample * sample,struct symbol ** parent,struct addr_location * root_al,int max_stack)3172 int thread__resolve_callchain(struct thread *thread,
3173 			      struct callchain_cursor *cursor,
3174 			      struct evsel *evsel,
3175 			      struct perf_sample *sample,
3176 			      struct symbol **parent,
3177 			      struct addr_location *root_al,
3178 			      int max_stack)
3179 {
3180 	int ret = 0;
3181 
3182 	if (cursor == NULL)
3183 		return -ENOMEM;
3184 
3185 	callchain_cursor_reset(cursor);
3186 
3187 	if (callchain_param.order == ORDER_CALLEE) {
3188 		ret = thread__resolve_callchain_sample(thread, cursor,
3189 						       evsel, sample,
3190 						       parent, root_al,
3191 						       max_stack);
3192 		if (ret)
3193 			return ret;
3194 		ret = thread__resolve_callchain_unwind(thread, cursor,
3195 						       evsel, sample,
3196 						       max_stack);
3197 	} else {
3198 		ret = thread__resolve_callchain_unwind(thread, cursor,
3199 						       evsel, sample,
3200 						       max_stack);
3201 		if (ret)
3202 			return ret;
3203 		ret = thread__resolve_callchain_sample(thread, cursor,
3204 						       evsel, sample,
3205 						       parent, root_al,
3206 						       max_stack);
3207 	}
3208 
3209 	return ret;
3210 }
3211 
machine__for_each_thread(struct machine * machine,int (* fn)(struct thread * thread,void * p),void * priv)3212 int machine__for_each_thread(struct machine *machine,
3213 			     int (*fn)(struct thread *thread, void *p),
3214 			     void *priv)
3215 {
3216 	struct threads *threads;
3217 	struct rb_node *nd;
3218 	int rc = 0;
3219 	int i;
3220 
3221 	for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3222 		threads = &machine->threads[i];
3223 		for (nd = rb_first_cached(&threads->entries); nd;
3224 		     nd = rb_next(nd)) {
3225 			struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
3226 
3227 			rc = fn(trb->thread, priv);
3228 			if (rc != 0)
3229 				return rc;
3230 		}
3231 	}
3232 	return rc;
3233 }
3234 
machines__for_each_thread(struct machines * machines,int (* fn)(struct thread * thread,void * p),void * priv)3235 int machines__for_each_thread(struct machines *machines,
3236 			      int (*fn)(struct thread *thread, void *p),
3237 			      void *priv)
3238 {
3239 	struct rb_node *nd;
3240 	int rc = 0;
3241 
3242 	rc = machine__for_each_thread(&machines->host, fn, priv);
3243 	if (rc != 0)
3244 		return rc;
3245 
3246 	for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3247 		struct machine *machine = rb_entry(nd, struct machine, rb_node);
3248 
3249 		rc = machine__for_each_thread(machine, fn, priv);
3250 		if (rc != 0)
3251 			return rc;
3252 	}
3253 	return rc;
3254 }
3255 
machine__get_current_tid(struct machine * machine,int cpu)3256 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3257 {
3258 	if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3259 		return -1;
3260 
3261 	return machine->current_tid[cpu];
3262 }
3263 
machine__set_current_tid(struct machine * machine,int cpu,pid_t pid,pid_t tid)3264 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3265 			     pid_t tid)
3266 {
3267 	struct thread *thread;
3268 	const pid_t init_val = -1;
3269 
3270 	if (cpu < 0)
3271 		return -EINVAL;
3272 
3273 	if (realloc_array_as_needed(machine->current_tid,
3274 				    machine->current_tid_sz,
3275 				    (unsigned int)cpu,
3276 				    &init_val))
3277 		return -ENOMEM;
3278 
3279 	machine->current_tid[cpu] = tid;
3280 
3281 	thread = machine__findnew_thread(machine, pid, tid);
3282 	if (!thread)
3283 		return -ENOMEM;
3284 
3285 	thread__set_cpu(thread, cpu);
3286 	thread__put(thread);
3287 
3288 	return 0;
3289 }
3290 
3291 /*
3292  * Compares the raw arch string. N.B. see instead perf_env__arch() or
3293  * machine__normalized_is() if a normalized arch is needed.
3294  */
machine__is(struct machine * machine,const char * arch)3295 bool machine__is(struct machine *machine, const char *arch)
3296 {
3297 	return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3298 }
3299 
machine__normalized_is(struct machine * machine,const char * arch)3300 bool machine__normalized_is(struct machine *machine, const char *arch)
3301 {
3302 	return machine && !strcmp(perf_env__arch(machine->env), arch);
3303 }
3304 
machine__nr_cpus_avail(struct machine * machine)3305 int machine__nr_cpus_avail(struct machine *machine)
3306 {
3307 	return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3308 }
3309 
machine__get_kernel_start(struct machine * machine)3310 int machine__get_kernel_start(struct machine *machine)
3311 {
3312 	struct map *map = machine__kernel_map(machine);
3313 	int err = 0;
3314 
3315 	/*
3316 	 * The only addresses above 2^63 are kernel addresses of a 64-bit
3317 	 * kernel.  Note that addresses are unsigned so that on a 32-bit system
3318 	 * all addresses including kernel addresses are less than 2^32.  In
3319 	 * that case (32-bit system), if the kernel mapping is unknown, all
3320 	 * addresses will be assumed to be in user space - see
3321 	 * machine__kernel_ip().
3322 	 */
3323 	machine->kernel_start = 1ULL << 63;
3324 	if (map) {
3325 		err = map__load(map);
3326 		/*
3327 		 * On x86_64, PTI entry trampolines are less than the
3328 		 * start of kernel text, but still above 2^63. So leave
3329 		 * kernel_start = 1ULL << 63 for x86_64.
3330 		 */
3331 		if (!err && !machine__is(machine, "x86_64"))
3332 			machine->kernel_start = map__start(map);
3333 	}
3334 	return err;
3335 }
3336 
machine__addr_cpumode(struct machine * machine,u8 cpumode,u64 addr)3337 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3338 {
3339 	u8 addr_cpumode = cpumode;
3340 	bool kernel_ip;
3341 
3342 	if (!machine->single_address_space)
3343 		goto out;
3344 
3345 	kernel_ip = machine__kernel_ip(machine, addr);
3346 	switch (cpumode) {
3347 	case PERF_RECORD_MISC_KERNEL:
3348 	case PERF_RECORD_MISC_USER:
3349 		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3350 					   PERF_RECORD_MISC_USER;
3351 		break;
3352 	case PERF_RECORD_MISC_GUEST_KERNEL:
3353 	case PERF_RECORD_MISC_GUEST_USER:
3354 		addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3355 					   PERF_RECORD_MISC_GUEST_USER;
3356 		break;
3357 	default:
3358 		break;
3359 	}
3360 out:
3361 	return addr_cpumode;
3362 }
3363 
machine__findnew_dso_id(struct machine * machine,const char * filename,struct dso_id * id)3364 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3365 {
3366 	return dsos__findnew_id(&machine->dsos, filename, id);
3367 }
3368 
machine__findnew_dso(struct machine * machine,const char * filename)3369 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3370 {
3371 	return machine__findnew_dso_id(machine, filename, NULL);
3372 }
3373 
machine__resolve_kernel_addr(void * vmachine,unsigned long long * addrp,char ** modp)3374 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3375 {
3376 	struct machine *machine = vmachine;
3377 	struct map *map;
3378 	struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3379 
3380 	if (sym == NULL)
3381 		return NULL;
3382 
3383 	*modp = __map__is_kmodule(map) ? (char *)map__dso(map)->short_name : NULL;
3384 	*addrp = map__unmap_ip(map, sym->start);
3385 	return sym->name;
3386 }
3387 
machine__for_each_dso(struct machine * machine,machine__dso_t fn,void * priv)3388 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3389 {
3390 	struct dso *pos;
3391 	int err = 0;
3392 
3393 	list_for_each_entry(pos, &machine->dsos.head, node) {
3394 		if (fn(pos, machine, priv))
3395 			err = -1;
3396 	}
3397 	return err;
3398 }
3399 
machine__for_each_kernel_map(struct machine * machine,machine__map_t fn,void * priv)3400 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3401 {
3402 	struct maps *maps = machine__kernel_maps(machine);
3403 	struct map_rb_node *pos;
3404 	int err = 0;
3405 
3406 	maps__for_each_entry(maps, pos) {
3407 		err = fn(pos->map, priv);
3408 		if (err != 0) {
3409 			break;
3410 		}
3411 	}
3412 	return err;
3413 }
3414 
machine__is_lock_function(struct machine * machine,u64 addr)3415 bool machine__is_lock_function(struct machine *machine, u64 addr)
3416 {
3417 	if (!machine->sched.text_start) {
3418 		struct map *kmap;
3419 		struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3420 
3421 		if (!sym) {
3422 			/* to avoid retry */
3423 			machine->sched.text_start = 1;
3424 			return false;
3425 		}
3426 
3427 		machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3428 
3429 		/* should not fail from here */
3430 		sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3431 		machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3432 
3433 		sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3434 		machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3435 
3436 		sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3437 		machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3438 	}
3439 
3440 	/* failed to get kernel symbols */
3441 	if (machine->sched.text_start == 1)
3442 		return false;
3443 
3444 	/* mutex and rwsem functions are in sched text section */
3445 	if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3446 		return true;
3447 
3448 	/* spinlock functions are in lock text section */
3449 	if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3450 		return true;
3451 
3452 	return false;
3453 }
3454