Loading...
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
46static void __machine__remove_thread(struct machine *machine, struct thread_rb_node *nd,
47 struct thread *th, bool lock);
48
49static struct dso *machine__kernel_dso(struct machine *machine)
50{
51 return map__dso(machine->vmlinux_map);
52}
53
54static 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
61static 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 threads->last_match = NULL;
71 }
72}
73
74static int thread_rb_node__cmp_tid(const void *key, const struct rb_node *nd)
75{
76 int to_find = (int) *((pid_t *)key);
77
78 return to_find - (int)thread__tid(rb_entry(nd, struct thread_rb_node, rb_node)->thread);
79}
80
81static struct thread_rb_node *thread_rb_node__find(const struct thread *th,
82 struct rb_root *tree)
83{
84 pid_t to_find = thread__tid(th);
85 struct rb_node *nd = rb_find(&to_find, tree, thread_rb_node__cmp_tid);
86
87 return rb_entry(nd, struct thread_rb_node, rb_node);
88}
89
90static int machine__set_mmap_name(struct machine *machine)
91{
92 if (machine__is_host(machine))
93 machine->mmap_name = strdup("[kernel.kallsyms]");
94 else if (machine__is_default_guest(machine))
95 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
96 else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
97 machine->pid) < 0)
98 machine->mmap_name = NULL;
99
100 return machine->mmap_name ? 0 : -ENOMEM;
101}
102
103static void thread__set_guest_comm(struct thread *thread, pid_t pid)
104{
105 char comm[64];
106
107 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
108 thread__set_comm(thread, comm, 0);
109}
110
111int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
112{
113 int err = -ENOMEM;
114
115 memset(machine, 0, sizeof(*machine));
116 machine->kmaps = maps__new(machine);
117 if (machine->kmaps == NULL)
118 return -ENOMEM;
119
120 RB_CLEAR_NODE(&machine->rb_node);
121 dsos__init(&machine->dsos);
122
123 machine__threads_init(machine);
124
125 machine->vdso_info = NULL;
126 machine->env = NULL;
127
128 machine->pid = pid;
129
130 machine->id_hdr_size = 0;
131 machine->kptr_restrict_warned = false;
132 machine->comm_exec = false;
133 machine->kernel_start = 0;
134 machine->vmlinux_map = NULL;
135
136 machine->root_dir = strdup(root_dir);
137 if (machine->root_dir == NULL)
138 goto out;
139
140 if (machine__set_mmap_name(machine))
141 goto out;
142
143 if (pid != HOST_KERNEL_ID) {
144 struct thread *thread = machine__findnew_thread(machine, -1,
145 pid);
146
147 if (thread == NULL)
148 goto out;
149
150 thread__set_guest_comm(thread, pid);
151 thread__put(thread);
152 }
153
154 machine->current_tid = NULL;
155 err = 0;
156
157out:
158 if (err) {
159 zfree(&machine->kmaps);
160 zfree(&machine->root_dir);
161 zfree(&machine->mmap_name);
162 }
163 return 0;
164}
165
166struct machine *machine__new_host(void)
167{
168 struct machine *machine = malloc(sizeof(*machine));
169
170 if (machine != NULL) {
171 machine__init(machine, "", HOST_KERNEL_ID);
172
173 if (machine__create_kernel_maps(machine) < 0)
174 goto out_delete;
175 }
176
177 return machine;
178out_delete:
179 free(machine);
180 return NULL;
181}
182
183struct machine *machine__new_kallsyms(void)
184{
185 struct machine *machine = machine__new_host();
186 /*
187 * FIXME:
188 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
189 * ask for not using the kcore parsing code, once this one is fixed
190 * to create a map per module.
191 */
192 if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
193 machine__delete(machine);
194 machine = NULL;
195 }
196
197 return machine;
198}
199
200static void dsos__purge(struct dsos *dsos)
201{
202 struct dso *pos, *n;
203
204 down_write(&dsos->lock);
205
206 list_for_each_entry_safe(pos, n, &dsos->head, node) {
207 RB_CLEAR_NODE(&pos->rb_node);
208 pos->root = NULL;
209 list_del_init(&pos->node);
210 dso__put(pos);
211 }
212
213 up_write(&dsos->lock);
214}
215
216static void dsos__exit(struct dsos *dsos)
217{
218 dsos__purge(dsos);
219 exit_rwsem(&dsos->lock);
220}
221
222void machine__delete_threads(struct machine *machine)
223{
224 struct rb_node *nd;
225 int i;
226
227 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
228 struct threads *threads = &machine->threads[i];
229 down_write(&threads->lock);
230 nd = rb_first_cached(&threads->entries);
231 while (nd) {
232 struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
233
234 nd = rb_next(nd);
235 __machine__remove_thread(machine, trb, trb->thread, false);
236 }
237 up_write(&threads->lock);
238 }
239}
240
241void machine__exit(struct machine *machine)
242{
243 int i;
244
245 if (machine == NULL)
246 return;
247
248 machine__destroy_kernel_maps(machine);
249 maps__zput(machine->kmaps);
250 dsos__exit(&machine->dsos);
251 machine__exit_vdso(machine);
252 zfree(&machine->root_dir);
253 zfree(&machine->mmap_name);
254 zfree(&machine->current_tid);
255 zfree(&machine->kallsyms_filename);
256
257 machine__delete_threads(machine);
258 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
259 struct threads *threads = &machine->threads[i];
260
261 exit_rwsem(&threads->lock);
262 }
263}
264
265void machine__delete(struct machine *machine)
266{
267 if (machine) {
268 machine__exit(machine);
269 free(machine);
270 }
271}
272
273void machines__init(struct machines *machines)
274{
275 machine__init(&machines->host, "", HOST_KERNEL_ID);
276 machines->guests = RB_ROOT_CACHED;
277}
278
279void machines__exit(struct machines *machines)
280{
281 machine__exit(&machines->host);
282 /* XXX exit guest */
283}
284
285struct machine *machines__add(struct machines *machines, pid_t pid,
286 const char *root_dir)
287{
288 struct rb_node **p = &machines->guests.rb_root.rb_node;
289 struct rb_node *parent = NULL;
290 struct machine *pos, *machine = malloc(sizeof(*machine));
291 bool leftmost = true;
292
293 if (machine == NULL)
294 return NULL;
295
296 if (machine__init(machine, root_dir, pid) != 0) {
297 free(machine);
298 return NULL;
299 }
300
301 while (*p != NULL) {
302 parent = *p;
303 pos = rb_entry(parent, struct machine, rb_node);
304 if (pid < pos->pid)
305 p = &(*p)->rb_left;
306 else {
307 p = &(*p)->rb_right;
308 leftmost = false;
309 }
310 }
311
312 rb_link_node(&machine->rb_node, parent, p);
313 rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
314
315 machine->machines = machines;
316
317 return machine;
318}
319
320void machines__set_comm_exec(struct machines *machines, bool comm_exec)
321{
322 struct rb_node *nd;
323
324 machines->host.comm_exec = comm_exec;
325
326 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
327 struct machine *machine = rb_entry(nd, struct machine, rb_node);
328
329 machine->comm_exec = comm_exec;
330 }
331}
332
333struct machine *machines__find(struct machines *machines, pid_t pid)
334{
335 struct rb_node **p = &machines->guests.rb_root.rb_node;
336 struct rb_node *parent = NULL;
337 struct machine *machine;
338 struct machine *default_machine = NULL;
339
340 if (pid == HOST_KERNEL_ID)
341 return &machines->host;
342
343 while (*p != NULL) {
344 parent = *p;
345 machine = rb_entry(parent, struct machine, rb_node);
346 if (pid < machine->pid)
347 p = &(*p)->rb_left;
348 else if (pid > machine->pid)
349 p = &(*p)->rb_right;
350 else
351 return machine;
352 if (!machine->pid)
353 default_machine = machine;
354 }
355
356 return default_machine;
357}
358
359struct machine *machines__findnew(struct machines *machines, pid_t pid)
360{
361 char path[PATH_MAX];
362 const char *root_dir = "";
363 struct machine *machine = machines__find(machines, pid);
364
365 if (machine && (machine->pid == pid))
366 goto out;
367
368 if ((pid != HOST_KERNEL_ID) &&
369 (pid != DEFAULT_GUEST_KERNEL_ID) &&
370 (symbol_conf.guestmount)) {
371 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
372 if (access(path, R_OK)) {
373 static struct strlist *seen;
374
375 if (!seen)
376 seen = strlist__new(NULL, NULL);
377
378 if (!strlist__has_entry(seen, path)) {
379 pr_err("Can't access file %s\n", path);
380 strlist__add(seen, path);
381 }
382 machine = NULL;
383 goto out;
384 }
385 root_dir = path;
386 }
387
388 machine = machines__add(machines, pid, root_dir);
389out:
390 return machine;
391}
392
393struct machine *machines__find_guest(struct machines *machines, pid_t pid)
394{
395 struct machine *machine = machines__find(machines, pid);
396
397 if (!machine)
398 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
399 return machine;
400}
401
402/*
403 * A common case for KVM test programs is that the test program acts as the
404 * hypervisor, creating, running and destroying the virtual machine, and
405 * providing the guest object code from its own object code. In this case,
406 * the VM is not running an OS, but only the functions loaded into it by the
407 * hypervisor test program, and conveniently, loaded at the same virtual
408 * addresses.
409 *
410 * Normally to resolve addresses, MMAP events are needed to map addresses
411 * back to the object code and debug symbols for that object code.
412 *
413 * Currently, there is no way to get such mapping information from guests
414 * but, in the scenario described above, the guest has the same mappings
415 * as the hypervisor, so support for that scenario can be achieved.
416 *
417 * To support that, copy the host thread's maps to the guest thread's maps.
418 * Note, we do not discover the guest until we encounter a guest event,
419 * which works well because it is not until then that we know that the host
420 * thread's maps have been set up.
421 *
422 * This function returns the guest thread. Apart from keeping the data
423 * structures sane, using a thread belonging to the guest machine, instead
424 * of the host thread, allows it to have its own comm (refer
425 * thread__set_guest_comm()).
426 */
427static struct thread *findnew_guest_code(struct machine *machine,
428 struct machine *host_machine,
429 pid_t pid)
430{
431 struct thread *host_thread;
432 struct thread *thread;
433 int err;
434
435 if (!machine)
436 return NULL;
437
438 thread = machine__findnew_thread(machine, -1, pid);
439 if (!thread)
440 return NULL;
441
442 /* Assume maps are set up if there are any */
443 if (maps__nr_maps(thread__maps(thread)))
444 return thread;
445
446 host_thread = machine__find_thread(host_machine, -1, pid);
447 if (!host_thread)
448 goto out_err;
449
450 thread__set_guest_comm(thread, pid);
451
452 /*
453 * Guest code can be found in hypervisor process at the same address
454 * so copy host maps.
455 */
456 err = maps__copy_from(thread__maps(thread), thread__maps(host_thread));
457 thread__put(host_thread);
458 if (err)
459 goto out_err;
460
461 return thread;
462
463out_err:
464 thread__zput(thread);
465 return NULL;
466}
467
468struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
469{
470 struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
471 struct machine *machine = machines__findnew(machines, pid);
472
473 return findnew_guest_code(machine, host_machine, pid);
474}
475
476struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
477{
478 struct machines *machines = machine->machines;
479 struct machine *host_machine;
480
481 if (!machines)
482 return NULL;
483
484 host_machine = machines__find(machines, HOST_KERNEL_ID);
485
486 return findnew_guest_code(machine, host_machine, pid);
487}
488
489void machines__process_guests(struct machines *machines,
490 machine__process_t process, void *data)
491{
492 struct rb_node *nd;
493
494 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
495 struct machine *pos = rb_entry(nd, struct machine, rb_node);
496 process(pos, data);
497 }
498}
499
500void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
501{
502 struct rb_node *node;
503 struct machine *machine;
504
505 machines->host.id_hdr_size = id_hdr_size;
506
507 for (node = rb_first_cached(&machines->guests); node;
508 node = rb_next(node)) {
509 machine = rb_entry(node, struct machine, rb_node);
510 machine->id_hdr_size = id_hdr_size;
511 }
512
513 return;
514}
515
516static void machine__update_thread_pid(struct machine *machine,
517 struct thread *th, pid_t pid)
518{
519 struct thread *leader;
520
521 if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
522 return;
523
524 thread__set_pid(th, pid);
525
526 if (thread__pid(th) == thread__tid(th))
527 return;
528
529 leader = __machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
530 if (!leader)
531 goto out_err;
532
533 if (!thread__maps(leader))
534 thread__set_maps(leader, maps__new(machine));
535
536 if (!thread__maps(leader))
537 goto out_err;
538
539 if (thread__maps(th) == thread__maps(leader))
540 goto out_put;
541
542 if (thread__maps(th)) {
543 /*
544 * Maps are created from MMAP events which provide the pid and
545 * tid. Consequently there never should be any maps on a thread
546 * with an unknown pid. Just print an error if there are.
547 */
548 if (!maps__empty(thread__maps(th)))
549 pr_err("Discarding thread maps for %d:%d\n",
550 thread__pid(th), thread__tid(th));
551 maps__put(thread__maps(th));
552 }
553
554 thread__set_maps(th, maps__get(thread__maps(leader)));
555out_put:
556 thread__put(leader);
557 return;
558out_err:
559 pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
560 goto out_put;
561}
562
563/*
564 * Front-end cache - TID lookups come in blocks,
565 * so most of the time we dont have to look up
566 * the full rbtree:
567 */
568static struct thread*
569__threads__get_last_match(struct threads *threads, struct machine *machine,
570 int pid, int tid)
571{
572 struct thread *th;
573
574 th = threads->last_match;
575 if (th != NULL) {
576 if (thread__tid(th) == tid) {
577 machine__update_thread_pid(machine, th, pid);
578 return thread__get(th);
579 }
580 thread__put(threads->last_match);
581 threads->last_match = NULL;
582 }
583
584 return NULL;
585}
586
587static struct thread*
588threads__get_last_match(struct threads *threads, struct machine *machine,
589 int pid, int tid)
590{
591 struct thread *th = NULL;
592
593 if (perf_singlethreaded)
594 th = __threads__get_last_match(threads, machine, pid, tid);
595
596 return th;
597}
598
599static void
600__threads__set_last_match(struct threads *threads, struct thread *th)
601{
602 thread__put(threads->last_match);
603 threads->last_match = thread__get(th);
604}
605
606static void
607threads__set_last_match(struct threads *threads, struct thread *th)
608{
609 if (perf_singlethreaded)
610 __threads__set_last_match(threads, th);
611}
612
613/*
614 * Caller must eventually drop thread->refcnt returned with a successful
615 * lookup/new thread inserted.
616 */
617static struct thread *____machine__findnew_thread(struct machine *machine,
618 struct threads *threads,
619 pid_t pid, pid_t tid,
620 bool create)
621{
622 struct rb_node **p = &threads->entries.rb_root.rb_node;
623 struct rb_node *parent = NULL;
624 struct thread *th;
625 struct thread_rb_node *nd;
626 bool leftmost = true;
627
628 th = threads__get_last_match(threads, machine, pid, tid);
629 if (th)
630 return th;
631
632 while (*p != NULL) {
633 parent = *p;
634 th = rb_entry(parent, struct thread_rb_node, rb_node)->thread;
635
636 if (thread__tid(th) == tid) {
637 threads__set_last_match(threads, th);
638 machine__update_thread_pid(machine, th, pid);
639 return thread__get(th);
640 }
641
642 if (tid < thread__tid(th))
643 p = &(*p)->rb_left;
644 else {
645 p = &(*p)->rb_right;
646 leftmost = false;
647 }
648 }
649
650 if (!create)
651 return NULL;
652
653 th = thread__new(pid, tid);
654 if (th == NULL)
655 return NULL;
656
657 nd = malloc(sizeof(*nd));
658 if (nd == NULL) {
659 thread__put(th);
660 return NULL;
661 }
662 nd->thread = th;
663
664 rb_link_node(&nd->rb_node, parent, p);
665 rb_insert_color_cached(&nd->rb_node, &threads->entries, leftmost);
666 /*
667 * We have to initialize maps separately after rb tree is updated.
668 *
669 * The reason is that we call machine__findnew_thread within
670 * thread__init_maps to find the thread leader and that would screwed
671 * the rb tree.
672 */
673 if (thread__init_maps(th, machine)) {
674 pr_err("Thread init failed thread %d\n", pid);
675 rb_erase_cached(&nd->rb_node, &threads->entries);
676 RB_CLEAR_NODE(&nd->rb_node);
677 free(nd);
678 thread__put(th);
679 return NULL;
680 }
681 /*
682 * It is now in the rbtree, get a ref
683 */
684 threads__set_last_match(threads, th);
685 ++threads->nr;
686
687 return thread__get(th);
688}
689
690struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
691{
692 return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
693}
694
695struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
696 pid_t tid)
697{
698 struct threads *threads = machine__threads(machine, tid);
699 struct thread *th;
700
701 down_write(&threads->lock);
702 th = __machine__findnew_thread(machine, pid, tid);
703 up_write(&threads->lock);
704 return th;
705}
706
707struct thread *machine__find_thread(struct machine *machine, pid_t pid,
708 pid_t tid)
709{
710 struct threads *threads = machine__threads(machine, tid);
711 struct thread *th;
712
713 down_read(&threads->lock);
714 th = ____machine__findnew_thread(machine, threads, pid, tid, false);
715 up_read(&threads->lock);
716 return th;
717}
718
719/*
720 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
721 * So here a single thread is created for that, but actually there is a separate
722 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
723 * is only 1. That causes problems for some tools, requiring workarounds. For
724 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
725 */
726struct thread *machine__idle_thread(struct machine *machine)
727{
728 struct thread *thread = machine__findnew_thread(machine, 0, 0);
729
730 if (!thread || thread__set_comm(thread, "swapper", 0) ||
731 thread__set_namespaces(thread, 0, NULL))
732 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
733
734 return thread;
735}
736
737struct comm *machine__thread_exec_comm(struct machine *machine,
738 struct thread *thread)
739{
740 if (machine->comm_exec)
741 return thread__exec_comm(thread);
742 else
743 return thread__comm(thread);
744}
745
746int machine__process_comm_event(struct machine *machine, union perf_event *event,
747 struct perf_sample *sample)
748{
749 struct thread *thread = machine__findnew_thread(machine,
750 event->comm.pid,
751 event->comm.tid);
752 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
753 int err = 0;
754
755 if (exec)
756 machine->comm_exec = true;
757
758 if (dump_trace)
759 perf_event__fprintf_comm(event, stdout);
760
761 if (thread == NULL ||
762 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
763 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
764 err = -1;
765 }
766
767 thread__put(thread);
768
769 return err;
770}
771
772int machine__process_namespaces_event(struct machine *machine __maybe_unused,
773 union perf_event *event,
774 struct perf_sample *sample __maybe_unused)
775{
776 struct thread *thread = machine__findnew_thread(machine,
777 event->namespaces.pid,
778 event->namespaces.tid);
779 int err = 0;
780
781 WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
782 "\nWARNING: kernel seems to support more namespaces than perf"
783 " tool.\nTry updating the perf tool..\n\n");
784
785 WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
786 "\nWARNING: perf tool seems to support more namespaces than"
787 " the kernel.\nTry updating the kernel..\n\n");
788
789 if (dump_trace)
790 perf_event__fprintf_namespaces(event, stdout);
791
792 if (thread == NULL ||
793 thread__set_namespaces(thread, sample->time, &event->namespaces)) {
794 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
795 err = -1;
796 }
797
798 thread__put(thread);
799
800 return err;
801}
802
803int machine__process_cgroup_event(struct machine *machine,
804 union perf_event *event,
805 struct perf_sample *sample __maybe_unused)
806{
807 struct cgroup *cgrp;
808
809 if (dump_trace)
810 perf_event__fprintf_cgroup(event, stdout);
811
812 cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
813 if (cgrp == NULL)
814 return -ENOMEM;
815
816 return 0;
817}
818
819int machine__process_lost_event(struct machine *machine __maybe_unused,
820 union perf_event *event, struct perf_sample *sample __maybe_unused)
821{
822 dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
823 event->lost.id, event->lost.lost);
824 return 0;
825}
826
827int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
828 union perf_event *event, struct perf_sample *sample)
829{
830 dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
831 sample->id, event->lost_samples.lost);
832 return 0;
833}
834
835static struct dso *machine__findnew_module_dso(struct machine *machine,
836 struct kmod_path *m,
837 const char *filename)
838{
839 struct dso *dso;
840
841 down_write(&machine->dsos.lock);
842
843 dso = __dsos__find(&machine->dsos, m->name, true);
844 if (!dso) {
845 dso = __dsos__addnew(&machine->dsos, m->name);
846 if (dso == NULL)
847 goto out_unlock;
848
849 dso__set_module_info(dso, m, machine);
850 dso__set_long_name(dso, strdup(filename), true);
851 dso->kernel = DSO_SPACE__KERNEL;
852 }
853
854 dso__get(dso);
855out_unlock:
856 up_write(&machine->dsos.lock);
857 return dso;
858}
859
860int machine__process_aux_event(struct machine *machine __maybe_unused,
861 union perf_event *event)
862{
863 if (dump_trace)
864 perf_event__fprintf_aux(event, stdout);
865 return 0;
866}
867
868int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
869 union perf_event *event)
870{
871 if (dump_trace)
872 perf_event__fprintf_itrace_start(event, stdout);
873 return 0;
874}
875
876int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
877 union perf_event *event)
878{
879 if (dump_trace)
880 perf_event__fprintf_aux_output_hw_id(event, stdout);
881 return 0;
882}
883
884int machine__process_switch_event(struct machine *machine __maybe_unused,
885 union perf_event *event)
886{
887 if (dump_trace)
888 perf_event__fprintf_switch(event, stdout);
889 return 0;
890}
891
892static int machine__process_ksymbol_register(struct machine *machine,
893 union perf_event *event,
894 struct perf_sample *sample __maybe_unused)
895{
896 struct symbol *sym;
897 struct dso *dso;
898 struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
899 bool put_map = false;
900 int err = 0;
901
902 if (!map) {
903 dso = dso__new(event->ksymbol.name);
904
905 if (!dso) {
906 err = -ENOMEM;
907 goto out;
908 }
909 dso->kernel = DSO_SPACE__KERNEL;
910 map = map__new2(0, dso);
911 dso__put(dso);
912 if (!map) {
913 err = -ENOMEM;
914 goto out;
915 }
916 /*
917 * The inserted map has a get on it, we need to put to release
918 * the reference count here, but do it after all accesses are
919 * done.
920 */
921 put_map = true;
922 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
923 dso->binary_type = DSO_BINARY_TYPE__OOL;
924 dso->data.file_size = event->ksymbol.len;
925 dso__set_loaded(dso);
926 }
927
928 map__set_start(map, event->ksymbol.addr);
929 map__set_end(map, map__start(map) + event->ksymbol.len);
930 err = maps__insert(machine__kernel_maps(machine), map);
931 if (err) {
932 err = -ENOMEM;
933 goto out;
934 }
935
936 dso__set_loaded(dso);
937
938 if (is_bpf_image(event->ksymbol.name)) {
939 dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
940 dso__set_long_name(dso, "", false);
941 }
942 } else {
943 dso = map__dso(map);
944 }
945
946 sym = symbol__new(map__map_ip(map, map__start(map)),
947 event->ksymbol.len,
948 0, 0, event->ksymbol.name);
949 if (!sym) {
950 err = -ENOMEM;
951 goto out;
952 }
953 dso__insert_symbol(dso, sym);
954out:
955 if (put_map)
956 map__put(map);
957 return err;
958}
959
960static int machine__process_ksymbol_unregister(struct machine *machine,
961 union perf_event *event,
962 struct perf_sample *sample __maybe_unused)
963{
964 struct symbol *sym;
965 struct map *map;
966
967 map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
968 if (!map)
969 return 0;
970
971 if (!RC_CHK_EQUAL(map, machine->vmlinux_map))
972 maps__remove(machine__kernel_maps(machine), map);
973 else {
974 struct dso *dso = map__dso(map);
975
976 sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
977 if (sym)
978 dso__delete_symbol(dso, sym);
979 }
980
981 return 0;
982}
983
984int machine__process_ksymbol(struct machine *machine __maybe_unused,
985 union perf_event *event,
986 struct perf_sample *sample)
987{
988 if (dump_trace)
989 perf_event__fprintf_ksymbol(event, stdout);
990
991 if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
992 return machine__process_ksymbol_unregister(machine, event,
993 sample);
994 return machine__process_ksymbol_register(machine, event, sample);
995}
996
997int machine__process_text_poke(struct machine *machine, union perf_event *event,
998 struct perf_sample *sample __maybe_unused)
999{
1000 struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
1001 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
1002 struct dso *dso = map ? map__dso(map) : NULL;
1003
1004 if (dump_trace)
1005 perf_event__fprintf_text_poke(event, machine, stdout);
1006
1007 if (!event->text_poke.new_len)
1008 return 0;
1009
1010 if (cpumode != PERF_RECORD_MISC_KERNEL) {
1011 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
1012 return 0;
1013 }
1014
1015 if (dso) {
1016 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
1017 int ret;
1018
1019 /*
1020 * Kernel maps might be changed when loading symbols so loading
1021 * must be done prior to using kernel maps.
1022 */
1023 map__load(map);
1024 ret = dso__data_write_cache_addr(dso, map, machine,
1025 event->text_poke.addr,
1026 new_bytes,
1027 event->text_poke.new_len);
1028 if (ret != event->text_poke.new_len)
1029 pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
1030 event->text_poke.addr);
1031 } else {
1032 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
1033 event->text_poke.addr);
1034 }
1035
1036 return 0;
1037}
1038
1039static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
1040 const char *filename)
1041{
1042 struct map *map = NULL;
1043 struct kmod_path m;
1044 struct dso *dso;
1045 int err;
1046
1047 if (kmod_path__parse_name(&m, filename))
1048 return NULL;
1049
1050 dso = machine__findnew_module_dso(machine, &m, filename);
1051 if (dso == NULL)
1052 goto out;
1053
1054 map = map__new2(start, dso);
1055 if (map == NULL)
1056 goto out;
1057
1058 err = maps__insert(machine__kernel_maps(machine), map);
1059 /* If maps__insert failed, return NULL. */
1060 if (err) {
1061 map__put(map);
1062 map = NULL;
1063 }
1064out:
1065 /* put the dso here, corresponding to machine__findnew_module_dso */
1066 dso__put(dso);
1067 zfree(&m.name);
1068 return map;
1069}
1070
1071size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
1072{
1073 struct rb_node *nd;
1074 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
1075
1076 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1077 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1078 ret += __dsos__fprintf(&pos->dsos.head, fp);
1079 }
1080
1081 return ret;
1082}
1083
1084size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
1085 bool (skip)(struct dso *dso, int parm), int parm)
1086{
1087 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
1088}
1089
1090size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
1091 bool (skip)(struct dso *dso, int parm), int parm)
1092{
1093 struct rb_node *nd;
1094 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
1095
1096 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1097 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1098 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
1099 }
1100 return ret;
1101}
1102
1103size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
1104{
1105 int i;
1106 size_t printed = 0;
1107 struct dso *kdso = machine__kernel_dso(machine);
1108
1109 if (kdso->has_build_id) {
1110 char filename[PATH_MAX];
1111 if (dso__build_id_filename(kdso, filename, sizeof(filename),
1112 false))
1113 printed += fprintf(fp, "[0] %s\n", filename);
1114 }
1115
1116 for (i = 0; i < vmlinux_path__nr_entries; ++i)
1117 printed += fprintf(fp, "[%d] %s\n",
1118 i + kdso->has_build_id, vmlinux_path[i]);
1119
1120 return printed;
1121}
1122
1123size_t machine__fprintf(struct machine *machine, FILE *fp)
1124{
1125 struct rb_node *nd;
1126 size_t ret;
1127 int i;
1128
1129 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
1130 struct threads *threads = &machine->threads[i];
1131
1132 down_read(&threads->lock);
1133
1134 ret = fprintf(fp, "Threads: %u\n", threads->nr);
1135
1136 for (nd = rb_first_cached(&threads->entries); nd;
1137 nd = rb_next(nd)) {
1138 struct thread *pos = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
1139
1140 ret += thread__fprintf(pos, fp);
1141 }
1142
1143 up_read(&threads->lock);
1144 }
1145 return ret;
1146}
1147
1148static struct dso *machine__get_kernel(struct machine *machine)
1149{
1150 const char *vmlinux_name = machine->mmap_name;
1151 struct dso *kernel;
1152
1153 if (machine__is_host(machine)) {
1154 if (symbol_conf.vmlinux_name)
1155 vmlinux_name = symbol_conf.vmlinux_name;
1156
1157 kernel = machine__findnew_kernel(machine, vmlinux_name,
1158 "[kernel]", DSO_SPACE__KERNEL);
1159 } else {
1160 if (symbol_conf.default_guest_vmlinux_name)
1161 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1162
1163 kernel = machine__findnew_kernel(machine, vmlinux_name,
1164 "[guest.kernel]",
1165 DSO_SPACE__KERNEL_GUEST);
1166 }
1167
1168 if (kernel != NULL && (!kernel->has_build_id))
1169 dso__read_running_kernel_build_id(kernel, machine);
1170
1171 return kernel;
1172}
1173
1174void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1175 size_t bufsz)
1176{
1177 if (machine__is_default_guest(machine))
1178 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1179 else
1180 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1181}
1182
1183const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1184
1185/* Figure out the start address of kernel map from /proc/kallsyms.
1186 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1187 * symbol_name if it's not that important.
1188 */
1189static int machine__get_running_kernel_start(struct machine *machine,
1190 const char **symbol_name,
1191 u64 *start, u64 *end)
1192{
1193 char filename[PATH_MAX];
1194 int i, err = -1;
1195 const char *name;
1196 u64 addr = 0;
1197
1198 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1199
1200 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1201 return 0;
1202
1203 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1204 err = kallsyms__get_function_start(filename, name, &addr);
1205 if (!err)
1206 break;
1207 }
1208
1209 if (err)
1210 return -1;
1211
1212 if (symbol_name)
1213 *symbol_name = name;
1214
1215 *start = addr;
1216
1217 err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1218 if (err)
1219 err = kallsyms__get_function_start(filename, "_etext", &addr);
1220 if (!err)
1221 *end = addr;
1222
1223 return 0;
1224}
1225
1226int machine__create_extra_kernel_map(struct machine *machine,
1227 struct dso *kernel,
1228 struct extra_kernel_map *xm)
1229{
1230 struct kmap *kmap;
1231 struct map *map;
1232 int err;
1233
1234 map = map__new2(xm->start, kernel);
1235 if (!map)
1236 return -ENOMEM;
1237
1238 map__set_end(map, xm->end);
1239 map__set_pgoff(map, xm->pgoff);
1240
1241 kmap = map__kmap(map);
1242
1243 strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1244
1245 err = maps__insert(machine__kernel_maps(machine), map);
1246
1247 if (!err) {
1248 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1249 kmap->name, map__start(map), map__end(map));
1250 }
1251
1252 map__put(map);
1253
1254 return err;
1255}
1256
1257static u64 find_entry_trampoline(struct dso *dso)
1258{
1259 /* Duplicates are removed so lookup all aliases */
1260 const char *syms[] = {
1261 "_entry_trampoline",
1262 "__entry_trampoline_start",
1263 "entry_SYSCALL_64_trampoline",
1264 };
1265 struct symbol *sym = dso__first_symbol(dso);
1266 unsigned int i;
1267
1268 for (; sym; sym = dso__next_symbol(sym)) {
1269 if (sym->binding != STB_GLOBAL)
1270 continue;
1271 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1272 if (!strcmp(sym->name, syms[i]))
1273 return sym->start;
1274 }
1275 }
1276
1277 return 0;
1278}
1279
1280/*
1281 * These values can be used for kernels that do not have symbols for the entry
1282 * trampolines in kallsyms.
1283 */
1284#define X86_64_CPU_ENTRY_AREA_PER_CPU 0xfffffe0000000000ULL
1285#define X86_64_CPU_ENTRY_AREA_SIZE 0x2c000
1286#define X86_64_ENTRY_TRAMPOLINE 0x6000
1287
1288struct machine__map_x86_64_entry_trampolines_args {
1289 struct maps *kmaps;
1290 bool found;
1291};
1292
1293static int machine__map_x86_64_entry_trampolines_cb(struct map *map, void *data)
1294{
1295 struct machine__map_x86_64_entry_trampolines_args *args = data;
1296 struct map *dest_map;
1297 struct kmap *kmap = __map__kmap(map);
1298
1299 if (!kmap || !is_entry_trampoline(kmap->name))
1300 return 0;
1301
1302 dest_map = maps__find(args->kmaps, map__pgoff(map));
1303 if (dest_map != map)
1304 map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1305
1306 args->found = true;
1307 return 0;
1308}
1309
1310/* Map x86_64 PTI entry trampolines */
1311int machine__map_x86_64_entry_trampolines(struct machine *machine,
1312 struct dso *kernel)
1313{
1314 struct machine__map_x86_64_entry_trampolines_args args = {
1315 .kmaps = machine__kernel_maps(machine),
1316 .found = false,
1317 };
1318 int nr_cpus_avail, cpu;
1319 u64 pgoff;
1320
1321 /*
1322 * In the vmlinux case, pgoff is a virtual address which must now be
1323 * mapped to a vmlinux offset.
1324 */
1325 maps__for_each_map(args.kmaps, machine__map_x86_64_entry_trampolines_cb, &args);
1326
1327 if (args.found || machine->trampolines_mapped)
1328 return 0;
1329
1330 pgoff = find_entry_trampoline(kernel);
1331 if (!pgoff)
1332 return 0;
1333
1334 nr_cpus_avail = machine__nr_cpus_avail(machine);
1335
1336 /* Add a 1 page map for each CPU's entry trampoline */
1337 for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1338 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1339 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1340 X86_64_ENTRY_TRAMPOLINE;
1341 struct extra_kernel_map xm = {
1342 .start = va,
1343 .end = va + page_size,
1344 .pgoff = pgoff,
1345 };
1346
1347 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1348
1349 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1350 return -1;
1351 }
1352
1353 machine->trampolines_mapped = nr_cpus_avail;
1354
1355 return 0;
1356}
1357
1358int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1359 struct dso *kernel __maybe_unused)
1360{
1361 return 0;
1362}
1363
1364static int
1365__machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1366{
1367 /* In case of renewal the kernel map, destroy previous one */
1368 machine__destroy_kernel_maps(machine);
1369
1370 map__put(machine->vmlinux_map);
1371 machine->vmlinux_map = map__new2(0, kernel);
1372 if (machine->vmlinux_map == NULL)
1373 return -ENOMEM;
1374
1375 map__set_mapping_type(machine->vmlinux_map, MAPPING_TYPE__IDENTITY);
1376 return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1377}
1378
1379void machine__destroy_kernel_maps(struct machine *machine)
1380{
1381 struct kmap *kmap;
1382 struct map *map = machine__kernel_map(machine);
1383
1384 if (map == NULL)
1385 return;
1386
1387 kmap = map__kmap(map);
1388 maps__remove(machine__kernel_maps(machine), map);
1389 if (kmap && kmap->ref_reloc_sym) {
1390 zfree((char **)&kmap->ref_reloc_sym->name);
1391 zfree(&kmap->ref_reloc_sym);
1392 }
1393
1394 map__zput(machine->vmlinux_map);
1395}
1396
1397int machines__create_guest_kernel_maps(struct machines *machines)
1398{
1399 int ret = 0;
1400 struct dirent **namelist = NULL;
1401 int i, items = 0;
1402 char path[PATH_MAX];
1403 pid_t pid;
1404 char *endp;
1405
1406 if (symbol_conf.default_guest_vmlinux_name ||
1407 symbol_conf.default_guest_modules ||
1408 symbol_conf.default_guest_kallsyms) {
1409 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1410 }
1411
1412 if (symbol_conf.guestmount) {
1413 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1414 if (items <= 0)
1415 return -ENOENT;
1416 for (i = 0; i < items; i++) {
1417 if (!isdigit(namelist[i]->d_name[0])) {
1418 /* Filter out . and .. */
1419 continue;
1420 }
1421 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1422 if ((*endp != '\0') ||
1423 (endp == namelist[i]->d_name) ||
1424 (errno == ERANGE)) {
1425 pr_debug("invalid directory (%s). Skipping.\n",
1426 namelist[i]->d_name);
1427 continue;
1428 }
1429 sprintf(path, "%s/%s/proc/kallsyms",
1430 symbol_conf.guestmount,
1431 namelist[i]->d_name);
1432 ret = access(path, R_OK);
1433 if (ret) {
1434 pr_debug("Can't access file %s\n", path);
1435 goto failure;
1436 }
1437 machines__create_kernel_maps(machines, pid);
1438 }
1439failure:
1440 free(namelist);
1441 }
1442
1443 return ret;
1444}
1445
1446void machines__destroy_kernel_maps(struct machines *machines)
1447{
1448 struct rb_node *next = rb_first_cached(&machines->guests);
1449
1450 machine__destroy_kernel_maps(&machines->host);
1451
1452 while (next) {
1453 struct machine *pos = rb_entry(next, struct machine, rb_node);
1454
1455 next = rb_next(&pos->rb_node);
1456 rb_erase_cached(&pos->rb_node, &machines->guests);
1457 machine__delete(pos);
1458 }
1459}
1460
1461int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1462{
1463 struct machine *machine = machines__findnew(machines, pid);
1464
1465 if (machine == NULL)
1466 return -1;
1467
1468 return machine__create_kernel_maps(machine);
1469}
1470
1471int machine__load_kallsyms(struct machine *machine, const char *filename)
1472{
1473 struct map *map = machine__kernel_map(machine);
1474 struct dso *dso = map__dso(map);
1475 int ret = __dso__load_kallsyms(dso, filename, map, true);
1476
1477 if (ret > 0) {
1478 dso__set_loaded(dso);
1479 /*
1480 * Since /proc/kallsyms will have multiple sessions for the
1481 * kernel, with modules between them, fixup the end of all
1482 * sections.
1483 */
1484 maps__fixup_end(machine__kernel_maps(machine));
1485 }
1486
1487 return ret;
1488}
1489
1490int machine__load_vmlinux_path(struct machine *machine)
1491{
1492 struct map *map = machine__kernel_map(machine);
1493 struct dso *dso = map__dso(map);
1494 int ret = dso__load_vmlinux_path(dso, map);
1495
1496 if (ret > 0)
1497 dso__set_loaded(dso);
1498
1499 return ret;
1500}
1501
1502static char *get_kernel_version(const char *root_dir)
1503{
1504 char version[PATH_MAX];
1505 FILE *file;
1506 char *name, *tmp;
1507 const char *prefix = "Linux version ";
1508
1509 sprintf(version, "%s/proc/version", root_dir);
1510 file = fopen(version, "r");
1511 if (!file)
1512 return NULL;
1513
1514 tmp = fgets(version, sizeof(version), file);
1515 fclose(file);
1516 if (!tmp)
1517 return NULL;
1518
1519 name = strstr(version, prefix);
1520 if (!name)
1521 return NULL;
1522 name += strlen(prefix);
1523 tmp = strchr(name, ' ');
1524 if (tmp)
1525 *tmp = '\0';
1526
1527 return strdup(name);
1528}
1529
1530static bool is_kmod_dso(struct dso *dso)
1531{
1532 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1533 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1534}
1535
1536static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1537{
1538 char *long_name;
1539 struct dso *dso;
1540 struct map *map = maps__find_by_name(maps, m->name);
1541
1542 if (map == NULL)
1543 return 0;
1544
1545 long_name = strdup(path);
1546 if (long_name == NULL)
1547 return -ENOMEM;
1548
1549 dso = map__dso(map);
1550 dso__set_long_name(dso, long_name, true);
1551 dso__kernel_module_get_build_id(dso, "");
1552
1553 /*
1554 * Full name could reveal us kmod compression, so
1555 * we need to update the symtab_type if needed.
1556 */
1557 if (m->comp && is_kmod_dso(dso)) {
1558 dso->symtab_type++;
1559 dso->comp = m->comp;
1560 }
1561
1562 return 0;
1563}
1564
1565static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1566{
1567 struct dirent *dent;
1568 DIR *dir = opendir(dir_name);
1569 int ret = 0;
1570
1571 if (!dir) {
1572 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1573 return -1;
1574 }
1575
1576 while ((dent = readdir(dir)) != NULL) {
1577 char path[PATH_MAX];
1578 struct stat st;
1579
1580 /*sshfs might return bad dent->d_type, so we have to stat*/
1581 path__join(path, sizeof(path), dir_name, dent->d_name);
1582 if (stat(path, &st))
1583 continue;
1584
1585 if (S_ISDIR(st.st_mode)) {
1586 if (!strcmp(dent->d_name, ".") ||
1587 !strcmp(dent->d_name, ".."))
1588 continue;
1589
1590 /* Do not follow top-level source and build symlinks */
1591 if (depth == 0) {
1592 if (!strcmp(dent->d_name, "source") ||
1593 !strcmp(dent->d_name, "build"))
1594 continue;
1595 }
1596
1597 ret = maps__set_modules_path_dir(maps, path, depth + 1);
1598 if (ret < 0)
1599 goto out;
1600 } else {
1601 struct kmod_path m;
1602
1603 ret = kmod_path__parse_name(&m, dent->d_name);
1604 if (ret)
1605 goto out;
1606
1607 if (m.kmod)
1608 ret = maps__set_module_path(maps, path, &m);
1609
1610 zfree(&m.name);
1611
1612 if (ret)
1613 goto out;
1614 }
1615 }
1616
1617out:
1618 closedir(dir);
1619 return ret;
1620}
1621
1622static int machine__set_modules_path(struct machine *machine)
1623{
1624 char *version;
1625 char modules_path[PATH_MAX];
1626
1627 version = get_kernel_version(machine->root_dir);
1628 if (!version)
1629 return -1;
1630
1631 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1632 machine->root_dir, version);
1633 free(version);
1634
1635 return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1636}
1637int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1638 u64 *size __maybe_unused,
1639 const char *name __maybe_unused)
1640{
1641 return 0;
1642}
1643
1644static int machine__create_module(void *arg, const char *name, u64 start,
1645 u64 size)
1646{
1647 struct machine *machine = arg;
1648 struct map *map;
1649
1650 if (arch__fix_module_text_start(&start, &size, name) < 0)
1651 return -1;
1652
1653 map = machine__addnew_module_map(machine, start, name);
1654 if (map == NULL)
1655 return -1;
1656 map__set_end(map, start + size);
1657
1658 dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1659 map__put(map);
1660 return 0;
1661}
1662
1663static int machine__create_modules(struct machine *machine)
1664{
1665 const char *modules;
1666 char path[PATH_MAX];
1667
1668 if (machine__is_default_guest(machine)) {
1669 modules = symbol_conf.default_guest_modules;
1670 } else {
1671 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1672 modules = path;
1673 }
1674
1675 if (symbol__restricted_filename(modules, "/proc/modules"))
1676 return -1;
1677
1678 if (modules__parse(modules, machine, machine__create_module))
1679 return -1;
1680
1681 if (!machine__set_modules_path(machine))
1682 return 0;
1683
1684 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1685
1686 return 0;
1687}
1688
1689static void machine__set_kernel_mmap(struct machine *machine,
1690 u64 start, u64 end)
1691{
1692 map__set_start(machine->vmlinux_map, start);
1693 map__set_end(machine->vmlinux_map, end);
1694 /*
1695 * Be a bit paranoid here, some perf.data file came with
1696 * a zero sized synthesized MMAP event for the kernel.
1697 */
1698 if (start == 0 && end == 0)
1699 map__set_end(machine->vmlinux_map, ~0ULL);
1700}
1701
1702static int machine__update_kernel_mmap(struct machine *machine,
1703 u64 start, u64 end)
1704{
1705 struct map *orig, *updated;
1706 int err;
1707
1708 orig = machine->vmlinux_map;
1709 updated = map__get(orig);
1710
1711 machine->vmlinux_map = updated;
1712 machine__set_kernel_mmap(machine, start, end);
1713 maps__remove(machine__kernel_maps(machine), orig);
1714 err = maps__insert(machine__kernel_maps(machine), updated);
1715 map__put(orig);
1716
1717 return err;
1718}
1719
1720int machine__create_kernel_maps(struct machine *machine)
1721{
1722 struct dso *kernel = machine__get_kernel(machine);
1723 const char *name = NULL;
1724 u64 start = 0, end = ~0ULL;
1725 int ret;
1726
1727 if (kernel == NULL)
1728 return -1;
1729
1730 ret = __machine__create_kernel_maps(machine, kernel);
1731 if (ret < 0)
1732 goto out_put;
1733
1734 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1735 if (machine__is_host(machine))
1736 pr_debug("Problems creating module maps, "
1737 "continuing anyway...\n");
1738 else
1739 pr_debug("Problems creating module maps for guest %d, "
1740 "continuing anyway...\n", machine->pid);
1741 }
1742
1743 if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1744 if (name &&
1745 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1746 machine__destroy_kernel_maps(machine);
1747 ret = -1;
1748 goto out_put;
1749 }
1750
1751 /*
1752 * we have a real start address now, so re-order the kmaps
1753 * assume it's the last in the kmaps
1754 */
1755 ret = machine__update_kernel_mmap(machine, start, end);
1756 if (ret < 0)
1757 goto out_put;
1758 }
1759
1760 if (machine__create_extra_kernel_maps(machine, kernel))
1761 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1762
1763 if (end == ~0ULL) {
1764 /* update end address of the kernel map using adjacent module address */
1765 struct map *next = maps__find_next_entry(machine__kernel_maps(machine),
1766 machine__kernel_map(machine));
1767
1768 if (next)
1769 machine__set_kernel_mmap(machine, start, map__start(next));
1770 }
1771
1772out_put:
1773 dso__put(kernel);
1774 return ret;
1775}
1776
1777static bool machine__uses_kcore(struct machine *machine)
1778{
1779 struct dso *dso;
1780
1781 list_for_each_entry(dso, &machine->dsos.head, node) {
1782 if (dso__is_kcore(dso))
1783 return true;
1784 }
1785
1786 return false;
1787}
1788
1789static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1790 struct extra_kernel_map *xm)
1791{
1792 return machine__is(machine, "x86_64") &&
1793 is_entry_trampoline(xm->name);
1794}
1795
1796static int machine__process_extra_kernel_map(struct machine *machine,
1797 struct extra_kernel_map *xm)
1798{
1799 struct dso *kernel = machine__kernel_dso(machine);
1800
1801 if (kernel == NULL)
1802 return -1;
1803
1804 return machine__create_extra_kernel_map(machine, kernel, xm);
1805}
1806
1807static int machine__process_kernel_mmap_event(struct machine *machine,
1808 struct extra_kernel_map *xm,
1809 struct build_id *bid)
1810{
1811 enum dso_space_type dso_space;
1812 bool is_kernel_mmap;
1813 const char *mmap_name = machine->mmap_name;
1814
1815 /* If we have maps from kcore then we do not need or want any others */
1816 if (machine__uses_kcore(machine))
1817 return 0;
1818
1819 if (machine__is_host(machine))
1820 dso_space = DSO_SPACE__KERNEL;
1821 else
1822 dso_space = DSO_SPACE__KERNEL_GUEST;
1823
1824 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1825 if (!is_kernel_mmap && !machine__is_host(machine)) {
1826 /*
1827 * If the event was recorded inside the guest and injected into
1828 * the host perf.data file, then it will match a host mmap_name,
1829 * so try that - see machine__set_mmap_name().
1830 */
1831 mmap_name = "[kernel.kallsyms]";
1832 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1833 }
1834 if (xm->name[0] == '/' ||
1835 (!is_kernel_mmap && xm->name[0] == '[')) {
1836 struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1837
1838 if (map == NULL)
1839 goto out_problem;
1840
1841 map__set_end(map, map__start(map) + xm->end - xm->start);
1842
1843 if (build_id__is_defined(bid))
1844 dso__set_build_id(map__dso(map), bid);
1845
1846 map__put(map);
1847 } else if (is_kernel_mmap) {
1848 const char *symbol_name = xm->name + strlen(mmap_name);
1849 /*
1850 * Should be there already, from the build-id table in
1851 * the header.
1852 */
1853 struct dso *kernel = NULL;
1854 struct dso *dso;
1855
1856 down_read(&machine->dsos.lock);
1857
1858 list_for_each_entry(dso, &machine->dsos.head, node) {
1859
1860 /*
1861 * The cpumode passed to is_kernel_module is not the
1862 * cpumode of *this* event. If we insist on passing
1863 * correct cpumode to is_kernel_module, we should
1864 * record the cpumode when we adding this dso to the
1865 * linked list.
1866 *
1867 * However we don't really need passing correct
1868 * cpumode. We know the correct cpumode must be kernel
1869 * mode (if not, we should not link it onto kernel_dsos
1870 * list).
1871 *
1872 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1873 * is_kernel_module() treats it as a kernel cpumode.
1874 */
1875
1876 if (!dso->kernel ||
1877 is_kernel_module(dso->long_name,
1878 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1879 continue;
1880
1881
1882 kernel = dso__get(dso);
1883 break;
1884 }
1885
1886 up_read(&machine->dsos.lock);
1887
1888 if (kernel == NULL)
1889 kernel = machine__findnew_dso(machine, machine->mmap_name);
1890 if (kernel == NULL)
1891 goto out_problem;
1892
1893 kernel->kernel = dso_space;
1894 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1895 dso__put(kernel);
1896 goto out_problem;
1897 }
1898
1899 if (strstr(kernel->long_name, "vmlinux"))
1900 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1901
1902 if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1903 dso__put(kernel);
1904 goto out_problem;
1905 }
1906
1907 if (build_id__is_defined(bid))
1908 dso__set_build_id(kernel, bid);
1909
1910 /*
1911 * Avoid using a zero address (kptr_restrict) for the ref reloc
1912 * symbol. Effectively having zero here means that at record
1913 * time /proc/sys/kernel/kptr_restrict was non zero.
1914 */
1915 if (xm->pgoff != 0) {
1916 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1917 symbol_name,
1918 xm->pgoff);
1919 }
1920
1921 if (machine__is_default_guest(machine)) {
1922 /*
1923 * preload dso of guest kernel and modules
1924 */
1925 dso__load(kernel, machine__kernel_map(machine));
1926 }
1927 dso__put(kernel);
1928 } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1929 return machine__process_extra_kernel_map(machine, xm);
1930 }
1931 return 0;
1932out_problem:
1933 return -1;
1934}
1935
1936int machine__process_mmap2_event(struct machine *machine,
1937 union perf_event *event,
1938 struct perf_sample *sample)
1939{
1940 struct thread *thread;
1941 struct map *map;
1942 struct dso_id dso_id = {
1943 .maj = event->mmap2.maj,
1944 .min = event->mmap2.min,
1945 .ino = event->mmap2.ino,
1946 .ino_generation = event->mmap2.ino_generation,
1947 };
1948 struct build_id __bid, *bid = NULL;
1949 int ret = 0;
1950
1951 if (dump_trace)
1952 perf_event__fprintf_mmap2(event, stdout);
1953
1954 if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1955 bid = &__bid;
1956 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1957 }
1958
1959 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1960 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1961 struct extra_kernel_map xm = {
1962 .start = event->mmap2.start,
1963 .end = event->mmap2.start + event->mmap2.len,
1964 .pgoff = event->mmap2.pgoff,
1965 };
1966
1967 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1968 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1969 if (ret < 0)
1970 goto out_problem;
1971 return 0;
1972 }
1973
1974 thread = machine__findnew_thread(machine, event->mmap2.pid,
1975 event->mmap2.tid);
1976 if (thread == NULL)
1977 goto out_problem;
1978
1979 map = map__new(machine, event->mmap2.start,
1980 event->mmap2.len, event->mmap2.pgoff,
1981 &dso_id, event->mmap2.prot,
1982 event->mmap2.flags, bid,
1983 event->mmap2.filename, thread);
1984
1985 if (map == NULL)
1986 goto out_problem_map;
1987
1988 ret = thread__insert_map(thread, map);
1989 if (ret)
1990 goto out_problem_insert;
1991
1992 thread__put(thread);
1993 map__put(map);
1994 return 0;
1995
1996out_problem_insert:
1997 map__put(map);
1998out_problem_map:
1999 thread__put(thread);
2000out_problem:
2001 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
2002 return 0;
2003}
2004
2005int machine__process_mmap_event(struct machine *machine, union perf_event *event,
2006 struct perf_sample *sample)
2007{
2008 struct thread *thread;
2009 struct map *map;
2010 u32 prot = 0;
2011 int ret = 0;
2012
2013 if (dump_trace)
2014 perf_event__fprintf_mmap(event, stdout);
2015
2016 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
2017 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
2018 struct extra_kernel_map xm = {
2019 .start = event->mmap.start,
2020 .end = event->mmap.start + event->mmap.len,
2021 .pgoff = event->mmap.pgoff,
2022 };
2023
2024 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
2025 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
2026 if (ret < 0)
2027 goto out_problem;
2028 return 0;
2029 }
2030
2031 thread = machine__findnew_thread(machine, event->mmap.pid,
2032 event->mmap.tid);
2033 if (thread == NULL)
2034 goto out_problem;
2035
2036 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
2037 prot = PROT_EXEC;
2038
2039 map = map__new(machine, event->mmap.start,
2040 event->mmap.len, event->mmap.pgoff,
2041 NULL, prot, 0, NULL, event->mmap.filename, thread);
2042
2043 if (map == NULL)
2044 goto out_problem_map;
2045
2046 ret = thread__insert_map(thread, map);
2047 if (ret)
2048 goto out_problem_insert;
2049
2050 thread__put(thread);
2051 map__put(map);
2052 return 0;
2053
2054out_problem_insert:
2055 map__put(map);
2056out_problem_map:
2057 thread__put(thread);
2058out_problem:
2059 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
2060 return 0;
2061}
2062
2063static void __machine__remove_thread(struct machine *machine, struct thread_rb_node *nd,
2064 struct thread *th, bool lock)
2065{
2066 struct threads *threads = machine__threads(machine, thread__tid(th));
2067
2068 if (!nd)
2069 nd = thread_rb_node__find(th, &threads->entries.rb_root);
2070
2071 if (threads->last_match && RC_CHK_EQUAL(threads->last_match, th))
2072 threads__set_last_match(threads, NULL);
2073
2074 if (lock)
2075 down_write(&threads->lock);
2076
2077 BUG_ON(refcount_read(thread__refcnt(th)) == 0);
2078
2079 thread__put(nd->thread);
2080 rb_erase_cached(&nd->rb_node, &threads->entries);
2081 RB_CLEAR_NODE(&nd->rb_node);
2082 --threads->nr;
2083
2084 free(nd);
2085
2086 if (lock)
2087 up_write(&threads->lock);
2088}
2089
2090void machine__remove_thread(struct machine *machine, struct thread *th)
2091{
2092 return __machine__remove_thread(machine, NULL, th, true);
2093}
2094
2095int machine__process_fork_event(struct machine *machine, union perf_event *event,
2096 struct perf_sample *sample)
2097{
2098 struct thread *thread = machine__find_thread(machine,
2099 event->fork.pid,
2100 event->fork.tid);
2101 struct thread *parent = machine__findnew_thread(machine,
2102 event->fork.ppid,
2103 event->fork.ptid);
2104 bool do_maps_clone = true;
2105 int err = 0;
2106
2107 if (dump_trace)
2108 perf_event__fprintf_task(event, stdout);
2109
2110 /*
2111 * There may be an existing thread that is not actually the parent,
2112 * either because we are processing events out of order, or because the
2113 * (fork) event that would have removed the thread was lost. Assume the
2114 * latter case and continue on as best we can.
2115 */
2116 if (thread__pid(parent) != (pid_t)event->fork.ppid) {
2117 dump_printf("removing erroneous parent thread %d/%d\n",
2118 thread__pid(parent), thread__tid(parent));
2119 machine__remove_thread(machine, parent);
2120 thread__put(parent);
2121 parent = machine__findnew_thread(machine, event->fork.ppid,
2122 event->fork.ptid);
2123 }
2124
2125 /* if a thread currently exists for the thread id remove it */
2126 if (thread != NULL) {
2127 machine__remove_thread(machine, thread);
2128 thread__put(thread);
2129 }
2130
2131 thread = machine__findnew_thread(machine, event->fork.pid,
2132 event->fork.tid);
2133 /*
2134 * When synthesizing FORK events, we are trying to create thread
2135 * objects for the already running tasks on the machine.
2136 *
2137 * Normally, for a kernel FORK event, we want to clone the parent's
2138 * maps because that is what the kernel just did.
2139 *
2140 * But when synthesizing, this should not be done. If we do, we end up
2141 * with overlapping maps as we process the synthesized MMAP2 events that
2142 * get delivered shortly thereafter.
2143 *
2144 * Use the FORK event misc flags in an internal way to signal this
2145 * situation, so we can elide the map clone when appropriate.
2146 */
2147 if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
2148 do_maps_clone = false;
2149
2150 if (thread == NULL || parent == NULL ||
2151 thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
2152 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
2153 err = -1;
2154 }
2155 thread__put(thread);
2156 thread__put(parent);
2157
2158 return err;
2159}
2160
2161int machine__process_exit_event(struct machine *machine, union perf_event *event,
2162 struct perf_sample *sample __maybe_unused)
2163{
2164 struct thread *thread = machine__find_thread(machine,
2165 event->fork.pid,
2166 event->fork.tid);
2167
2168 if (dump_trace)
2169 perf_event__fprintf_task(event, stdout);
2170
2171 if (thread != NULL) {
2172 if (symbol_conf.keep_exited_threads)
2173 thread__set_exited(thread, /*exited=*/true);
2174 else
2175 machine__remove_thread(machine, thread);
2176 }
2177 thread__put(thread);
2178 return 0;
2179}
2180
2181int machine__process_event(struct machine *machine, union perf_event *event,
2182 struct perf_sample *sample)
2183{
2184 int ret;
2185
2186 switch (event->header.type) {
2187 case PERF_RECORD_COMM:
2188 ret = machine__process_comm_event(machine, event, sample); break;
2189 case PERF_RECORD_MMAP:
2190 ret = machine__process_mmap_event(machine, event, sample); break;
2191 case PERF_RECORD_NAMESPACES:
2192 ret = machine__process_namespaces_event(machine, event, sample); break;
2193 case PERF_RECORD_CGROUP:
2194 ret = machine__process_cgroup_event(machine, event, sample); break;
2195 case PERF_RECORD_MMAP2:
2196 ret = machine__process_mmap2_event(machine, event, sample); break;
2197 case PERF_RECORD_FORK:
2198 ret = machine__process_fork_event(machine, event, sample); break;
2199 case PERF_RECORD_EXIT:
2200 ret = machine__process_exit_event(machine, event, sample); break;
2201 case PERF_RECORD_LOST:
2202 ret = machine__process_lost_event(machine, event, sample); break;
2203 case PERF_RECORD_AUX:
2204 ret = machine__process_aux_event(machine, event); break;
2205 case PERF_RECORD_ITRACE_START:
2206 ret = machine__process_itrace_start_event(machine, event); break;
2207 case PERF_RECORD_LOST_SAMPLES:
2208 ret = machine__process_lost_samples_event(machine, event, sample); break;
2209 case PERF_RECORD_SWITCH:
2210 case PERF_RECORD_SWITCH_CPU_WIDE:
2211 ret = machine__process_switch_event(machine, event); break;
2212 case PERF_RECORD_KSYMBOL:
2213 ret = machine__process_ksymbol(machine, event, sample); break;
2214 case PERF_RECORD_BPF_EVENT:
2215 ret = machine__process_bpf(machine, event, sample); break;
2216 case PERF_RECORD_TEXT_POKE:
2217 ret = machine__process_text_poke(machine, event, sample); break;
2218 case PERF_RECORD_AUX_OUTPUT_HW_ID:
2219 ret = machine__process_aux_output_hw_id_event(machine, event); break;
2220 default:
2221 ret = -1;
2222 break;
2223 }
2224
2225 return ret;
2226}
2227
2228static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2229{
2230 return regexec(regex, sym->name, 0, NULL, 0) == 0;
2231}
2232
2233static void ip__resolve_ams(struct thread *thread,
2234 struct addr_map_symbol *ams,
2235 u64 ip)
2236{
2237 struct addr_location al;
2238
2239 addr_location__init(&al);
2240 /*
2241 * We cannot use the header.misc hint to determine whether a
2242 * branch stack address is user, kernel, guest, hypervisor.
2243 * Branches may straddle the kernel/user/hypervisor boundaries.
2244 * Thus, we have to try consecutively until we find a match
2245 * or else, the symbol is unknown
2246 */
2247 thread__find_cpumode_addr_location(thread, ip, &al);
2248
2249 ams->addr = ip;
2250 ams->al_addr = al.addr;
2251 ams->al_level = al.level;
2252 ams->ms.maps = maps__get(al.maps);
2253 ams->ms.sym = al.sym;
2254 ams->ms.map = map__get(al.map);
2255 ams->phys_addr = 0;
2256 ams->data_page_size = 0;
2257 addr_location__exit(&al);
2258}
2259
2260static void ip__resolve_data(struct thread *thread,
2261 u8 m, struct addr_map_symbol *ams,
2262 u64 addr, u64 phys_addr, u64 daddr_page_size)
2263{
2264 struct addr_location al;
2265
2266 addr_location__init(&al);
2267
2268 thread__find_symbol(thread, m, addr, &al);
2269
2270 ams->addr = addr;
2271 ams->al_addr = al.addr;
2272 ams->al_level = al.level;
2273 ams->ms.maps = maps__get(al.maps);
2274 ams->ms.sym = al.sym;
2275 ams->ms.map = map__get(al.map);
2276 ams->phys_addr = phys_addr;
2277 ams->data_page_size = daddr_page_size;
2278 addr_location__exit(&al);
2279}
2280
2281struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2282 struct addr_location *al)
2283{
2284 struct mem_info *mi = mem_info__new();
2285
2286 if (!mi)
2287 return NULL;
2288
2289 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2290 ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2291 sample->addr, sample->phys_addr,
2292 sample->data_page_size);
2293 mi->data_src.val = sample->data_src;
2294
2295 return mi;
2296}
2297
2298static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2299{
2300 struct map *map = ms->map;
2301 char *srcline = NULL;
2302 struct dso *dso;
2303
2304 if (!map || callchain_param.key == CCKEY_FUNCTION)
2305 return srcline;
2306
2307 dso = map__dso(map);
2308 srcline = srcline__tree_find(&dso->srclines, ip);
2309 if (!srcline) {
2310 bool show_sym = false;
2311 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2312
2313 srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2314 ms->sym, show_sym, show_addr, ip);
2315 srcline__tree_insert(&dso->srclines, ip, srcline);
2316 }
2317
2318 return srcline;
2319}
2320
2321struct iterations {
2322 int nr_loop_iter;
2323 u64 cycles;
2324};
2325
2326static int add_callchain_ip(struct thread *thread,
2327 struct callchain_cursor *cursor,
2328 struct symbol **parent,
2329 struct addr_location *root_al,
2330 u8 *cpumode,
2331 u64 ip,
2332 bool branch,
2333 struct branch_flags *flags,
2334 struct iterations *iter,
2335 u64 branch_from)
2336{
2337 struct map_symbol ms = {};
2338 struct addr_location al;
2339 int nr_loop_iter = 0, err = 0;
2340 u64 iter_cycles = 0;
2341 const char *srcline = NULL;
2342
2343 addr_location__init(&al);
2344 al.filtered = 0;
2345 al.sym = NULL;
2346 al.srcline = NULL;
2347 if (!cpumode) {
2348 thread__find_cpumode_addr_location(thread, ip, &al);
2349 } else {
2350 if (ip >= PERF_CONTEXT_MAX) {
2351 switch (ip) {
2352 case PERF_CONTEXT_HV:
2353 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2354 break;
2355 case PERF_CONTEXT_KERNEL:
2356 *cpumode = PERF_RECORD_MISC_KERNEL;
2357 break;
2358 case PERF_CONTEXT_USER:
2359 *cpumode = PERF_RECORD_MISC_USER;
2360 break;
2361 default:
2362 pr_debug("invalid callchain context: "
2363 "%"PRId64"\n", (s64) ip);
2364 /*
2365 * It seems the callchain is corrupted.
2366 * Discard all.
2367 */
2368 callchain_cursor_reset(cursor);
2369 err = 1;
2370 goto out;
2371 }
2372 goto out;
2373 }
2374 thread__find_symbol(thread, *cpumode, ip, &al);
2375 }
2376
2377 if (al.sym != NULL) {
2378 if (perf_hpp_list.parent && !*parent &&
2379 symbol__match_regex(al.sym, &parent_regex))
2380 *parent = al.sym;
2381 else if (have_ignore_callees && root_al &&
2382 symbol__match_regex(al.sym, &ignore_callees_regex)) {
2383 /* Treat this symbol as the root,
2384 forgetting its callees. */
2385 addr_location__copy(root_al, &al);
2386 callchain_cursor_reset(cursor);
2387 }
2388 }
2389
2390 if (symbol_conf.hide_unresolved && al.sym == NULL)
2391 goto out;
2392
2393 if (iter) {
2394 nr_loop_iter = iter->nr_loop_iter;
2395 iter_cycles = iter->cycles;
2396 }
2397
2398 ms.maps = maps__get(al.maps);
2399 ms.map = map__get(al.map);
2400 ms.sym = al.sym;
2401 srcline = callchain_srcline(&ms, al.addr);
2402 err = callchain_cursor_append(cursor, ip, &ms,
2403 branch, flags, nr_loop_iter,
2404 iter_cycles, branch_from, srcline);
2405out:
2406 addr_location__exit(&al);
2407 map_symbol__exit(&ms);
2408 return err;
2409}
2410
2411struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2412 struct addr_location *al)
2413{
2414 unsigned int i;
2415 const struct branch_stack *bs = sample->branch_stack;
2416 struct branch_entry *entries = perf_sample__branch_entries(sample);
2417 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2418
2419 if (!bi)
2420 return NULL;
2421
2422 for (i = 0; i < bs->nr; i++) {
2423 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2424 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2425 bi[i].flags = entries[i].flags;
2426 }
2427 return bi;
2428}
2429
2430static void save_iterations(struct iterations *iter,
2431 struct branch_entry *be, int nr)
2432{
2433 int i;
2434
2435 iter->nr_loop_iter++;
2436 iter->cycles = 0;
2437
2438 for (i = 0; i < nr; i++)
2439 iter->cycles += be[i].flags.cycles;
2440}
2441
2442#define CHASHSZ 127
2443#define CHASHBITS 7
2444#define NO_ENTRY 0xff
2445
2446#define PERF_MAX_BRANCH_DEPTH 127
2447
2448/* Remove loops. */
2449static int remove_loops(struct branch_entry *l, int nr,
2450 struct iterations *iter)
2451{
2452 int i, j, off;
2453 unsigned char chash[CHASHSZ];
2454
2455 memset(chash, NO_ENTRY, sizeof(chash));
2456
2457 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2458
2459 for (i = 0; i < nr; i++) {
2460 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2461
2462 /* no collision handling for now */
2463 if (chash[h] == NO_ENTRY) {
2464 chash[h] = i;
2465 } else if (l[chash[h]].from == l[i].from) {
2466 bool is_loop = true;
2467 /* check if it is a real loop */
2468 off = 0;
2469 for (j = chash[h]; j < i && i + off < nr; j++, off++)
2470 if (l[j].from != l[i + off].from) {
2471 is_loop = false;
2472 break;
2473 }
2474 if (is_loop) {
2475 j = nr - (i + off);
2476 if (j > 0) {
2477 save_iterations(iter + i + off,
2478 l + i, off);
2479
2480 memmove(iter + i, iter + i + off,
2481 j * sizeof(*iter));
2482
2483 memmove(l + i, l + i + off,
2484 j * sizeof(*l));
2485 }
2486
2487 nr -= off;
2488 }
2489 }
2490 }
2491 return nr;
2492}
2493
2494static int lbr_callchain_add_kernel_ip(struct thread *thread,
2495 struct callchain_cursor *cursor,
2496 struct perf_sample *sample,
2497 struct symbol **parent,
2498 struct addr_location *root_al,
2499 u64 branch_from,
2500 bool callee, int end)
2501{
2502 struct ip_callchain *chain = sample->callchain;
2503 u8 cpumode = PERF_RECORD_MISC_USER;
2504 int err, i;
2505
2506 if (callee) {
2507 for (i = 0; i < end + 1; i++) {
2508 err = add_callchain_ip(thread, cursor, parent,
2509 root_al, &cpumode, chain->ips[i],
2510 false, NULL, NULL, branch_from);
2511 if (err)
2512 return err;
2513 }
2514 return 0;
2515 }
2516
2517 for (i = end; i >= 0; i--) {
2518 err = add_callchain_ip(thread, cursor, parent,
2519 root_al, &cpumode, chain->ips[i],
2520 false, NULL, NULL, branch_from);
2521 if (err)
2522 return err;
2523 }
2524
2525 return 0;
2526}
2527
2528static void save_lbr_cursor_node(struct thread *thread,
2529 struct callchain_cursor *cursor,
2530 int idx)
2531{
2532 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2533
2534 if (!lbr_stitch)
2535 return;
2536
2537 if (cursor->pos == cursor->nr) {
2538 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2539 return;
2540 }
2541
2542 if (!cursor->curr)
2543 cursor->curr = cursor->first;
2544 else
2545 cursor->curr = cursor->curr->next;
2546 memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2547 sizeof(struct callchain_cursor_node));
2548
2549 lbr_stitch->prev_lbr_cursor[idx].valid = true;
2550 cursor->pos++;
2551}
2552
2553static int lbr_callchain_add_lbr_ip(struct thread *thread,
2554 struct callchain_cursor *cursor,
2555 struct perf_sample *sample,
2556 struct symbol **parent,
2557 struct addr_location *root_al,
2558 u64 *branch_from,
2559 bool callee)
2560{
2561 struct branch_stack *lbr_stack = sample->branch_stack;
2562 struct branch_entry *entries = perf_sample__branch_entries(sample);
2563 u8 cpumode = PERF_RECORD_MISC_USER;
2564 int lbr_nr = lbr_stack->nr;
2565 struct branch_flags *flags;
2566 int err, i;
2567 u64 ip;
2568
2569 /*
2570 * The curr and pos are not used in writing session. They are cleared
2571 * in callchain_cursor_commit() when the writing session is closed.
2572 * Using curr and pos to track the current cursor node.
2573 */
2574 if (thread__lbr_stitch(thread)) {
2575 cursor->curr = NULL;
2576 cursor->pos = cursor->nr;
2577 if (cursor->nr) {
2578 cursor->curr = cursor->first;
2579 for (i = 0; i < (int)(cursor->nr - 1); i++)
2580 cursor->curr = cursor->curr->next;
2581 }
2582 }
2583
2584 if (callee) {
2585 /* Add LBR ip from first entries.to */
2586 ip = entries[0].to;
2587 flags = &entries[0].flags;
2588 *branch_from = entries[0].from;
2589 err = add_callchain_ip(thread, cursor, parent,
2590 root_al, &cpumode, ip,
2591 true, flags, NULL,
2592 *branch_from);
2593 if (err)
2594 return err;
2595
2596 /*
2597 * The number of cursor node increases.
2598 * Move the current cursor node.
2599 * But does not need to save current cursor node for entry 0.
2600 * It's impossible to stitch the whole LBRs of previous sample.
2601 */
2602 if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2603 if (!cursor->curr)
2604 cursor->curr = cursor->first;
2605 else
2606 cursor->curr = cursor->curr->next;
2607 cursor->pos++;
2608 }
2609
2610 /* Add LBR ip from entries.from one by one. */
2611 for (i = 0; i < lbr_nr; i++) {
2612 ip = entries[i].from;
2613 flags = &entries[i].flags;
2614 err = add_callchain_ip(thread, cursor, parent,
2615 root_al, &cpumode, ip,
2616 true, flags, NULL,
2617 *branch_from);
2618 if (err)
2619 return err;
2620 save_lbr_cursor_node(thread, cursor, i);
2621 }
2622 return 0;
2623 }
2624
2625 /* Add LBR ip from entries.from one by one. */
2626 for (i = lbr_nr - 1; i >= 0; i--) {
2627 ip = entries[i].from;
2628 flags = &entries[i].flags;
2629 err = add_callchain_ip(thread, cursor, parent,
2630 root_al, &cpumode, ip,
2631 true, flags, NULL,
2632 *branch_from);
2633 if (err)
2634 return err;
2635 save_lbr_cursor_node(thread, cursor, i);
2636 }
2637
2638 if (lbr_nr > 0) {
2639 /* Add LBR ip from first entries.to */
2640 ip = entries[0].to;
2641 flags = &entries[0].flags;
2642 *branch_from = entries[0].from;
2643 err = add_callchain_ip(thread, cursor, parent,
2644 root_al, &cpumode, ip,
2645 true, flags, NULL,
2646 *branch_from);
2647 if (err)
2648 return err;
2649 }
2650
2651 return 0;
2652}
2653
2654static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2655 struct callchain_cursor *cursor)
2656{
2657 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2658 struct callchain_cursor_node *cnode;
2659 struct stitch_list *stitch_node;
2660 int err;
2661
2662 list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2663 cnode = &stitch_node->cursor;
2664
2665 err = callchain_cursor_append(cursor, cnode->ip,
2666 &cnode->ms,
2667 cnode->branch,
2668 &cnode->branch_flags,
2669 cnode->nr_loop_iter,
2670 cnode->iter_cycles,
2671 cnode->branch_from,
2672 cnode->srcline);
2673 if (err)
2674 return err;
2675 }
2676 return 0;
2677}
2678
2679static struct stitch_list *get_stitch_node(struct thread *thread)
2680{
2681 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2682 struct stitch_list *stitch_node;
2683
2684 if (!list_empty(&lbr_stitch->free_lists)) {
2685 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2686 struct stitch_list, node);
2687 list_del(&stitch_node->node);
2688
2689 return stitch_node;
2690 }
2691
2692 return malloc(sizeof(struct stitch_list));
2693}
2694
2695static bool has_stitched_lbr(struct thread *thread,
2696 struct perf_sample *cur,
2697 struct perf_sample *prev,
2698 unsigned int max_lbr,
2699 bool callee)
2700{
2701 struct branch_stack *cur_stack = cur->branch_stack;
2702 struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2703 struct branch_stack *prev_stack = prev->branch_stack;
2704 struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2705 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2706 int i, j, nr_identical_branches = 0;
2707 struct stitch_list *stitch_node;
2708 u64 cur_base, distance;
2709
2710 if (!cur_stack || !prev_stack)
2711 return false;
2712
2713 /* Find the physical index of the base-of-stack for current sample. */
2714 cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2715
2716 distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2717 (max_lbr + prev_stack->hw_idx - cur_base);
2718 /* Previous sample has shorter stack. Nothing can be stitched. */
2719 if (distance + 1 > prev_stack->nr)
2720 return false;
2721
2722 /*
2723 * Check if there are identical LBRs between two samples.
2724 * Identical LBRs must have same from, to and flags values. Also,
2725 * they have to be saved in the same LBR registers (same physical
2726 * index).
2727 *
2728 * Starts from the base-of-stack of current sample.
2729 */
2730 for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2731 if ((prev_entries[i].from != cur_entries[j].from) ||
2732 (prev_entries[i].to != cur_entries[j].to) ||
2733 (prev_entries[i].flags.value != cur_entries[j].flags.value))
2734 break;
2735 nr_identical_branches++;
2736 }
2737
2738 if (!nr_identical_branches)
2739 return false;
2740
2741 /*
2742 * Save the LBRs between the base-of-stack of previous sample
2743 * and the base-of-stack of current sample into lbr_stitch->lists.
2744 * These LBRs will be stitched later.
2745 */
2746 for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2747
2748 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2749 continue;
2750
2751 stitch_node = get_stitch_node(thread);
2752 if (!stitch_node)
2753 return false;
2754
2755 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2756 sizeof(struct callchain_cursor_node));
2757
2758 if (callee)
2759 list_add(&stitch_node->node, &lbr_stitch->lists);
2760 else
2761 list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2762 }
2763
2764 return true;
2765}
2766
2767static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2768{
2769 if (thread__lbr_stitch(thread))
2770 return true;
2771
2772 thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2773 if (!thread__lbr_stitch(thread))
2774 goto err;
2775
2776 thread__lbr_stitch(thread)->prev_lbr_cursor =
2777 calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2778 if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2779 goto free_lbr_stitch;
2780
2781 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2782 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2783
2784 return true;
2785
2786free_lbr_stitch:
2787 free(thread__lbr_stitch(thread));
2788 thread__set_lbr_stitch(thread, NULL);
2789err:
2790 pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2791 thread__set_lbr_stitch_enable(thread, false);
2792 return false;
2793}
2794
2795/*
2796 * Resolve LBR callstack chain sample
2797 * Return:
2798 * 1 on success get LBR callchain information
2799 * 0 no available LBR callchain information, should try fp
2800 * negative error code on other errors.
2801 */
2802static int resolve_lbr_callchain_sample(struct thread *thread,
2803 struct callchain_cursor *cursor,
2804 struct perf_sample *sample,
2805 struct symbol **parent,
2806 struct addr_location *root_al,
2807 int max_stack,
2808 unsigned int max_lbr)
2809{
2810 bool callee = (callchain_param.order == ORDER_CALLEE);
2811 struct ip_callchain *chain = sample->callchain;
2812 int chain_nr = min(max_stack, (int)chain->nr), i;
2813 struct lbr_stitch *lbr_stitch;
2814 bool stitched_lbr = false;
2815 u64 branch_from = 0;
2816 int err;
2817
2818 for (i = 0; i < chain_nr; i++) {
2819 if (chain->ips[i] == PERF_CONTEXT_USER)
2820 break;
2821 }
2822
2823 /* LBR only affects the user callchain */
2824 if (i == chain_nr)
2825 return 0;
2826
2827 if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2828 (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2829 lbr_stitch = thread__lbr_stitch(thread);
2830
2831 stitched_lbr = has_stitched_lbr(thread, sample,
2832 &lbr_stitch->prev_sample,
2833 max_lbr, callee);
2834
2835 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2836 list_replace_init(&lbr_stitch->lists,
2837 &lbr_stitch->free_lists);
2838 }
2839 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2840 }
2841
2842 if (callee) {
2843 /* Add kernel ip */
2844 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2845 parent, root_al, branch_from,
2846 true, i);
2847 if (err)
2848 goto error;
2849
2850 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2851 root_al, &branch_from, true);
2852 if (err)
2853 goto error;
2854
2855 if (stitched_lbr) {
2856 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2857 if (err)
2858 goto error;
2859 }
2860
2861 } else {
2862 if (stitched_lbr) {
2863 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2864 if (err)
2865 goto error;
2866 }
2867 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2868 root_al, &branch_from, false);
2869 if (err)
2870 goto error;
2871
2872 /* Add kernel ip */
2873 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2874 parent, root_al, branch_from,
2875 false, i);
2876 if (err)
2877 goto error;
2878 }
2879 return 1;
2880
2881error:
2882 return (err < 0) ? err : 0;
2883}
2884
2885static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2886 struct callchain_cursor *cursor,
2887 struct symbol **parent,
2888 struct addr_location *root_al,
2889 u8 *cpumode, int ent)
2890{
2891 int err = 0;
2892
2893 while (--ent >= 0) {
2894 u64 ip = chain->ips[ent];
2895
2896 if (ip >= PERF_CONTEXT_MAX) {
2897 err = add_callchain_ip(thread, cursor, parent,
2898 root_al, cpumode, ip,
2899 false, NULL, NULL, 0);
2900 break;
2901 }
2902 }
2903 return err;
2904}
2905
2906static u64 get_leaf_frame_caller(struct perf_sample *sample,
2907 struct thread *thread, int usr_idx)
2908{
2909 if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2910 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2911 else
2912 return 0;
2913}
2914
2915static int thread__resolve_callchain_sample(struct thread *thread,
2916 struct callchain_cursor *cursor,
2917 struct evsel *evsel,
2918 struct perf_sample *sample,
2919 struct symbol **parent,
2920 struct addr_location *root_al,
2921 int max_stack)
2922{
2923 struct branch_stack *branch = sample->branch_stack;
2924 struct branch_entry *entries = perf_sample__branch_entries(sample);
2925 struct ip_callchain *chain = sample->callchain;
2926 int chain_nr = 0;
2927 u8 cpumode = PERF_RECORD_MISC_USER;
2928 int i, j, err, nr_entries, usr_idx;
2929 int skip_idx = -1;
2930 int first_call = 0;
2931 u64 leaf_frame_caller;
2932
2933 if (chain)
2934 chain_nr = chain->nr;
2935
2936 if (evsel__has_branch_callstack(evsel)) {
2937 struct perf_env *env = evsel__env(evsel);
2938
2939 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2940 root_al, max_stack,
2941 !env ? 0 : env->max_branches);
2942 if (err)
2943 return (err < 0) ? err : 0;
2944 }
2945
2946 /*
2947 * Based on DWARF debug information, some architectures skip
2948 * a callchain entry saved by the kernel.
2949 */
2950 skip_idx = arch_skip_callchain_idx(thread, chain);
2951
2952 /*
2953 * Add branches to call stack for easier browsing. This gives
2954 * more context for a sample than just the callers.
2955 *
2956 * This uses individual histograms of paths compared to the
2957 * aggregated histograms the normal LBR mode uses.
2958 *
2959 * Limitations for now:
2960 * - No extra filters
2961 * - No annotations (should annotate somehow)
2962 */
2963
2964 if (branch && callchain_param.branch_callstack) {
2965 int nr = min(max_stack, (int)branch->nr);
2966 struct branch_entry be[nr];
2967 struct iterations iter[nr];
2968
2969 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2970 pr_warning("corrupted branch chain. skipping...\n");
2971 goto check_calls;
2972 }
2973
2974 for (i = 0; i < nr; i++) {
2975 if (callchain_param.order == ORDER_CALLEE) {
2976 be[i] = entries[i];
2977
2978 if (chain == NULL)
2979 continue;
2980
2981 /*
2982 * Check for overlap into the callchain.
2983 * The return address is one off compared to
2984 * the branch entry. To adjust for this
2985 * assume the calling instruction is not longer
2986 * than 8 bytes.
2987 */
2988 if (i == skip_idx ||
2989 chain->ips[first_call] >= PERF_CONTEXT_MAX)
2990 first_call++;
2991 else if (be[i].from < chain->ips[first_call] &&
2992 be[i].from >= chain->ips[first_call] - 8)
2993 first_call++;
2994 } else
2995 be[i] = entries[branch->nr - i - 1];
2996 }
2997
2998 memset(iter, 0, sizeof(struct iterations) * nr);
2999 nr = remove_loops(be, nr, iter);
3000
3001 for (i = 0; i < nr; i++) {
3002 err = add_callchain_ip(thread, cursor, parent,
3003 root_al,
3004 NULL, be[i].to,
3005 true, &be[i].flags,
3006 NULL, be[i].from);
3007
3008 if (!err)
3009 err = add_callchain_ip(thread, cursor, parent, root_al,
3010 NULL, be[i].from,
3011 true, &be[i].flags,
3012 &iter[i], 0);
3013 if (err == -EINVAL)
3014 break;
3015 if (err)
3016 return err;
3017 }
3018
3019 if (chain_nr == 0)
3020 return 0;
3021
3022 chain_nr -= nr;
3023 }
3024
3025check_calls:
3026 if (chain && callchain_param.order != ORDER_CALLEE) {
3027 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
3028 &cpumode, chain->nr - first_call);
3029 if (err)
3030 return (err < 0) ? err : 0;
3031 }
3032 for (i = first_call, nr_entries = 0;
3033 i < chain_nr && nr_entries < max_stack; i++) {
3034 u64 ip;
3035
3036 if (callchain_param.order == ORDER_CALLEE)
3037 j = i;
3038 else
3039 j = chain->nr - i - 1;
3040
3041#ifdef HAVE_SKIP_CALLCHAIN_IDX
3042 if (j == skip_idx)
3043 continue;
3044#endif
3045 ip = chain->ips[j];
3046 if (ip < PERF_CONTEXT_MAX)
3047 ++nr_entries;
3048 else if (callchain_param.order != ORDER_CALLEE) {
3049 err = find_prev_cpumode(chain, thread, cursor, parent,
3050 root_al, &cpumode, j);
3051 if (err)
3052 return (err < 0) ? err : 0;
3053 continue;
3054 }
3055
3056 /*
3057 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
3058 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
3059 * the index will be different in order to add the missing frame
3060 * at the right place.
3061 */
3062
3063 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
3064
3065 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
3066
3067 leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
3068
3069 /*
3070 * check if leaf_frame_Caller != ip to not add the same
3071 * value twice.
3072 */
3073
3074 if (leaf_frame_caller && leaf_frame_caller != ip) {
3075
3076 err = add_callchain_ip(thread, cursor, parent,
3077 root_al, &cpumode, leaf_frame_caller,
3078 false, NULL, NULL, 0);
3079 if (err)
3080 return (err < 0) ? err : 0;
3081 }
3082 }
3083
3084 err = add_callchain_ip(thread, cursor, parent,
3085 root_al, &cpumode, ip,
3086 false, NULL, NULL, 0);
3087
3088 if (err)
3089 return (err < 0) ? err : 0;
3090 }
3091
3092 return 0;
3093}
3094
3095static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
3096{
3097 struct symbol *sym = ms->sym;
3098 struct map *map = ms->map;
3099 struct inline_node *inline_node;
3100 struct inline_list *ilist;
3101 struct dso *dso;
3102 u64 addr;
3103 int ret = 1;
3104 struct map_symbol ilist_ms;
3105
3106 if (!symbol_conf.inline_name || !map || !sym)
3107 return ret;
3108
3109 addr = map__dso_map_ip(map, ip);
3110 addr = map__rip_2objdump(map, addr);
3111 dso = map__dso(map);
3112
3113 inline_node = inlines__tree_find(&dso->inlined_nodes, addr);
3114 if (!inline_node) {
3115 inline_node = dso__parse_addr_inlines(dso, addr, sym);
3116 if (!inline_node)
3117 return ret;
3118 inlines__tree_insert(&dso->inlined_nodes, inline_node);
3119 }
3120
3121 ilist_ms = (struct map_symbol) {
3122 .maps = maps__get(ms->maps),
3123 .map = map__get(map),
3124 };
3125 list_for_each_entry(ilist, &inline_node->val, list) {
3126 ilist_ms.sym = ilist->symbol;
3127 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
3128 NULL, 0, 0, 0, ilist->srcline);
3129
3130 if (ret != 0)
3131 return ret;
3132 }
3133 map_symbol__exit(&ilist_ms);
3134
3135 return ret;
3136}
3137
3138static int unwind_entry(struct unwind_entry *entry, void *arg)
3139{
3140 struct callchain_cursor *cursor = arg;
3141 const char *srcline = NULL;
3142 u64 addr = entry->ip;
3143
3144 if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
3145 return 0;
3146
3147 if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
3148 return 0;
3149
3150 /*
3151 * Convert entry->ip from a virtual address to an offset in
3152 * its corresponding binary.
3153 */
3154 if (entry->ms.map)
3155 addr = map__dso_map_ip(entry->ms.map, entry->ip);
3156
3157 srcline = callchain_srcline(&entry->ms, addr);
3158 return callchain_cursor_append(cursor, entry->ip, &entry->ms,
3159 false, NULL, 0, 0, 0, srcline);
3160}
3161
3162static int thread__resolve_callchain_unwind(struct thread *thread,
3163 struct callchain_cursor *cursor,
3164 struct evsel *evsel,
3165 struct perf_sample *sample,
3166 int max_stack)
3167{
3168 /* Can we do dwarf post unwind? */
3169 if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
3170 (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
3171 return 0;
3172
3173 /* Bail out if nothing was captured. */
3174 if ((!sample->user_regs.regs) ||
3175 (!sample->user_stack.size))
3176 return 0;
3177
3178 return unwind__get_entries(unwind_entry, cursor,
3179 thread, sample, max_stack, false);
3180}
3181
3182int thread__resolve_callchain(struct thread *thread,
3183 struct callchain_cursor *cursor,
3184 struct evsel *evsel,
3185 struct perf_sample *sample,
3186 struct symbol **parent,
3187 struct addr_location *root_al,
3188 int max_stack)
3189{
3190 int ret = 0;
3191
3192 if (cursor == NULL)
3193 return -ENOMEM;
3194
3195 callchain_cursor_reset(cursor);
3196
3197 if (callchain_param.order == ORDER_CALLEE) {
3198 ret = thread__resolve_callchain_sample(thread, cursor,
3199 evsel, sample,
3200 parent, root_al,
3201 max_stack);
3202 if (ret)
3203 return ret;
3204 ret = thread__resolve_callchain_unwind(thread, cursor,
3205 evsel, sample,
3206 max_stack);
3207 } else {
3208 ret = thread__resolve_callchain_unwind(thread, cursor,
3209 evsel, sample,
3210 max_stack);
3211 if (ret)
3212 return ret;
3213 ret = thread__resolve_callchain_sample(thread, cursor,
3214 evsel, sample,
3215 parent, root_al,
3216 max_stack);
3217 }
3218
3219 return ret;
3220}
3221
3222int machine__for_each_thread(struct machine *machine,
3223 int (*fn)(struct thread *thread, void *p),
3224 void *priv)
3225{
3226 struct threads *threads;
3227 struct rb_node *nd;
3228 int rc = 0;
3229 int i;
3230
3231 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3232 threads = &machine->threads[i];
3233 for (nd = rb_first_cached(&threads->entries); nd;
3234 nd = rb_next(nd)) {
3235 struct thread_rb_node *trb = rb_entry(nd, struct thread_rb_node, rb_node);
3236
3237 rc = fn(trb->thread, priv);
3238 if (rc != 0)
3239 return rc;
3240 }
3241 }
3242 return rc;
3243}
3244
3245int machines__for_each_thread(struct machines *machines,
3246 int (*fn)(struct thread *thread, void *p),
3247 void *priv)
3248{
3249 struct rb_node *nd;
3250 int rc = 0;
3251
3252 rc = machine__for_each_thread(&machines->host, fn, priv);
3253 if (rc != 0)
3254 return rc;
3255
3256 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3257 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3258
3259 rc = machine__for_each_thread(machine, fn, priv);
3260 if (rc != 0)
3261 return rc;
3262 }
3263 return rc;
3264}
3265
3266pid_t machine__get_current_tid(struct machine *machine, int cpu)
3267{
3268 if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3269 return -1;
3270
3271 return machine->current_tid[cpu];
3272}
3273
3274int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3275 pid_t tid)
3276{
3277 struct thread *thread;
3278 const pid_t init_val = -1;
3279
3280 if (cpu < 0)
3281 return -EINVAL;
3282
3283 if (realloc_array_as_needed(machine->current_tid,
3284 machine->current_tid_sz,
3285 (unsigned int)cpu,
3286 &init_val))
3287 return -ENOMEM;
3288
3289 machine->current_tid[cpu] = tid;
3290
3291 thread = machine__findnew_thread(machine, pid, tid);
3292 if (!thread)
3293 return -ENOMEM;
3294
3295 thread__set_cpu(thread, cpu);
3296 thread__put(thread);
3297
3298 return 0;
3299}
3300
3301/*
3302 * Compares the raw arch string. N.B. see instead perf_env__arch() or
3303 * machine__normalized_is() if a normalized arch is needed.
3304 */
3305bool machine__is(struct machine *machine, const char *arch)
3306{
3307 return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3308}
3309
3310bool machine__normalized_is(struct machine *machine, const char *arch)
3311{
3312 return machine && !strcmp(perf_env__arch(machine->env), arch);
3313}
3314
3315int machine__nr_cpus_avail(struct machine *machine)
3316{
3317 return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3318}
3319
3320int machine__get_kernel_start(struct machine *machine)
3321{
3322 struct map *map = machine__kernel_map(machine);
3323 int err = 0;
3324
3325 /*
3326 * The only addresses above 2^63 are kernel addresses of a 64-bit
3327 * kernel. Note that addresses are unsigned so that on a 32-bit system
3328 * all addresses including kernel addresses are less than 2^32. In
3329 * that case (32-bit system), if the kernel mapping is unknown, all
3330 * addresses will be assumed to be in user space - see
3331 * machine__kernel_ip().
3332 */
3333 machine->kernel_start = 1ULL << 63;
3334 if (map) {
3335 err = map__load(map);
3336 /*
3337 * On x86_64, PTI entry trampolines are less than the
3338 * start of kernel text, but still above 2^63. So leave
3339 * kernel_start = 1ULL << 63 for x86_64.
3340 */
3341 if (!err && !machine__is(machine, "x86_64"))
3342 machine->kernel_start = map__start(map);
3343 }
3344 return err;
3345}
3346
3347u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3348{
3349 u8 addr_cpumode = cpumode;
3350 bool kernel_ip;
3351
3352 if (!machine->single_address_space)
3353 goto out;
3354
3355 kernel_ip = machine__kernel_ip(machine, addr);
3356 switch (cpumode) {
3357 case PERF_RECORD_MISC_KERNEL:
3358 case PERF_RECORD_MISC_USER:
3359 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3360 PERF_RECORD_MISC_USER;
3361 break;
3362 case PERF_RECORD_MISC_GUEST_KERNEL:
3363 case PERF_RECORD_MISC_GUEST_USER:
3364 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3365 PERF_RECORD_MISC_GUEST_USER;
3366 break;
3367 default:
3368 break;
3369 }
3370out:
3371 return addr_cpumode;
3372}
3373
3374struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3375{
3376 return dsos__findnew_id(&machine->dsos, filename, id);
3377}
3378
3379struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3380{
3381 return machine__findnew_dso_id(machine, filename, NULL);
3382}
3383
3384char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3385{
3386 struct machine *machine = vmachine;
3387 struct map *map;
3388 struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3389
3390 if (sym == NULL)
3391 return NULL;
3392
3393 *modp = __map__is_kmodule(map) ? (char *)map__dso(map)->short_name : NULL;
3394 *addrp = map__unmap_ip(map, sym->start);
3395 return sym->name;
3396}
3397
3398int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3399{
3400 struct dso *pos;
3401 int err = 0;
3402
3403 list_for_each_entry(pos, &machine->dsos.head, node) {
3404 if (fn(pos, machine, priv))
3405 err = -1;
3406 }
3407 return err;
3408}
3409
3410int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3411{
3412 struct maps *maps = machine__kernel_maps(machine);
3413
3414 return maps__for_each_map(maps, fn, priv);
3415}
3416
3417bool machine__is_lock_function(struct machine *machine, u64 addr)
3418{
3419 if (!machine->sched.text_start) {
3420 struct map *kmap;
3421 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3422
3423 if (!sym) {
3424 /* to avoid retry */
3425 machine->sched.text_start = 1;
3426 return false;
3427 }
3428
3429 machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3430
3431 /* should not fail from here */
3432 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3433 machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3434
3435 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3436 machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3437
3438 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3439 machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3440 }
3441
3442 /* failed to get kernel symbols */
3443 if (machine->sched.text_start == 1)
3444 return false;
3445
3446 /* mutex and rwsem functions are in sched text section */
3447 if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3448 return true;
3449
3450 /* spinlock functions are in lock text section */
3451 if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3452 return true;
3453
3454 return false;
3455}
1// SPDX-License-Identifier: GPL-2.0
2#include <dirent.h>
3#include <errno.h>
4#include <inttypes.h>
5#include <regex.h>
6#include <stdlib.h>
7#include "callchain.h"
8#include "debug.h"
9#include "dso.h"
10#include "env.h"
11#include "event.h"
12#include "evsel.h"
13#include "hist.h"
14#include "machine.h"
15#include "map.h"
16#include "map_symbol.h"
17#include "branch.h"
18#include "mem-events.h"
19#include "srcline.h"
20#include "symbol.h"
21#include "sort.h"
22#include "strlist.h"
23#include "target.h"
24#include "thread.h"
25#include "util.h"
26#include "vdso.h"
27#include <stdbool.h>
28#include <sys/types.h>
29#include <sys/stat.h>
30#include <unistd.h>
31#include "unwind.h"
32#include "linux/hash.h"
33#include "asm/bug.h"
34#include "bpf-event.h"
35#include <internal/lib.h> // page_size
36#include "cgroup.h"
37
38#include <linux/ctype.h>
39#include <symbol/kallsyms.h>
40#include <linux/mman.h>
41#include <linux/string.h>
42#include <linux/zalloc.h>
43
44static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
45
46static struct dso *machine__kernel_dso(struct machine *machine)
47{
48 return machine->vmlinux_map->dso;
49}
50
51static void dsos__init(struct dsos *dsos)
52{
53 INIT_LIST_HEAD(&dsos->head);
54 dsos->root = RB_ROOT;
55 init_rwsem(&dsos->lock);
56}
57
58static void machine__threads_init(struct machine *machine)
59{
60 int i;
61
62 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
63 struct threads *threads = &machine->threads[i];
64 threads->entries = RB_ROOT_CACHED;
65 init_rwsem(&threads->lock);
66 threads->nr = 0;
67 INIT_LIST_HEAD(&threads->dead);
68 threads->last_match = NULL;
69 }
70}
71
72static int machine__set_mmap_name(struct machine *machine)
73{
74 if (machine__is_host(machine))
75 machine->mmap_name = strdup("[kernel.kallsyms]");
76 else if (machine__is_default_guest(machine))
77 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
78 else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
79 machine->pid) < 0)
80 machine->mmap_name = NULL;
81
82 return machine->mmap_name ? 0 : -ENOMEM;
83}
84
85int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
86{
87 int err = -ENOMEM;
88
89 memset(machine, 0, sizeof(*machine));
90 maps__init(&machine->kmaps, machine);
91 RB_CLEAR_NODE(&machine->rb_node);
92 dsos__init(&machine->dsos);
93
94 machine__threads_init(machine);
95
96 machine->vdso_info = NULL;
97 machine->env = NULL;
98
99 machine->pid = pid;
100
101 machine->id_hdr_size = 0;
102 machine->kptr_restrict_warned = false;
103 machine->comm_exec = false;
104 machine->kernel_start = 0;
105 machine->vmlinux_map = NULL;
106
107 machine->root_dir = strdup(root_dir);
108 if (machine->root_dir == NULL)
109 return -ENOMEM;
110
111 if (machine__set_mmap_name(machine))
112 goto out;
113
114 if (pid != HOST_KERNEL_ID) {
115 struct thread *thread = machine__findnew_thread(machine, -1,
116 pid);
117 char comm[64];
118
119 if (thread == NULL)
120 goto out;
121
122 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
123 thread__set_comm(thread, comm, 0);
124 thread__put(thread);
125 }
126
127 machine->current_tid = NULL;
128 err = 0;
129
130out:
131 if (err) {
132 zfree(&machine->root_dir);
133 zfree(&machine->mmap_name);
134 }
135 return 0;
136}
137
138struct machine *machine__new_host(void)
139{
140 struct machine *machine = malloc(sizeof(*machine));
141
142 if (machine != NULL) {
143 machine__init(machine, "", HOST_KERNEL_ID);
144
145 if (machine__create_kernel_maps(machine) < 0)
146 goto out_delete;
147 }
148
149 return machine;
150out_delete:
151 free(machine);
152 return NULL;
153}
154
155struct machine *machine__new_kallsyms(void)
156{
157 struct machine *machine = machine__new_host();
158 /*
159 * FIXME:
160 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
161 * ask for not using the kcore parsing code, once this one is fixed
162 * to create a map per module.
163 */
164 if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
165 machine__delete(machine);
166 machine = NULL;
167 }
168
169 return machine;
170}
171
172static void dsos__purge(struct dsos *dsos)
173{
174 struct dso *pos, *n;
175
176 down_write(&dsos->lock);
177
178 list_for_each_entry_safe(pos, n, &dsos->head, node) {
179 RB_CLEAR_NODE(&pos->rb_node);
180 pos->root = NULL;
181 list_del_init(&pos->node);
182 dso__put(pos);
183 }
184
185 up_write(&dsos->lock);
186}
187
188static void dsos__exit(struct dsos *dsos)
189{
190 dsos__purge(dsos);
191 exit_rwsem(&dsos->lock);
192}
193
194void machine__delete_threads(struct machine *machine)
195{
196 struct rb_node *nd;
197 int i;
198
199 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
200 struct threads *threads = &machine->threads[i];
201 down_write(&threads->lock);
202 nd = rb_first_cached(&threads->entries);
203 while (nd) {
204 struct thread *t = rb_entry(nd, struct thread, rb_node);
205
206 nd = rb_next(nd);
207 __machine__remove_thread(machine, t, false);
208 }
209 up_write(&threads->lock);
210 }
211}
212
213void machine__exit(struct machine *machine)
214{
215 int i;
216
217 if (machine == NULL)
218 return;
219
220 machine__destroy_kernel_maps(machine);
221 maps__exit(&machine->kmaps);
222 dsos__exit(&machine->dsos);
223 machine__exit_vdso(machine);
224 zfree(&machine->root_dir);
225 zfree(&machine->mmap_name);
226 zfree(&machine->current_tid);
227
228 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
229 struct threads *threads = &machine->threads[i];
230 struct thread *thread, *n;
231 /*
232 * Forget about the dead, at this point whatever threads were
233 * left in the dead lists better have a reference count taken
234 * by who is using them, and then, when they drop those references
235 * and it finally hits zero, thread__put() will check and see that
236 * its not in the dead threads list and will not try to remove it
237 * from there, just calling thread__delete() straight away.
238 */
239 list_for_each_entry_safe(thread, n, &threads->dead, node)
240 list_del_init(&thread->node);
241
242 exit_rwsem(&threads->lock);
243 }
244}
245
246void machine__delete(struct machine *machine)
247{
248 if (machine) {
249 machine__exit(machine);
250 free(machine);
251 }
252}
253
254void machines__init(struct machines *machines)
255{
256 machine__init(&machines->host, "", HOST_KERNEL_ID);
257 machines->guests = RB_ROOT_CACHED;
258}
259
260void machines__exit(struct machines *machines)
261{
262 machine__exit(&machines->host);
263 /* XXX exit guest */
264}
265
266struct machine *machines__add(struct machines *machines, pid_t pid,
267 const char *root_dir)
268{
269 struct rb_node **p = &machines->guests.rb_root.rb_node;
270 struct rb_node *parent = NULL;
271 struct machine *pos, *machine = malloc(sizeof(*machine));
272 bool leftmost = true;
273
274 if (machine == NULL)
275 return NULL;
276
277 if (machine__init(machine, root_dir, pid) != 0) {
278 free(machine);
279 return NULL;
280 }
281
282 while (*p != NULL) {
283 parent = *p;
284 pos = rb_entry(parent, struct machine, rb_node);
285 if (pid < pos->pid)
286 p = &(*p)->rb_left;
287 else {
288 p = &(*p)->rb_right;
289 leftmost = false;
290 }
291 }
292
293 rb_link_node(&machine->rb_node, parent, p);
294 rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
295
296 return machine;
297}
298
299void machines__set_comm_exec(struct machines *machines, bool comm_exec)
300{
301 struct rb_node *nd;
302
303 machines->host.comm_exec = comm_exec;
304
305 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
306 struct machine *machine = rb_entry(nd, struct machine, rb_node);
307
308 machine->comm_exec = comm_exec;
309 }
310}
311
312struct machine *machines__find(struct machines *machines, pid_t pid)
313{
314 struct rb_node **p = &machines->guests.rb_root.rb_node;
315 struct rb_node *parent = NULL;
316 struct machine *machine;
317 struct machine *default_machine = NULL;
318
319 if (pid == HOST_KERNEL_ID)
320 return &machines->host;
321
322 while (*p != NULL) {
323 parent = *p;
324 machine = rb_entry(parent, struct machine, rb_node);
325 if (pid < machine->pid)
326 p = &(*p)->rb_left;
327 else if (pid > machine->pid)
328 p = &(*p)->rb_right;
329 else
330 return machine;
331 if (!machine->pid)
332 default_machine = machine;
333 }
334
335 return default_machine;
336}
337
338struct machine *machines__findnew(struct machines *machines, pid_t pid)
339{
340 char path[PATH_MAX];
341 const char *root_dir = "";
342 struct machine *machine = machines__find(machines, pid);
343
344 if (machine && (machine->pid == pid))
345 goto out;
346
347 if ((pid != HOST_KERNEL_ID) &&
348 (pid != DEFAULT_GUEST_KERNEL_ID) &&
349 (symbol_conf.guestmount)) {
350 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
351 if (access(path, R_OK)) {
352 static struct strlist *seen;
353
354 if (!seen)
355 seen = strlist__new(NULL, NULL);
356
357 if (!strlist__has_entry(seen, path)) {
358 pr_err("Can't access file %s\n", path);
359 strlist__add(seen, path);
360 }
361 machine = NULL;
362 goto out;
363 }
364 root_dir = path;
365 }
366
367 machine = machines__add(machines, pid, root_dir);
368out:
369 return machine;
370}
371
372void machines__process_guests(struct machines *machines,
373 machine__process_t process, void *data)
374{
375 struct rb_node *nd;
376
377 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
378 struct machine *pos = rb_entry(nd, struct machine, rb_node);
379 process(pos, data);
380 }
381}
382
383void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
384{
385 struct rb_node *node;
386 struct machine *machine;
387
388 machines->host.id_hdr_size = id_hdr_size;
389
390 for (node = rb_first_cached(&machines->guests); node;
391 node = rb_next(node)) {
392 machine = rb_entry(node, struct machine, rb_node);
393 machine->id_hdr_size = id_hdr_size;
394 }
395
396 return;
397}
398
399static void machine__update_thread_pid(struct machine *machine,
400 struct thread *th, pid_t pid)
401{
402 struct thread *leader;
403
404 if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
405 return;
406
407 th->pid_ = pid;
408
409 if (th->pid_ == th->tid)
410 return;
411
412 leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
413 if (!leader)
414 goto out_err;
415
416 if (!leader->maps)
417 leader->maps = maps__new(machine);
418
419 if (!leader->maps)
420 goto out_err;
421
422 if (th->maps == leader->maps)
423 return;
424
425 if (th->maps) {
426 /*
427 * Maps are created from MMAP events which provide the pid and
428 * tid. Consequently there never should be any maps on a thread
429 * with an unknown pid. Just print an error if there are.
430 */
431 if (!maps__empty(th->maps))
432 pr_err("Discarding thread maps for %d:%d\n",
433 th->pid_, th->tid);
434 maps__put(th->maps);
435 }
436
437 th->maps = maps__get(leader->maps);
438out_put:
439 thread__put(leader);
440 return;
441out_err:
442 pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
443 goto out_put;
444}
445
446/*
447 * Front-end cache - TID lookups come in blocks,
448 * so most of the time we dont have to look up
449 * the full rbtree:
450 */
451static struct thread*
452__threads__get_last_match(struct threads *threads, struct machine *machine,
453 int pid, int tid)
454{
455 struct thread *th;
456
457 th = threads->last_match;
458 if (th != NULL) {
459 if (th->tid == tid) {
460 machine__update_thread_pid(machine, th, pid);
461 return thread__get(th);
462 }
463
464 threads->last_match = NULL;
465 }
466
467 return NULL;
468}
469
470static struct thread*
471threads__get_last_match(struct threads *threads, struct machine *machine,
472 int pid, int tid)
473{
474 struct thread *th = NULL;
475
476 if (perf_singlethreaded)
477 th = __threads__get_last_match(threads, machine, pid, tid);
478
479 return th;
480}
481
482static void
483__threads__set_last_match(struct threads *threads, struct thread *th)
484{
485 threads->last_match = th;
486}
487
488static void
489threads__set_last_match(struct threads *threads, struct thread *th)
490{
491 if (perf_singlethreaded)
492 __threads__set_last_match(threads, th);
493}
494
495/*
496 * Caller must eventually drop thread->refcnt returned with a successful
497 * lookup/new thread inserted.
498 */
499static struct thread *____machine__findnew_thread(struct machine *machine,
500 struct threads *threads,
501 pid_t pid, pid_t tid,
502 bool create)
503{
504 struct rb_node **p = &threads->entries.rb_root.rb_node;
505 struct rb_node *parent = NULL;
506 struct thread *th;
507 bool leftmost = true;
508
509 th = threads__get_last_match(threads, machine, pid, tid);
510 if (th)
511 return th;
512
513 while (*p != NULL) {
514 parent = *p;
515 th = rb_entry(parent, struct thread, rb_node);
516
517 if (th->tid == tid) {
518 threads__set_last_match(threads, th);
519 machine__update_thread_pid(machine, th, pid);
520 return thread__get(th);
521 }
522
523 if (tid < th->tid)
524 p = &(*p)->rb_left;
525 else {
526 p = &(*p)->rb_right;
527 leftmost = false;
528 }
529 }
530
531 if (!create)
532 return NULL;
533
534 th = thread__new(pid, tid);
535 if (th != NULL) {
536 rb_link_node(&th->rb_node, parent, p);
537 rb_insert_color_cached(&th->rb_node, &threads->entries, leftmost);
538
539 /*
540 * We have to initialize maps separately after rb tree is updated.
541 *
542 * The reason is that we call machine__findnew_thread
543 * within thread__init_maps to find the thread
544 * leader and that would screwed the rb tree.
545 */
546 if (thread__init_maps(th, machine)) {
547 rb_erase_cached(&th->rb_node, &threads->entries);
548 RB_CLEAR_NODE(&th->rb_node);
549 thread__put(th);
550 return NULL;
551 }
552 /*
553 * It is now in the rbtree, get a ref
554 */
555 thread__get(th);
556 threads__set_last_match(threads, th);
557 ++threads->nr;
558 }
559
560 return th;
561}
562
563struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
564{
565 return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
566}
567
568struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
569 pid_t tid)
570{
571 struct threads *threads = machine__threads(machine, tid);
572 struct thread *th;
573
574 down_write(&threads->lock);
575 th = __machine__findnew_thread(machine, pid, tid);
576 up_write(&threads->lock);
577 return th;
578}
579
580struct thread *machine__find_thread(struct machine *machine, pid_t pid,
581 pid_t tid)
582{
583 struct threads *threads = machine__threads(machine, tid);
584 struct thread *th;
585
586 down_read(&threads->lock);
587 th = ____machine__findnew_thread(machine, threads, pid, tid, false);
588 up_read(&threads->lock);
589 return th;
590}
591
592struct comm *machine__thread_exec_comm(struct machine *machine,
593 struct thread *thread)
594{
595 if (machine->comm_exec)
596 return thread__exec_comm(thread);
597 else
598 return thread__comm(thread);
599}
600
601int machine__process_comm_event(struct machine *machine, union perf_event *event,
602 struct perf_sample *sample)
603{
604 struct thread *thread = machine__findnew_thread(machine,
605 event->comm.pid,
606 event->comm.tid);
607 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
608 int err = 0;
609
610 if (exec)
611 machine->comm_exec = true;
612
613 if (dump_trace)
614 perf_event__fprintf_comm(event, stdout);
615
616 if (thread == NULL ||
617 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
618 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
619 err = -1;
620 }
621
622 thread__put(thread);
623
624 return err;
625}
626
627int machine__process_namespaces_event(struct machine *machine __maybe_unused,
628 union perf_event *event,
629 struct perf_sample *sample __maybe_unused)
630{
631 struct thread *thread = machine__findnew_thread(machine,
632 event->namespaces.pid,
633 event->namespaces.tid);
634 int err = 0;
635
636 WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
637 "\nWARNING: kernel seems to support more namespaces than perf"
638 " tool.\nTry updating the perf tool..\n\n");
639
640 WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
641 "\nWARNING: perf tool seems to support more namespaces than"
642 " the kernel.\nTry updating the kernel..\n\n");
643
644 if (dump_trace)
645 perf_event__fprintf_namespaces(event, stdout);
646
647 if (thread == NULL ||
648 thread__set_namespaces(thread, sample->time, &event->namespaces)) {
649 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
650 err = -1;
651 }
652
653 thread__put(thread);
654
655 return err;
656}
657
658int machine__process_cgroup_event(struct machine *machine,
659 union perf_event *event,
660 struct perf_sample *sample __maybe_unused)
661{
662 struct cgroup *cgrp;
663
664 if (dump_trace)
665 perf_event__fprintf_cgroup(event, stdout);
666
667 cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
668 if (cgrp == NULL)
669 return -ENOMEM;
670
671 return 0;
672}
673
674int machine__process_lost_event(struct machine *machine __maybe_unused,
675 union perf_event *event, struct perf_sample *sample __maybe_unused)
676{
677 dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
678 event->lost.id, event->lost.lost);
679 return 0;
680}
681
682int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
683 union perf_event *event, struct perf_sample *sample)
684{
685 dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
686 sample->id, event->lost_samples.lost);
687 return 0;
688}
689
690static struct dso *machine__findnew_module_dso(struct machine *machine,
691 struct kmod_path *m,
692 const char *filename)
693{
694 struct dso *dso;
695
696 down_write(&machine->dsos.lock);
697
698 dso = __dsos__find(&machine->dsos, m->name, true);
699 if (!dso) {
700 dso = __dsos__addnew(&machine->dsos, m->name);
701 if (dso == NULL)
702 goto out_unlock;
703
704 dso__set_module_info(dso, m, machine);
705 dso__set_long_name(dso, strdup(filename), true);
706 dso->kernel = DSO_SPACE__KERNEL;
707 }
708
709 dso__get(dso);
710out_unlock:
711 up_write(&machine->dsos.lock);
712 return dso;
713}
714
715int machine__process_aux_event(struct machine *machine __maybe_unused,
716 union perf_event *event)
717{
718 if (dump_trace)
719 perf_event__fprintf_aux(event, stdout);
720 return 0;
721}
722
723int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
724 union perf_event *event)
725{
726 if (dump_trace)
727 perf_event__fprintf_itrace_start(event, stdout);
728 return 0;
729}
730
731int machine__process_switch_event(struct machine *machine __maybe_unused,
732 union perf_event *event)
733{
734 if (dump_trace)
735 perf_event__fprintf_switch(event, stdout);
736 return 0;
737}
738
739static int machine__process_ksymbol_register(struct machine *machine,
740 union perf_event *event,
741 struct perf_sample *sample __maybe_unused)
742{
743 struct symbol *sym;
744 struct map *map = maps__find(&machine->kmaps, event->ksymbol.addr);
745
746 if (!map) {
747 struct dso *dso = dso__new(event->ksymbol.name);
748
749 if (dso) {
750 dso->kernel = DSO_SPACE__KERNEL;
751 map = map__new2(0, dso);
752 }
753
754 if (!dso || !map) {
755 dso__put(dso);
756 return -ENOMEM;
757 }
758
759 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
760 map->dso->binary_type = DSO_BINARY_TYPE__OOL;
761 map->dso->data.file_size = event->ksymbol.len;
762 dso__set_loaded(map->dso);
763 }
764
765 map->start = event->ksymbol.addr;
766 map->end = map->start + event->ksymbol.len;
767 maps__insert(&machine->kmaps, map);
768 dso__set_loaded(dso);
769
770 if (is_bpf_image(event->ksymbol.name)) {
771 dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
772 dso__set_long_name(dso, "", false);
773 }
774 }
775
776 sym = symbol__new(map->map_ip(map, map->start),
777 event->ksymbol.len,
778 0, 0, event->ksymbol.name);
779 if (!sym)
780 return -ENOMEM;
781 dso__insert_symbol(map->dso, sym);
782 return 0;
783}
784
785static int machine__process_ksymbol_unregister(struct machine *machine,
786 union perf_event *event,
787 struct perf_sample *sample __maybe_unused)
788{
789 struct map *map;
790
791 map = maps__find(&machine->kmaps, event->ksymbol.addr);
792 if (map)
793 maps__remove(&machine->kmaps, map);
794
795 return 0;
796}
797
798int machine__process_ksymbol(struct machine *machine __maybe_unused,
799 union perf_event *event,
800 struct perf_sample *sample)
801{
802 if (dump_trace)
803 perf_event__fprintf_ksymbol(event, stdout);
804
805 if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
806 return machine__process_ksymbol_unregister(machine, event,
807 sample);
808 return machine__process_ksymbol_register(machine, event, sample);
809}
810
811int machine__process_text_poke(struct machine *machine, union perf_event *event,
812 struct perf_sample *sample __maybe_unused)
813{
814 struct map *map = maps__find(&machine->kmaps, event->text_poke.addr);
815 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
816
817 if (dump_trace)
818 perf_event__fprintf_text_poke(event, machine, stdout);
819
820 if (!event->text_poke.new_len)
821 return 0;
822
823 if (cpumode != PERF_RECORD_MISC_KERNEL) {
824 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
825 return 0;
826 }
827
828 if (map && map->dso) {
829 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
830 int ret;
831
832 /*
833 * Kernel maps might be changed when loading symbols so loading
834 * must be done prior to using kernel maps.
835 */
836 map__load(map);
837 ret = dso__data_write_cache_addr(map->dso, map, machine,
838 event->text_poke.addr,
839 new_bytes,
840 event->text_poke.new_len);
841 if (ret != event->text_poke.new_len)
842 pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
843 event->text_poke.addr);
844 } else {
845 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
846 event->text_poke.addr);
847 }
848
849 return 0;
850}
851
852static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
853 const char *filename)
854{
855 struct map *map = NULL;
856 struct kmod_path m;
857 struct dso *dso;
858
859 if (kmod_path__parse_name(&m, filename))
860 return NULL;
861
862 dso = machine__findnew_module_dso(machine, &m, filename);
863 if (dso == NULL)
864 goto out;
865
866 map = map__new2(start, dso);
867 if (map == NULL)
868 goto out;
869
870 maps__insert(&machine->kmaps, map);
871
872 /* Put the map here because maps__insert alread got it */
873 map__put(map);
874out:
875 /* put the dso here, corresponding to machine__findnew_module_dso */
876 dso__put(dso);
877 zfree(&m.name);
878 return map;
879}
880
881size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
882{
883 struct rb_node *nd;
884 size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
885
886 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
887 struct machine *pos = rb_entry(nd, struct machine, rb_node);
888 ret += __dsos__fprintf(&pos->dsos.head, fp);
889 }
890
891 return ret;
892}
893
894size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
895 bool (skip)(struct dso *dso, int parm), int parm)
896{
897 return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
898}
899
900size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
901 bool (skip)(struct dso *dso, int parm), int parm)
902{
903 struct rb_node *nd;
904 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
905
906 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
907 struct machine *pos = rb_entry(nd, struct machine, rb_node);
908 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
909 }
910 return ret;
911}
912
913size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
914{
915 int i;
916 size_t printed = 0;
917 struct dso *kdso = machine__kernel_dso(machine);
918
919 if (kdso->has_build_id) {
920 char filename[PATH_MAX];
921 if (dso__build_id_filename(kdso, filename, sizeof(filename),
922 false))
923 printed += fprintf(fp, "[0] %s\n", filename);
924 }
925
926 for (i = 0; i < vmlinux_path__nr_entries; ++i)
927 printed += fprintf(fp, "[%d] %s\n",
928 i + kdso->has_build_id, vmlinux_path[i]);
929
930 return printed;
931}
932
933size_t machine__fprintf(struct machine *machine, FILE *fp)
934{
935 struct rb_node *nd;
936 size_t ret;
937 int i;
938
939 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
940 struct threads *threads = &machine->threads[i];
941
942 down_read(&threads->lock);
943
944 ret = fprintf(fp, "Threads: %u\n", threads->nr);
945
946 for (nd = rb_first_cached(&threads->entries); nd;
947 nd = rb_next(nd)) {
948 struct thread *pos = rb_entry(nd, struct thread, rb_node);
949
950 ret += thread__fprintf(pos, fp);
951 }
952
953 up_read(&threads->lock);
954 }
955 return ret;
956}
957
958static struct dso *machine__get_kernel(struct machine *machine)
959{
960 const char *vmlinux_name = machine->mmap_name;
961 struct dso *kernel;
962
963 if (machine__is_host(machine)) {
964 if (symbol_conf.vmlinux_name)
965 vmlinux_name = symbol_conf.vmlinux_name;
966
967 kernel = machine__findnew_kernel(machine, vmlinux_name,
968 "[kernel]", DSO_SPACE__KERNEL);
969 } else {
970 if (symbol_conf.default_guest_vmlinux_name)
971 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
972
973 kernel = machine__findnew_kernel(machine, vmlinux_name,
974 "[guest.kernel]",
975 DSO_SPACE__KERNEL_GUEST);
976 }
977
978 if (kernel != NULL && (!kernel->has_build_id))
979 dso__read_running_kernel_build_id(kernel, machine);
980
981 return kernel;
982}
983
984struct process_args {
985 u64 start;
986};
987
988void machine__get_kallsyms_filename(struct machine *machine, char *buf,
989 size_t bufsz)
990{
991 if (machine__is_default_guest(machine))
992 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
993 else
994 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
995}
996
997const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
998
999/* Figure out the start address of kernel map from /proc/kallsyms.
1000 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1001 * symbol_name if it's not that important.
1002 */
1003static int machine__get_running_kernel_start(struct machine *machine,
1004 const char **symbol_name,
1005 u64 *start, u64 *end)
1006{
1007 char filename[PATH_MAX];
1008 int i, err = -1;
1009 const char *name;
1010 u64 addr = 0;
1011
1012 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1013
1014 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1015 return 0;
1016
1017 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1018 err = kallsyms__get_function_start(filename, name, &addr);
1019 if (!err)
1020 break;
1021 }
1022
1023 if (err)
1024 return -1;
1025
1026 if (symbol_name)
1027 *symbol_name = name;
1028
1029 *start = addr;
1030
1031 err = kallsyms__get_function_start(filename, "_etext", &addr);
1032 if (!err)
1033 *end = addr;
1034
1035 return 0;
1036}
1037
1038int machine__create_extra_kernel_map(struct machine *machine,
1039 struct dso *kernel,
1040 struct extra_kernel_map *xm)
1041{
1042 struct kmap *kmap;
1043 struct map *map;
1044
1045 map = map__new2(xm->start, kernel);
1046 if (!map)
1047 return -1;
1048
1049 map->end = xm->end;
1050 map->pgoff = xm->pgoff;
1051
1052 kmap = map__kmap(map);
1053
1054 strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1055
1056 maps__insert(&machine->kmaps, map);
1057
1058 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1059 kmap->name, map->start, map->end);
1060
1061 map__put(map);
1062
1063 return 0;
1064}
1065
1066static u64 find_entry_trampoline(struct dso *dso)
1067{
1068 /* Duplicates are removed so lookup all aliases */
1069 const char *syms[] = {
1070 "_entry_trampoline",
1071 "__entry_trampoline_start",
1072 "entry_SYSCALL_64_trampoline",
1073 };
1074 struct symbol *sym = dso__first_symbol(dso);
1075 unsigned int i;
1076
1077 for (; sym; sym = dso__next_symbol(sym)) {
1078 if (sym->binding != STB_GLOBAL)
1079 continue;
1080 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1081 if (!strcmp(sym->name, syms[i]))
1082 return sym->start;
1083 }
1084 }
1085
1086 return 0;
1087}
1088
1089/*
1090 * These values can be used for kernels that do not have symbols for the entry
1091 * trampolines in kallsyms.
1092 */
1093#define X86_64_CPU_ENTRY_AREA_PER_CPU 0xfffffe0000000000ULL
1094#define X86_64_CPU_ENTRY_AREA_SIZE 0x2c000
1095#define X86_64_ENTRY_TRAMPOLINE 0x6000
1096
1097/* Map x86_64 PTI entry trampolines */
1098int machine__map_x86_64_entry_trampolines(struct machine *machine,
1099 struct dso *kernel)
1100{
1101 struct maps *kmaps = &machine->kmaps;
1102 int nr_cpus_avail, cpu;
1103 bool found = false;
1104 struct map *map;
1105 u64 pgoff;
1106
1107 /*
1108 * In the vmlinux case, pgoff is a virtual address which must now be
1109 * mapped to a vmlinux offset.
1110 */
1111 maps__for_each_entry(kmaps, map) {
1112 struct kmap *kmap = __map__kmap(map);
1113 struct map *dest_map;
1114
1115 if (!kmap || !is_entry_trampoline(kmap->name))
1116 continue;
1117
1118 dest_map = maps__find(kmaps, map->pgoff);
1119 if (dest_map != map)
1120 map->pgoff = dest_map->map_ip(dest_map, map->pgoff);
1121 found = true;
1122 }
1123 if (found || machine->trampolines_mapped)
1124 return 0;
1125
1126 pgoff = find_entry_trampoline(kernel);
1127 if (!pgoff)
1128 return 0;
1129
1130 nr_cpus_avail = machine__nr_cpus_avail(machine);
1131
1132 /* Add a 1 page map for each CPU's entry trampoline */
1133 for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1134 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1135 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1136 X86_64_ENTRY_TRAMPOLINE;
1137 struct extra_kernel_map xm = {
1138 .start = va,
1139 .end = va + page_size,
1140 .pgoff = pgoff,
1141 };
1142
1143 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1144
1145 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1146 return -1;
1147 }
1148
1149 machine->trampolines_mapped = nr_cpus_avail;
1150
1151 return 0;
1152}
1153
1154int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1155 struct dso *kernel __maybe_unused)
1156{
1157 return 0;
1158}
1159
1160static int
1161__machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1162{
1163 /* In case of renewal the kernel map, destroy previous one */
1164 machine__destroy_kernel_maps(machine);
1165
1166 machine->vmlinux_map = map__new2(0, kernel);
1167 if (machine->vmlinux_map == NULL)
1168 return -1;
1169
1170 machine->vmlinux_map->map_ip = machine->vmlinux_map->unmap_ip = identity__map_ip;
1171 maps__insert(&machine->kmaps, machine->vmlinux_map);
1172 return 0;
1173}
1174
1175void machine__destroy_kernel_maps(struct machine *machine)
1176{
1177 struct kmap *kmap;
1178 struct map *map = machine__kernel_map(machine);
1179
1180 if (map == NULL)
1181 return;
1182
1183 kmap = map__kmap(map);
1184 maps__remove(&machine->kmaps, map);
1185 if (kmap && kmap->ref_reloc_sym) {
1186 zfree((char **)&kmap->ref_reloc_sym->name);
1187 zfree(&kmap->ref_reloc_sym);
1188 }
1189
1190 map__zput(machine->vmlinux_map);
1191}
1192
1193int machines__create_guest_kernel_maps(struct machines *machines)
1194{
1195 int ret = 0;
1196 struct dirent **namelist = NULL;
1197 int i, items = 0;
1198 char path[PATH_MAX];
1199 pid_t pid;
1200 char *endp;
1201
1202 if (symbol_conf.default_guest_vmlinux_name ||
1203 symbol_conf.default_guest_modules ||
1204 symbol_conf.default_guest_kallsyms) {
1205 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1206 }
1207
1208 if (symbol_conf.guestmount) {
1209 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1210 if (items <= 0)
1211 return -ENOENT;
1212 for (i = 0; i < items; i++) {
1213 if (!isdigit(namelist[i]->d_name[0])) {
1214 /* Filter out . and .. */
1215 continue;
1216 }
1217 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1218 if ((*endp != '\0') ||
1219 (endp == namelist[i]->d_name) ||
1220 (errno == ERANGE)) {
1221 pr_debug("invalid directory (%s). Skipping.\n",
1222 namelist[i]->d_name);
1223 continue;
1224 }
1225 sprintf(path, "%s/%s/proc/kallsyms",
1226 symbol_conf.guestmount,
1227 namelist[i]->d_name);
1228 ret = access(path, R_OK);
1229 if (ret) {
1230 pr_debug("Can't access file %s\n", path);
1231 goto failure;
1232 }
1233 machines__create_kernel_maps(machines, pid);
1234 }
1235failure:
1236 free(namelist);
1237 }
1238
1239 return ret;
1240}
1241
1242void machines__destroy_kernel_maps(struct machines *machines)
1243{
1244 struct rb_node *next = rb_first_cached(&machines->guests);
1245
1246 machine__destroy_kernel_maps(&machines->host);
1247
1248 while (next) {
1249 struct machine *pos = rb_entry(next, struct machine, rb_node);
1250
1251 next = rb_next(&pos->rb_node);
1252 rb_erase_cached(&pos->rb_node, &machines->guests);
1253 machine__delete(pos);
1254 }
1255}
1256
1257int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1258{
1259 struct machine *machine = machines__findnew(machines, pid);
1260
1261 if (machine == NULL)
1262 return -1;
1263
1264 return machine__create_kernel_maps(machine);
1265}
1266
1267int machine__load_kallsyms(struct machine *machine, const char *filename)
1268{
1269 struct map *map = machine__kernel_map(machine);
1270 int ret = __dso__load_kallsyms(map->dso, filename, map, true);
1271
1272 if (ret > 0) {
1273 dso__set_loaded(map->dso);
1274 /*
1275 * Since /proc/kallsyms will have multiple sessions for the
1276 * kernel, with modules between them, fixup the end of all
1277 * sections.
1278 */
1279 maps__fixup_end(&machine->kmaps);
1280 }
1281
1282 return ret;
1283}
1284
1285int machine__load_vmlinux_path(struct machine *machine)
1286{
1287 struct map *map = machine__kernel_map(machine);
1288 int ret = dso__load_vmlinux_path(map->dso, map);
1289
1290 if (ret > 0)
1291 dso__set_loaded(map->dso);
1292
1293 return ret;
1294}
1295
1296static char *get_kernel_version(const char *root_dir)
1297{
1298 char version[PATH_MAX];
1299 FILE *file;
1300 char *name, *tmp;
1301 const char *prefix = "Linux version ";
1302
1303 sprintf(version, "%s/proc/version", root_dir);
1304 file = fopen(version, "r");
1305 if (!file)
1306 return NULL;
1307
1308 tmp = fgets(version, sizeof(version), file);
1309 fclose(file);
1310 if (!tmp)
1311 return NULL;
1312
1313 name = strstr(version, prefix);
1314 if (!name)
1315 return NULL;
1316 name += strlen(prefix);
1317 tmp = strchr(name, ' ');
1318 if (tmp)
1319 *tmp = '\0';
1320
1321 return strdup(name);
1322}
1323
1324static bool is_kmod_dso(struct dso *dso)
1325{
1326 return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1327 dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1328}
1329
1330static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1331{
1332 char *long_name;
1333 struct map *map = maps__find_by_name(maps, m->name);
1334
1335 if (map == NULL)
1336 return 0;
1337
1338 long_name = strdup(path);
1339 if (long_name == NULL)
1340 return -ENOMEM;
1341
1342 dso__set_long_name(map->dso, long_name, true);
1343 dso__kernel_module_get_build_id(map->dso, "");
1344
1345 /*
1346 * Full name could reveal us kmod compression, so
1347 * we need to update the symtab_type if needed.
1348 */
1349 if (m->comp && is_kmod_dso(map->dso)) {
1350 map->dso->symtab_type++;
1351 map->dso->comp = m->comp;
1352 }
1353
1354 return 0;
1355}
1356
1357static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1358{
1359 struct dirent *dent;
1360 DIR *dir = opendir(dir_name);
1361 int ret = 0;
1362
1363 if (!dir) {
1364 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1365 return -1;
1366 }
1367
1368 while ((dent = readdir(dir)) != NULL) {
1369 char path[PATH_MAX];
1370 struct stat st;
1371
1372 /*sshfs might return bad dent->d_type, so we have to stat*/
1373 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1374 if (stat(path, &st))
1375 continue;
1376
1377 if (S_ISDIR(st.st_mode)) {
1378 if (!strcmp(dent->d_name, ".") ||
1379 !strcmp(dent->d_name, ".."))
1380 continue;
1381
1382 /* Do not follow top-level source and build symlinks */
1383 if (depth == 0) {
1384 if (!strcmp(dent->d_name, "source") ||
1385 !strcmp(dent->d_name, "build"))
1386 continue;
1387 }
1388
1389 ret = maps__set_modules_path_dir(maps, path, depth + 1);
1390 if (ret < 0)
1391 goto out;
1392 } else {
1393 struct kmod_path m;
1394
1395 ret = kmod_path__parse_name(&m, dent->d_name);
1396 if (ret)
1397 goto out;
1398
1399 if (m.kmod)
1400 ret = maps__set_module_path(maps, path, &m);
1401
1402 zfree(&m.name);
1403
1404 if (ret)
1405 goto out;
1406 }
1407 }
1408
1409out:
1410 closedir(dir);
1411 return ret;
1412}
1413
1414static int machine__set_modules_path(struct machine *machine)
1415{
1416 char *version;
1417 char modules_path[PATH_MAX];
1418
1419 version = get_kernel_version(machine->root_dir);
1420 if (!version)
1421 return -1;
1422
1423 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1424 machine->root_dir, version);
1425 free(version);
1426
1427 return maps__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1428}
1429int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1430 u64 *size __maybe_unused,
1431 const char *name __maybe_unused)
1432{
1433 return 0;
1434}
1435
1436static int machine__create_module(void *arg, const char *name, u64 start,
1437 u64 size)
1438{
1439 struct machine *machine = arg;
1440 struct map *map;
1441
1442 if (arch__fix_module_text_start(&start, &size, name) < 0)
1443 return -1;
1444
1445 map = machine__addnew_module_map(machine, start, name);
1446 if (map == NULL)
1447 return -1;
1448 map->end = start + size;
1449
1450 dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1451
1452 return 0;
1453}
1454
1455static int machine__create_modules(struct machine *machine)
1456{
1457 const char *modules;
1458 char path[PATH_MAX];
1459
1460 if (machine__is_default_guest(machine)) {
1461 modules = symbol_conf.default_guest_modules;
1462 } else {
1463 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1464 modules = path;
1465 }
1466
1467 if (symbol__restricted_filename(modules, "/proc/modules"))
1468 return -1;
1469
1470 if (modules__parse(modules, machine, machine__create_module))
1471 return -1;
1472
1473 if (!machine__set_modules_path(machine))
1474 return 0;
1475
1476 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1477
1478 return 0;
1479}
1480
1481static void machine__set_kernel_mmap(struct machine *machine,
1482 u64 start, u64 end)
1483{
1484 machine->vmlinux_map->start = start;
1485 machine->vmlinux_map->end = end;
1486 /*
1487 * Be a bit paranoid here, some perf.data file came with
1488 * a zero sized synthesized MMAP event for the kernel.
1489 */
1490 if (start == 0 && end == 0)
1491 machine->vmlinux_map->end = ~0ULL;
1492}
1493
1494static void machine__update_kernel_mmap(struct machine *machine,
1495 u64 start, u64 end)
1496{
1497 struct map *map = machine__kernel_map(machine);
1498
1499 map__get(map);
1500 maps__remove(&machine->kmaps, map);
1501
1502 machine__set_kernel_mmap(machine, start, end);
1503
1504 maps__insert(&machine->kmaps, map);
1505 map__put(map);
1506}
1507
1508int machine__create_kernel_maps(struct machine *machine)
1509{
1510 struct dso *kernel = machine__get_kernel(machine);
1511 const char *name = NULL;
1512 struct map *map;
1513 u64 start = 0, end = ~0ULL;
1514 int ret;
1515
1516 if (kernel == NULL)
1517 return -1;
1518
1519 ret = __machine__create_kernel_maps(machine, kernel);
1520 if (ret < 0)
1521 goto out_put;
1522
1523 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1524 if (machine__is_host(machine))
1525 pr_debug("Problems creating module maps, "
1526 "continuing anyway...\n");
1527 else
1528 pr_debug("Problems creating module maps for guest %d, "
1529 "continuing anyway...\n", machine->pid);
1530 }
1531
1532 if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1533 if (name &&
1534 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1535 machine__destroy_kernel_maps(machine);
1536 ret = -1;
1537 goto out_put;
1538 }
1539
1540 /*
1541 * we have a real start address now, so re-order the kmaps
1542 * assume it's the last in the kmaps
1543 */
1544 machine__update_kernel_mmap(machine, start, end);
1545 }
1546
1547 if (machine__create_extra_kernel_maps(machine, kernel))
1548 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1549
1550 if (end == ~0ULL) {
1551 /* update end address of the kernel map using adjacent module address */
1552 map = map__next(machine__kernel_map(machine));
1553 if (map)
1554 machine__set_kernel_mmap(machine, start, map->start);
1555 }
1556
1557out_put:
1558 dso__put(kernel);
1559 return ret;
1560}
1561
1562static bool machine__uses_kcore(struct machine *machine)
1563{
1564 struct dso *dso;
1565
1566 list_for_each_entry(dso, &machine->dsos.head, node) {
1567 if (dso__is_kcore(dso))
1568 return true;
1569 }
1570
1571 return false;
1572}
1573
1574static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1575 union perf_event *event)
1576{
1577 return machine__is(machine, "x86_64") &&
1578 is_entry_trampoline(event->mmap.filename);
1579}
1580
1581static int machine__process_extra_kernel_map(struct machine *machine,
1582 union perf_event *event)
1583{
1584 struct dso *kernel = machine__kernel_dso(machine);
1585 struct extra_kernel_map xm = {
1586 .start = event->mmap.start,
1587 .end = event->mmap.start + event->mmap.len,
1588 .pgoff = event->mmap.pgoff,
1589 };
1590
1591 if (kernel == NULL)
1592 return -1;
1593
1594 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1595
1596 return machine__create_extra_kernel_map(machine, kernel, &xm);
1597}
1598
1599static int machine__process_kernel_mmap_event(struct machine *machine,
1600 union perf_event *event)
1601{
1602 struct map *map;
1603 enum dso_space_type dso_space;
1604 bool is_kernel_mmap;
1605
1606 /* If we have maps from kcore then we do not need or want any others */
1607 if (machine__uses_kcore(machine))
1608 return 0;
1609
1610 if (machine__is_host(machine))
1611 dso_space = DSO_SPACE__KERNEL;
1612 else
1613 dso_space = DSO_SPACE__KERNEL_GUEST;
1614
1615 is_kernel_mmap = memcmp(event->mmap.filename,
1616 machine->mmap_name,
1617 strlen(machine->mmap_name) - 1) == 0;
1618 if (event->mmap.filename[0] == '/' ||
1619 (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1620 map = machine__addnew_module_map(machine, event->mmap.start,
1621 event->mmap.filename);
1622 if (map == NULL)
1623 goto out_problem;
1624
1625 map->end = map->start + event->mmap.len;
1626 } else if (is_kernel_mmap) {
1627 const char *symbol_name = (event->mmap.filename +
1628 strlen(machine->mmap_name));
1629 /*
1630 * Should be there already, from the build-id table in
1631 * the header.
1632 */
1633 struct dso *kernel = NULL;
1634 struct dso *dso;
1635
1636 down_read(&machine->dsos.lock);
1637
1638 list_for_each_entry(dso, &machine->dsos.head, node) {
1639
1640 /*
1641 * The cpumode passed to is_kernel_module is not the
1642 * cpumode of *this* event. If we insist on passing
1643 * correct cpumode to is_kernel_module, we should
1644 * record the cpumode when we adding this dso to the
1645 * linked list.
1646 *
1647 * However we don't really need passing correct
1648 * cpumode. We know the correct cpumode must be kernel
1649 * mode (if not, we should not link it onto kernel_dsos
1650 * list).
1651 *
1652 * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1653 * is_kernel_module() treats it as a kernel cpumode.
1654 */
1655
1656 if (!dso->kernel ||
1657 is_kernel_module(dso->long_name,
1658 PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1659 continue;
1660
1661
1662 kernel = dso;
1663 break;
1664 }
1665
1666 up_read(&machine->dsos.lock);
1667
1668 if (kernel == NULL)
1669 kernel = machine__findnew_dso(machine, machine->mmap_name);
1670 if (kernel == NULL)
1671 goto out_problem;
1672
1673 kernel->kernel = dso_space;
1674 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1675 dso__put(kernel);
1676 goto out_problem;
1677 }
1678
1679 if (strstr(kernel->long_name, "vmlinux"))
1680 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1681
1682 machine__update_kernel_mmap(machine, event->mmap.start,
1683 event->mmap.start + event->mmap.len);
1684
1685 /*
1686 * Avoid using a zero address (kptr_restrict) for the ref reloc
1687 * symbol. Effectively having zero here means that at record
1688 * time /proc/sys/kernel/kptr_restrict was non zero.
1689 */
1690 if (event->mmap.pgoff != 0) {
1691 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1692 symbol_name,
1693 event->mmap.pgoff);
1694 }
1695
1696 if (machine__is_default_guest(machine)) {
1697 /*
1698 * preload dso of guest kernel and modules
1699 */
1700 dso__load(kernel, machine__kernel_map(machine));
1701 }
1702 } else if (perf_event__is_extra_kernel_mmap(machine, event)) {
1703 return machine__process_extra_kernel_map(machine, event);
1704 }
1705 return 0;
1706out_problem:
1707 return -1;
1708}
1709
1710int machine__process_mmap2_event(struct machine *machine,
1711 union perf_event *event,
1712 struct perf_sample *sample)
1713{
1714 struct thread *thread;
1715 struct map *map;
1716 struct dso_id dso_id = {
1717 .maj = event->mmap2.maj,
1718 .min = event->mmap2.min,
1719 .ino = event->mmap2.ino,
1720 .ino_generation = event->mmap2.ino_generation,
1721 };
1722 int ret = 0;
1723
1724 if (dump_trace)
1725 perf_event__fprintf_mmap2(event, stdout);
1726
1727 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1728 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1729 ret = machine__process_kernel_mmap_event(machine, event);
1730 if (ret < 0)
1731 goto out_problem;
1732 return 0;
1733 }
1734
1735 thread = machine__findnew_thread(machine, event->mmap2.pid,
1736 event->mmap2.tid);
1737 if (thread == NULL)
1738 goto out_problem;
1739
1740 map = map__new(machine, event->mmap2.start,
1741 event->mmap2.len, event->mmap2.pgoff,
1742 &dso_id, event->mmap2.prot,
1743 event->mmap2.flags,
1744 event->mmap2.filename, thread);
1745
1746 if (map == NULL)
1747 goto out_problem_map;
1748
1749 ret = thread__insert_map(thread, map);
1750 if (ret)
1751 goto out_problem_insert;
1752
1753 thread__put(thread);
1754 map__put(map);
1755 return 0;
1756
1757out_problem_insert:
1758 map__put(map);
1759out_problem_map:
1760 thread__put(thread);
1761out_problem:
1762 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1763 return 0;
1764}
1765
1766int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1767 struct perf_sample *sample)
1768{
1769 struct thread *thread;
1770 struct map *map;
1771 u32 prot = 0;
1772 int ret = 0;
1773
1774 if (dump_trace)
1775 perf_event__fprintf_mmap(event, stdout);
1776
1777 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1778 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1779 ret = machine__process_kernel_mmap_event(machine, event);
1780 if (ret < 0)
1781 goto out_problem;
1782 return 0;
1783 }
1784
1785 thread = machine__findnew_thread(machine, event->mmap.pid,
1786 event->mmap.tid);
1787 if (thread == NULL)
1788 goto out_problem;
1789
1790 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1791 prot = PROT_EXEC;
1792
1793 map = map__new(machine, event->mmap.start,
1794 event->mmap.len, event->mmap.pgoff,
1795 NULL, prot, 0, event->mmap.filename, thread);
1796
1797 if (map == NULL)
1798 goto out_problem_map;
1799
1800 ret = thread__insert_map(thread, map);
1801 if (ret)
1802 goto out_problem_insert;
1803
1804 thread__put(thread);
1805 map__put(map);
1806 return 0;
1807
1808out_problem_insert:
1809 map__put(map);
1810out_problem_map:
1811 thread__put(thread);
1812out_problem:
1813 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1814 return 0;
1815}
1816
1817static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1818{
1819 struct threads *threads = machine__threads(machine, th->tid);
1820
1821 if (threads->last_match == th)
1822 threads__set_last_match(threads, NULL);
1823
1824 if (lock)
1825 down_write(&threads->lock);
1826
1827 BUG_ON(refcount_read(&th->refcnt) == 0);
1828
1829 rb_erase_cached(&th->rb_node, &threads->entries);
1830 RB_CLEAR_NODE(&th->rb_node);
1831 --threads->nr;
1832 /*
1833 * Move it first to the dead_threads list, then drop the reference,
1834 * if this is the last reference, then the thread__delete destructor
1835 * will be called and we will remove it from the dead_threads list.
1836 */
1837 list_add_tail(&th->node, &threads->dead);
1838
1839 /*
1840 * We need to do the put here because if this is the last refcount,
1841 * then we will be touching the threads->dead head when removing the
1842 * thread.
1843 */
1844 thread__put(th);
1845
1846 if (lock)
1847 up_write(&threads->lock);
1848}
1849
1850void machine__remove_thread(struct machine *machine, struct thread *th)
1851{
1852 return __machine__remove_thread(machine, th, true);
1853}
1854
1855int machine__process_fork_event(struct machine *machine, union perf_event *event,
1856 struct perf_sample *sample)
1857{
1858 struct thread *thread = machine__find_thread(machine,
1859 event->fork.pid,
1860 event->fork.tid);
1861 struct thread *parent = machine__findnew_thread(machine,
1862 event->fork.ppid,
1863 event->fork.ptid);
1864 bool do_maps_clone = true;
1865 int err = 0;
1866
1867 if (dump_trace)
1868 perf_event__fprintf_task(event, stdout);
1869
1870 /*
1871 * There may be an existing thread that is not actually the parent,
1872 * either because we are processing events out of order, or because the
1873 * (fork) event that would have removed the thread was lost. Assume the
1874 * latter case and continue on as best we can.
1875 */
1876 if (parent->pid_ != (pid_t)event->fork.ppid) {
1877 dump_printf("removing erroneous parent thread %d/%d\n",
1878 parent->pid_, parent->tid);
1879 machine__remove_thread(machine, parent);
1880 thread__put(parent);
1881 parent = machine__findnew_thread(machine, event->fork.ppid,
1882 event->fork.ptid);
1883 }
1884
1885 /* if a thread currently exists for the thread id remove it */
1886 if (thread != NULL) {
1887 machine__remove_thread(machine, thread);
1888 thread__put(thread);
1889 }
1890
1891 thread = machine__findnew_thread(machine, event->fork.pid,
1892 event->fork.tid);
1893 /*
1894 * When synthesizing FORK events, we are trying to create thread
1895 * objects for the already running tasks on the machine.
1896 *
1897 * Normally, for a kernel FORK event, we want to clone the parent's
1898 * maps because that is what the kernel just did.
1899 *
1900 * But when synthesizing, this should not be done. If we do, we end up
1901 * with overlapping maps as we process the sythesized MMAP2 events that
1902 * get delivered shortly thereafter.
1903 *
1904 * Use the FORK event misc flags in an internal way to signal this
1905 * situation, so we can elide the map clone when appropriate.
1906 */
1907 if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1908 do_maps_clone = false;
1909
1910 if (thread == NULL || parent == NULL ||
1911 thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1912 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1913 err = -1;
1914 }
1915 thread__put(thread);
1916 thread__put(parent);
1917
1918 return err;
1919}
1920
1921int machine__process_exit_event(struct machine *machine, union perf_event *event,
1922 struct perf_sample *sample __maybe_unused)
1923{
1924 struct thread *thread = machine__find_thread(machine,
1925 event->fork.pid,
1926 event->fork.tid);
1927
1928 if (dump_trace)
1929 perf_event__fprintf_task(event, stdout);
1930
1931 if (thread != NULL) {
1932 thread__exited(thread);
1933 thread__put(thread);
1934 }
1935
1936 return 0;
1937}
1938
1939int machine__process_event(struct machine *machine, union perf_event *event,
1940 struct perf_sample *sample)
1941{
1942 int ret;
1943
1944 switch (event->header.type) {
1945 case PERF_RECORD_COMM:
1946 ret = machine__process_comm_event(machine, event, sample); break;
1947 case PERF_RECORD_MMAP:
1948 ret = machine__process_mmap_event(machine, event, sample); break;
1949 case PERF_RECORD_NAMESPACES:
1950 ret = machine__process_namespaces_event(machine, event, sample); break;
1951 case PERF_RECORD_CGROUP:
1952 ret = machine__process_cgroup_event(machine, event, sample); break;
1953 case PERF_RECORD_MMAP2:
1954 ret = machine__process_mmap2_event(machine, event, sample); break;
1955 case PERF_RECORD_FORK:
1956 ret = machine__process_fork_event(machine, event, sample); break;
1957 case PERF_RECORD_EXIT:
1958 ret = machine__process_exit_event(machine, event, sample); break;
1959 case PERF_RECORD_LOST:
1960 ret = machine__process_lost_event(machine, event, sample); break;
1961 case PERF_RECORD_AUX:
1962 ret = machine__process_aux_event(machine, event); break;
1963 case PERF_RECORD_ITRACE_START:
1964 ret = machine__process_itrace_start_event(machine, event); break;
1965 case PERF_RECORD_LOST_SAMPLES:
1966 ret = machine__process_lost_samples_event(machine, event, sample); break;
1967 case PERF_RECORD_SWITCH:
1968 case PERF_RECORD_SWITCH_CPU_WIDE:
1969 ret = machine__process_switch_event(machine, event); break;
1970 case PERF_RECORD_KSYMBOL:
1971 ret = machine__process_ksymbol(machine, event, sample); break;
1972 case PERF_RECORD_BPF_EVENT:
1973 ret = machine__process_bpf(machine, event, sample); break;
1974 case PERF_RECORD_TEXT_POKE:
1975 ret = machine__process_text_poke(machine, event, sample); break;
1976 default:
1977 ret = -1;
1978 break;
1979 }
1980
1981 return ret;
1982}
1983
1984static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1985{
1986 if (!regexec(regex, sym->name, 0, NULL, 0))
1987 return 1;
1988 return 0;
1989}
1990
1991static void ip__resolve_ams(struct thread *thread,
1992 struct addr_map_symbol *ams,
1993 u64 ip)
1994{
1995 struct addr_location al;
1996
1997 memset(&al, 0, sizeof(al));
1998 /*
1999 * We cannot use the header.misc hint to determine whether a
2000 * branch stack address is user, kernel, guest, hypervisor.
2001 * Branches may straddle the kernel/user/hypervisor boundaries.
2002 * Thus, we have to try consecutively until we find a match
2003 * or else, the symbol is unknown
2004 */
2005 thread__find_cpumode_addr_location(thread, ip, &al);
2006
2007 ams->addr = ip;
2008 ams->al_addr = al.addr;
2009 ams->ms.maps = al.maps;
2010 ams->ms.sym = al.sym;
2011 ams->ms.map = al.map;
2012 ams->phys_addr = 0;
2013}
2014
2015static void ip__resolve_data(struct thread *thread,
2016 u8 m, struct addr_map_symbol *ams,
2017 u64 addr, u64 phys_addr)
2018{
2019 struct addr_location al;
2020
2021 memset(&al, 0, sizeof(al));
2022
2023 thread__find_symbol(thread, m, addr, &al);
2024
2025 ams->addr = addr;
2026 ams->al_addr = al.addr;
2027 ams->ms.maps = al.maps;
2028 ams->ms.sym = al.sym;
2029 ams->ms.map = al.map;
2030 ams->phys_addr = phys_addr;
2031}
2032
2033struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2034 struct addr_location *al)
2035{
2036 struct mem_info *mi = mem_info__new();
2037
2038 if (!mi)
2039 return NULL;
2040
2041 ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2042 ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2043 sample->addr, sample->phys_addr);
2044 mi->data_src.val = sample->data_src;
2045
2046 return mi;
2047}
2048
2049static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2050{
2051 struct map *map = ms->map;
2052 char *srcline = NULL;
2053
2054 if (!map || callchain_param.key == CCKEY_FUNCTION)
2055 return srcline;
2056
2057 srcline = srcline__tree_find(&map->dso->srclines, ip);
2058 if (!srcline) {
2059 bool show_sym = false;
2060 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2061
2062 srcline = get_srcline(map->dso, map__rip_2objdump(map, ip),
2063 ms->sym, show_sym, show_addr, ip);
2064 srcline__tree_insert(&map->dso->srclines, ip, srcline);
2065 }
2066
2067 return srcline;
2068}
2069
2070struct iterations {
2071 int nr_loop_iter;
2072 u64 cycles;
2073};
2074
2075static int add_callchain_ip(struct thread *thread,
2076 struct callchain_cursor *cursor,
2077 struct symbol **parent,
2078 struct addr_location *root_al,
2079 u8 *cpumode,
2080 u64 ip,
2081 bool branch,
2082 struct branch_flags *flags,
2083 struct iterations *iter,
2084 u64 branch_from)
2085{
2086 struct map_symbol ms;
2087 struct addr_location al;
2088 int nr_loop_iter = 0;
2089 u64 iter_cycles = 0;
2090 const char *srcline = NULL;
2091
2092 al.filtered = 0;
2093 al.sym = NULL;
2094 if (!cpumode) {
2095 thread__find_cpumode_addr_location(thread, ip, &al);
2096 } else {
2097 if (ip >= PERF_CONTEXT_MAX) {
2098 switch (ip) {
2099 case PERF_CONTEXT_HV:
2100 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2101 break;
2102 case PERF_CONTEXT_KERNEL:
2103 *cpumode = PERF_RECORD_MISC_KERNEL;
2104 break;
2105 case PERF_CONTEXT_USER:
2106 *cpumode = PERF_RECORD_MISC_USER;
2107 break;
2108 default:
2109 pr_debug("invalid callchain context: "
2110 "%"PRId64"\n", (s64) ip);
2111 /*
2112 * It seems the callchain is corrupted.
2113 * Discard all.
2114 */
2115 callchain_cursor_reset(cursor);
2116 return 1;
2117 }
2118 return 0;
2119 }
2120 thread__find_symbol(thread, *cpumode, ip, &al);
2121 }
2122
2123 if (al.sym != NULL) {
2124 if (perf_hpp_list.parent && !*parent &&
2125 symbol__match_regex(al.sym, &parent_regex))
2126 *parent = al.sym;
2127 else if (have_ignore_callees && root_al &&
2128 symbol__match_regex(al.sym, &ignore_callees_regex)) {
2129 /* Treat this symbol as the root,
2130 forgetting its callees. */
2131 *root_al = al;
2132 callchain_cursor_reset(cursor);
2133 }
2134 }
2135
2136 if (symbol_conf.hide_unresolved && al.sym == NULL)
2137 return 0;
2138
2139 if (iter) {
2140 nr_loop_iter = iter->nr_loop_iter;
2141 iter_cycles = iter->cycles;
2142 }
2143
2144 ms.maps = al.maps;
2145 ms.map = al.map;
2146 ms.sym = al.sym;
2147 srcline = callchain_srcline(&ms, al.addr);
2148 return callchain_cursor_append(cursor, ip, &ms,
2149 branch, flags, nr_loop_iter,
2150 iter_cycles, branch_from, srcline);
2151}
2152
2153struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2154 struct addr_location *al)
2155{
2156 unsigned int i;
2157 const struct branch_stack *bs = sample->branch_stack;
2158 struct branch_entry *entries = perf_sample__branch_entries(sample);
2159 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2160
2161 if (!bi)
2162 return NULL;
2163
2164 for (i = 0; i < bs->nr; i++) {
2165 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2166 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2167 bi[i].flags = entries[i].flags;
2168 }
2169 return bi;
2170}
2171
2172static void save_iterations(struct iterations *iter,
2173 struct branch_entry *be, int nr)
2174{
2175 int i;
2176
2177 iter->nr_loop_iter++;
2178 iter->cycles = 0;
2179
2180 for (i = 0; i < nr; i++)
2181 iter->cycles += be[i].flags.cycles;
2182}
2183
2184#define CHASHSZ 127
2185#define CHASHBITS 7
2186#define NO_ENTRY 0xff
2187
2188#define PERF_MAX_BRANCH_DEPTH 127
2189
2190/* Remove loops. */
2191static int remove_loops(struct branch_entry *l, int nr,
2192 struct iterations *iter)
2193{
2194 int i, j, off;
2195 unsigned char chash[CHASHSZ];
2196
2197 memset(chash, NO_ENTRY, sizeof(chash));
2198
2199 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2200
2201 for (i = 0; i < nr; i++) {
2202 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2203
2204 /* no collision handling for now */
2205 if (chash[h] == NO_ENTRY) {
2206 chash[h] = i;
2207 } else if (l[chash[h]].from == l[i].from) {
2208 bool is_loop = true;
2209 /* check if it is a real loop */
2210 off = 0;
2211 for (j = chash[h]; j < i && i + off < nr; j++, off++)
2212 if (l[j].from != l[i + off].from) {
2213 is_loop = false;
2214 break;
2215 }
2216 if (is_loop) {
2217 j = nr - (i + off);
2218 if (j > 0) {
2219 save_iterations(iter + i + off,
2220 l + i, off);
2221
2222 memmove(iter + i, iter + i + off,
2223 j * sizeof(*iter));
2224
2225 memmove(l + i, l + i + off,
2226 j * sizeof(*l));
2227 }
2228
2229 nr -= off;
2230 }
2231 }
2232 }
2233 return nr;
2234}
2235
2236static int lbr_callchain_add_kernel_ip(struct thread *thread,
2237 struct callchain_cursor *cursor,
2238 struct perf_sample *sample,
2239 struct symbol **parent,
2240 struct addr_location *root_al,
2241 u64 branch_from,
2242 bool callee, int end)
2243{
2244 struct ip_callchain *chain = sample->callchain;
2245 u8 cpumode = PERF_RECORD_MISC_USER;
2246 int err, i;
2247
2248 if (callee) {
2249 for (i = 0; i < end + 1; i++) {
2250 err = add_callchain_ip(thread, cursor, parent,
2251 root_al, &cpumode, chain->ips[i],
2252 false, NULL, NULL, branch_from);
2253 if (err)
2254 return err;
2255 }
2256 return 0;
2257 }
2258
2259 for (i = end; i >= 0; i--) {
2260 err = add_callchain_ip(thread, cursor, parent,
2261 root_al, &cpumode, chain->ips[i],
2262 false, NULL, NULL, branch_from);
2263 if (err)
2264 return err;
2265 }
2266
2267 return 0;
2268}
2269
2270static void save_lbr_cursor_node(struct thread *thread,
2271 struct callchain_cursor *cursor,
2272 int idx)
2273{
2274 struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2275
2276 if (!lbr_stitch)
2277 return;
2278
2279 if (cursor->pos == cursor->nr) {
2280 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2281 return;
2282 }
2283
2284 if (!cursor->curr)
2285 cursor->curr = cursor->first;
2286 else
2287 cursor->curr = cursor->curr->next;
2288 memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2289 sizeof(struct callchain_cursor_node));
2290
2291 lbr_stitch->prev_lbr_cursor[idx].valid = true;
2292 cursor->pos++;
2293}
2294
2295static int lbr_callchain_add_lbr_ip(struct thread *thread,
2296 struct callchain_cursor *cursor,
2297 struct perf_sample *sample,
2298 struct symbol **parent,
2299 struct addr_location *root_al,
2300 u64 *branch_from,
2301 bool callee)
2302{
2303 struct branch_stack *lbr_stack = sample->branch_stack;
2304 struct branch_entry *entries = perf_sample__branch_entries(sample);
2305 u8 cpumode = PERF_RECORD_MISC_USER;
2306 int lbr_nr = lbr_stack->nr;
2307 struct branch_flags *flags;
2308 int err, i;
2309 u64 ip;
2310
2311 /*
2312 * The curr and pos are not used in writing session. They are cleared
2313 * in callchain_cursor_commit() when the writing session is closed.
2314 * Using curr and pos to track the current cursor node.
2315 */
2316 if (thread->lbr_stitch) {
2317 cursor->curr = NULL;
2318 cursor->pos = cursor->nr;
2319 if (cursor->nr) {
2320 cursor->curr = cursor->first;
2321 for (i = 0; i < (int)(cursor->nr - 1); i++)
2322 cursor->curr = cursor->curr->next;
2323 }
2324 }
2325
2326 if (callee) {
2327 /* Add LBR ip from first entries.to */
2328 ip = entries[0].to;
2329 flags = &entries[0].flags;
2330 *branch_from = entries[0].from;
2331 err = add_callchain_ip(thread, cursor, parent,
2332 root_al, &cpumode, ip,
2333 true, flags, NULL,
2334 *branch_from);
2335 if (err)
2336 return err;
2337
2338 /*
2339 * The number of cursor node increases.
2340 * Move the current cursor node.
2341 * But does not need to save current cursor node for entry 0.
2342 * It's impossible to stitch the whole LBRs of previous sample.
2343 */
2344 if (thread->lbr_stitch && (cursor->pos != cursor->nr)) {
2345 if (!cursor->curr)
2346 cursor->curr = cursor->first;
2347 else
2348 cursor->curr = cursor->curr->next;
2349 cursor->pos++;
2350 }
2351
2352 /* Add LBR ip from entries.from one by one. */
2353 for (i = 0; i < lbr_nr; i++) {
2354 ip = entries[i].from;
2355 flags = &entries[i].flags;
2356 err = add_callchain_ip(thread, cursor, parent,
2357 root_al, &cpumode, ip,
2358 true, flags, NULL,
2359 *branch_from);
2360 if (err)
2361 return err;
2362 save_lbr_cursor_node(thread, cursor, i);
2363 }
2364 return 0;
2365 }
2366
2367 /* Add LBR ip from entries.from one by one. */
2368 for (i = lbr_nr - 1; i >= 0; i--) {
2369 ip = entries[i].from;
2370 flags = &entries[i].flags;
2371 err = add_callchain_ip(thread, cursor, parent,
2372 root_al, &cpumode, ip,
2373 true, flags, NULL,
2374 *branch_from);
2375 if (err)
2376 return err;
2377 save_lbr_cursor_node(thread, cursor, i);
2378 }
2379
2380 /* Add LBR ip from first entries.to */
2381 ip = entries[0].to;
2382 flags = &entries[0].flags;
2383 *branch_from = entries[0].from;
2384 err = add_callchain_ip(thread, cursor, parent,
2385 root_al, &cpumode, ip,
2386 true, flags, NULL,
2387 *branch_from);
2388 if (err)
2389 return err;
2390
2391 return 0;
2392}
2393
2394static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2395 struct callchain_cursor *cursor)
2396{
2397 struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2398 struct callchain_cursor_node *cnode;
2399 struct stitch_list *stitch_node;
2400 int err;
2401
2402 list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2403 cnode = &stitch_node->cursor;
2404
2405 err = callchain_cursor_append(cursor, cnode->ip,
2406 &cnode->ms,
2407 cnode->branch,
2408 &cnode->branch_flags,
2409 cnode->nr_loop_iter,
2410 cnode->iter_cycles,
2411 cnode->branch_from,
2412 cnode->srcline);
2413 if (err)
2414 return err;
2415 }
2416 return 0;
2417}
2418
2419static struct stitch_list *get_stitch_node(struct thread *thread)
2420{
2421 struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2422 struct stitch_list *stitch_node;
2423
2424 if (!list_empty(&lbr_stitch->free_lists)) {
2425 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2426 struct stitch_list, node);
2427 list_del(&stitch_node->node);
2428
2429 return stitch_node;
2430 }
2431
2432 return malloc(sizeof(struct stitch_list));
2433}
2434
2435static bool has_stitched_lbr(struct thread *thread,
2436 struct perf_sample *cur,
2437 struct perf_sample *prev,
2438 unsigned int max_lbr,
2439 bool callee)
2440{
2441 struct branch_stack *cur_stack = cur->branch_stack;
2442 struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2443 struct branch_stack *prev_stack = prev->branch_stack;
2444 struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2445 struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2446 int i, j, nr_identical_branches = 0;
2447 struct stitch_list *stitch_node;
2448 u64 cur_base, distance;
2449
2450 if (!cur_stack || !prev_stack)
2451 return false;
2452
2453 /* Find the physical index of the base-of-stack for current sample. */
2454 cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2455
2456 distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2457 (max_lbr + prev_stack->hw_idx - cur_base);
2458 /* Previous sample has shorter stack. Nothing can be stitched. */
2459 if (distance + 1 > prev_stack->nr)
2460 return false;
2461
2462 /*
2463 * Check if there are identical LBRs between two samples.
2464 * Identicall LBRs must have same from, to and flags values. Also,
2465 * they have to be saved in the same LBR registers (same physical
2466 * index).
2467 *
2468 * Starts from the base-of-stack of current sample.
2469 */
2470 for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2471 if ((prev_entries[i].from != cur_entries[j].from) ||
2472 (prev_entries[i].to != cur_entries[j].to) ||
2473 (prev_entries[i].flags.value != cur_entries[j].flags.value))
2474 break;
2475 nr_identical_branches++;
2476 }
2477
2478 if (!nr_identical_branches)
2479 return false;
2480
2481 /*
2482 * Save the LBRs between the base-of-stack of previous sample
2483 * and the base-of-stack of current sample into lbr_stitch->lists.
2484 * These LBRs will be stitched later.
2485 */
2486 for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2487
2488 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2489 continue;
2490
2491 stitch_node = get_stitch_node(thread);
2492 if (!stitch_node)
2493 return false;
2494
2495 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2496 sizeof(struct callchain_cursor_node));
2497
2498 if (callee)
2499 list_add(&stitch_node->node, &lbr_stitch->lists);
2500 else
2501 list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2502 }
2503
2504 return true;
2505}
2506
2507static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2508{
2509 if (thread->lbr_stitch)
2510 return true;
2511
2512 thread->lbr_stitch = zalloc(sizeof(*thread->lbr_stitch));
2513 if (!thread->lbr_stitch)
2514 goto err;
2515
2516 thread->lbr_stitch->prev_lbr_cursor = calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2517 if (!thread->lbr_stitch->prev_lbr_cursor)
2518 goto free_lbr_stitch;
2519
2520 INIT_LIST_HEAD(&thread->lbr_stitch->lists);
2521 INIT_LIST_HEAD(&thread->lbr_stitch->free_lists);
2522
2523 return true;
2524
2525free_lbr_stitch:
2526 zfree(&thread->lbr_stitch);
2527err:
2528 pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2529 thread->lbr_stitch_enable = false;
2530 return false;
2531}
2532
2533/*
2534 * Recolve LBR callstack chain sample
2535 * Return:
2536 * 1 on success get LBR callchain information
2537 * 0 no available LBR callchain information, should try fp
2538 * negative error code on other errors.
2539 */
2540static int resolve_lbr_callchain_sample(struct thread *thread,
2541 struct callchain_cursor *cursor,
2542 struct perf_sample *sample,
2543 struct symbol **parent,
2544 struct addr_location *root_al,
2545 int max_stack,
2546 unsigned int max_lbr)
2547{
2548 bool callee = (callchain_param.order == ORDER_CALLEE);
2549 struct ip_callchain *chain = sample->callchain;
2550 int chain_nr = min(max_stack, (int)chain->nr), i;
2551 struct lbr_stitch *lbr_stitch;
2552 bool stitched_lbr = false;
2553 u64 branch_from = 0;
2554 int err;
2555
2556 for (i = 0; i < chain_nr; i++) {
2557 if (chain->ips[i] == PERF_CONTEXT_USER)
2558 break;
2559 }
2560
2561 /* LBR only affects the user callchain */
2562 if (i == chain_nr)
2563 return 0;
2564
2565 if (thread->lbr_stitch_enable && !sample->no_hw_idx &&
2566 (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2567 lbr_stitch = thread->lbr_stitch;
2568
2569 stitched_lbr = has_stitched_lbr(thread, sample,
2570 &lbr_stitch->prev_sample,
2571 max_lbr, callee);
2572
2573 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2574 list_replace_init(&lbr_stitch->lists,
2575 &lbr_stitch->free_lists);
2576 }
2577 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2578 }
2579
2580 if (callee) {
2581 /* Add kernel ip */
2582 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2583 parent, root_al, branch_from,
2584 true, i);
2585 if (err)
2586 goto error;
2587
2588 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2589 root_al, &branch_from, true);
2590 if (err)
2591 goto error;
2592
2593 if (stitched_lbr) {
2594 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2595 if (err)
2596 goto error;
2597 }
2598
2599 } else {
2600 if (stitched_lbr) {
2601 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2602 if (err)
2603 goto error;
2604 }
2605 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2606 root_al, &branch_from, false);
2607 if (err)
2608 goto error;
2609
2610 /* Add kernel ip */
2611 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2612 parent, root_al, branch_from,
2613 false, i);
2614 if (err)
2615 goto error;
2616 }
2617 return 1;
2618
2619error:
2620 return (err < 0) ? err : 0;
2621}
2622
2623static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2624 struct callchain_cursor *cursor,
2625 struct symbol **parent,
2626 struct addr_location *root_al,
2627 u8 *cpumode, int ent)
2628{
2629 int err = 0;
2630
2631 while (--ent >= 0) {
2632 u64 ip = chain->ips[ent];
2633
2634 if (ip >= PERF_CONTEXT_MAX) {
2635 err = add_callchain_ip(thread, cursor, parent,
2636 root_al, cpumode, ip,
2637 false, NULL, NULL, 0);
2638 break;
2639 }
2640 }
2641 return err;
2642}
2643
2644static int thread__resolve_callchain_sample(struct thread *thread,
2645 struct callchain_cursor *cursor,
2646 struct evsel *evsel,
2647 struct perf_sample *sample,
2648 struct symbol **parent,
2649 struct addr_location *root_al,
2650 int max_stack)
2651{
2652 struct branch_stack *branch = sample->branch_stack;
2653 struct branch_entry *entries = perf_sample__branch_entries(sample);
2654 struct ip_callchain *chain = sample->callchain;
2655 int chain_nr = 0;
2656 u8 cpumode = PERF_RECORD_MISC_USER;
2657 int i, j, err, nr_entries;
2658 int skip_idx = -1;
2659 int first_call = 0;
2660
2661 if (chain)
2662 chain_nr = chain->nr;
2663
2664 if (evsel__has_branch_callstack(evsel)) {
2665 struct perf_env *env = evsel__env(evsel);
2666
2667 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2668 root_al, max_stack,
2669 !env ? 0 : env->max_branches);
2670 if (err)
2671 return (err < 0) ? err : 0;
2672 }
2673
2674 /*
2675 * Based on DWARF debug information, some architectures skip
2676 * a callchain entry saved by the kernel.
2677 */
2678 skip_idx = arch_skip_callchain_idx(thread, chain);
2679
2680 /*
2681 * Add branches to call stack for easier browsing. This gives
2682 * more context for a sample than just the callers.
2683 *
2684 * This uses individual histograms of paths compared to the
2685 * aggregated histograms the normal LBR mode uses.
2686 *
2687 * Limitations for now:
2688 * - No extra filters
2689 * - No annotations (should annotate somehow)
2690 */
2691
2692 if (branch && callchain_param.branch_callstack) {
2693 int nr = min(max_stack, (int)branch->nr);
2694 struct branch_entry be[nr];
2695 struct iterations iter[nr];
2696
2697 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2698 pr_warning("corrupted branch chain. skipping...\n");
2699 goto check_calls;
2700 }
2701
2702 for (i = 0; i < nr; i++) {
2703 if (callchain_param.order == ORDER_CALLEE) {
2704 be[i] = entries[i];
2705
2706 if (chain == NULL)
2707 continue;
2708
2709 /*
2710 * Check for overlap into the callchain.
2711 * The return address is one off compared to
2712 * the branch entry. To adjust for this
2713 * assume the calling instruction is not longer
2714 * than 8 bytes.
2715 */
2716 if (i == skip_idx ||
2717 chain->ips[first_call] >= PERF_CONTEXT_MAX)
2718 first_call++;
2719 else if (be[i].from < chain->ips[first_call] &&
2720 be[i].from >= chain->ips[first_call] - 8)
2721 first_call++;
2722 } else
2723 be[i] = entries[branch->nr - i - 1];
2724 }
2725
2726 memset(iter, 0, sizeof(struct iterations) * nr);
2727 nr = remove_loops(be, nr, iter);
2728
2729 for (i = 0; i < nr; i++) {
2730 err = add_callchain_ip(thread, cursor, parent,
2731 root_al,
2732 NULL, be[i].to,
2733 true, &be[i].flags,
2734 NULL, be[i].from);
2735
2736 if (!err)
2737 err = add_callchain_ip(thread, cursor, parent, root_al,
2738 NULL, be[i].from,
2739 true, &be[i].flags,
2740 &iter[i], 0);
2741 if (err == -EINVAL)
2742 break;
2743 if (err)
2744 return err;
2745 }
2746
2747 if (chain_nr == 0)
2748 return 0;
2749
2750 chain_nr -= nr;
2751 }
2752
2753check_calls:
2754 if (chain && callchain_param.order != ORDER_CALLEE) {
2755 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2756 &cpumode, chain->nr - first_call);
2757 if (err)
2758 return (err < 0) ? err : 0;
2759 }
2760 for (i = first_call, nr_entries = 0;
2761 i < chain_nr && nr_entries < max_stack; i++) {
2762 u64 ip;
2763
2764 if (callchain_param.order == ORDER_CALLEE)
2765 j = i;
2766 else
2767 j = chain->nr - i - 1;
2768
2769#ifdef HAVE_SKIP_CALLCHAIN_IDX
2770 if (j == skip_idx)
2771 continue;
2772#endif
2773 ip = chain->ips[j];
2774 if (ip < PERF_CONTEXT_MAX)
2775 ++nr_entries;
2776 else if (callchain_param.order != ORDER_CALLEE) {
2777 err = find_prev_cpumode(chain, thread, cursor, parent,
2778 root_al, &cpumode, j);
2779 if (err)
2780 return (err < 0) ? err : 0;
2781 continue;
2782 }
2783
2784 err = add_callchain_ip(thread, cursor, parent,
2785 root_al, &cpumode, ip,
2786 false, NULL, NULL, 0);
2787
2788 if (err)
2789 return (err < 0) ? err : 0;
2790 }
2791
2792 return 0;
2793}
2794
2795static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2796{
2797 struct symbol *sym = ms->sym;
2798 struct map *map = ms->map;
2799 struct inline_node *inline_node;
2800 struct inline_list *ilist;
2801 u64 addr;
2802 int ret = 1;
2803
2804 if (!symbol_conf.inline_name || !map || !sym)
2805 return ret;
2806
2807 addr = map__map_ip(map, ip);
2808 addr = map__rip_2objdump(map, addr);
2809
2810 inline_node = inlines__tree_find(&map->dso->inlined_nodes, addr);
2811 if (!inline_node) {
2812 inline_node = dso__parse_addr_inlines(map->dso, addr, sym);
2813 if (!inline_node)
2814 return ret;
2815 inlines__tree_insert(&map->dso->inlined_nodes, inline_node);
2816 }
2817
2818 list_for_each_entry(ilist, &inline_node->val, list) {
2819 struct map_symbol ilist_ms = {
2820 .maps = ms->maps,
2821 .map = map,
2822 .sym = ilist->symbol,
2823 };
2824 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2825 NULL, 0, 0, 0, ilist->srcline);
2826
2827 if (ret != 0)
2828 return ret;
2829 }
2830
2831 return ret;
2832}
2833
2834static int unwind_entry(struct unwind_entry *entry, void *arg)
2835{
2836 struct callchain_cursor *cursor = arg;
2837 const char *srcline = NULL;
2838 u64 addr = entry->ip;
2839
2840 if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2841 return 0;
2842
2843 if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2844 return 0;
2845
2846 /*
2847 * Convert entry->ip from a virtual address to an offset in
2848 * its corresponding binary.
2849 */
2850 if (entry->ms.map)
2851 addr = map__map_ip(entry->ms.map, entry->ip);
2852
2853 srcline = callchain_srcline(&entry->ms, addr);
2854 return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2855 false, NULL, 0, 0, 0, srcline);
2856}
2857
2858static int thread__resolve_callchain_unwind(struct thread *thread,
2859 struct callchain_cursor *cursor,
2860 struct evsel *evsel,
2861 struct perf_sample *sample,
2862 int max_stack)
2863{
2864 /* Can we do dwarf post unwind? */
2865 if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2866 (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2867 return 0;
2868
2869 /* Bail out if nothing was captured. */
2870 if ((!sample->user_regs.regs) ||
2871 (!sample->user_stack.size))
2872 return 0;
2873
2874 return unwind__get_entries(unwind_entry, cursor,
2875 thread, sample, max_stack);
2876}
2877
2878int thread__resolve_callchain(struct thread *thread,
2879 struct callchain_cursor *cursor,
2880 struct evsel *evsel,
2881 struct perf_sample *sample,
2882 struct symbol **parent,
2883 struct addr_location *root_al,
2884 int max_stack)
2885{
2886 int ret = 0;
2887
2888 callchain_cursor_reset(cursor);
2889
2890 if (callchain_param.order == ORDER_CALLEE) {
2891 ret = thread__resolve_callchain_sample(thread, cursor,
2892 evsel, sample,
2893 parent, root_al,
2894 max_stack);
2895 if (ret)
2896 return ret;
2897 ret = thread__resolve_callchain_unwind(thread, cursor,
2898 evsel, sample,
2899 max_stack);
2900 } else {
2901 ret = thread__resolve_callchain_unwind(thread, cursor,
2902 evsel, sample,
2903 max_stack);
2904 if (ret)
2905 return ret;
2906 ret = thread__resolve_callchain_sample(thread, cursor,
2907 evsel, sample,
2908 parent, root_al,
2909 max_stack);
2910 }
2911
2912 return ret;
2913}
2914
2915int machine__for_each_thread(struct machine *machine,
2916 int (*fn)(struct thread *thread, void *p),
2917 void *priv)
2918{
2919 struct threads *threads;
2920 struct rb_node *nd;
2921 struct thread *thread;
2922 int rc = 0;
2923 int i;
2924
2925 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
2926 threads = &machine->threads[i];
2927 for (nd = rb_first_cached(&threads->entries); nd;
2928 nd = rb_next(nd)) {
2929 thread = rb_entry(nd, struct thread, rb_node);
2930 rc = fn(thread, priv);
2931 if (rc != 0)
2932 return rc;
2933 }
2934
2935 list_for_each_entry(thread, &threads->dead, node) {
2936 rc = fn(thread, priv);
2937 if (rc != 0)
2938 return rc;
2939 }
2940 }
2941 return rc;
2942}
2943
2944int machines__for_each_thread(struct machines *machines,
2945 int (*fn)(struct thread *thread, void *p),
2946 void *priv)
2947{
2948 struct rb_node *nd;
2949 int rc = 0;
2950
2951 rc = machine__for_each_thread(&machines->host, fn, priv);
2952 if (rc != 0)
2953 return rc;
2954
2955 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
2956 struct machine *machine = rb_entry(nd, struct machine, rb_node);
2957
2958 rc = machine__for_each_thread(machine, fn, priv);
2959 if (rc != 0)
2960 return rc;
2961 }
2962 return rc;
2963}
2964
2965pid_t machine__get_current_tid(struct machine *machine, int cpu)
2966{
2967 int nr_cpus = min(machine->env->nr_cpus_online, MAX_NR_CPUS);
2968
2969 if (cpu < 0 || cpu >= nr_cpus || !machine->current_tid)
2970 return -1;
2971
2972 return machine->current_tid[cpu];
2973}
2974
2975int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
2976 pid_t tid)
2977{
2978 struct thread *thread;
2979 int nr_cpus = min(machine->env->nr_cpus_online, MAX_NR_CPUS);
2980
2981 if (cpu < 0)
2982 return -EINVAL;
2983
2984 if (!machine->current_tid) {
2985 int i;
2986
2987 machine->current_tid = calloc(nr_cpus, sizeof(pid_t));
2988 if (!machine->current_tid)
2989 return -ENOMEM;
2990 for (i = 0; i < nr_cpus; i++)
2991 machine->current_tid[i] = -1;
2992 }
2993
2994 if (cpu >= nr_cpus) {
2995 pr_err("Requested CPU %d too large. ", cpu);
2996 pr_err("Consider raising MAX_NR_CPUS\n");
2997 return -EINVAL;
2998 }
2999
3000 machine->current_tid[cpu] = tid;
3001
3002 thread = machine__findnew_thread(machine, pid, tid);
3003 if (!thread)
3004 return -ENOMEM;
3005
3006 thread->cpu = cpu;
3007 thread__put(thread);
3008
3009 return 0;
3010}
3011
3012/*
3013 * Compares the raw arch string. N.B. see instead perf_env__arch() if a
3014 * normalized arch is needed.
3015 */
3016bool machine__is(struct machine *machine, const char *arch)
3017{
3018 return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3019}
3020
3021int machine__nr_cpus_avail(struct machine *machine)
3022{
3023 return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3024}
3025
3026int machine__get_kernel_start(struct machine *machine)
3027{
3028 struct map *map = machine__kernel_map(machine);
3029 int err = 0;
3030
3031 /*
3032 * The only addresses above 2^63 are kernel addresses of a 64-bit
3033 * kernel. Note that addresses are unsigned so that on a 32-bit system
3034 * all addresses including kernel addresses are less than 2^32. In
3035 * that case (32-bit system), if the kernel mapping is unknown, all
3036 * addresses will be assumed to be in user space - see
3037 * machine__kernel_ip().
3038 */
3039 machine->kernel_start = 1ULL << 63;
3040 if (map) {
3041 err = map__load(map);
3042 /*
3043 * On x86_64, PTI entry trampolines are less than the
3044 * start of kernel text, but still above 2^63. So leave
3045 * kernel_start = 1ULL << 63 for x86_64.
3046 */
3047 if (!err && !machine__is(machine, "x86_64"))
3048 machine->kernel_start = map->start;
3049 }
3050 return err;
3051}
3052
3053u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3054{
3055 u8 addr_cpumode = cpumode;
3056 bool kernel_ip;
3057
3058 if (!machine->single_address_space)
3059 goto out;
3060
3061 kernel_ip = machine__kernel_ip(machine, addr);
3062 switch (cpumode) {
3063 case PERF_RECORD_MISC_KERNEL:
3064 case PERF_RECORD_MISC_USER:
3065 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3066 PERF_RECORD_MISC_USER;
3067 break;
3068 case PERF_RECORD_MISC_GUEST_KERNEL:
3069 case PERF_RECORD_MISC_GUEST_USER:
3070 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3071 PERF_RECORD_MISC_GUEST_USER;
3072 break;
3073 default:
3074 break;
3075 }
3076out:
3077 return addr_cpumode;
3078}
3079
3080struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3081{
3082 return dsos__findnew_id(&machine->dsos, filename, id);
3083}
3084
3085struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3086{
3087 return machine__findnew_dso_id(machine, filename, NULL);
3088}
3089
3090char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3091{
3092 struct machine *machine = vmachine;
3093 struct map *map;
3094 struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3095
3096 if (sym == NULL)
3097 return NULL;
3098
3099 *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
3100 *addrp = map->unmap_ip(map, sym->start);
3101 return sym->name;
3102}