?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'trace-printf-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/trace...
[drm/drm-misc.git] / tools / perf / util / machine.c
blob4f0ac998b0ccfd7a6dfd2e1ac9e9ca95d0b59438
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 "mem-info.h"
20 #include "path.h"
21 #include "srcline.h"
22 #include "symbol.h"
23 #include "sort.h"
24 #include "strlist.h"
25 #include "target.h"
26 #include "thread.h"
27 #include "util.h"
28 #include "vdso.h"
29 #include <stdbool.h>
30 #include <sys/types.h>
31 #include <sys/stat.h>
32 #include <unistd.h>
33 #include "unwind.h"
34 #include "linux/hash.h"
35 #include "asm/bug.h"
36 #include "bpf-event.h"
37 #include <internal/lib.h> // page_size
38 #include "cgroup.h"
39 #include "arm64-frame-pointer-unwind-support.h"
40
41 #include <linux/ctype.h>
42 #include <symbol/kallsyms.h>
43 #include <linux/mman.h>
44 #include <linux/string.h>
45 #include <linux/zalloc.h>
46
47 static struct dso *machine__kernel_dso(struct machine *machine)
48 {
49 return map__dso(machine->vmlinux_map);
50 }
51
52 static int machine__set_mmap_name(struct machine *machine)
53 {
54 if (machine__is_host(machine))
55 machine->mmap_name = strdup("[kernel.kallsyms]");
56 else if (machine__is_default_guest(machine))
57 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
58 else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
59 machine->pid) < 0)
60 machine->mmap_name = NULL;
61
62 return machine->mmap_name ? 0 : -ENOMEM;
63 }
64
65 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
66 {
67 char comm[64];
68
69 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
70 thread__set_comm(thread, comm, 0);
71 }
72
73 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
74 {
75 int err = -ENOMEM;
76
77 memset(machine, 0, sizeof(*machine));
78 machine->kmaps = maps__new(machine);
79 if (machine->kmaps == NULL)
80 return -ENOMEM;
81
82 RB_CLEAR_NODE(&machine->rb_node);
83 dsos__init(&machine->dsos);
84
85 threads__init(&machine->threads);
86
87 machine->vdso_info = NULL;
88 machine->env = NULL;
89
90 machine->pid = pid;
91
92 machine->id_hdr_size = 0;
93 machine->kptr_restrict_warned = false;
94 machine->comm_exec = false;
95 machine->kernel_start = 0;
96 machine->vmlinux_map = NULL;
97
98 machine->root_dir = strdup(root_dir);
99 if (machine->root_dir == NULL)
100 goto out;
101
102 if (machine__set_mmap_name(machine))
103 goto out;
104
105 if (pid != HOST_KERNEL_ID) {
106 struct thread *thread = machine__findnew_thread(machine, -1,
107 pid);
108
109 if (thread == NULL)
110 goto out;
111
112 thread__set_guest_comm(thread, pid);
113 thread__put(thread);
114 }
115
116 machine->current_tid = NULL;
117 err = 0;
118
119 out:
120 if (err) {
121 zfree(&machine->kmaps);
122 zfree(&machine->root_dir);
123 zfree(&machine->mmap_name);
124 }
125 return 0;
126 }
127
128 struct machine *machine__new_host(void)
129 {
130 struct machine *machine = malloc(sizeof(*machine));
131
132 if (machine != NULL) {
133 machine__init(machine, "", HOST_KERNEL_ID);
134
135 if (machine__create_kernel_maps(machine) < 0)
136 goto out_delete;
137 }
138
139 return machine;
140 out_delete:
141 free(machine);
142 return NULL;
143 }
144
145 struct machine *machine__new_kallsyms(void)
146 {
147 struct machine *machine = machine__new_host();
148 /*
149 * FIXME:
150 * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
151 * ask for not using the kcore parsing code, once this one is fixed
152 * to create a map per module.
153 */
154 if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
155 machine__delete(machine);
156 machine = NULL;
157 }
158
159 return machine;
160 }
161
162 void machine__delete_threads(struct machine *machine)
163 {
164 threads__remove_all_threads(&machine->threads);
165 }
166
167 void machine__exit(struct machine *machine)
168 {
169 if (machine == NULL)
170 return;
171
172 machine__destroy_kernel_maps(machine);
173 maps__zput(machine->kmaps);
174 dsos__exit(&machine->dsos);
175 machine__exit_vdso(machine);
176 zfree(&machine->root_dir);
177 zfree(&machine->mmap_name);
178 zfree(&machine->current_tid);
179 zfree(&machine->kallsyms_filename);
180
181 threads__exit(&machine->threads);
182 }
183
184 void machine__delete(struct machine *machine)
185 {
186 if (machine) {
187 machine__exit(machine);
188 free(machine);
189 }
190 }
191
192 void machines__init(struct machines *machines)
193 {
194 machine__init(&machines->host, "", HOST_KERNEL_ID);
195 machines->guests = RB_ROOT_CACHED;
196 }
197
198 void machines__exit(struct machines *machines)
199 {
200 machine__exit(&machines->host);
201 /* XXX exit guest */
202 }
203
204 struct machine *machines__add(struct machines *machines, pid_t pid,
205 const char *root_dir)
206 {
207 struct rb_node **p = &machines->guests.rb_root.rb_node;
208 struct rb_node *parent = NULL;
209 struct machine *pos, *machine = malloc(sizeof(*machine));
210 bool leftmost = true;
211
212 if (machine == NULL)
213 return NULL;
214
215 if (machine__init(machine, root_dir, pid) != 0) {
216 free(machine);
217 return NULL;
218 }
219
220 while (*p != NULL) {
221 parent = *p;
222 pos = rb_entry(parent, struct machine, rb_node);
223 if (pid < pos->pid)
224 p = &(*p)->rb_left;
225 else {
226 p = &(*p)->rb_right;
227 leftmost = false;
228 }
229 }
230
231 rb_link_node(&machine->rb_node, parent, p);
232 rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
233
234 machine->machines = machines;
235
236 return machine;
237 }
238
239 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
240 {
241 struct rb_node *nd;
242
243 machines->host.comm_exec = comm_exec;
244
245 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
246 struct machine *machine = rb_entry(nd, struct machine, rb_node);
247
248 machine->comm_exec = comm_exec;
249 }
250 }
251
252 struct machine *machines__find(struct machines *machines, pid_t pid)
253 {
254 struct rb_node **p = &machines->guests.rb_root.rb_node;
255 struct rb_node *parent = NULL;
256 struct machine *machine;
257 struct machine *default_machine = NULL;
258
259 if (pid == HOST_KERNEL_ID)
260 return &machines->host;
261
262 while (*p != NULL) {
263 parent = *p;
264 machine = rb_entry(parent, struct machine, rb_node);
265 if (pid < machine->pid)
266 p = &(*p)->rb_left;
267 else if (pid > machine->pid)
268 p = &(*p)->rb_right;
269 else
270 return machine;
271 if (!machine->pid)
272 default_machine = machine;
273 }
274
275 return default_machine;
276 }
277
278 struct machine *machines__findnew(struct machines *machines, pid_t pid)
279 {
280 char path[PATH_MAX];
281 const char *root_dir = "";
282 struct machine *machine = machines__find(machines, pid);
283
284 if (machine && (machine->pid == pid))
285 goto out;
286
287 if ((pid != HOST_KERNEL_ID) &&
288 (pid != DEFAULT_GUEST_KERNEL_ID) &&
289 (symbol_conf.guestmount)) {
290 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
291 if (access(path, R_OK)) {
292 static struct strlist *seen;
293
294 if (!seen)
295 seen = strlist__new(NULL, NULL);
296
297 if (!strlist__has_entry(seen, path)) {
298 pr_err("Can't access file %s\n", path);
299 strlist__add(seen, path);
300 }
301 machine = NULL;
302 goto out;
303 }
304 root_dir = path;
305 }
306
307 machine = machines__add(machines, pid, root_dir);
308 out:
309 return machine;
310 }
311
312 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
313 {
314 struct machine *machine = machines__find(machines, pid);
315
316 if (!machine)
317 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
318 return machine;
319 }
320
321 /*
322 * A common case for KVM test programs is that the test program acts as the
323 * hypervisor, creating, running and destroying the virtual machine, and
324 * providing the guest object code from its own object code. In this case,
325 * the VM is not running an OS, but only the functions loaded into it by the
326 * hypervisor test program, and conveniently, loaded at the same virtual
327 * addresses.
328 *
329 * Normally to resolve addresses, MMAP events are needed to map addresses
330 * back to the object code and debug symbols for that object code.
331 *
332 * Currently, there is no way to get such mapping information from guests
333 * but, in the scenario described above, the guest has the same mappings
334 * as the hypervisor, so support for that scenario can be achieved.
335 *
336 * To support that, copy the host thread's maps to the guest thread's maps.
337 * Note, we do not discover the guest until we encounter a guest event,
338 * which works well because it is not until then that we know that the host
339 * thread's maps have been set up.
340 *
341 * This function returns the guest thread. Apart from keeping the data
342 * structures sane, using a thread belonging to the guest machine, instead
343 * of the host thread, allows it to have its own comm (refer
344 * thread__set_guest_comm()).
345 */
346 static struct thread *findnew_guest_code(struct machine *machine,
347 struct machine *host_machine,
348 pid_t pid)
349 {
350 struct thread *host_thread;
351 struct thread *thread;
352 int err;
353
354 if (!machine)
355 return NULL;
356
357 thread = machine__findnew_thread(machine, -1, pid);
358 if (!thread)
359 return NULL;
360
361 /* Assume maps are set up if there are any */
362 if (!maps__empty(thread__maps(thread)))
363 return thread;
364
365 host_thread = machine__find_thread(host_machine, -1, pid);
366 if (!host_thread)
367 goto out_err;
368
369 thread__set_guest_comm(thread, pid);
370
371 /*
372 * Guest code can be found in hypervisor process at the same address
373 * so copy host maps.
374 */
375 err = maps__copy_from(thread__maps(thread), thread__maps(host_thread));
376 thread__put(host_thread);
377 if (err)
378 goto out_err;
379
380 return thread;
381
382 out_err:
383 thread__zput(thread);
384 return NULL;
385 }
386
387 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
388 {
389 struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
390 struct machine *machine = machines__findnew(machines, pid);
391
392 return findnew_guest_code(machine, host_machine, pid);
393 }
394
395 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
396 {
397 struct machines *machines = machine->machines;
398 struct machine *host_machine;
399
400 if (!machines)
401 return NULL;
402
403 host_machine = machines__find(machines, HOST_KERNEL_ID);
404
405 return findnew_guest_code(machine, host_machine, pid);
406 }
407
408 void machines__process_guests(struct machines *machines,
409 machine__process_t process, void *data)
410 {
411 struct rb_node *nd;
412
413 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
414 struct machine *pos = rb_entry(nd, struct machine, rb_node);
415 process(pos, data);
416 }
417 }
418
419 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
420 {
421 struct rb_node *node;
422 struct machine *machine;
423
424 machines->host.id_hdr_size = id_hdr_size;
425
426 for (node = rb_first_cached(&machines->guests); node;
427 node = rb_next(node)) {
428 machine = rb_entry(node, struct machine, rb_node);
429 machine->id_hdr_size = id_hdr_size;
430 }
431
432 return;
433 }
434
435 static void machine__update_thread_pid(struct machine *machine,
436 struct thread *th, pid_t pid)
437 {
438 struct thread *leader;
439
440 if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
441 return;
442
443 thread__set_pid(th, pid);
444
445 if (thread__pid(th) == thread__tid(th))
446 return;
447
448 leader = machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
449 if (!leader)
450 goto out_err;
451
452 if (!thread__maps(leader))
453 thread__set_maps(leader, maps__new(machine));
454
455 if (!thread__maps(leader))
456 goto out_err;
457
458 if (thread__maps(th) == thread__maps(leader))
459 goto out_put;
460
461 if (thread__maps(th)) {
462 /*
463 * Maps are created from MMAP events which provide the pid and
464 * tid. Consequently there never should be any maps on a thread
465 * with an unknown pid. Just print an error if there are.
466 */
467 if (!maps__empty(thread__maps(th)))
468 pr_err("Discarding thread maps for %d:%d\n",
469 thread__pid(th), thread__tid(th));
470 maps__put(thread__maps(th));
471 }
472
473 thread__set_maps(th, maps__get(thread__maps(leader)));
474 out_put:
475 thread__put(leader);
476 return;
477 out_err:
478 pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
479 goto out_put;
480 }
481
482 /*
483 * Caller must eventually drop thread->refcnt returned with a successful
484 * lookup/new thread inserted.
485 */
486 static struct thread *__machine__findnew_thread(struct machine *machine,
487 pid_t pid,
488 pid_t tid,
489 bool create)
490 {
491 struct thread *th = threads__find(&machine->threads, tid);
492 bool created;
493
494 if (th) {
495 machine__update_thread_pid(machine, th, pid);
496 return th;
497 }
498 if (!create)
499 return NULL;
500
501 th = threads__findnew(&machine->threads, pid, tid, &created);
502 if (created) {
503 /*
504 * We have to initialize maps separately after rb tree is
505 * updated.
506 *
507 * The reason is that we call machine__findnew_thread within
508 * thread__init_maps to find the thread leader and that would
509 * screwed the rb tree.
510 */
511 if (thread__init_maps(th, machine)) {
512 pr_err("Thread init failed thread %d\n", pid);
513 threads__remove(&machine->threads, th);
514 thread__put(th);
515 return NULL;
516 }
517 } else
518 machine__update_thread_pid(machine, th, pid);
519
520 return th;
521 }
522
523 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
524 {
525 return __machine__findnew_thread(machine, pid, tid, /*create=*/true);
526 }
527
528 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
529 pid_t tid)
530 {
531 return __machine__findnew_thread(machine, pid, tid, /*create=*/false);
532 }
533
534 /*
535 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
536 * So here a single thread is created for that, but actually there is a separate
537 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
538 * is only 1. That causes problems for some tools, requiring workarounds. For
539 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
540 */
541 struct thread *machine__idle_thread(struct machine *machine)
542 {
543 struct thread *thread = machine__findnew_thread(machine, 0, 0);
544
545 if (!thread || thread__set_comm(thread, "swapper", 0) ||
546 thread__set_namespaces(thread, 0, NULL))
547 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
548
549 return thread;
550 }
551
552 struct comm *machine__thread_exec_comm(struct machine *machine,
553 struct thread *thread)
554 {
555 if (machine->comm_exec)
556 return thread__exec_comm(thread);
557 else
558 return thread__comm(thread);
559 }
560
561 int machine__process_comm_event(struct machine *machine, union perf_event *event,
562 struct perf_sample *sample)
563 {
564 struct thread *thread = machine__findnew_thread(machine,
565 event->comm.pid,
566 event->comm.tid);
567 bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
568 int err = 0;
569
570 if (exec)
571 machine->comm_exec = true;
572
573 if (dump_trace)
574 perf_event__fprintf_comm(event, stdout);
575
576 if (thread == NULL ||
577 __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
578 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
579 err = -1;
580 }
581
582 thread__put(thread);
583
584 return err;
585 }
586
587 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
588 union perf_event *event,
589 struct perf_sample *sample __maybe_unused)
590 {
591 struct thread *thread = machine__findnew_thread(machine,
592 event->namespaces.pid,
593 event->namespaces.tid);
594 int err = 0;
595
596 WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
597 "\nWARNING: kernel seems to support more namespaces than perf"
598 " tool.\nTry updating the perf tool..\n\n");
599
600 WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
601 "\nWARNING: perf tool seems to support more namespaces than"
602 " the kernel.\nTry updating the kernel..\n\n");
603
604 if (dump_trace)
605 perf_event__fprintf_namespaces(event, stdout);
606
607 if (thread == NULL ||
608 thread__set_namespaces(thread, sample->time, &event->namespaces)) {
609 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
610 err = -1;
611 }
612
613 thread__put(thread);
614
615 return err;
616 }
617
618 int machine__process_cgroup_event(struct machine *machine,
619 union perf_event *event,
620 struct perf_sample *sample __maybe_unused)
621 {
622 struct cgroup *cgrp;
623
624 if (dump_trace)
625 perf_event__fprintf_cgroup(event, stdout);
626
627 cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
628 if (cgrp == NULL)
629 return -ENOMEM;
630
631 return 0;
632 }
633
634 int machine__process_lost_event(struct machine *machine __maybe_unused,
635 union perf_event *event, struct perf_sample *sample __maybe_unused)
636 {
637 dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
638 event->lost.id, event->lost.lost);
639 return 0;
640 }
641
642 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
643 union perf_event *event, struct perf_sample *sample)
644 {
645 dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "%s\n",
646 sample->id, event->lost_samples.lost,
647 event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF ? " (BPF)" : "");
648 return 0;
649 }
650
651 int machine__process_aux_event(struct machine *machine __maybe_unused,
652 union perf_event *event)
653 {
654 if (dump_trace)
655 perf_event__fprintf_aux(event, stdout);
656 return 0;
657 }
658
659 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
660 union perf_event *event)
661 {
662 if (dump_trace)
663 perf_event__fprintf_itrace_start(event, stdout);
664 return 0;
665 }
666
667 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
668 union perf_event *event)
669 {
670 if (dump_trace)
671 perf_event__fprintf_aux_output_hw_id(event, stdout);
672 return 0;
673 }
674
675 int machine__process_switch_event(struct machine *machine __maybe_unused,
676 union perf_event *event)
677 {
678 if (dump_trace)
679 perf_event__fprintf_switch(event, stdout);
680 return 0;
681 }
682
683 static int machine__process_ksymbol_register(struct machine *machine,
684 union perf_event *event,
685 struct perf_sample *sample __maybe_unused)
686 {
687 struct symbol *sym;
688 struct dso *dso = NULL;
689 struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
690 int err = 0;
691
692 if (!map) {
693 dso = dso__new(event->ksymbol.name);
694
695 if (!dso) {
696 err = -ENOMEM;
697 goto out;
698 }
699 dso__set_kernel(dso, DSO_SPACE__KERNEL);
700 map = map__new2(0, dso);
701 if (!map) {
702 err = -ENOMEM;
703 goto out;
704 }
705 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
706 dso__set_binary_type(dso, DSO_BINARY_TYPE__OOL);
707 dso__data(dso)->file_size = event->ksymbol.len;
708 dso__set_loaded(dso);
709 }
710
711 map__set_start(map, event->ksymbol.addr);
712 map__set_end(map, map__start(map) + event->ksymbol.len);
713 err = maps__insert(machine__kernel_maps(machine), map);
714 if (err) {
715 err = -ENOMEM;
716 goto out;
717 }
718
719 dso__set_loaded(dso);
720
721 if (is_bpf_image(event->ksymbol.name)) {
722 dso__set_binary_type(dso, DSO_BINARY_TYPE__BPF_IMAGE);
723 dso__set_long_name(dso, "", false);
724 }
725 } else {
726 dso = dso__get(map__dso(map));
727 }
728
729 sym = symbol__new(map__map_ip(map, map__start(map)),
730 event->ksymbol.len,
731 0, 0, event->ksymbol.name);
732 if (!sym) {
733 err = -ENOMEM;
734 goto out;
735 }
736 dso__insert_symbol(dso, sym);
737 out:
738 map__put(map);
739 dso__put(dso);
740 return err;
741 }
742
743 static int machine__process_ksymbol_unregister(struct machine *machine,
744 union perf_event *event,
745 struct perf_sample *sample __maybe_unused)
746 {
747 struct symbol *sym;
748 struct map *map;
749
750 map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
751 if (!map)
752 return 0;
753
754 if (!RC_CHK_EQUAL(map, machine->vmlinux_map))
755 maps__remove(machine__kernel_maps(machine), map);
756 else {
757 struct dso *dso = map__dso(map);
758
759 sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
760 if (sym)
761 dso__delete_symbol(dso, sym);
762 }
763 map__put(map);
764 return 0;
765 }
766
767 int machine__process_ksymbol(struct machine *machine __maybe_unused,
768 union perf_event *event,
769 struct perf_sample *sample)
770 {
771 if (dump_trace)
772 perf_event__fprintf_ksymbol(event, stdout);
773
774 if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
775 return machine__process_ksymbol_unregister(machine, event,
776 sample);
777 return machine__process_ksymbol_register(machine, event, sample);
778 }
779
780 int machine__process_text_poke(struct machine *machine, union perf_event *event,
781 struct perf_sample *sample __maybe_unused)
782 {
783 struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
784 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
785 struct dso *dso = map ? map__dso(map) : NULL;
786
787 if (dump_trace)
788 perf_event__fprintf_text_poke(event, machine, stdout);
789
790 if (!event->text_poke.new_len)
791 goto out;
792
793 if (cpumode != PERF_RECORD_MISC_KERNEL) {
794 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
795 goto out;
796 }
797
798 if (dso) {
799 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
800 int ret;
801
802 /*
803 * Kernel maps might be changed when loading symbols so loading
804 * must be done prior to using kernel maps.
805 */
806 map__load(map);
807 ret = dso__data_write_cache_addr(dso, map, machine,
808 event->text_poke.addr,
809 new_bytes,
810 event->text_poke.new_len);
811 if (ret != event->text_poke.new_len)
812 pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
813 event->text_poke.addr);
814 } else {
815 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
816 event->text_poke.addr);
817 }
818 out:
819 map__put(map);
820 return 0;
821 }
822
823 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
824 const char *filename)
825 {
826 struct map *map = NULL;
827 struct kmod_path m;
828 struct dso *dso;
829 int err;
830
831 if (kmod_path__parse_name(&m, filename))
832 return NULL;
833
834 dso = dsos__findnew_module_dso(&machine->dsos, machine, &m, filename);
835 if (dso == NULL)
836 goto out;
837
838 map = map__new2(start, dso);
839 if (map == NULL)
840 goto out;
841
842 err = maps__insert(machine__kernel_maps(machine), map);
843 /* If maps__insert failed, return NULL. */
844 if (err) {
845 map__put(map);
846 map = NULL;
847 }
848 out:
849 /* put the dso here, corresponding to machine__findnew_module_dso */
850 dso__put(dso);
851 zfree(&m.name);
852 return map;
853 }
854
855 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
856 {
857 struct rb_node *nd;
858 size_t ret = dsos__fprintf(&machines->host.dsos, fp);
859
860 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
861 struct machine *pos = rb_entry(nd, struct machine, rb_node);
862 ret += dsos__fprintf(&pos->dsos, fp);
863 }
864
865 return ret;
866 }
867
868 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
869 bool (skip)(struct dso *dso, int parm), int parm)
870 {
871 return dsos__fprintf_buildid(&m->dsos, fp, skip, parm);
872 }
873
874 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
875 bool (skip)(struct dso *dso, int parm), int parm)
876 {
877 struct rb_node *nd;
878 size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
879
880 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
881 struct machine *pos = rb_entry(nd, struct machine, rb_node);
882 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
883 }
884 return ret;
885 }
886
887 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
888 {
889 int i;
890 size_t printed = 0;
891 struct dso *kdso = machine__kernel_dso(machine);
892
893 if (dso__has_build_id(kdso)) {
894 char filename[PATH_MAX];
895
896 if (dso__build_id_filename(kdso, filename, sizeof(filename), false))
897 printed += fprintf(fp, "[0] %s\n", filename);
898 }
899
900 for (i = 0; i < vmlinux_path__nr_entries; ++i) {
901 printed += fprintf(fp, "[%d] %s\n", i + dso__has_build_id(kdso),
902 vmlinux_path[i]);
903 }
904 return printed;
905 }
906
907 struct machine_fprintf_cb_args {
908 FILE *fp;
909 size_t printed;
910 };
911
912 static int machine_fprintf_cb(struct thread *thread, void *data)
913 {
914 struct machine_fprintf_cb_args *args = data;
915
916 /* TODO: handle fprintf errors. */
917 args->printed += thread__fprintf(thread, args->fp);
918 return 0;
919 }
920
921 size_t machine__fprintf(struct machine *machine, FILE *fp)
922 {
923 struct machine_fprintf_cb_args args = {
924 .fp = fp,
925 .printed = 0,
926 };
927 size_t ret = fprintf(fp, "Threads: %zu\n", threads__nr(&machine->threads));
928
929 machine__for_each_thread(machine, machine_fprintf_cb, &args);
930 return ret + args.printed;
931 }
932
933 static struct dso *machine__get_kernel(struct machine *machine)
934 {
935 const char *vmlinux_name = machine->mmap_name;
936 struct dso *kernel;
937
938 if (machine__is_host(machine)) {
939 if (symbol_conf.vmlinux_name)
940 vmlinux_name = symbol_conf.vmlinux_name;
941
942 kernel = machine__findnew_kernel(machine, vmlinux_name,
943 "[kernel]", DSO_SPACE__KERNEL);
944 } else {
945 if (symbol_conf.default_guest_vmlinux_name)
946 vmlinux_name = symbol_conf.default_guest_vmlinux_name;
947
948 kernel = machine__findnew_kernel(machine, vmlinux_name,
949 "[guest.kernel]",
950 DSO_SPACE__KERNEL_GUEST);
951 }
952
953 if (kernel != NULL && (!dso__has_build_id(kernel)))
954 dso__read_running_kernel_build_id(kernel, machine);
955
956 return kernel;
957 }
958
959 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
960 size_t bufsz)
961 {
962 if (machine__is_default_guest(machine))
963 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
964 else
965 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
966 }
967
968 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
969
970 /* Figure out the start address of kernel map from /proc/kallsyms.
971 * Returns the name of the start symbol in *symbol_name. Pass in NULL as
972 * symbol_name if it's not that important.
973 */
974 static int machine__get_running_kernel_start(struct machine *machine,
975 const char **symbol_name,
976 u64 *start, u64 *end)
977 {
978 char filename[PATH_MAX];
979 int i, err = -1;
980 const char *name;
981 u64 addr = 0;
982
983 machine__get_kallsyms_filename(machine, filename, PATH_MAX);
984
985 if (symbol__restricted_filename(filename, "/proc/kallsyms"))
986 return 0;
987
988 for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
989 err = kallsyms__get_function_start(filename, name, &addr);
990 if (!err)
991 break;
992 }
993
994 if (err)
995 return -1;
996
997 if (symbol_name)
998 *symbol_name = name;
999
1000 *start = addr;
1001
1002 err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1003 if (err)
1004 err = kallsyms__get_function_start(filename, "_etext", &addr);
1005 if (!err)
1006 *end = addr;
1007
1008 return 0;
1009 }
1010
1011 int machine__create_extra_kernel_map(struct machine *machine,
1012 struct dso *kernel,
1013 struct extra_kernel_map *xm)
1014 {
1015 struct kmap *kmap;
1016 struct map *map;
1017 int err;
1018
1019 map = map__new2(xm->start, kernel);
1020 if (!map)
1021 return -ENOMEM;
1022
1023 map__set_end(map, xm->end);
1024 map__set_pgoff(map, xm->pgoff);
1025
1026 kmap = map__kmap(map);
1027
1028 strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1029
1030 err = maps__insert(machine__kernel_maps(machine), map);
1031
1032 if (!err) {
1033 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1034 kmap->name, map__start(map), map__end(map));
1035 }
1036
1037 map__put(map);
1038
1039 return err;
1040 }
1041
1042 static u64 find_entry_trampoline(struct dso *dso)
1043 {
1044 /* Duplicates are removed so lookup all aliases */
1045 const char *syms[] = {
1046 "_entry_trampoline",
1047 "__entry_trampoline_start",
1048 "entry_SYSCALL_64_trampoline",
1049 };
1050 struct symbol *sym = dso__first_symbol(dso);
1051 unsigned int i;
1052
1053 for (; sym; sym = dso__next_symbol(sym)) {
1054 if (sym->binding != STB_GLOBAL)
1055 continue;
1056 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1057 if (!strcmp(sym->name, syms[i]))
1058 return sym->start;
1059 }
1060 }
1061
1062 return 0;
1063 }
1064
1065 /*
1066 * These values can be used for kernels that do not have symbols for the entry
1067 * trampolines in kallsyms.
1068 */
1069 #define X86_64_CPU_ENTRY_AREA_PER_CPU 0xfffffe0000000000ULL
1070 #define X86_64_CPU_ENTRY_AREA_SIZE 0x2c000
1071 #define X86_64_ENTRY_TRAMPOLINE 0x6000
1072
1073 struct machine__map_x86_64_entry_trampolines_args {
1074 struct maps *kmaps;
1075 bool found;
1076 };
1077
1078 static int machine__map_x86_64_entry_trampolines_cb(struct map *map, void *data)
1079 {
1080 struct machine__map_x86_64_entry_trampolines_args *args = data;
1081 struct map *dest_map;
1082 struct kmap *kmap = __map__kmap(map);
1083
1084 if (!kmap || !is_entry_trampoline(kmap->name))
1085 return 0;
1086
1087 dest_map = maps__find(args->kmaps, map__pgoff(map));
1088 if (RC_CHK_ACCESS(dest_map) != RC_CHK_ACCESS(map))
1089 map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1090
1091 map__put(dest_map);
1092 args->found = true;
1093 return 0;
1094 }
1095
1096 /* Map x86_64 PTI entry trampolines */
1097 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1098 struct dso *kernel)
1099 {
1100 struct machine__map_x86_64_entry_trampolines_args args = {
1101 .kmaps = machine__kernel_maps(machine),
1102 .found = false,
1103 };
1104 int nr_cpus_avail, cpu;
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_map(args.kmaps, machine__map_x86_64_entry_trampolines_cb, &args);
1112
1113 if (args.found || machine->trampolines_mapped)
1114 return 0;
1115
1116 pgoff = find_entry_trampoline(kernel);
1117 if (!pgoff)
1118 return 0;
1119
1120 nr_cpus_avail = machine__nr_cpus_avail(machine);
1121
1122 /* Add a 1 page map for each CPU's entry trampoline */
1123 for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1124 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1125 cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1126 X86_64_ENTRY_TRAMPOLINE;
1127 struct extra_kernel_map xm = {
1128 .start = va,
1129 .end = va + page_size,
1130 .pgoff = pgoff,
1131 };
1132
1133 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1134
1135 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1136 return -1;
1137 }
1138
1139 machine->trampolines_mapped = nr_cpus_avail;
1140
1141 return 0;
1142 }
1143
1144 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1145 struct dso *kernel __maybe_unused)
1146 {
1147 return 0;
1148 }
1149
1150 static int
1151 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1152 {
1153 /* In case of renewal the kernel map, destroy previous one */
1154 machine__destroy_kernel_maps(machine);
1155
1156 map__put(machine->vmlinux_map);
1157 machine->vmlinux_map = map__new2(0, kernel);
1158 if (machine->vmlinux_map == NULL)
1159 return -ENOMEM;
1160
1161 map__set_mapping_type(machine->vmlinux_map, MAPPING_TYPE__IDENTITY);
1162 return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1163 }
1164
1165 void machine__destroy_kernel_maps(struct machine *machine)
1166 {
1167 struct kmap *kmap;
1168 struct map *map = machine__kernel_map(machine);
1169
1170 if (map == NULL)
1171 return;
1172
1173 kmap = map__kmap(map);
1174 maps__remove(machine__kernel_maps(machine), map);
1175 if (kmap && kmap->ref_reloc_sym) {
1176 zfree((char **)&kmap->ref_reloc_sym->name);
1177 zfree(&kmap->ref_reloc_sym);
1178 }
1179
1180 map__zput(machine->vmlinux_map);
1181 }
1182
1183 int machines__create_guest_kernel_maps(struct machines *machines)
1184 {
1185 int ret = 0;
1186 struct dirent **namelist = NULL;
1187 int i, items = 0;
1188 char path[PATH_MAX];
1189 pid_t pid;
1190 char *endp;
1191
1192 if (symbol_conf.default_guest_vmlinux_name ||
1193 symbol_conf.default_guest_modules ||
1194 symbol_conf.default_guest_kallsyms) {
1195 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1196 }
1197
1198 if (symbol_conf.guestmount) {
1199 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1200 if (items <= 0)
1201 return -ENOENT;
1202 for (i = 0; i < items; i++) {
1203 if (!isdigit(namelist[i]->d_name[0])) {
1204 /* Filter out . and .. */
1205 continue;
1206 }
1207 pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1208 if ((*endp != '\0') ||
1209 (endp == namelist[i]->d_name) ||
1210 (errno == ERANGE)) {
1211 pr_debug("invalid directory (%s). Skipping.\n",
1212 namelist[i]->d_name);
1213 continue;
1214 }
1215 sprintf(path, "%s/%s/proc/kallsyms",
1216 symbol_conf.guestmount,
1217 namelist[i]->d_name);
1218 ret = access(path, R_OK);
1219 if (ret) {
1220 pr_debug("Can't access file %s\n", path);
1221 goto failure;
1222 }
1223 machines__create_kernel_maps(machines, pid);
1224 }
1225 failure:
1226 free(namelist);
1227 }
1228
1229 return ret;
1230 }
1231
1232 void machines__destroy_kernel_maps(struct machines *machines)
1233 {
1234 struct rb_node *next = rb_first_cached(&machines->guests);
1235
1236 machine__destroy_kernel_maps(&machines->host);
1237
1238 while (next) {
1239 struct machine *pos = rb_entry(next, struct machine, rb_node);
1240
1241 next = rb_next(&pos->rb_node);
1242 rb_erase_cached(&pos->rb_node, &machines->guests);
1243 machine__delete(pos);
1244 }
1245 }
1246
1247 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1248 {
1249 struct machine *machine = machines__findnew(machines, pid);
1250
1251 if (machine == NULL)
1252 return -1;
1253
1254 return machine__create_kernel_maps(machine);
1255 }
1256
1257 int machine__load_kallsyms(struct machine *machine, const char *filename)
1258 {
1259 struct map *map = machine__kernel_map(machine);
1260 struct dso *dso = map__dso(map);
1261 int ret = __dso__load_kallsyms(dso, filename, map, true);
1262
1263 if (ret > 0) {
1264 dso__set_loaded(dso);
1265 /*
1266 * Since /proc/kallsyms will have multiple sessions for the
1267 * kernel, with modules between them, fixup the end of all
1268 * sections.
1269 */
1270 maps__fixup_end(machine__kernel_maps(machine));
1271 }
1272
1273 return ret;
1274 }
1275
1276 int machine__load_vmlinux_path(struct machine *machine)
1277 {
1278 struct map *map = machine__kernel_map(machine);
1279 struct dso *dso = map__dso(map);
1280 int ret = dso__load_vmlinux_path(dso, map);
1281
1282 if (ret > 0)
1283 dso__set_loaded(dso);
1284
1285 return ret;
1286 }
1287
1288 static char *get_kernel_version(const char *root_dir)
1289 {
1290 char version[PATH_MAX];
1291 FILE *file;
1292 char *name, *tmp;
1293 const char *prefix = "Linux version ";
1294
1295 sprintf(version, "%s/proc/version", root_dir);
1296 file = fopen(version, "r");
1297 if (!file)
1298 return NULL;
1299
1300 tmp = fgets(version, sizeof(version), file);
1301 fclose(file);
1302 if (!tmp)
1303 return NULL;
1304
1305 name = strstr(version, prefix);
1306 if (!name)
1307 return NULL;
1308 name += strlen(prefix);
1309 tmp = strchr(name, ' ');
1310 if (tmp)
1311 *tmp = '\0';
1312
1313 return strdup(name);
1314 }
1315
1316 static bool is_kmod_dso(struct dso *dso)
1317 {
1318 return dso__symtab_type(dso) == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1319 dso__symtab_type(dso) == DSO_BINARY_TYPE__GUEST_KMODULE;
1320 }
1321
1322 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1323 {
1324 char *long_name;
1325 struct dso *dso;
1326 struct map *map = maps__find_by_name(maps, m->name);
1327
1328 if (map == NULL)
1329 return 0;
1330
1331 long_name = strdup(path);
1332 if (long_name == NULL) {
1333 map__put(map);
1334 return -ENOMEM;
1335 }
1336
1337 dso = map__dso(map);
1338 dso__set_long_name(dso, long_name, true);
1339 dso__kernel_module_get_build_id(dso, "");
1340
1341 /*
1342 * Full name could reveal us kmod compression, so
1343 * we need to update the symtab_type if needed.
1344 */
1345 if (m->comp && is_kmod_dso(dso)) {
1346 dso__set_symtab_type(dso, dso__symtab_type(dso)+1);
1347 dso__set_comp(dso, m->comp);
1348 }
1349 map__put(map);
1350 return 0;
1351 }
1352
1353 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1354 {
1355 struct dirent *dent;
1356 DIR *dir = opendir(dir_name);
1357 int ret = 0;
1358
1359 if (!dir) {
1360 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1361 return -1;
1362 }
1363
1364 while ((dent = readdir(dir)) != NULL) {
1365 char path[PATH_MAX];
1366 struct stat st;
1367
1368 /*sshfs might return bad dent->d_type, so we have to stat*/
1369 path__join(path, sizeof(path), dir_name, dent->d_name);
1370 if (stat(path, &st))
1371 continue;
1372
1373 if (S_ISDIR(st.st_mode)) {
1374 if (!strcmp(dent->d_name, ".") ||
1375 !strcmp(dent->d_name, ".."))
1376 continue;
1377
1378 /* Do not follow top-level source and build symlinks */
1379 if (depth == 0) {
1380 if (!strcmp(dent->d_name, "source") ||
1381 !strcmp(dent->d_name, "build"))
1382 continue;
1383 }
1384
1385 ret = maps__set_modules_path_dir(maps, path, depth + 1);
1386 if (ret < 0)
1387 goto out;
1388 } else {
1389 struct kmod_path m;
1390
1391 ret = kmod_path__parse_name(&m, dent->d_name);
1392 if (ret)
1393 goto out;
1394
1395 if (m.kmod)
1396 ret = maps__set_module_path(maps, path, &m);
1397
1398 zfree(&m.name);
1399
1400 if (ret)
1401 goto out;
1402 }
1403 }
1404
1405 out:
1406 closedir(dir);
1407 return ret;
1408 }
1409
1410 static int machine__set_modules_path(struct machine *machine)
1411 {
1412 char *version;
1413 char modules_path[PATH_MAX];
1414
1415 version = get_kernel_version(machine->root_dir);
1416 if (!version)
1417 return -1;
1418
1419 snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1420 machine->root_dir, version);
1421 free(version);
1422
1423 return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1424 }
1425 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1426 u64 *size __maybe_unused,
1427 const char *name __maybe_unused)
1428 {
1429 return 0;
1430 }
1431
1432 static int machine__create_module(void *arg, const char *name, u64 start,
1433 u64 size)
1434 {
1435 struct machine *machine = arg;
1436 struct map *map;
1437
1438 if (arch__fix_module_text_start(&start, &size, name) < 0)
1439 return -1;
1440
1441 map = machine__addnew_module_map(machine, start, name);
1442 if (map == NULL)
1443 return -1;
1444 map__set_end(map, start + size);
1445
1446 dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1447 map__put(map);
1448 return 0;
1449 }
1450
1451 static int machine__create_modules(struct machine *machine)
1452 {
1453 const char *modules;
1454 char path[PATH_MAX];
1455
1456 if (machine__is_default_guest(machine)) {
1457 modules = symbol_conf.default_guest_modules;
1458 } else {
1459 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1460 modules = path;
1461 }
1462
1463 if (symbol__restricted_filename(modules, "/proc/modules"))
1464 return -1;
1465
1466 if (modules__parse(modules, machine, machine__create_module))
1467 return -1;
1468
1469 if (!machine__set_modules_path(machine))
1470 return 0;
1471
1472 pr_debug("Problems setting modules path maps, continuing anyway...\n");
1473
1474 return 0;
1475 }
1476
1477 static void machine__set_kernel_mmap(struct machine *machine,
1478 u64 start, u64 end)
1479 {
1480 map__set_start(machine->vmlinux_map, start);
1481 map__set_end(machine->vmlinux_map, end);
1482 /*
1483 * Be a bit paranoid here, some perf.data file came with
1484 * a zero sized synthesized MMAP event for the kernel.
1485 */
1486 if (start == 0 && end == 0)
1487 map__set_end(machine->vmlinux_map, ~0ULL);
1488 }
1489
1490 static int machine__update_kernel_mmap(struct machine *machine,
1491 u64 start, u64 end)
1492 {
1493 struct map *orig, *updated;
1494 int err;
1495
1496 orig = machine->vmlinux_map;
1497 updated = map__get(orig);
1498
1499 machine->vmlinux_map = updated;
1500 maps__remove(machine__kernel_maps(machine), orig);
1501 machine__set_kernel_mmap(machine, start, end);
1502 err = maps__insert(machine__kernel_maps(machine), updated);
1503 map__put(orig);
1504
1505 return err;
1506 }
1507
1508 int machine__create_kernel_maps(struct machine *machine)
1509 {
1510 struct dso *kernel = machine__get_kernel(machine);
1511 const char *name = NULL;
1512 u64 start = 0, end = ~0ULL;
1513 int ret;
1514
1515 if (kernel == NULL)
1516 return -1;
1517
1518 ret = __machine__create_kernel_maps(machine, kernel);
1519 if (ret < 0)
1520 goto out_put;
1521
1522 if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1523 if (machine__is_host(machine))
1524 pr_debug("Problems creating module maps, "
1525 "continuing anyway...\n");
1526 else
1527 pr_debug("Problems creating module maps for guest %d, "
1528 "continuing anyway...\n", machine->pid);
1529 }
1530
1531 if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1532 if (name &&
1533 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1534 machine__destroy_kernel_maps(machine);
1535 ret = -1;
1536 goto out_put;
1537 }
1538
1539 /*
1540 * we have a real start address now, so re-order the kmaps
1541 * assume it's the last in the kmaps
1542 */
1543 ret = machine__update_kernel_mmap(machine, start, end);
1544 if (ret < 0)
1545 goto out_put;
1546 }
1547
1548 if (machine__create_extra_kernel_maps(machine, kernel))
1549 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1550
1551 if (end == ~0ULL) {
1552 /* update end address of the kernel map using adjacent module address */
1553 struct map *next = maps__find_next_entry(machine__kernel_maps(machine),
1554 machine__kernel_map(machine));
1555
1556 if (next) {
1557 machine__set_kernel_mmap(machine, start, map__start(next));
1558 map__put(next);
1559 }
1560 }
1561
1562 out_put:
1563 dso__put(kernel);
1564 return ret;
1565 }
1566
1567 static int machine__uses_kcore_cb(struct dso *dso, void *data __maybe_unused)
1568 {
1569 return dso__is_kcore(dso) ? 1 : 0;
1570 }
1571
1572 static bool machine__uses_kcore(struct machine *machine)
1573 {
1574 return dsos__for_each_dso(&machine->dsos, machine__uses_kcore_cb, NULL) != 0 ? true : false;
1575 }
1576
1577 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1578 struct extra_kernel_map *xm)
1579 {
1580 return machine__is(machine, "x86_64") &&
1581 is_entry_trampoline(xm->name);
1582 }
1583
1584 static int machine__process_extra_kernel_map(struct machine *machine,
1585 struct extra_kernel_map *xm)
1586 {
1587 struct dso *kernel = machine__kernel_dso(machine);
1588
1589 if (kernel == NULL)
1590 return -1;
1591
1592 return machine__create_extra_kernel_map(machine, kernel, xm);
1593 }
1594
1595 static int machine__process_kernel_mmap_event(struct machine *machine,
1596 struct extra_kernel_map *xm,
1597 struct build_id *bid)
1598 {
1599 enum dso_space_type dso_space;
1600 bool is_kernel_mmap;
1601 const char *mmap_name = machine->mmap_name;
1602
1603 /* If we have maps from kcore then we do not need or want any others */
1604 if (machine__uses_kcore(machine))
1605 return 0;
1606
1607 if (machine__is_host(machine))
1608 dso_space = DSO_SPACE__KERNEL;
1609 else
1610 dso_space = DSO_SPACE__KERNEL_GUEST;
1611
1612 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1613 if (!is_kernel_mmap && !machine__is_host(machine)) {
1614 /*
1615 * If the event was recorded inside the guest and injected into
1616 * the host perf.data file, then it will match a host mmap_name,
1617 * so try that - see machine__set_mmap_name().
1618 */
1619 mmap_name = "[kernel.kallsyms]";
1620 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1621 }
1622 if (xm->name[0] == '/' ||
1623 (!is_kernel_mmap && xm->name[0] == '[')) {
1624 struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1625
1626 if (map == NULL)
1627 goto out_problem;
1628
1629 map__set_end(map, map__start(map) + xm->end - xm->start);
1630
1631 if (build_id__is_defined(bid))
1632 dso__set_build_id(map__dso(map), bid);
1633
1634 map__put(map);
1635 } else if (is_kernel_mmap) {
1636 const char *symbol_name = xm->name + strlen(mmap_name);
1637 /*
1638 * Should be there already, from the build-id table in
1639 * the header.
1640 */
1641 struct dso *kernel = dsos__find_kernel_dso(&machine->dsos);
1642
1643 if (kernel == NULL)
1644 kernel = machine__findnew_dso(machine, machine->mmap_name);
1645 if (kernel == NULL)
1646 goto out_problem;
1647
1648 dso__set_kernel(kernel, dso_space);
1649 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1650 dso__put(kernel);
1651 goto out_problem;
1652 }
1653
1654 if (strstr(dso__long_name(kernel), "vmlinux"))
1655 dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1656
1657 if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1658 dso__put(kernel);
1659 goto out_problem;
1660 }
1661
1662 if (build_id__is_defined(bid))
1663 dso__set_build_id(kernel, bid);
1664
1665 /*
1666 * Avoid using a zero address (kptr_restrict) for the ref reloc
1667 * symbol. Effectively having zero here means that at record
1668 * time /proc/sys/kernel/kptr_restrict was non zero.
1669 */
1670 if (xm->pgoff != 0) {
1671 map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1672 symbol_name,
1673 xm->pgoff);
1674 }
1675
1676 if (machine__is_default_guest(machine)) {
1677 /*
1678 * preload dso of guest kernel and modules
1679 */
1680 dso__load(kernel, machine__kernel_map(machine));
1681 }
1682 dso__put(kernel);
1683 } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1684 return machine__process_extra_kernel_map(machine, xm);
1685 }
1686 return 0;
1687 out_problem:
1688 return -1;
1689 }
1690
1691 int machine__process_mmap2_event(struct machine *machine,
1692 union perf_event *event,
1693 struct perf_sample *sample)
1694 {
1695 struct thread *thread;
1696 struct map *map;
1697 struct dso_id dso_id = {
1698 .maj = event->mmap2.maj,
1699 .min = event->mmap2.min,
1700 .ino = event->mmap2.ino,
1701 .ino_generation = event->mmap2.ino_generation,
1702 };
1703 struct build_id __bid, *bid = NULL;
1704 int ret = 0;
1705
1706 if (dump_trace)
1707 perf_event__fprintf_mmap2(event, stdout);
1708
1709 if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1710 bid = &__bid;
1711 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1712 }
1713
1714 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1715 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1716 struct extra_kernel_map xm = {
1717 .start = event->mmap2.start,
1718 .end = event->mmap2.start + event->mmap2.len,
1719 .pgoff = event->mmap2.pgoff,
1720 };
1721
1722 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1723 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1724 if (ret < 0)
1725 goto out_problem;
1726 return 0;
1727 }
1728
1729 thread = machine__findnew_thread(machine, event->mmap2.pid,
1730 event->mmap2.tid);
1731 if (thread == NULL)
1732 goto out_problem;
1733
1734 map = map__new(machine, event->mmap2.start,
1735 event->mmap2.len, event->mmap2.pgoff,
1736 &dso_id, event->mmap2.prot,
1737 event->mmap2.flags, bid,
1738 event->mmap2.filename, thread);
1739
1740 if (map == NULL)
1741 goto out_problem_map;
1742
1743 ret = thread__insert_map(thread, map);
1744 if (ret)
1745 goto out_problem_insert;
1746
1747 thread__put(thread);
1748 map__put(map);
1749 return 0;
1750
1751 out_problem_insert:
1752 map__put(map);
1753 out_problem_map:
1754 thread__put(thread);
1755 out_problem:
1756 dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1757 return 0;
1758 }
1759
1760 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1761 struct perf_sample *sample)
1762 {
1763 struct thread *thread;
1764 struct map *map;
1765 u32 prot = 0;
1766 int ret = 0;
1767
1768 if (dump_trace)
1769 perf_event__fprintf_mmap(event, stdout);
1770
1771 if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1772 sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1773 struct extra_kernel_map xm = {
1774 .start = event->mmap.start,
1775 .end = event->mmap.start + event->mmap.len,
1776 .pgoff = event->mmap.pgoff,
1777 };
1778
1779 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1780 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1781 if (ret < 0)
1782 goto out_problem;
1783 return 0;
1784 }
1785
1786 thread = machine__findnew_thread(machine, event->mmap.pid,
1787 event->mmap.tid);
1788 if (thread == NULL)
1789 goto out_problem;
1790
1791 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1792 prot = PROT_EXEC;
1793
1794 map = map__new(machine, event->mmap.start,
1795 event->mmap.len, event->mmap.pgoff,
1796 NULL, prot, 0, NULL, event->mmap.filename, thread);
1797
1798 if (map == NULL)
1799 goto out_problem_map;
1800
1801 ret = thread__insert_map(thread, map);
1802 if (ret)
1803 goto out_problem_insert;
1804
1805 thread__put(thread);
1806 map__put(map);
1807 return 0;
1808
1809 out_problem_insert:
1810 map__put(map);
1811 out_problem_map:
1812 thread__put(thread);
1813 out_problem:
1814 dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1815 return 0;
1816 }
1817
1818 void machine__remove_thread(struct machine *machine, struct thread *th)
1819 {
1820 return threads__remove(&machine->threads, th);
1821 }
1822
1823 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1824 struct perf_sample *sample)
1825 {
1826 struct thread *thread = machine__find_thread(machine,
1827 event->fork.pid,
1828 event->fork.tid);
1829 struct thread *parent = machine__findnew_thread(machine,
1830 event->fork.ppid,
1831 event->fork.ptid);
1832 bool do_maps_clone = true;
1833 int err = 0;
1834
1835 if (dump_trace)
1836 perf_event__fprintf_task(event, stdout);
1837
1838 /*
1839 * There may be an existing thread that is not actually the parent,
1840 * either because we are processing events out of order, or because the
1841 * (fork) event that would have removed the thread was lost. Assume the
1842 * latter case and continue on as best we can.
1843 */
1844 if (thread__pid(parent) != (pid_t)event->fork.ppid) {
1845 dump_printf("removing erroneous parent thread %d/%d\n",
1846 thread__pid(parent), thread__tid(parent));
1847 machine__remove_thread(machine, parent);
1848 thread__put(parent);
1849 parent = machine__findnew_thread(machine, event->fork.ppid,
1850 event->fork.ptid);
1851 }
1852
1853 /* if a thread currently exists for the thread id remove it */
1854 if (thread != NULL) {
1855 machine__remove_thread(machine, thread);
1856 thread__put(thread);
1857 }
1858
1859 thread = machine__findnew_thread(machine, event->fork.pid,
1860 event->fork.tid);
1861 /*
1862 * When synthesizing FORK events, we are trying to create thread
1863 * objects for the already running tasks on the machine.
1864 *
1865 * Normally, for a kernel FORK event, we want to clone the parent's
1866 * maps because that is what the kernel just did.
1867 *
1868 * But when synthesizing, this should not be done. If we do, we end up
1869 * with overlapping maps as we process the synthesized MMAP2 events that
1870 * get delivered shortly thereafter.
1871 *
1872 * Use the FORK event misc flags in an internal way to signal this
1873 * situation, so we can elide the map clone when appropriate.
1874 */
1875 if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1876 do_maps_clone = false;
1877
1878 if (thread == NULL || parent == NULL ||
1879 thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1880 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1881 err = -1;
1882 }
1883 thread__put(thread);
1884 thread__put(parent);
1885
1886 return err;
1887 }
1888
1889 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1890 struct perf_sample *sample __maybe_unused)
1891 {
1892 struct thread *thread = machine__find_thread(machine,
1893 event->fork.pid,
1894 event->fork.tid);
1895
1896 if (dump_trace)
1897 perf_event__fprintf_task(event, stdout);
1898
1899 if (thread != NULL) {
1900 if (symbol_conf.keep_exited_threads)
1901 thread__set_exited(thread, /*exited=*/true);
1902 else
1903 machine__remove_thread(machine, thread);
1904 }
1905 thread__put(thread);
1906 return 0;
1907 }
1908
1909 int machine__process_event(struct machine *machine, union perf_event *event,
1910 struct perf_sample *sample)
1911 {
1912 int ret;
1913
1914 switch (event->header.type) {
1915 case PERF_RECORD_COMM:
1916 ret = machine__process_comm_event(machine, event, sample); break;
1917 case PERF_RECORD_MMAP:
1918 ret = machine__process_mmap_event(machine, event, sample); break;
1919 case PERF_RECORD_NAMESPACES:
1920 ret = machine__process_namespaces_event(machine, event, sample); break;
1921 case PERF_RECORD_CGROUP:
1922 ret = machine__process_cgroup_event(machine, event, sample); break;
1923 case PERF_RECORD_MMAP2:
1924 ret = machine__process_mmap2_event(machine, event, sample); break;
1925 case PERF_RECORD_FORK:
1926 ret = machine__process_fork_event(machine, event, sample); break;
1927 case PERF_RECORD_EXIT:
1928 ret = machine__process_exit_event(machine, event, sample); break;
1929 case PERF_RECORD_LOST:
1930 ret = machine__process_lost_event(machine, event, sample); break;
1931 case PERF_RECORD_AUX:
1932 ret = machine__process_aux_event(machine, event); break;
1933 case PERF_RECORD_ITRACE_START:
1934 ret = machine__process_itrace_start_event(machine, event); break;
1935 case PERF_RECORD_LOST_SAMPLES:
1936 ret = machine__process_lost_samples_event(machine, event, sample); break;
1937 case PERF_RECORD_SWITCH:
1938 case PERF_RECORD_SWITCH_CPU_WIDE:
1939 ret = machine__process_switch_event(machine, event); break;
1940 case PERF_RECORD_KSYMBOL:
1941 ret = machine__process_ksymbol(machine, event, sample); break;
1942 case PERF_RECORD_BPF_EVENT:
1943 ret = machine__process_bpf(machine, event, sample); break;
1944 case PERF_RECORD_TEXT_POKE:
1945 ret = machine__process_text_poke(machine, event, sample); break;
1946 case PERF_RECORD_AUX_OUTPUT_HW_ID:
1947 ret = machine__process_aux_output_hw_id_event(machine, event); break;
1948 default:
1949 ret = -1;
1950 break;
1951 }
1952
1953 return ret;
1954 }
1955
1956 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1957 {
1958 return regexec(regex, sym->name, 0, NULL, 0) == 0;
1959 }
1960
1961 static void ip__resolve_ams(struct thread *thread,
1962 struct addr_map_symbol *ams,
1963 u64 ip)
1964 {
1965 struct addr_location al;
1966
1967 addr_location__init(&al);
1968 /*
1969 * We cannot use the header.misc hint to determine whether a
1970 * branch stack address is user, kernel, guest, hypervisor.
1971 * Branches may straddle the kernel/user/hypervisor boundaries.
1972 * Thus, we have to try consecutively until we find a match
1973 * or else, the symbol is unknown
1974 */
1975 thread__find_cpumode_addr_location(thread, ip, &al);
1976
1977 ams->addr = ip;
1978 ams->al_addr = al.addr;
1979 ams->al_level = al.level;
1980 ams->ms.maps = maps__get(al.maps);
1981 ams->ms.sym = al.sym;
1982 ams->ms.map = map__get(al.map);
1983 ams->phys_addr = 0;
1984 ams->data_page_size = 0;
1985 addr_location__exit(&al);
1986 }
1987
1988 static void ip__resolve_data(struct thread *thread,
1989 u8 m, struct addr_map_symbol *ams,
1990 u64 addr, u64 phys_addr, u64 daddr_page_size)
1991 {
1992 struct addr_location al;
1993
1994 addr_location__init(&al);
1995
1996 thread__find_symbol(thread, m, addr, &al);
1997
1998 ams->addr = addr;
1999 ams->al_addr = al.addr;
2000 ams->al_level = al.level;
2001 ams->ms.maps = maps__get(al.maps);
2002 ams->ms.sym = al.sym;
2003 ams->ms.map = map__get(al.map);
2004 ams->phys_addr = phys_addr;
2005 ams->data_page_size = daddr_page_size;
2006 addr_location__exit(&al);
2007 }
2008
2009 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2010 struct addr_location *al)
2011 {
2012 struct mem_info *mi = mem_info__new();
2013
2014 if (!mi)
2015 return NULL;
2016
2017 ip__resolve_ams(al->thread, mem_info__iaddr(mi), sample->ip);
2018 ip__resolve_data(al->thread, al->cpumode, mem_info__daddr(mi),
2019 sample->addr, sample->phys_addr,
2020 sample->data_page_size);
2021 mem_info__data_src(mi)->val = sample->data_src;
2022
2023 return mi;
2024 }
2025
2026 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2027 {
2028 struct map *map = ms->map;
2029 char *srcline = NULL;
2030 struct dso *dso;
2031
2032 if (!map || callchain_param.key == CCKEY_FUNCTION)
2033 return srcline;
2034
2035 dso = map__dso(map);
2036 srcline = srcline__tree_find(dso__srclines(dso), ip);
2037 if (!srcline) {
2038 bool show_sym = false;
2039 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2040
2041 srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2042 ms->sym, show_sym, show_addr, ip);
2043 srcline__tree_insert(dso__srclines(dso), ip, srcline);
2044 }
2045
2046 return srcline;
2047 }
2048
2049 struct iterations {
2050 int nr_loop_iter;
2051 u64 cycles;
2052 };
2053
2054 static int add_callchain_ip(struct thread *thread,
2055 struct callchain_cursor *cursor,
2056 struct symbol **parent,
2057 struct addr_location *root_al,
2058 u8 *cpumode,
2059 u64 ip,
2060 bool branch,
2061 struct branch_flags *flags,
2062 struct iterations *iter,
2063 u64 branch_from,
2064 bool symbols)
2065 {
2066 struct map_symbol ms = {};
2067 struct addr_location al;
2068 int nr_loop_iter = 0, err = 0;
2069 u64 iter_cycles = 0;
2070 const char *srcline = NULL;
2071
2072 addr_location__init(&al);
2073 al.filtered = 0;
2074 al.sym = NULL;
2075 al.srcline = NULL;
2076 if (!cpumode) {
2077 thread__find_cpumode_addr_location(thread, ip, &al);
2078 } else {
2079 if (ip >= PERF_CONTEXT_MAX) {
2080 switch (ip) {
2081 case PERF_CONTEXT_HV:
2082 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2083 break;
2084 case PERF_CONTEXT_KERNEL:
2085 *cpumode = PERF_RECORD_MISC_KERNEL;
2086 break;
2087 case PERF_CONTEXT_USER:
2088 *cpumode = PERF_RECORD_MISC_USER;
2089 break;
2090 default:
2091 pr_debug("invalid callchain context: "
2092 "%"PRId64"\n", (s64) ip);
2093 /*
2094 * It seems the callchain is corrupted.
2095 * Discard all.
2096 */
2097 callchain_cursor_reset(cursor);
2098 err = 1;
2099 goto out;
2100 }
2101 goto out;
2102 }
2103 if (symbols)
2104 thread__find_symbol(thread, *cpumode, ip, &al);
2105 }
2106
2107 if (al.sym != NULL) {
2108 if (perf_hpp_list.parent && !*parent &&
2109 symbol__match_regex(al.sym, &parent_regex))
2110 *parent = al.sym;
2111 else if (have_ignore_callees && root_al &&
2112 symbol__match_regex(al.sym, &ignore_callees_regex)) {
2113 /* Treat this symbol as the root,
2114 forgetting its callees. */
2115 addr_location__copy(root_al, &al);
2116 callchain_cursor_reset(cursor);
2117 }
2118 }
2119
2120 if (symbol_conf.hide_unresolved && al.sym == NULL)
2121 goto out;
2122
2123 if (iter) {
2124 nr_loop_iter = iter->nr_loop_iter;
2125 iter_cycles = iter->cycles;
2126 }
2127
2128 ms.maps = maps__get(al.maps);
2129 ms.map = map__get(al.map);
2130 ms.sym = al.sym;
2131 srcline = callchain_srcline(&ms, al.addr);
2132 err = callchain_cursor_append(cursor, ip, &ms,
2133 branch, flags, nr_loop_iter,
2134 iter_cycles, branch_from, srcline);
2135 out:
2136 addr_location__exit(&al);
2137 map_symbol__exit(&ms);
2138 return err;
2139 }
2140
2141 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2142 struct addr_location *al)
2143 {
2144 unsigned int i;
2145 const struct branch_stack *bs = sample->branch_stack;
2146 struct branch_entry *entries = perf_sample__branch_entries(sample);
2147 u64 *branch_stack_cntr = sample->branch_stack_cntr;
2148 struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2149
2150 if (!bi)
2151 return NULL;
2152
2153 for (i = 0; i < bs->nr; i++) {
2154 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2155 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2156 bi[i].flags = entries[i].flags;
2157 if (branch_stack_cntr)
2158 bi[i].branch_stack_cntr = branch_stack_cntr[i];
2159 }
2160 return bi;
2161 }
2162
2163 static void save_iterations(struct iterations *iter,
2164 struct branch_entry *be, int nr)
2165 {
2166 int i;
2167
2168 iter->nr_loop_iter++;
2169 iter->cycles = 0;
2170
2171 for (i = 0; i < nr; i++)
2172 iter->cycles += be[i].flags.cycles;
2173 }
2174
2175 #define CHASHSZ 127
2176 #define CHASHBITS 7
2177 #define NO_ENTRY 0xff
2178
2179 #define PERF_MAX_BRANCH_DEPTH 127
2180
2181 /* Remove loops. */
2182 static int remove_loops(struct branch_entry *l, int nr,
2183 struct iterations *iter)
2184 {
2185 int i, j, off;
2186 unsigned char chash[CHASHSZ];
2187
2188 memset(chash, NO_ENTRY, sizeof(chash));
2189
2190 BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2191
2192 for (i = 0; i < nr; i++) {
2193 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2194
2195 /* no collision handling for now */
2196 if (chash[h] == NO_ENTRY) {
2197 chash[h] = i;
2198 } else if (l[chash[h]].from == l[i].from) {
2199 bool is_loop = true;
2200 /* check if it is a real loop */
2201 off = 0;
2202 for (j = chash[h]; j < i && i + off < nr; j++, off++)
2203 if (l[j].from != l[i + off].from) {
2204 is_loop = false;
2205 break;
2206 }
2207 if (is_loop) {
2208 j = nr - (i + off);
2209 if (j > 0) {
2210 save_iterations(iter + i + off,
2211 l + i, off);
2212
2213 memmove(iter + i, iter + i + off,
2214 j * sizeof(*iter));
2215
2216 memmove(l + i, l + i + off,
2217 j * sizeof(*l));
2218 }
2219
2220 nr -= off;
2221 }
2222 }
2223 }
2224 return nr;
2225 }
2226
2227 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2228 struct callchain_cursor *cursor,
2229 struct perf_sample *sample,
2230 struct symbol **parent,
2231 struct addr_location *root_al,
2232 u64 branch_from,
2233 bool callee, int end,
2234 bool symbols)
2235 {
2236 struct ip_callchain *chain = sample->callchain;
2237 u8 cpumode = PERF_RECORD_MISC_USER;
2238 int err, i;
2239
2240 if (callee) {
2241 for (i = 0; i < end + 1; i++) {
2242 err = add_callchain_ip(thread, cursor, parent,
2243 root_al, &cpumode, chain->ips[i],
2244 false, NULL, NULL, branch_from,
2245 symbols);
2246 if (err)
2247 return err;
2248 }
2249 return 0;
2250 }
2251
2252 for (i = end; i >= 0; i--) {
2253 err = add_callchain_ip(thread, cursor, parent,
2254 root_al, &cpumode, chain->ips[i],
2255 false, NULL, NULL, branch_from,
2256 symbols);
2257 if (err)
2258 return err;
2259 }
2260
2261 return 0;
2262 }
2263
2264 static void save_lbr_cursor_node(struct thread *thread,
2265 struct callchain_cursor *cursor,
2266 int idx)
2267 {
2268 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2269
2270 if (!lbr_stitch)
2271 return;
2272
2273 if (cursor->pos == cursor->nr) {
2274 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2275 return;
2276 }
2277
2278 if (!cursor->curr)
2279 cursor->curr = cursor->first;
2280 else
2281 cursor->curr = cursor->curr->next;
2282
2283 map_symbol__exit(&lbr_stitch->prev_lbr_cursor[idx].ms);
2284 memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2285 sizeof(struct callchain_cursor_node));
2286 lbr_stitch->prev_lbr_cursor[idx].ms.maps = maps__get(cursor->curr->ms.maps);
2287 lbr_stitch->prev_lbr_cursor[idx].ms.map = map__get(cursor->curr->ms.map);
2288
2289 lbr_stitch->prev_lbr_cursor[idx].valid = true;
2290 cursor->pos++;
2291 }
2292
2293 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2294 struct callchain_cursor *cursor,
2295 struct perf_sample *sample,
2296 struct symbol **parent,
2297 struct addr_location *root_al,
2298 u64 *branch_from,
2299 bool callee,
2300 bool symbols)
2301 {
2302 struct branch_stack *lbr_stack = sample->branch_stack;
2303 struct branch_entry *entries = perf_sample__branch_entries(sample);
2304 u8 cpumode = PERF_RECORD_MISC_USER;
2305 int lbr_nr = lbr_stack->nr;
2306 struct branch_flags *flags;
2307 int err, i;
2308 u64 ip;
2309
2310 /*
2311 * The curr and pos are not used in writing session. They are cleared
2312 * in callchain_cursor_commit() when the writing session is closed.
2313 * Using curr and pos to track the current cursor node.
2314 */
2315 if (thread__lbr_stitch(thread)) {
2316 cursor->curr = NULL;
2317 cursor->pos = cursor->nr;
2318 if (cursor->nr) {
2319 cursor->curr = cursor->first;
2320 for (i = 0; i < (int)(cursor->nr - 1); i++)
2321 cursor->curr = cursor->curr->next;
2322 }
2323 }
2324
2325 if (callee) {
2326 /* Add LBR ip from first entries.to */
2327 ip = entries[0].to;
2328 flags = &entries[0].flags;
2329 *branch_from = entries[0].from;
2330 err = add_callchain_ip(thread, cursor, parent,
2331 root_al, &cpumode, ip,
2332 true, flags, NULL,
2333 *branch_from, symbols);
2334 if (err)
2335 return err;
2336
2337 /*
2338 * The number of cursor node increases.
2339 * Move the current cursor node.
2340 * But does not need to save current cursor node for entry 0.
2341 * It's impossible to stitch the whole LBRs of previous sample.
2342 */
2343 if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2344 if (!cursor->curr)
2345 cursor->curr = cursor->first;
2346 else
2347 cursor->curr = cursor->curr->next;
2348 cursor->pos++;
2349 }
2350
2351 /* Add LBR ip from entries.from one by one. */
2352 for (i = 0; i < lbr_nr; i++) {
2353 ip = entries[i].from;
2354 flags = &entries[i].flags;
2355 err = add_callchain_ip(thread, cursor, parent,
2356 root_al, &cpumode, ip,
2357 true, flags, NULL,
2358 *branch_from, symbols);
2359 if (err)
2360 return err;
2361 save_lbr_cursor_node(thread, cursor, i);
2362 }
2363 return 0;
2364 }
2365
2366 /* Add LBR ip from entries.from one by one. */
2367 for (i = lbr_nr - 1; i >= 0; i--) {
2368 ip = entries[i].from;
2369 flags = &entries[i].flags;
2370 err = add_callchain_ip(thread, cursor, parent,
2371 root_al, &cpumode, ip,
2372 true, flags, NULL,
2373 *branch_from, symbols);
2374 if (err)
2375 return err;
2376 save_lbr_cursor_node(thread, cursor, i);
2377 }
2378
2379 if (lbr_nr > 0) {
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, symbols);
2388 if (err)
2389 return err;
2390 }
2391
2392 return 0;
2393 }
2394
2395 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2396 struct callchain_cursor *cursor)
2397 {
2398 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2399 struct callchain_cursor_node *cnode;
2400 struct stitch_list *stitch_node;
2401 int err;
2402
2403 list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2404 cnode = &stitch_node->cursor;
2405
2406 err = callchain_cursor_append(cursor, cnode->ip,
2407 &cnode->ms,
2408 cnode->branch,
2409 &cnode->branch_flags,
2410 cnode->nr_loop_iter,
2411 cnode->iter_cycles,
2412 cnode->branch_from,
2413 cnode->srcline);
2414 if (err)
2415 return err;
2416 }
2417 return 0;
2418 }
2419
2420 static struct stitch_list *get_stitch_node(struct thread *thread)
2421 {
2422 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2423 struct stitch_list *stitch_node;
2424
2425 if (!list_empty(&lbr_stitch->free_lists)) {
2426 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2427 struct stitch_list, node);
2428 list_del(&stitch_node->node);
2429
2430 return stitch_node;
2431 }
2432
2433 return malloc(sizeof(struct stitch_list));
2434 }
2435
2436 static bool has_stitched_lbr(struct thread *thread,
2437 struct perf_sample *cur,
2438 struct perf_sample *prev,
2439 unsigned int max_lbr,
2440 bool callee)
2441 {
2442 struct branch_stack *cur_stack = cur->branch_stack;
2443 struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2444 struct branch_stack *prev_stack = prev->branch_stack;
2445 struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2446 struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2447 int i, j, nr_identical_branches = 0;
2448 struct stitch_list *stitch_node;
2449 u64 cur_base, distance;
2450
2451 if (!cur_stack || !prev_stack)
2452 return false;
2453
2454 /* Find the physical index of the base-of-stack for current sample. */
2455 cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2456
2457 distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2458 (max_lbr + prev_stack->hw_idx - cur_base);
2459 /* Previous sample has shorter stack. Nothing can be stitched. */
2460 if (distance + 1 > prev_stack->nr)
2461 return false;
2462
2463 /*
2464 * Check if there are identical LBRs between two samples.
2465 * Identical LBRs must have same from, to and flags values. Also,
2466 * they have to be saved in the same LBR registers (same physical
2467 * index).
2468 *
2469 * Starts from the base-of-stack of current sample.
2470 */
2471 for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2472 if ((prev_entries[i].from != cur_entries[j].from) ||
2473 (prev_entries[i].to != cur_entries[j].to) ||
2474 (prev_entries[i].flags.value != cur_entries[j].flags.value))
2475 break;
2476 nr_identical_branches++;
2477 }
2478
2479 if (!nr_identical_branches)
2480 return false;
2481
2482 /*
2483 * Save the LBRs between the base-of-stack of previous sample
2484 * and the base-of-stack of current sample into lbr_stitch->lists.
2485 * These LBRs will be stitched later.
2486 */
2487 for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2488
2489 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2490 continue;
2491
2492 stitch_node = get_stitch_node(thread);
2493 if (!stitch_node)
2494 return false;
2495
2496 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2497 sizeof(struct callchain_cursor_node));
2498
2499 stitch_node->cursor.ms.maps = maps__get(lbr_stitch->prev_lbr_cursor[i].ms.maps);
2500 stitch_node->cursor.ms.map = map__get(lbr_stitch->prev_lbr_cursor[i].ms.map);
2501
2502 if (callee)
2503 list_add(&stitch_node->node, &lbr_stitch->lists);
2504 else
2505 list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2506 }
2507
2508 return true;
2509 }
2510
2511 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2512 {
2513 if (thread__lbr_stitch(thread))
2514 return true;
2515
2516 thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2517 if (!thread__lbr_stitch(thread))
2518 goto err;
2519
2520 thread__lbr_stitch(thread)->prev_lbr_cursor =
2521 calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2522 if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2523 goto free_lbr_stitch;
2524
2525 thread__lbr_stitch(thread)->prev_lbr_cursor_size = max_lbr + 1;
2526
2527 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2528 INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2529
2530 return true;
2531
2532 free_lbr_stitch:
2533 free(thread__lbr_stitch(thread));
2534 thread__set_lbr_stitch(thread, NULL);
2535 err:
2536 pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2537 thread__set_lbr_stitch_enable(thread, false);
2538 return false;
2539 }
2540
2541 /*
2542 * Resolve LBR callstack chain sample
2543 * Return:
2544 * 1 on success get LBR callchain information
2545 * 0 no available LBR callchain information, should try fp
2546 * negative error code on other errors.
2547 */
2548 static int resolve_lbr_callchain_sample(struct thread *thread,
2549 struct callchain_cursor *cursor,
2550 struct perf_sample *sample,
2551 struct symbol **parent,
2552 struct addr_location *root_al,
2553 int max_stack,
2554 unsigned int max_lbr,
2555 bool symbols)
2556 {
2557 bool callee = (callchain_param.order == ORDER_CALLEE);
2558 struct ip_callchain *chain = sample->callchain;
2559 int chain_nr = min(max_stack, (int)chain->nr), i;
2560 struct lbr_stitch *lbr_stitch;
2561 bool stitched_lbr = false;
2562 u64 branch_from = 0;
2563 int err;
2564
2565 for (i = 0; i < chain_nr; i++) {
2566 if (chain->ips[i] == PERF_CONTEXT_USER)
2567 break;
2568 }
2569
2570 /* LBR only affects the user callchain */
2571 if (i == chain_nr)
2572 return 0;
2573
2574 if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2575 (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2576 lbr_stitch = thread__lbr_stitch(thread);
2577
2578 stitched_lbr = has_stitched_lbr(thread, sample,
2579 &lbr_stitch->prev_sample,
2580 max_lbr, callee);
2581
2582 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2583 struct stitch_list *stitch_node;
2584
2585 list_for_each_entry(stitch_node, &lbr_stitch->lists, node)
2586 map_symbol__exit(&stitch_node->cursor.ms);
2587
2588 list_splice_init(&lbr_stitch->lists, &lbr_stitch->free_lists);
2589 }
2590 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2591 }
2592
2593 if (callee) {
2594 /* Add kernel ip */
2595 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2596 parent, root_al, branch_from,
2597 true, i, symbols);
2598 if (err)
2599 goto error;
2600
2601 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2602 root_al, &branch_from, true, symbols);
2603 if (err)
2604 goto error;
2605
2606 if (stitched_lbr) {
2607 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2608 if (err)
2609 goto error;
2610 }
2611
2612 } else {
2613 if (stitched_lbr) {
2614 err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2615 if (err)
2616 goto error;
2617 }
2618 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2619 root_al, &branch_from, false, symbols);
2620 if (err)
2621 goto error;
2622
2623 /* Add kernel ip */
2624 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2625 parent, root_al, branch_from,
2626 false, i, symbols);
2627 if (err)
2628 goto error;
2629 }
2630 return 1;
2631
2632 error:
2633 return (err < 0) ? err : 0;
2634 }
2635
2636 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2637 struct callchain_cursor *cursor,
2638 struct symbol **parent,
2639 struct addr_location *root_al,
2640 u8 *cpumode, int ent, bool symbols)
2641 {
2642 int err = 0;
2643
2644 while (--ent >= 0) {
2645 u64 ip = chain->ips[ent];
2646
2647 if (ip >= PERF_CONTEXT_MAX) {
2648 err = add_callchain_ip(thread, cursor, parent,
2649 root_al, cpumode, ip,
2650 false, NULL, NULL, 0, symbols);
2651 break;
2652 }
2653 }
2654 return err;
2655 }
2656
2657 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2658 struct thread *thread, int usr_idx)
2659 {
2660 if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2661 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2662 else
2663 return 0;
2664 }
2665
2666 static int thread__resolve_callchain_sample(struct thread *thread,
2667 struct callchain_cursor *cursor,
2668 struct evsel *evsel,
2669 struct perf_sample *sample,
2670 struct symbol **parent,
2671 struct addr_location *root_al,
2672 int max_stack,
2673 bool symbols)
2674 {
2675 struct branch_stack *branch = sample->branch_stack;
2676 struct branch_entry *entries = perf_sample__branch_entries(sample);
2677 struct ip_callchain *chain = sample->callchain;
2678 int chain_nr = 0;
2679 u8 cpumode = PERF_RECORD_MISC_USER;
2680 int i, j, err, nr_entries, usr_idx;
2681 int skip_idx = -1;
2682 int first_call = 0;
2683 u64 leaf_frame_caller;
2684
2685 if (chain)
2686 chain_nr = chain->nr;
2687
2688 if (evsel__has_branch_callstack(evsel)) {
2689 struct perf_env *env = evsel__env(evsel);
2690
2691 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2692 root_al, max_stack,
2693 !env ? 0 : env->max_branches,
2694 symbols);
2695 if (err)
2696 return (err < 0) ? err : 0;
2697 }
2698
2699 /*
2700 * Based on DWARF debug information, some architectures skip
2701 * a callchain entry saved by the kernel.
2702 */
2703 skip_idx = arch_skip_callchain_idx(thread, chain);
2704
2705 /*
2706 * Add branches to call stack for easier browsing. This gives
2707 * more context for a sample than just the callers.
2708 *
2709 * This uses individual histograms of paths compared to the
2710 * aggregated histograms the normal LBR mode uses.
2711 *
2712 * Limitations for now:
2713 * - No extra filters
2714 * - No annotations (should annotate somehow)
2715 */
2716
2717 if (branch && callchain_param.branch_callstack) {
2718 int nr = min(max_stack, (int)branch->nr);
2719 struct branch_entry be[nr];
2720 struct iterations iter[nr];
2721
2722 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2723 pr_warning("corrupted branch chain. skipping...\n");
2724 goto check_calls;
2725 }
2726
2727 for (i = 0; i < nr; i++) {
2728 if (callchain_param.order == ORDER_CALLEE) {
2729 be[i] = entries[i];
2730
2731 if (chain == NULL)
2732 continue;
2733
2734 /*
2735 * Check for overlap into the callchain.
2736 * The return address is one off compared to
2737 * the branch entry. To adjust for this
2738 * assume the calling instruction is not longer
2739 * than 8 bytes.
2740 */
2741 if (i == skip_idx ||
2742 chain->ips[first_call] >= PERF_CONTEXT_MAX)
2743 first_call++;
2744 else if (be[i].from < chain->ips[first_call] &&
2745 be[i].from >= chain->ips[first_call] - 8)
2746 first_call++;
2747 } else
2748 be[i] = entries[branch->nr - i - 1];
2749 }
2750
2751 memset(iter, 0, sizeof(struct iterations) * nr);
2752 nr = remove_loops(be, nr, iter);
2753
2754 for (i = 0; i < nr; i++) {
2755 err = add_callchain_ip(thread, cursor, parent,
2756 root_al,
2757 NULL, be[i].to,
2758 true, &be[i].flags,
2759 NULL, be[i].from, symbols);
2760
2761 if (!err) {
2762 err = add_callchain_ip(thread, cursor, parent, root_al,
2763 NULL, be[i].from,
2764 true, &be[i].flags,
2765 &iter[i], 0, symbols);
2766 }
2767 if (err == -EINVAL)
2768 break;
2769 if (err)
2770 return err;
2771 }
2772
2773 if (chain_nr == 0)
2774 return 0;
2775
2776 chain_nr -= nr;
2777 }
2778
2779 check_calls:
2780 if (chain && callchain_param.order != ORDER_CALLEE) {
2781 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2782 &cpumode, chain->nr - first_call, symbols);
2783 if (err)
2784 return (err < 0) ? err : 0;
2785 }
2786 for (i = first_call, nr_entries = 0;
2787 i < chain_nr && nr_entries < max_stack; i++) {
2788 u64 ip;
2789
2790 if (callchain_param.order == ORDER_CALLEE)
2791 j = i;
2792 else
2793 j = chain->nr - i - 1;
2794
2795 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2796 if (j == skip_idx)
2797 continue;
2798 #endif
2799 ip = chain->ips[j];
2800 if (ip < PERF_CONTEXT_MAX)
2801 ++nr_entries;
2802 else if (callchain_param.order != ORDER_CALLEE) {
2803 err = find_prev_cpumode(chain, thread, cursor, parent,
2804 root_al, &cpumode, j, symbols);
2805 if (err)
2806 return (err < 0) ? err : 0;
2807 continue;
2808 }
2809
2810 /*
2811 * PERF_CONTEXT_USER allows us to locate where the user stack ends.
2812 * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
2813 * the index will be different in order to add the missing frame
2814 * at the right place.
2815 */
2816
2817 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
2818
2819 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
2820
2821 leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
2822
2823 /*
2824 * check if leaf_frame_Caller != ip to not add the same
2825 * value twice.
2826 */
2827
2828 if (leaf_frame_caller && leaf_frame_caller != ip) {
2829
2830 err = add_callchain_ip(thread, cursor, parent,
2831 root_al, &cpumode, leaf_frame_caller,
2832 false, NULL, NULL, 0, symbols);
2833 if (err)
2834 return (err < 0) ? err : 0;
2835 }
2836 }
2837
2838 err = add_callchain_ip(thread, cursor, parent,
2839 root_al, &cpumode, ip,
2840 false, NULL, NULL, 0, symbols);
2841
2842 if (err)
2843 return (err < 0) ? err : 0;
2844 }
2845
2846 return 0;
2847 }
2848
2849 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2850 {
2851 struct symbol *sym = ms->sym;
2852 struct map *map = ms->map;
2853 struct inline_node *inline_node;
2854 struct inline_list *ilist;
2855 struct dso *dso;
2856 u64 addr;
2857 int ret = 1;
2858 struct map_symbol ilist_ms;
2859
2860 if (!symbol_conf.inline_name || !map || !sym)
2861 return ret;
2862
2863 addr = map__dso_map_ip(map, ip);
2864 addr = map__rip_2objdump(map, addr);
2865 dso = map__dso(map);
2866
2867 inline_node = inlines__tree_find(dso__inlined_nodes(dso), addr);
2868 if (!inline_node) {
2869 inline_node = dso__parse_addr_inlines(dso, addr, sym);
2870 if (!inline_node)
2871 return ret;
2872 inlines__tree_insert(dso__inlined_nodes(dso), inline_node);
2873 }
2874
2875 ilist_ms = (struct map_symbol) {
2876 .maps = maps__get(ms->maps),
2877 .map = map__get(map),
2878 };
2879 list_for_each_entry(ilist, &inline_node->val, list) {
2880 ilist_ms.sym = ilist->symbol;
2881 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2882 NULL, 0, 0, 0, ilist->srcline);
2883
2884 if (ret != 0)
2885 return ret;
2886 }
2887 map_symbol__exit(&ilist_ms);
2888
2889 return ret;
2890 }
2891
2892 static int unwind_entry(struct unwind_entry *entry, void *arg)
2893 {
2894 struct callchain_cursor *cursor = arg;
2895 const char *srcline = NULL;
2896 u64 addr = entry->ip;
2897
2898 if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2899 return 0;
2900
2901 if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2902 return 0;
2903
2904 /*
2905 * Convert entry->ip from a virtual address to an offset in
2906 * its corresponding binary.
2907 */
2908 if (entry->ms.map)
2909 addr = map__dso_map_ip(entry->ms.map, entry->ip);
2910
2911 srcline = callchain_srcline(&entry->ms, addr);
2912 return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2913 false, NULL, 0, 0, 0, srcline);
2914 }
2915
2916 static int thread__resolve_callchain_unwind(struct thread *thread,
2917 struct callchain_cursor *cursor,
2918 struct evsel *evsel,
2919 struct perf_sample *sample,
2920 int max_stack, bool symbols)
2921 {
2922 /* Can we do dwarf post unwind? */
2923 if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2924 (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2925 return 0;
2926
2927 /* Bail out if nothing was captured. */
2928 if ((!sample->user_regs.regs) ||
2929 (!sample->user_stack.size))
2930 return 0;
2931
2932 if (!symbols)
2933 pr_debug("Not resolving symbols with an unwinder isn't currently supported\n");
2934
2935 return unwind__get_entries(unwind_entry, cursor,
2936 thread, sample, max_stack, false);
2937 }
2938
2939 int __thread__resolve_callchain(struct thread *thread,
2940 struct callchain_cursor *cursor,
2941 struct evsel *evsel,
2942 struct perf_sample *sample,
2943 struct symbol **parent,
2944 struct addr_location *root_al,
2945 int max_stack,
2946 bool symbols)
2947 {
2948 int ret = 0;
2949
2950 if (cursor == NULL)
2951 return -ENOMEM;
2952
2953 callchain_cursor_reset(cursor);
2954
2955 if (callchain_param.order == ORDER_CALLEE) {
2956 ret = thread__resolve_callchain_sample(thread, cursor,
2957 evsel, sample,
2958 parent, root_al,
2959 max_stack, symbols);
2960 if (ret)
2961 return ret;
2962 ret = thread__resolve_callchain_unwind(thread, cursor,
2963 evsel, sample,
2964 max_stack, symbols);
2965 } else {
2966 ret = thread__resolve_callchain_unwind(thread, cursor,
2967 evsel, sample,
2968 max_stack, symbols);
2969 if (ret)
2970 return ret;
2971 ret = thread__resolve_callchain_sample(thread, cursor,
2972 evsel, sample,
2973 parent, root_al,
2974 max_stack, symbols);
2975 }
2976
2977 return ret;
2978 }
2979
2980 int machine__for_each_thread(struct machine *machine,
2981 int (*fn)(struct thread *thread, void *p),
2982 void *priv)
2983 {
2984 return threads__for_each_thread(&machine->threads, fn, priv);
2985 }
2986
2987 int machines__for_each_thread(struct machines *machines,
2988 int (*fn)(struct thread *thread, void *p),
2989 void *priv)
2990 {
2991 struct rb_node *nd;
2992 int rc = 0;
2993
2994 rc = machine__for_each_thread(&machines->host, fn, priv);
2995 if (rc != 0)
2996 return rc;
2997
2998 for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
2999 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3000
3001 rc = machine__for_each_thread(machine, fn, priv);
3002 if (rc != 0)
3003 return rc;
3004 }
3005 return rc;
3006 }
3007
3008
3009 static int thread_list_cb(struct thread *thread, void *data)
3010 {
3011 struct list_head *list = data;
3012 struct thread_list *entry = malloc(sizeof(*entry));
3013
3014 if (!entry)
3015 return -ENOMEM;
3016
3017 entry->thread = thread__get(thread);
3018 list_add_tail(&entry->list, list);
3019 return 0;
3020 }
3021
3022 int machine__thread_list(struct machine *machine, struct list_head *list)
3023 {
3024 return machine__for_each_thread(machine, thread_list_cb, list);
3025 }
3026
3027 void thread_list__delete(struct list_head *list)
3028 {
3029 struct thread_list *pos, *next;
3030
3031 list_for_each_entry_safe(pos, next, list, list) {
3032 thread__zput(pos->thread);
3033 list_del(&pos->list);
3034 free(pos);
3035 }
3036 }
3037
3038 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3039 {
3040 if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3041 return -1;
3042
3043 return machine->current_tid[cpu];
3044 }
3045
3046 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3047 pid_t tid)
3048 {
3049 struct thread *thread;
3050 const pid_t init_val = -1;
3051
3052 if (cpu < 0)
3053 return -EINVAL;
3054
3055 if (realloc_array_as_needed(machine->current_tid,
3056 machine->current_tid_sz,
3057 (unsigned int)cpu,
3058 &init_val))
3059 return -ENOMEM;
3060
3061 machine->current_tid[cpu] = tid;
3062
3063 thread = machine__findnew_thread(machine, pid, tid);
3064 if (!thread)
3065 return -ENOMEM;
3066
3067 thread__set_cpu(thread, cpu);
3068 thread__put(thread);
3069
3070 return 0;
3071 }
3072
3073 /*
3074 * Compares the raw arch string. N.B. see instead perf_env__arch() or
3075 * machine__normalized_is() if a normalized arch is needed.
3076 */
3077 bool machine__is(struct machine *machine, const char *arch)
3078 {
3079 return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3080 }
3081
3082 bool machine__normalized_is(struct machine *machine, const char *arch)
3083 {
3084 return machine && !strcmp(perf_env__arch(machine->env), arch);
3085 }
3086
3087 int machine__nr_cpus_avail(struct machine *machine)
3088 {
3089 return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3090 }
3091
3092 int machine__get_kernel_start(struct machine *machine)
3093 {
3094 struct map *map = machine__kernel_map(machine);
3095 int err = 0;
3096
3097 /*
3098 * The only addresses above 2^63 are kernel addresses of a 64-bit
3099 * kernel. Note that addresses are unsigned so that on a 32-bit system
3100 * all addresses including kernel addresses are less than 2^32. In
3101 * that case (32-bit system), if the kernel mapping is unknown, all
3102 * addresses will be assumed to be in user space - see
3103 * machine__kernel_ip().
3104 */
3105 machine->kernel_start = 1ULL << 63;
3106 if (map) {
3107 err = map__load(map);
3108 /*
3109 * On x86_64, PTI entry trampolines are less than the
3110 * start of kernel text, but still above 2^63. So leave
3111 * kernel_start = 1ULL << 63 for x86_64.
3112 */
3113 if (!err && !machine__is(machine, "x86_64"))
3114 machine->kernel_start = map__start(map);
3115 }
3116 return err;
3117 }
3118
3119 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3120 {
3121 u8 addr_cpumode = cpumode;
3122 bool kernel_ip;
3123
3124 if (!machine->single_address_space)
3125 goto out;
3126
3127 kernel_ip = machine__kernel_ip(machine, addr);
3128 switch (cpumode) {
3129 case PERF_RECORD_MISC_KERNEL:
3130 case PERF_RECORD_MISC_USER:
3131 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3132 PERF_RECORD_MISC_USER;
3133 break;
3134 case PERF_RECORD_MISC_GUEST_KERNEL:
3135 case PERF_RECORD_MISC_GUEST_USER:
3136 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3137 PERF_RECORD_MISC_GUEST_USER;
3138 break;
3139 default:
3140 break;
3141 }
3142 out:
3143 return addr_cpumode;
3144 }
3145
3146 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename,
3147 const struct dso_id *id)
3148 {
3149 return dsos__findnew_id(&machine->dsos, filename, id);
3150 }
3151
3152 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3153 {
3154 return machine__findnew_dso_id(machine, filename, NULL);
3155 }
3156
3157 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3158 {
3159 struct machine *machine = vmachine;
3160 struct map *map;
3161 struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3162
3163 if (sym == NULL)
3164 return NULL;
3165
3166 *modp = __map__is_kmodule(map) ? (char *)dso__short_name(map__dso(map)) : NULL;
3167 *addrp = map__unmap_ip(map, sym->start);
3168 return sym->name;
3169 }
3170
3171 struct machine__for_each_dso_cb_args {
3172 struct machine *machine;
3173 machine__dso_t fn;
3174 void *priv;
3175 };
3176
3177 static int machine__for_each_dso_cb(struct dso *dso, void *data)
3178 {
3179 struct machine__for_each_dso_cb_args *args = data;
3180
3181 return args->fn(dso, args->machine, args->priv);
3182 }
3183
3184 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3185 {
3186 struct machine__for_each_dso_cb_args args = {
3187 .machine = machine,
3188 .fn = fn,
3189 .priv = priv,
3190 };
3191
3192 return dsos__for_each_dso(&machine->dsos, machine__for_each_dso_cb, &args);
3193 }
3194
3195 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3196 {
3197 struct maps *maps = machine__kernel_maps(machine);
3198
3199 return maps__for_each_map(maps, fn, priv);
3200 }
3201
3202 bool machine__is_lock_function(struct machine *machine, u64 addr)
3203 {
3204 if (!machine->sched.text_start) {
3205 struct map *kmap;
3206 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3207
3208 if (!sym) {
3209 /* to avoid retry */
3210 machine->sched.text_start = 1;
3211 return false;
3212 }
3213
3214 machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3215
3216 /* should not fail from here */
3217 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3218 machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3219
3220 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3221 machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3222
3223 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3224 machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3225
3226 sym = machine__find_kernel_symbol_by_name(machine, "__traceiter_contention_begin", &kmap);
3227 if (sym) {
3228 machine->traceiter.text_start = map__unmap_ip(kmap, sym->start);
3229 machine->traceiter.text_end = map__unmap_ip(kmap, sym->end);
3230 }
3231 sym = machine__find_kernel_symbol_by_name(machine, "trace_contention_begin", &kmap);
3232 if (sym) {
3233 machine->trace.text_start = map__unmap_ip(kmap, sym->start);
3234 machine->trace.text_end = map__unmap_ip(kmap, sym->end);
3235 }
3236 }
3237
3238 /* failed to get kernel symbols */
3239 if (machine->sched.text_start == 1)
3240 return false;
3241
3242 /* mutex and rwsem functions are in sched text section */
3243 if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3244 return true;
3245
3246 /* spinlock functions are in lock text section */
3247 if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3248 return true;
3249
3250 /* traceiter functions currently don't have their own section
3251 * but we consider them lock functions
3252 */
3253 if (machine->traceiter.text_start != 0) {
3254 if (machine->traceiter.text_start <= addr && addr < machine->traceiter.text_end)
3255 return true;
3256 }
3257
3258 if (machine->trace.text_start != 0) {
3259 if (machine->trace.text_start <= addr && addr < machine->trace.text_end)
3260 return true;
3261 }
3262
3263 return false;
3264 }
3265
3266 int machine__hit_all_dsos(struct machine *machine)
3267 {
3268 return dsos__hit_all(&machine->dsos);
3269 }
Kernel DRM miscellaneous fixes and cross-tree changes