Linux-libre 4.11.5-gnu
[librecmc/linux-libre.git] / tools / perf / util / machine.c
1 #include "callchain.h"
2 #include "debug.h"
3 #include "event.h"
4 #include "evsel.h"
5 #include "hist.h"
6 #include "machine.h"
7 #include "map.h"
8 #include "sort.h"
9 #include "strlist.h"
10 #include "thread.h"
11 #include "vdso.h"
12 #include <stdbool.h>
13 #include <symbol/kallsyms.h>
14 #include "unwind.h"
15 #include "linux/hash.h"
16
17 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
18
19 static void dsos__init(struct dsos *dsos)
20 {
21         INIT_LIST_HEAD(&dsos->head);
22         dsos->root = RB_ROOT;
23         pthread_rwlock_init(&dsos->lock, NULL);
24 }
25
26 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
27 {
28         memset(machine, 0, sizeof(*machine));
29         map_groups__init(&machine->kmaps, machine);
30         RB_CLEAR_NODE(&machine->rb_node);
31         dsos__init(&machine->dsos);
32
33         machine->threads = RB_ROOT;
34         pthread_rwlock_init(&machine->threads_lock, NULL);
35         machine->nr_threads = 0;
36         INIT_LIST_HEAD(&machine->dead_threads);
37         machine->last_match = NULL;
38
39         machine->vdso_info = NULL;
40         machine->env = NULL;
41
42         machine->pid = pid;
43
44         machine->id_hdr_size = 0;
45         machine->kptr_restrict_warned = false;
46         machine->comm_exec = false;
47         machine->kernel_start = 0;
48
49         memset(machine->vmlinux_maps, 0, sizeof(machine->vmlinux_maps));
50
51         machine->root_dir = strdup(root_dir);
52         if (machine->root_dir == NULL)
53                 return -ENOMEM;
54
55         if (pid != HOST_KERNEL_ID) {
56                 struct thread *thread = machine__findnew_thread(machine, -1,
57                                                                 pid);
58                 char comm[64];
59
60                 if (thread == NULL)
61                         return -ENOMEM;
62
63                 snprintf(comm, sizeof(comm), "[guest/%d]", pid);
64                 thread__set_comm(thread, comm, 0);
65                 thread__put(thread);
66         }
67
68         machine->current_tid = NULL;
69
70         return 0;
71 }
72
73 struct machine *machine__new_host(void)
74 {
75         struct machine *machine = malloc(sizeof(*machine));
76
77         if (machine != NULL) {
78                 machine__init(machine, "", HOST_KERNEL_ID);
79
80                 if (machine__create_kernel_maps(machine) < 0)
81                         goto out_delete;
82         }
83
84         return machine;
85 out_delete:
86         free(machine);
87         return NULL;
88 }
89
90 struct machine *machine__new_kallsyms(void)
91 {
92         struct machine *machine = machine__new_host();
93         /*
94          * FIXME:
95          * 1) MAP__FUNCTION will go away when we stop loading separate maps for
96          *    functions and data objects.
97          * 2) We should switch to machine__load_kallsyms(), i.e. not explicitely
98          *    ask for not using the kcore parsing code, once this one is fixed
99          *    to create a map per module.
100          */
101         if (machine && __machine__load_kallsyms(machine, "/proc/kallsyms", MAP__FUNCTION, true) <= 0) {
102                 machine__delete(machine);
103                 machine = NULL;
104         }
105
106         return machine;
107 }
108
109 static void dsos__purge(struct dsos *dsos)
110 {
111         struct dso *pos, *n;
112
113         pthread_rwlock_wrlock(&dsos->lock);
114
115         list_for_each_entry_safe(pos, n, &dsos->head, node) {
116                 RB_CLEAR_NODE(&pos->rb_node);
117                 pos->root = NULL;
118                 list_del_init(&pos->node);
119                 dso__put(pos);
120         }
121
122         pthread_rwlock_unlock(&dsos->lock);
123 }
124
125 static void dsos__exit(struct dsos *dsos)
126 {
127         dsos__purge(dsos);
128         pthread_rwlock_destroy(&dsos->lock);
129 }
130
131 void machine__delete_threads(struct machine *machine)
132 {
133         struct rb_node *nd;
134
135         pthread_rwlock_wrlock(&machine->threads_lock);
136         nd = rb_first(&machine->threads);
137         while (nd) {
138                 struct thread *t = rb_entry(nd, struct thread, rb_node);
139
140                 nd = rb_next(nd);
141                 __machine__remove_thread(machine, t, false);
142         }
143         pthread_rwlock_unlock(&machine->threads_lock);
144 }
145
146 void machine__exit(struct machine *machine)
147 {
148         machine__destroy_kernel_maps(machine);
149         map_groups__exit(&machine->kmaps);
150         dsos__exit(&machine->dsos);
151         machine__exit_vdso(machine);
152         zfree(&machine->root_dir);
153         zfree(&machine->current_tid);
154         pthread_rwlock_destroy(&machine->threads_lock);
155 }
156
157 void machine__delete(struct machine *machine)
158 {
159         if (machine) {
160                 machine__exit(machine);
161                 free(machine);
162         }
163 }
164
165 void machines__init(struct machines *machines)
166 {
167         machine__init(&machines->host, "", HOST_KERNEL_ID);
168         machines->guests = RB_ROOT;
169 }
170
171 void machines__exit(struct machines *machines)
172 {
173         machine__exit(&machines->host);
174         /* XXX exit guest */
175 }
176
177 struct machine *machines__add(struct machines *machines, pid_t pid,
178                               const char *root_dir)
179 {
180         struct rb_node **p = &machines->guests.rb_node;
181         struct rb_node *parent = NULL;
182         struct machine *pos, *machine = malloc(sizeof(*machine));
183
184         if (machine == NULL)
185                 return NULL;
186
187         if (machine__init(machine, root_dir, pid) != 0) {
188                 free(machine);
189                 return NULL;
190         }
191
192         while (*p != NULL) {
193                 parent = *p;
194                 pos = rb_entry(parent, struct machine, rb_node);
195                 if (pid < pos->pid)
196                         p = &(*p)->rb_left;
197                 else
198                         p = &(*p)->rb_right;
199         }
200
201         rb_link_node(&machine->rb_node, parent, p);
202         rb_insert_color(&machine->rb_node, &machines->guests);
203
204         return machine;
205 }
206
207 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
208 {
209         struct rb_node *nd;
210
211         machines->host.comm_exec = comm_exec;
212
213         for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
214                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
215
216                 machine->comm_exec = comm_exec;
217         }
218 }
219
220 struct machine *machines__find(struct machines *machines, pid_t pid)
221 {
222         struct rb_node **p = &machines->guests.rb_node;
223         struct rb_node *parent = NULL;
224         struct machine *machine;
225         struct machine *default_machine = NULL;
226
227         if (pid == HOST_KERNEL_ID)
228                 return &machines->host;
229
230         while (*p != NULL) {
231                 parent = *p;
232                 machine = rb_entry(parent, struct machine, rb_node);
233                 if (pid < machine->pid)
234                         p = &(*p)->rb_left;
235                 else if (pid > machine->pid)
236                         p = &(*p)->rb_right;
237                 else
238                         return machine;
239                 if (!machine->pid)
240                         default_machine = machine;
241         }
242
243         return default_machine;
244 }
245
246 struct machine *machines__findnew(struct machines *machines, pid_t pid)
247 {
248         char path[PATH_MAX];
249         const char *root_dir = "";
250         struct machine *machine = machines__find(machines, pid);
251
252         if (machine && (machine->pid == pid))
253                 goto out;
254
255         if ((pid != HOST_KERNEL_ID) &&
256             (pid != DEFAULT_GUEST_KERNEL_ID) &&
257             (symbol_conf.guestmount)) {
258                 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
259                 if (access(path, R_OK)) {
260                         static struct strlist *seen;
261
262                         if (!seen)
263                                 seen = strlist__new(NULL, NULL);
264
265                         if (!strlist__has_entry(seen, path)) {
266                                 pr_err("Can't access file %s\n", path);
267                                 strlist__add(seen, path);
268                         }
269                         machine = NULL;
270                         goto out;
271                 }
272                 root_dir = path;
273         }
274
275         machine = machines__add(machines, pid, root_dir);
276 out:
277         return machine;
278 }
279
280 void machines__process_guests(struct machines *machines,
281                               machine__process_t process, void *data)
282 {
283         struct rb_node *nd;
284
285         for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
286                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
287                 process(pos, data);
288         }
289 }
290
291 char *machine__mmap_name(struct machine *machine, char *bf, size_t size)
292 {
293         if (machine__is_host(machine))
294                 snprintf(bf, size, "[%s]", "kernel.kallsyms");
295         else if (machine__is_default_guest(machine))
296                 snprintf(bf, size, "[%s]", "guest.kernel.kallsyms");
297         else {
298                 snprintf(bf, size, "[%s.%d]", "guest.kernel.kallsyms",
299                          machine->pid);
300         }
301
302         return bf;
303 }
304
305 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
306 {
307         struct rb_node *node;
308         struct machine *machine;
309
310         machines->host.id_hdr_size = id_hdr_size;
311
312         for (node = rb_first(&machines->guests); node; node = rb_next(node)) {
313                 machine = rb_entry(node, struct machine, rb_node);
314                 machine->id_hdr_size = id_hdr_size;
315         }
316
317         return;
318 }
319
320 static void machine__update_thread_pid(struct machine *machine,
321                                        struct thread *th, pid_t pid)
322 {
323         struct thread *leader;
324
325         if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
326                 return;
327
328         th->pid_ = pid;
329
330         if (th->pid_ == th->tid)
331                 return;
332
333         leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
334         if (!leader)
335                 goto out_err;
336
337         if (!leader->mg)
338                 leader->mg = map_groups__new(machine);
339
340         if (!leader->mg)
341                 goto out_err;
342
343         if (th->mg == leader->mg)
344                 return;
345
346         if (th->mg) {
347                 /*
348                  * Maps are created from MMAP events which provide the pid and
349                  * tid.  Consequently there never should be any maps on a thread
350                  * with an unknown pid.  Just print an error if there are.
351                  */
352                 if (!map_groups__empty(th->mg))
353                         pr_err("Discarding thread maps for %d:%d\n",
354                                th->pid_, th->tid);
355                 map_groups__put(th->mg);
356         }
357
358         th->mg = map_groups__get(leader->mg);
359 out_put:
360         thread__put(leader);
361         return;
362 out_err:
363         pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
364         goto out_put;
365 }
366
367 /*
368  * Caller must eventually drop thread->refcnt returned with a successful
369  * lookup/new thread inserted.
370  */
371 static struct thread *____machine__findnew_thread(struct machine *machine,
372                                                   pid_t pid, pid_t tid,
373                                                   bool create)
374 {
375         struct rb_node **p = &machine->threads.rb_node;
376         struct rb_node *parent = NULL;
377         struct thread *th;
378
379         /*
380          * Front-end cache - TID lookups come in blocks,
381          * so most of the time we dont have to look up
382          * the full rbtree:
383          */
384         th = machine->last_match;
385         if (th != NULL) {
386                 if (th->tid == tid) {
387                         machine__update_thread_pid(machine, th, pid);
388                         return thread__get(th);
389                 }
390
391                 machine->last_match = NULL;
392         }
393
394         while (*p != NULL) {
395                 parent = *p;
396                 th = rb_entry(parent, struct thread, rb_node);
397
398                 if (th->tid == tid) {
399                         machine->last_match = th;
400                         machine__update_thread_pid(machine, th, pid);
401                         return thread__get(th);
402                 }
403
404                 if (tid < th->tid)
405                         p = &(*p)->rb_left;
406                 else
407                         p = &(*p)->rb_right;
408         }
409
410         if (!create)
411                 return NULL;
412
413         th = thread__new(pid, tid);
414         if (th != NULL) {
415                 rb_link_node(&th->rb_node, parent, p);
416                 rb_insert_color(&th->rb_node, &machine->threads);
417
418                 /*
419                  * We have to initialize map_groups separately
420                  * after rb tree is updated.
421                  *
422                  * The reason is that we call machine__findnew_thread
423                  * within thread__init_map_groups to find the thread
424                  * leader and that would screwed the rb tree.
425                  */
426                 if (thread__init_map_groups(th, machine)) {
427                         rb_erase_init(&th->rb_node, &machine->threads);
428                         RB_CLEAR_NODE(&th->rb_node);
429                         thread__put(th);
430                         return NULL;
431                 }
432                 /*
433                  * It is now in the rbtree, get a ref
434                  */
435                 thread__get(th);
436                 machine->last_match = th;
437                 ++machine->nr_threads;
438         }
439
440         return th;
441 }
442
443 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
444 {
445         return ____machine__findnew_thread(machine, pid, tid, true);
446 }
447
448 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
449                                        pid_t tid)
450 {
451         struct thread *th;
452
453         pthread_rwlock_wrlock(&machine->threads_lock);
454         th = __machine__findnew_thread(machine, pid, tid);
455         pthread_rwlock_unlock(&machine->threads_lock);
456         return th;
457 }
458
459 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
460                                     pid_t tid)
461 {
462         struct thread *th;
463         pthread_rwlock_rdlock(&machine->threads_lock);
464         th =  ____machine__findnew_thread(machine, pid, tid, false);
465         pthread_rwlock_unlock(&machine->threads_lock);
466         return th;
467 }
468
469 struct comm *machine__thread_exec_comm(struct machine *machine,
470                                        struct thread *thread)
471 {
472         if (machine->comm_exec)
473                 return thread__exec_comm(thread);
474         else
475                 return thread__comm(thread);
476 }
477
478 int machine__process_comm_event(struct machine *machine, union perf_event *event,
479                                 struct perf_sample *sample)
480 {
481         struct thread *thread = machine__findnew_thread(machine,
482                                                         event->comm.pid,
483                                                         event->comm.tid);
484         bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
485         int err = 0;
486
487         if (exec)
488                 machine->comm_exec = true;
489
490         if (dump_trace)
491                 perf_event__fprintf_comm(event, stdout);
492
493         if (thread == NULL ||
494             __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
495                 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
496                 err = -1;
497         }
498
499         thread__put(thread);
500
501         return err;
502 }
503
504 int machine__process_lost_event(struct machine *machine __maybe_unused,
505                                 union perf_event *event, struct perf_sample *sample __maybe_unused)
506 {
507         dump_printf(": id:%" PRIu64 ": lost:%" PRIu64 "\n",
508                     event->lost.id, event->lost.lost);
509         return 0;
510 }
511
512 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
513                                         union perf_event *event, struct perf_sample *sample)
514 {
515         dump_printf(": id:%" PRIu64 ": lost samples :%" PRIu64 "\n",
516                     sample->id, event->lost_samples.lost);
517         return 0;
518 }
519
520 static struct dso *machine__findnew_module_dso(struct machine *machine,
521                                                struct kmod_path *m,
522                                                const char *filename)
523 {
524         struct dso *dso;
525
526         pthread_rwlock_wrlock(&machine->dsos.lock);
527
528         dso = __dsos__find(&machine->dsos, m->name, true);
529         if (!dso) {
530                 dso = __dsos__addnew(&machine->dsos, m->name);
531                 if (dso == NULL)
532                         goto out_unlock;
533
534                 if (machine__is_host(machine))
535                         dso->symtab_type = DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE;
536                 else
537                         dso->symtab_type = DSO_BINARY_TYPE__GUEST_KMODULE;
538
539                 /* _KMODULE_COMP should be next to _KMODULE */
540                 if (m->kmod && m->comp)
541                         dso->symtab_type++;
542
543                 dso__set_short_name(dso, strdup(m->name), true);
544                 dso__set_long_name(dso, strdup(filename), true);
545         }
546
547         dso__get(dso);
548 out_unlock:
549         pthread_rwlock_unlock(&machine->dsos.lock);
550         return dso;
551 }
552
553 int machine__process_aux_event(struct machine *machine __maybe_unused,
554                                union perf_event *event)
555 {
556         if (dump_trace)
557                 perf_event__fprintf_aux(event, stdout);
558         return 0;
559 }
560
561 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
562                                         union perf_event *event)
563 {
564         if (dump_trace)
565                 perf_event__fprintf_itrace_start(event, stdout);
566         return 0;
567 }
568
569 int machine__process_switch_event(struct machine *machine __maybe_unused,
570                                   union perf_event *event)
571 {
572         if (dump_trace)
573                 perf_event__fprintf_switch(event, stdout);
574         return 0;
575 }
576
577 static void dso__adjust_kmod_long_name(struct dso *dso, const char *filename)
578 {
579         const char *dup_filename;
580
581         if (!filename || !dso || !dso->long_name)
582                 return;
583         if (dso->long_name[0] != '[')
584                 return;
585         if (!strchr(filename, '/'))
586                 return;
587
588         dup_filename = strdup(filename);
589         if (!dup_filename)
590                 return;
591
592         dso__set_long_name(dso, dup_filename, true);
593 }
594
595 struct map *machine__findnew_module_map(struct machine *machine, u64 start,
596                                         const char *filename)
597 {
598         struct map *map = NULL;
599         struct dso *dso = NULL;
600         struct kmod_path m;
601
602         if (kmod_path__parse_name(&m, filename))
603                 return NULL;
604
605         map = map_groups__find_by_name(&machine->kmaps, MAP__FUNCTION,
606                                        m.name);
607         if (map) {
608                 /*
609                  * If the map's dso is an offline module, give dso__load()
610                  * a chance to find the file path of that module by fixing
611                  * long_name.
612                  */
613                 dso__adjust_kmod_long_name(map->dso, filename);
614                 goto out;
615         }
616
617         dso = machine__findnew_module_dso(machine, &m, filename);
618         if (dso == NULL)
619                 goto out;
620
621         map = map__new2(start, dso, MAP__FUNCTION);
622         if (map == NULL)
623                 goto out;
624
625         map_groups__insert(&machine->kmaps, map);
626
627         /* Put the map here because map_groups__insert alread got it */
628         map__put(map);
629 out:
630         /* put the dso here, corresponding to  machine__findnew_module_dso */
631         dso__put(dso);
632         free(m.name);
633         return map;
634 }
635
636 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
637 {
638         struct rb_node *nd;
639         size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
640
641         for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
642                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
643                 ret += __dsos__fprintf(&pos->dsos.head, fp);
644         }
645
646         return ret;
647 }
648
649 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
650                                      bool (skip)(struct dso *dso, int parm), int parm)
651 {
652         return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
653 }
654
655 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
656                                      bool (skip)(struct dso *dso, int parm), int parm)
657 {
658         struct rb_node *nd;
659         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
660
661         for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
662                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
663                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
664         }
665         return ret;
666 }
667
668 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
669 {
670         int i;
671         size_t printed = 0;
672         struct dso *kdso = machine__kernel_map(machine)->dso;
673
674         if (kdso->has_build_id) {
675                 char filename[PATH_MAX];
676                 if (dso__build_id_filename(kdso, filename, sizeof(filename)))
677                         printed += fprintf(fp, "[0] %s\n", filename);
678         }
679
680         for (i = 0; i < vmlinux_path__nr_entries; ++i)
681                 printed += fprintf(fp, "[%d] %s\n",
682                                    i + kdso->has_build_id, vmlinux_path[i]);
683
684         return printed;
685 }
686
687 size_t machine__fprintf(struct machine *machine, FILE *fp)
688 {
689         size_t ret;
690         struct rb_node *nd;
691
692         pthread_rwlock_rdlock(&machine->threads_lock);
693
694         ret = fprintf(fp, "Threads: %u\n", machine->nr_threads);
695
696         for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
697                 struct thread *pos = rb_entry(nd, struct thread, rb_node);
698
699                 ret += thread__fprintf(pos, fp);
700         }
701
702         pthread_rwlock_unlock(&machine->threads_lock);
703
704         return ret;
705 }
706
707 static struct dso *machine__get_kernel(struct machine *machine)
708 {
709         const char *vmlinux_name = NULL;
710         struct dso *kernel;
711
712         if (machine__is_host(machine)) {
713                 vmlinux_name = symbol_conf.vmlinux_name;
714                 if (!vmlinux_name)
715                         vmlinux_name = DSO__NAME_KALLSYMS;
716
717                 kernel = machine__findnew_kernel(machine, vmlinux_name,
718                                                  "[kernel]", DSO_TYPE_KERNEL);
719         } else {
720                 char bf[PATH_MAX];
721
722                 if (machine__is_default_guest(machine))
723                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
724                 if (!vmlinux_name)
725                         vmlinux_name = machine__mmap_name(machine, bf,
726                                                           sizeof(bf));
727
728                 kernel = machine__findnew_kernel(machine, vmlinux_name,
729                                                  "[guest.kernel]",
730                                                  DSO_TYPE_GUEST_KERNEL);
731         }
732
733         if (kernel != NULL && (!kernel->has_build_id))
734                 dso__read_running_kernel_build_id(kernel, machine);
735
736         return kernel;
737 }
738
739 struct process_args {
740         u64 start;
741 };
742
743 static void machine__get_kallsyms_filename(struct machine *machine, char *buf,
744                                            size_t bufsz)
745 {
746         if (machine__is_default_guest(machine))
747                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
748         else
749                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
750 }
751
752 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
753
754 /* Figure out the start address of kernel map from /proc/kallsyms.
755  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
756  * symbol_name if it's not that important.
757  */
758 static u64 machine__get_running_kernel_start(struct machine *machine,
759                                              const char **symbol_name)
760 {
761         char filename[PATH_MAX];
762         int i;
763         const char *name;
764         u64 addr = 0;
765
766         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
767
768         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
769                 return 0;
770
771         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
772                 addr = kallsyms__get_function_start(filename, name);
773                 if (addr)
774                         break;
775         }
776
777         if (symbol_name)
778                 *symbol_name = name;
779
780         return addr;
781 }
782
783 int __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
784 {
785         int type;
786         u64 start = machine__get_running_kernel_start(machine, NULL);
787
788         /* In case of renewal the kernel map, destroy previous one */
789         machine__destroy_kernel_maps(machine);
790
791         for (type = 0; type < MAP__NR_TYPES; ++type) {
792                 struct kmap *kmap;
793                 struct map *map;
794
795                 machine->vmlinux_maps[type] = map__new2(start, kernel, type);
796                 if (machine->vmlinux_maps[type] == NULL)
797                         return -1;
798
799                 machine->vmlinux_maps[type]->map_ip =
800                         machine->vmlinux_maps[type]->unmap_ip =
801                                 identity__map_ip;
802                 map = __machine__kernel_map(machine, type);
803                 kmap = map__kmap(map);
804                 if (!kmap)
805                         return -1;
806
807                 kmap->kmaps = &machine->kmaps;
808                 map_groups__insert(&machine->kmaps, map);
809         }
810
811         return 0;
812 }
813
814 void machine__destroy_kernel_maps(struct machine *machine)
815 {
816         int type;
817
818         for (type = 0; type < MAP__NR_TYPES; ++type) {
819                 struct kmap *kmap;
820                 struct map *map = __machine__kernel_map(machine, type);
821
822                 if (map == NULL)
823                         continue;
824
825                 kmap = map__kmap(map);
826                 map_groups__remove(&machine->kmaps, map);
827                 if (kmap && kmap->ref_reloc_sym) {
828                         /*
829                          * ref_reloc_sym is shared among all maps, so free just
830                          * on one of them.
831                          */
832                         if (type == MAP__FUNCTION) {
833                                 zfree((char **)&kmap->ref_reloc_sym->name);
834                                 zfree(&kmap->ref_reloc_sym);
835                         } else
836                                 kmap->ref_reloc_sym = NULL;
837                 }
838
839                 map__put(machine->vmlinux_maps[type]);
840                 machine->vmlinux_maps[type] = NULL;
841         }
842 }
843
844 int machines__create_guest_kernel_maps(struct machines *machines)
845 {
846         int ret = 0;
847         struct dirent **namelist = NULL;
848         int i, items = 0;
849         char path[PATH_MAX];
850         pid_t pid;
851         char *endp;
852
853         if (symbol_conf.default_guest_vmlinux_name ||
854             symbol_conf.default_guest_modules ||
855             symbol_conf.default_guest_kallsyms) {
856                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
857         }
858
859         if (symbol_conf.guestmount) {
860                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
861                 if (items <= 0)
862                         return -ENOENT;
863                 for (i = 0; i < items; i++) {
864                         if (!isdigit(namelist[i]->d_name[0])) {
865                                 /* Filter out . and .. */
866                                 continue;
867                         }
868                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
869                         if ((*endp != '\0') ||
870                             (endp == namelist[i]->d_name) ||
871                             (errno == ERANGE)) {
872                                 pr_debug("invalid directory (%s). Skipping.\n",
873                                          namelist[i]->d_name);
874                                 continue;
875                         }
876                         sprintf(path, "%s/%s/proc/kallsyms",
877                                 symbol_conf.guestmount,
878                                 namelist[i]->d_name);
879                         ret = access(path, R_OK);
880                         if (ret) {
881                                 pr_debug("Can't access file %s\n", path);
882                                 goto failure;
883                         }
884                         machines__create_kernel_maps(machines, pid);
885                 }
886 failure:
887                 free(namelist);
888         }
889
890         return ret;
891 }
892
893 void machines__destroy_kernel_maps(struct machines *machines)
894 {
895         struct rb_node *next = rb_first(&machines->guests);
896
897         machine__destroy_kernel_maps(&machines->host);
898
899         while (next) {
900                 struct machine *pos = rb_entry(next, struct machine, rb_node);
901
902                 next = rb_next(&pos->rb_node);
903                 rb_erase(&pos->rb_node, &machines->guests);
904                 machine__delete(pos);
905         }
906 }
907
908 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
909 {
910         struct machine *machine = machines__findnew(machines, pid);
911
912         if (machine == NULL)
913                 return -1;
914
915         return machine__create_kernel_maps(machine);
916 }
917
918 int __machine__load_kallsyms(struct machine *machine, const char *filename,
919                              enum map_type type, bool no_kcore)
920 {
921         struct map *map = machine__kernel_map(machine);
922         int ret = __dso__load_kallsyms(map->dso, filename, map, no_kcore);
923
924         if (ret > 0) {
925                 dso__set_loaded(map->dso, type);
926                 /*
927                  * Since /proc/kallsyms will have multiple sessions for the
928                  * kernel, with modules between them, fixup the end of all
929                  * sections.
930                  */
931                 __map_groups__fixup_end(&machine->kmaps, type);
932         }
933
934         return ret;
935 }
936
937 int machine__load_kallsyms(struct machine *machine, const char *filename,
938                            enum map_type type)
939 {
940         return __machine__load_kallsyms(machine, filename, type, false);
941 }
942
943 int machine__load_vmlinux_path(struct machine *machine, enum map_type type)
944 {
945         struct map *map = machine__kernel_map(machine);
946         int ret = dso__load_vmlinux_path(map->dso, map);
947
948         if (ret > 0)
949                 dso__set_loaded(map->dso, type);
950
951         return ret;
952 }
953
954 static void map_groups__fixup_end(struct map_groups *mg)
955 {
956         int i;
957         for (i = 0; i < MAP__NR_TYPES; ++i)
958                 __map_groups__fixup_end(mg, i);
959 }
960
961 static char *get_kernel_version(const char *root_dir)
962 {
963         char version[PATH_MAX];
964         FILE *file;
965         char *name, *tmp;
966         const char *prefix = "Linux version ";
967
968         sprintf(version, "%s/proc/version", root_dir);
969         file = fopen(version, "r");
970         if (!file)
971                 return NULL;
972
973         version[0] = '\0';
974         tmp = fgets(version, sizeof(version), file);
975         fclose(file);
976
977         name = strstr(version, prefix);
978         if (!name)
979                 return NULL;
980         name += strlen(prefix);
981         tmp = strchr(name, ' ');
982         if (tmp)
983                 *tmp = '\0';
984
985         return strdup(name);
986 }
987
988 static bool is_kmod_dso(struct dso *dso)
989 {
990         return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
991                dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
992 }
993
994 static int map_groups__set_module_path(struct map_groups *mg, const char *path,
995                                        struct kmod_path *m)
996 {
997         struct map *map;
998         char *long_name;
999
1000         map = map_groups__find_by_name(mg, MAP__FUNCTION, m->name);
1001         if (map == NULL)
1002                 return 0;
1003
1004         long_name = strdup(path);
1005         if (long_name == NULL)
1006                 return -ENOMEM;
1007
1008         dso__set_long_name(map->dso, long_name, true);
1009         dso__kernel_module_get_build_id(map->dso, "");
1010
1011         /*
1012          * Full name could reveal us kmod compression, so
1013          * we need to update the symtab_type if needed.
1014          */
1015         if (m->comp && is_kmod_dso(map->dso))
1016                 map->dso->symtab_type++;
1017
1018         return 0;
1019 }
1020
1021 static int map_groups__set_modules_path_dir(struct map_groups *mg,
1022                                 const char *dir_name, int depth)
1023 {
1024         struct dirent *dent;
1025         DIR *dir = opendir(dir_name);
1026         int ret = 0;
1027
1028         if (!dir) {
1029                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1030                 return -1;
1031         }
1032
1033         while ((dent = readdir(dir)) != NULL) {
1034                 char path[PATH_MAX];
1035                 struct stat st;
1036
1037                 /*sshfs might return bad dent->d_type, so we have to stat*/
1038                 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1039                 if (stat(path, &st))
1040                         continue;
1041
1042                 if (S_ISDIR(st.st_mode)) {
1043                         if (!strcmp(dent->d_name, ".") ||
1044                             !strcmp(dent->d_name, ".."))
1045                                 continue;
1046
1047                         /* Do not follow top-level source and build symlinks */
1048                         if (depth == 0) {
1049                                 if (!strcmp(dent->d_name, "source") ||
1050                                     !strcmp(dent->d_name, "build"))
1051                                         continue;
1052                         }
1053
1054                         ret = map_groups__set_modules_path_dir(mg, path,
1055                                                                depth + 1);
1056                         if (ret < 0)
1057                                 goto out;
1058                 } else {
1059                         struct kmod_path m;
1060
1061                         ret = kmod_path__parse_name(&m, dent->d_name);
1062                         if (ret)
1063                                 goto out;
1064
1065                         if (m.kmod)
1066                                 ret = map_groups__set_module_path(mg, path, &m);
1067
1068                         free(m.name);
1069
1070                         if (ret)
1071                                 goto out;
1072                 }
1073         }
1074
1075 out:
1076         closedir(dir);
1077         return ret;
1078 }
1079
1080 static int machine__set_modules_path(struct machine *machine)
1081 {
1082         char *version;
1083         char modules_path[PATH_MAX];
1084
1085         version = get_kernel_version(machine->root_dir);
1086         if (!version)
1087                 return -1;
1088
1089         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1090                  machine->root_dir, version);
1091         free(version);
1092
1093         return map_groups__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1094 }
1095 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1096                                 const char *name __maybe_unused)
1097 {
1098         return 0;
1099 }
1100
1101 static int machine__create_module(void *arg, const char *name, u64 start)
1102 {
1103         struct machine *machine = arg;
1104         struct map *map;
1105
1106         if (arch__fix_module_text_start(&start, name) < 0)
1107                 return -1;
1108
1109         map = machine__findnew_module_map(machine, start, name);
1110         if (map == NULL)
1111                 return -1;
1112
1113         dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1114
1115         return 0;
1116 }
1117
1118 static int machine__create_modules(struct machine *machine)
1119 {
1120         const char *modules;
1121         char path[PATH_MAX];
1122
1123         if (machine__is_default_guest(machine)) {
1124                 modules = symbol_conf.default_guest_modules;
1125         } else {
1126                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1127                 modules = path;
1128         }
1129
1130         if (symbol__restricted_filename(modules, "/proc/modules"))
1131                 return -1;
1132
1133         if (modules__parse(modules, machine, machine__create_module))
1134                 return -1;
1135
1136         if (!machine__set_modules_path(machine))
1137                 return 0;
1138
1139         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1140
1141         return 0;
1142 }
1143
1144 int machine__create_kernel_maps(struct machine *machine)
1145 {
1146         struct dso *kernel = machine__get_kernel(machine);
1147         const char *name;
1148         u64 addr;
1149         int ret;
1150
1151         if (kernel == NULL)
1152                 return -1;
1153
1154         ret = __machine__create_kernel_maps(machine, kernel);
1155         dso__put(kernel);
1156         if (ret < 0)
1157                 return -1;
1158
1159         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1160                 if (machine__is_host(machine))
1161                         pr_debug("Problems creating module maps, "
1162                                  "continuing anyway...\n");
1163                 else
1164                         pr_debug("Problems creating module maps for guest %d, "
1165                                  "continuing anyway...\n", machine->pid);
1166         }
1167
1168         /*
1169          * Now that we have all the maps created, just set the ->end of them:
1170          */
1171         map_groups__fixup_end(&machine->kmaps);
1172
1173         addr = machine__get_running_kernel_start(machine, &name);
1174         if (!addr) {
1175         } else if (maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps, name, addr)) {
1176                 machine__destroy_kernel_maps(machine);
1177                 return -1;
1178         }
1179
1180         return 0;
1181 }
1182
1183 static void machine__set_kernel_mmap_len(struct machine *machine,
1184                                          union perf_event *event)
1185 {
1186         int i;
1187
1188         for (i = 0; i < MAP__NR_TYPES; i++) {
1189                 machine->vmlinux_maps[i]->start = event->mmap.start;
1190                 machine->vmlinux_maps[i]->end   = (event->mmap.start +
1191                                                    event->mmap.len);
1192                 /*
1193                  * Be a bit paranoid here, some perf.data file came with
1194                  * a zero sized synthesized MMAP event for the kernel.
1195                  */
1196                 if (machine->vmlinux_maps[i]->end == 0)
1197                         machine->vmlinux_maps[i]->end = ~0ULL;
1198         }
1199 }
1200
1201 static bool machine__uses_kcore(struct machine *machine)
1202 {
1203         struct dso *dso;
1204
1205         list_for_each_entry(dso, &machine->dsos.head, node) {
1206                 if (dso__is_kcore(dso))
1207                         return true;
1208         }
1209
1210         return false;
1211 }
1212
1213 static int machine__process_kernel_mmap_event(struct machine *machine,
1214                                               union perf_event *event)
1215 {
1216         struct map *map;
1217         char kmmap_prefix[PATH_MAX];
1218         enum dso_kernel_type kernel_type;
1219         bool is_kernel_mmap;
1220
1221         /* If we have maps from kcore then we do not need or want any others */
1222         if (machine__uses_kcore(machine))
1223                 return 0;
1224
1225         machine__mmap_name(machine, kmmap_prefix, sizeof(kmmap_prefix));
1226         if (machine__is_host(machine))
1227                 kernel_type = DSO_TYPE_KERNEL;
1228         else
1229                 kernel_type = DSO_TYPE_GUEST_KERNEL;
1230
1231         is_kernel_mmap = memcmp(event->mmap.filename,
1232                                 kmmap_prefix,
1233                                 strlen(kmmap_prefix) - 1) == 0;
1234         if (event->mmap.filename[0] == '/' ||
1235             (!is_kernel_mmap && event->mmap.filename[0] == '[')) {
1236                 map = machine__findnew_module_map(machine, event->mmap.start,
1237                                                   event->mmap.filename);
1238                 if (map == NULL)
1239                         goto out_problem;
1240
1241                 map->end = map->start + event->mmap.len;
1242         } else if (is_kernel_mmap) {
1243                 const char *symbol_name = (event->mmap.filename +
1244                                 strlen(kmmap_prefix));
1245                 /*
1246                  * Should be there already, from the build-id table in
1247                  * the header.
1248                  */
1249                 struct dso *kernel = NULL;
1250                 struct dso *dso;
1251
1252                 pthread_rwlock_rdlock(&machine->dsos.lock);
1253
1254                 list_for_each_entry(dso, &machine->dsos.head, node) {
1255
1256                         /*
1257                          * The cpumode passed to is_kernel_module is not the
1258                          * cpumode of *this* event. If we insist on passing
1259                          * correct cpumode to is_kernel_module, we should
1260                          * record the cpumode when we adding this dso to the
1261                          * linked list.
1262                          *
1263                          * However we don't really need passing correct
1264                          * cpumode.  We know the correct cpumode must be kernel
1265                          * mode (if not, we should not link it onto kernel_dsos
1266                          * list).
1267                          *
1268                          * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1269                          * is_kernel_module() treats it as a kernel cpumode.
1270                          */
1271
1272                         if (!dso->kernel ||
1273                             is_kernel_module(dso->long_name,
1274                                              PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1275                                 continue;
1276
1277
1278                         kernel = dso;
1279                         break;
1280                 }
1281
1282                 pthread_rwlock_unlock(&machine->dsos.lock);
1283
1284                 if (kernel == NULL)
1285                         kernel = machine__findnew_dso(machine, kmmap_prefix);
1286                 if (kernel == NULL)
1287                         goto out_problem;
1288
1289                 kernel->kernel = kernel_type;
1290                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1291                         dso__put(kernel);
1292                         goto out_problem;
1293                 }
1294
1295                 if (strstr(kernel->long_name, "vmlinux"))
1296                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1297
1298                 machine__set_kernel_mmap_len(machine, event);
1299
1300                 /*
1301                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1302                  * symbol. Effectively having zero here means that at record
1303                  * time /proc/sys/kernel/kptr_restrict was non zero.
1304                  */
1305                 if (event->mmap.pgoff != 0) {
1306                         maps__set_kallsyms_ref_reloc_sym(machine->vmlinux_maps,
1307                                                          symbol_name,
1308                                                          event->mmap.pgoff);
1309                 }
1310
1311                 if (machine__is_default_guest(machine)) {
1312                         /*
1313                          * preload dso of guest kernel and modules
1314                          */
1315                         dso__load(kernel, machine__kernel_map(machine));
1316                 }
1317         }
1318         return 0;
1319 out_problem:
1320         return -1;
1321 }
1322
1323 int machine__process_mmap2_event(struct machine *machine,
1324                                  union perf_event *event,
1325                                  struct perf_sample *sample)
1326 {
1327         struct thread *thread;
1328         struct map *map;
1329         enum map_type type;
1330         int ret = 0;
1331
1332         if (dump_trace)
1333                 perf_event__fprintf_mmap2(event, stdout);
1334
1335         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1336             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1337                 ret = machine__process_kernel_mmap_event(machine, event);
1338                 if (ret < 0)
1339                         goto out_problem;
1340                 return 0;
1341         }
1342
1343         thread = machine__findnew_thread(machine, event->mmap2.pid,
1344                                         event->mmap2.tid);
1345         if (thread == NULL)
1346                 goto out_problem;
1347
1348         if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1349                 type = MAP__VARIABLE;
1350         else
1351                 type = MAP__FUNCTION;
1352
1353         map = map__new(machine, event->mmap2.start,
1354                         event->mmap2.len, event->mmap2.pgoff,
1355                         event->mmap2.pid, event->mmap2.maj,
1356                         event->mmap2.min, event->mmap2.ino,
1357                         event->mmap2.ino_generation,
1358                         event->mmap2.prot,
1359                         event->mmap2.flags,
1360                         event->mmap2.filename, type, thread);
1361
1362         if (map == NULL)
1363                 goto out_problem_map;
1364
1365         ret = thread__insert_map(thread, map);
1366         if (ret)
1367                 goto out_problem_insert;
1368
1369         thread__put(thread);
1370         map__put(map);
1371         return 0;
1372
1373 out_problem_insert:
1374         map__put(map);
1375 out_problem_map:
1376         thread__put(thread);
1377 out_problem:
1378         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1379         return 0;
1380 }
1381
1382 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1383                                 struct perf_sample *sample)
1384 {
1385         struct thread *thread;
1386         struct map *map;
1387         enum map_type type;
1388         int ret = 0;
1389
1390         if (dump_trace)
1391                 perf_event__fprintf_mmap(event, stdout);
1392
1393         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1394             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1395                 ret = machine__process_kernel_mmap_event(machine, event);
1396                 if (ret < 0)
1397                         goto out_problem;
1398                 return 0;
1399         }
1400
1401         thread = machine__findnew_thread(machine, event->mmap.pid,
1402                                          event->mmap.tid);
1403         if (thread == NULL)
1404                 goto out_problem;
1405
1406         if (event->header.misc & PERF_RECORD_MISC_MMAP_DATA)
1407                 type = MAP__VARIABLE;
1408         else
1409                 type = MAP__FUNCTION;
1410
1411         map = map__new(machine, event->mmap.start,
1412                         event->mmap.len, event->mmap.pgoff,
1413                         event->mmap.pid, 0, 0, 0, 0, 0, 0,
1414                         event->mmap.filename,
1415                         type, thread);
1416
1417         if (map == NULL)
1418                 goto out_problem_map;
1419
1420         ret = thread__insert_map(thread, map);
1421         if (ret)
1422                 goto out_problem_insert;
1423
1424         thread__put(thread);
1425         map__put(map);
1426         return 0;
1427
1428 out_problem_insert:
1429         map__put(map);
1430 out_problem_map:
1431         thread__put(thread);
1432 out_problem:
1433         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1434         return 0;
1435 }
1436
1437 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1438 {
1439         if (machine->last_match == th)
1440                 machine->last_match = NULL;
1441
1442         BUG_ON(atomic_read(&th->refcnt) == 0);
1443         if (lock)
1444                 pthread_rwlock_wrlock(&machine->threads_lock);
1445         rb_erase_init(&th->rb_node, &machine->threads);
1446         RB_CLEAR_NODE(&th->rb_node);
1447         --machine->nr_threads;
1448         /*
1449          * Move it first to the dead_threads list, then drop the reference,
1450          * if this is the last reference, then the thread__delete destructor
1451          * will be called and we will remove it from the dead_threads list.
1452          */
1453         list_add_tail(&th->node, &machine->dead_threads);
1454         if (lock)
1455                 pthread_rwlock_unlock(&machine->threads_lock);
1456         thread__put(th);
1457 }
1458
1459 void machine__remove_thread(struct machine *machine, struct thread *th)
1460 {
1461         return __machine__remove_thread(machine, th, true);
1462 }
1463
1464 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1465                                 struct perf_sample *sample)
1466 {
1467         struct thread *thread = machine__find_thread(machine,
1468                                                      event->fork.pid,
1469                                                      event->fork.tid);
1470         struct thread *parent = machine__findnew_thread(machine,
1471                                                         event->fork.ppid,
1472                                                         event->fork.ptid);
1473         int err = 0;
1474
1475         if (dump_trace)
1476                 perf_event__fprintf_task(event, stdout);
1477
1478         /*
1479          * There may be an existing thread that is not actually the parent,
1480          * either because we are processing events out of order, or because the
1481          * (fork) event that would have removed the thread was lost. Assume the
1482          * latter case and continue on as best we can.
1483          */
1484         if (parent->pid_ != (pid_t)event->fork.ppid) {
1485                 dump_printf("removing erroneous parent thread %d/%d\n",
1486                             parent->pid_, parent->tid);
1487                 machine__remove_thread(machine, parent);
1488                 thread__put(parent);
1489                 parent = machine__findnew_thread(machine, event->fork.ppid,
1490                                                  event->fork.ptid);
1491         }
1492
1493         /* if a thread currently exists for the thread id remove it */
1494         if (thread != NULL) {
1495                 machine__remove_thread(machine, thread);
1496                 thread__put(thread);
1497         }
1498
1499         thread = machine__findnew_thread(machine, event->fork.pid,
1500                                          event->fork.tid);
1501
1502         if (thread == NULL || parent == NULL ||
1503             thread__fork(thread, parent, sample->time) < 0) {
1504                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1505                 err = -1;
1506         }
1507         thread__put(thread);
1508         thread__put(parent);
1509
1510         return err;
1511 }
1512
1513 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1514                                 struct perf_sample *sample __maybe_unused)
1515 {
1516         struct thread *thread = machine__find_thread(machine,
1517                                                      event->fork.pid,
1518                                                      event->fork.tid);
1519
1520         if (dump_trace)
1521                 perf_event__fprintf_task(event, stdout);
1522
1523         if (thread != NULL) {
1524                 thread__exited(thread);
1525                 thread__put(thread);
1526         }
1527
1528         return 0;
1529 }
1530
1531 int machine__process_event(struct machine *machine, union perf_event *event,
1532                            struct perf_sample *sample)
1533 {
1534         int ret;
1535
1536         switch (event->header.type) {
1537         case PERF_RECORD_COMM:
1538                 ret = machine__process_comm_event(machine, event, sample); break;
1539         case PERF_RECORD_MMAP:
1540                 ret = machine__process_mmap_event(machine, event, sample); break;
1541         case PERF_RECORD_MMAP2:
1542                 ret = machine__process_mmap2_event(machine, event, sample); break;
1543         case PERF_RECORD_FORK:
1544                 ret = machine__process_fork_event(machine, event, sample); break;
1545         case PERF_RECORD_EXIT:
1546                 ret = machine__process_exit_event(machine, event, sample); break;
1547         case PERF_RECORD_LOST:
1548                 ret = machine__process_lost_event(machine, event, sample); break;
1549         case PERF_RECORD_AUX:
1550                 ret = machine__process_aux_event(machine, event); break;
1551         case PERF_RECORD_ITRACE_START:
1552                 ret = machine__process_itrace_start_event(machine, event); break;
1553         case PERF_RECORD_LOST_SAMPLES:
1554                 ret = machine__process_lost_samples_event(machine, event, sample); break;
1555         case PERF_RECORD_SWITCH:
1556         case PERF_RECORD_SWITCH_CPU_WIDE:
1557                 ret = machine__process_switch_event(machine, event); break;
1558         default:
1559                 ret = -1;
1560                 break;
1561         }
1562
1563         return ret;
1564 }
1565
1566 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1567 {
1568         if (!regexec(regex, sym->name, 0, NULL, 0))
1569                 return 1;
1570         return 0;
1571 }
1572
1573 static void ip__resolve_ams(struct thread *thread,
1574                             struct addr_map_symbol *ams,
1575                             u64 ip)
1576 {
1577         struct addr_location al;
1578
1579         memset(&al, 0, sizeof(al));
1580         /*
1581          * We cannot use the header.misc hint to determine whether a
1582          * branch stack address is user, kernel, guest, hypervisor.
1583          * Branches may straddle the kernel/user/hypervisor boundaries.
1584          * Thus, we have to try consecutively until we find a match
1585          * or else, the symbol is unknown
1586          */
1587         thread__find_cpumode_addr_location(thread, MAP__FUNCTION, ip, &al);
1588
1589         ams->addr = ip;
1590         ams->al_addr = al.addr;
1591         ams->sym = al.sym;
1592         ams->map = al.map;
1593 }
1594
1595 static void ip__resolve_data(struct thread *thread,
1596                              u8 m, struct addr_map_symbol *ams, u64 addr)
1597 {
1598         struct addr_location al;
1599
1600         memset(&al, 0, sizeof(al));
1601
1602         thread__find_addr_location(thread, m, MAP__VARIABLE, addr, &al);
1603         if (al.map == NULL) {
1604                 /*
1605                  * some shared data regions have execute bit set which puts
1606                  * their mapping in the MAP__FUNCTION type array.
1607                  * Check there as a fallback option before dropping the sample.
1608                  */
1609                 thread__find_addr_location(thread, m, MAP__FUNCTION, addr, &al);
1610         }
1611
1612         ams->addr = addr;
1613         ams->al_addr = al.addr;
1614         ams->sym = al.sym;
1615         ams->map = al.map;
1616 }
1617
1618 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
1619                                      struct addr_location *al)
1620 {
1621         struct mem_info *mi = zalloc(sizeof(*mi));
1622
1623         if (!mi)
1624                 return NULL;
1625
1626         ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
1627         ip__resolve_data(al->thread, al->cpumode, &mi->daddr, sample->addr);
1628         mi->data_src.val = sample->data_src;
1629
1630         return mi;
1631 }
1632
1633 static int add_callchain_ip(struct thread *thread,
1634                             struct callchain_cursor *cursor,
1635                             struct symbol **parent,
1636                             struct addr_location *root_al,
1637                             u8 *cpumode,
1638                             u64 ip,
1639                             bool branch,
1640                             struct branch_flags *flags,
1641                             int nr_loop_iter,
1642                             int samples)
1643 {
1644         struct addr_location al;
1645
1646         al.filtered = 0;
1647         al.sym = NULL;
1648         if (!cpumode) {
1649                 thread__find_cpumode_addr_location(thread, MAP__FUNCTION,
1650                                                    ip, &al);
1651         } else {
1652                 if (ip >= PERF_CONTEXT_MAX) {
1653                         switch (ip) {
1654                         case PERF_CONTEXT_HV:
1655                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
1656                                 break;
1657                         case PERF_CONTEXT_KERNEL:
1658                                 *cpumode = PERF_RECORD_MISC_KERNEL;
1659                                 break;
1660                         case PERF_CONTEXT_USER:
1661                                 *cpumode = PERF_RECORD_MISC_USER;
1662                                 break;
1663                         default:
1664                                 pr_debug("invalid callchain context: "
1665                                          "%"PRId64"\n", (s64) ip);
1666                                 /*
1667                                  * It seems the callchain is corrupted.
1668                                  * Discard all.
1669                                  */
1670                                 callchain_cursor_reset(cursor);
1671                                 return 1;
1672                         }
1673                         return 0;
1674                 }
1675                 thread__find_addr_location(thread, *cpumode, MAP__FUNCTION,
1676                                            ip, &al);
1677         }
1678
1679         if (al.sym != NULL) {
1680                 if (perf_hpp_list.parent && !*parent &&
1681                     symbol__match_regex(al.sym, &parent_regex))
1682                         *parent = al.sym;
1683                 else if (have_ignore_callees && root_al &&
1684                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
1685                         /* Treat this symbol as the root,
1686                            forgetting its callees. */
1687                         *root_al = al;
1688                         callchain_cursor_reset(cursor);
1689                 }
1690         }
1691
1692         if (symbol_conf.hide_unresolved && al.sym == NULL)
1693                 return 0;
1694         return callchain_cursor_append(cursor, al.addr, al.map, al.sym,
1695                                        branch, flags, nr_loop_iter, samples);
1696 }
1697
1698 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
1699                                            struct addr_location *al)
1700 {
1701         unsigned int i;
1702         const struct branch_stack *bs = sample->branch_stack;
1703         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
1704
1705         if (!bi)
1706                 return NULL;
1707
1708         for (i = 0; i < bs->nr; i++) {
1709                 ip__resolve_ams(al->thread, &bi[i].to, bs->entries[i].to);
1710                 ip__resolve_ams(al->thread, &bi[i].from, bs->entries[i].from);
1711                 bi[i].flags = bs->entries[i].flags;
1712         }
1713         return bi;
1714 }
1715
1716 #define CHASHSZ 127
1717 #define CHASHBITS 7
1718 #define NO_ENTRY 0xff
1719
1720 #define PERF_MAX_BRANCH_DEPTH 127
1721
1722 /* Remove loops. */
1723 static int remove_loops(struct branch_entry *l, int nr)
1724 {
1725         int i, j, off;
1726         unsigned char chash[CHASHSZ];
1727
1728         memset(chash, NO_ENTRY, sizeof(chash));
1729
1730         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
1731
1732         for (i = 0; i < nr; i++) {
1733                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
1734
1735                 /* no collision handling for now */
1736                 if (chash[h] == NO_ENTRY) {
1737                         chash[h] = i;
1738                 } else if (l[chash[h]].from == l[i].from) {
1739                         bool is_loop = true;
1740                         /* check if it is a real loop */
1741                         off = 0;
1742                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
1743                                 if (l[j].from != l[i + off].from) {
1744                                         is_loop = false;
1745                                         break;
1746                                 }
1747                         if (is_loop) {
1748                                 memmove(l + i, l + i + off,
1749                                         (nr - (i + off)) * sizeof(*l));
1750                                 nr -= off;
1751                         }
1752                 }
1753         }
1754         return nr;
1755 }
1756
1757 /*
1758  * Recolve LBR callstack chain sample
1759  * Return:
1760  * 1 on success get LBR callchain information
1761  * 0 no available LBR callchain information, should try fp
1762  * negative error code on other errors.
1763  */
1764 static int resolve_lbr_callchain_sample(struct thread *thread,
1765                                         struct callchain_cursor *cursor,
1766                                         struct perf_sample *sample,
1767                                         struct symbol **parent,
1768                                         struct addr_location *root_al,
1769                                         int max_stack)
1770 {
1771         struct ip_callchain *chain = sample->callchain;
1772         int chain_nr = min(max_stack, (int)chain->nr), i;
1773         u8 cpumode = PERF_RECORD_MISC_USER;
1774         u64 ip;
1775
1776         for (i = 0; i < chain_nr; i++) {
1777                 if (chain->ips[i] == PERF_CONTEXT_USER)
1778                         break;
1779         }
1780
1781         /* LBR only affects the user callchain */
1782         if (i != chain_nr) {
1783                 struct branch_stack *lbr_stack = sample->branch_stack;
1784                 int lbr_nr = lbr_stack->nr, j, k;
1785                 bool branch;
1786                 struct branch_flags *flags;
1787                 /*
1788                  * LBR callstack can only get user call chain.
1789                  * The mix_chain_nr is kernel call chain
1790                  * number plus LBR user call chain number.
1791                  * i is kernel call chain number,
1792                  * 1 is PERF_CONTEXT_USER,
1793                  * lbr_nr + 1 is the user call chain number.
1794                  * For details, please refer to the comments
1795                  * in callchain__printf
1796                  */
1797                 int mix_chain_nr = i + 1 + lbr_nr + 1;
1798
1799                 for (j = 0; j < mix_chain_nr; j++) {
1800                         int err;
1801                         branch = false;
1802                         flags = NULL;
1803
1804                         if (callchain_param.order == ORDER_CALLEE) {
1805                                 if (j < i + 1)
1806                                         ip = chain->ips[j];
1807                                 else if (j > i + 1) {
1808                                         k = j - i - 2;
1809                                         ip = lbr_stack->entries[k].from;
1810                                         branch = true;
1811                                         flags = &lbr_stack->entries[k].flags;
1812                                 } else {
1813                                         ip = lbr_stack->entries[0].to;
1814                                         branch = true;
1815                                         flags = &lbr_stack->entries[0].flags;
1816                                 }
1817                         } else {
1818                                 if (j < lbr_nr) {
1819                                         k = lbr_nr - j - 1;
1820                                         ip = lbr_stack->entries[k].from;
1821                                         branch = true;
1822                                         flags = &lbr_stack->entries[k].flags;
1823                                 }
1824                                 else if (j > lbr_nr)
1825                                         ip = chain->ips[i + 1 - (j - lbr_nr)];
1826                                 else {
1827                                         ip = lbr_stack->entries[0].to;
1828                                         branch = true;
1829                                         flags = &lbr_stack->entries[0].flags;
1830                                 }
1831                         }
1832
1833                         err = add_callchain_ip(thread, cursor, parent,
1834                                                root_al, &cpumode, ip,
1835                                                branch, flags, 0, 0);
1836                         if (err)
1837                                 return (err < 0) ? err : 0;
1838                 }
1839                 return 1;
1840         }
1841
1842         return 0;
1843 }
1844
1845 static int thread__resolve_callchain_sample(struct thread *thread,
1846                                             struct callchain_cursor *cursor,
1847                                             struct perf_evsel *evsel,
1848                                             struct perf_sample *sample,
1849                                             struct symbol **parent,
1850                                             struct addr_location *root_al,
1851                                             int max_stack)
1852 {
1853         struct branch_stack *branch = sample->branch_stack;
1854         struct ip_callchain *chain = sample->callchain;
1855         int chain_nr = chain->nr;
1856         u8 cpumode = PERF_RECORD_MISC_USER;
1857         int i, j, err, nr_entries;
1858         int skip_idx = -1;
1859         int first_call = 0;
1860         int nr_loop_iter;
1861
1862         if (perf_evsel__has_branch_callstack(evsel)) {
1863                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
1864                                                    root_al, max_stack);
1865                 if (err)
1866                         return (err < 0) ? err : 0;
1867         }
1868
1869         /*
1870          * Based on DWARF debug information, some architectures skip
1871          * a callchain entry saved by the kernel.
1872          */
1873         skip_idx = arch_skip_callchain_idx(thread, chain);
1874
1875         /*
1876          * Add branches to call stack for easier browsing. This gives
1877          * more context for a sample than just the callers.
1878          *
1879          * This uses individual histograms of paths compared to the
1880          * aggregated histograms the normal LBR mode uses.
1881          *
1882          * Limitations for now:
1883          * - No extra filters
1884          * - No annotations (should annotate somehow)
1885          */
1886
1887         if (branch && callchain_param.branch_callstack) {
1888                 int nr = min(max_stack, (int)branch->nr);
1889                 struct branch_entry be[nr];
1890
1891                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
1892                         pr_warning("corrupted branch chain. skipping...\n");
1893                         goto check_calls;
1894                 }
1895
1896                 for (i = 0; i < nr; i++) {
1897                         if (callchain_param.order == ORDER_CALLEE) {
1898                                 be[i] = branch->entries[i];
1899                                 /*
1900                                  * Check for overlap into the callchain.
1901                                  * The return address is one off compared to
1902                                  * the branch entry. To adjust for this
1903                                  * assume the calling instruction is not longer
1904                                  * than 8 bytes.
1905                                  */
1906                                 if (i == skip_idx ||
1907                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
1908                                         first_call++;
1909                                 else if (be[i].from < chain->ips[first_call] &&
1910                                     be[i].from >= chain->ips[first_call] - 8)
1911                                         first_call++;
1912                         } else
1913                                 be[i] = branch->entries[branch->nr - i - 1];
1914                 }
1915
1916                 nr_loop_iter = nr;
1917                 nr = remove_loops(be, nr);
1918
1919                 /*
1920                  * Get the number of iterations.
1921                  * It's only approximation, but good enough in practice.
1922                  */
1923                 if (nr_loop_iter > nr)
1924                         nr_loop_iter = nr_loop_iter - nr + 1;
1925                 else
1926                         nr_loop_iter = 0;
1927
1928                 for (i = 0; i < nr; i++) {
1929                         if (i == nr - 1)
1930                                 err = add_callchain_ip(thread, cursor, parent,
1931                                                        root_al,
1932                                                        NULL, be[i].to,
1933                                                        true, &be[i].flags,
1934                                                        nr_loop_iter, 1);
1935                         else
1936                                 err = add_callchain_ip(thread, cursor, parent,
1937                                                        root_al,
1938                                                        NULL, be[i].to,
1939                                                        true, &be[i].flags,
1940                                                        0, 0);
1941
1942                         if (!err)
1943                                 err = add_callchain_ip(thread, cursor, parent, root_al,
1944                                                        NULL, be[i].from,
1945                                                        true, &be[i].flags,
1946                                                        0, 0);
1947                         if (err == -EINVAL)
1948                                 break;
1949                         if (err)
1950                                 return err;
1951                 }
1952                 chain_nr -= nr;
1953         }
1954
1955 check_calls:
1956         for (i = first_call, nr_entries = 0;
1957              i < chain_nr && nr_entries < max_stack; i++) {
1958                 u64 ip;
1959
1960                 if (callchain_param.order == ORDER_CALLEE)
1961                         j = i;
1962                 else
1963                         j = chain->nr - i - 1;
1964
1965 #ifdef HAVE_SKIP_CALLCHAIN_IDX
1966                 if (j == skip_idx)
1967                         continue;
1968 #endif
1969                 ip = chain->ips[j];
1970
1971                 if (ip < PERF_CONTEXT_MAX)
1972                        ++nr_entries;
1973
1974                 err = add_callchain_ip(thread, cursor, parent,
1975                                        root_al, &cpumode, ip,
1976                                        false, NULL, 0, 0);
1977
1978                 if (err)
1979                         return (err < 0) ? err : 0;
1980         }
1981
1982         return 0;
1983 }
1984
1985 static int unwind_entry(struct unwind_entry *entry, void *arg)
1986 {
1987         struct callchain_cursor *cursor = arg;
1988
1989         if (symbol_conf.hide_unresolved && entry->sym == NULL)
1990                 return 0;
1991         return callchain_cursor_append(cursor, entry->ip,
1992                                        entry->map, entry->sym,
1993                                        false, NULL, 0, 0);
1994 }
1995
1996 static int thread__resolve_callchain_unwind(struct thread *thread,
1997                                             struct callchain_cursor *cursor,
1998                                             struct perf_evsel *evsel,
1999                                             struct perf_sample *sample,
2000                                             int max_stack)
2001 {
2002         /* Can we do dwarf post unwind? */
2003         if (!((evsel->attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2004               (evsel->attr.sample_type & PERF_SAMPLE_STACK_USER)))
2005                 return 0;
2006
2007         /* Bail out if nothing was captured. */
2008         if ((!sample->user_regs.regs) ||
2009             (!sample->user_stack.size))
2010                 return 0;
2011
2012         return unwind__get_entries(unwind_entry, cursor,
2013                                    thread, sample, max_stack);
2014 }
2015
2016 int thread__resolve_callchain(struct thread *thread,
2017                               struct callchain_cursor *cursor,
2018                               struct perf_evsel *evsel,
2019                               struct perf_sample *sample,
2020                               struct symbol **parent,
2021                               struct addr_location *root_al,
2022                               int max_stack)
2023 {
2024         int ret = 0;
2025
2026         callchain_cursor_reset(&callchain_cursor);
2027
2028         if (callchain_param.order == ORDER_CALLEE) {
2029                 ret = thread__resolve_callchain_sample(thread, cursor,
2030                                                        evsel, sample,
2031                                                        parent, root_al,
2032                                                        max_stack);
2033                 if (ret)
2034                         return ret;
2035                 ret = thread__resolve_callchain_unwind(thread, cursor,
2036                                                        evsel, sample,
2037                                                        max_stack);
2038         } else {
2039                 ret = thread__resolve_callchain_unwind(thread, cursor,
2040                                                        evsel, sample,
2041                                                        max_stack);
2042                 if (ret)
2043                         return ret;
2044                 ret = thread__resolve_callchain_sample(thread, cursor,
2045                                                        evsel, sample,
2046                                                        parent, root_al,
2047                                                        max_stack);
2048         }
2049
2050         return ret;
2051 }
2052
2053 int machine__for_each_thread(struct machine *machine,
2054                              int (*fn)(struct thread *thread, void *p),
2055                              void *priv)
2056 {
2057         struct rb_node *nd;
2058         struct thread *thread;
2059         int rc = 0;
2060
2061         for (nd = rb_first(&machine->threads); nd; nd = rb_next(nd)) {
2062                 thread = rb_entry(nd, struct thread, rb_node);
2063                 rc = fn(thread, priv);
2064                 if (rc != 0)
2065                         return rc;
2066         }
2067
2068         list_for_each_entry(thread, &machine->dead_threads, node) {
2069                 rc = fn(thread, priv);
2070                 if (rc != 0)
2071                         return rc;
2072         }
2073         return rc;
2074 }
2075
2076 int machines__for_each_thread(struct machines *machines,
2077                               int (*fn)(struct thread *thread, void *p),
2078                               void *priv)
2079 {
2080         struct rb_node *nd;
2081         int rc = 0;
2082
2083         rc = machine__for_each_thread(&machines->host, fn, priv);
2084         if (rc != 0)
2085                 return rc;
2086
2087         for (nd = rb_first(&machines->guests); nd; nd = rb_next(nd)) {
2088                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
2089
2090                 rc = machine__for_each_thread(machine, fn, priv);
2091                 if (rc != 0)
2092                         return rc;
2093         }
2094         return rc;
2095 }
2096
2097 int __machine__synthesize_threads(struct machine *machine, struct perf_tool *tool,
2098                                   struct target *target, struct thread_map *threads,
2099                                   perf_event__handler_t process, bool data_mmap,
2100                                   unsigned int proc_map_timeout)
2101 {
2102         if (target__has_task(target))
2103                 return perf_event__synthesize_thread_map(tool, threads, process, machine, data_mmap, proc_map_timeout);
2104         else if (target__has_cpu(target))
2105                 return perf_event__synthesize_threads(tool, process, machine, data_mmap, proc_map_timeout);
2106         /* command specified */
2107         return 0;
2108 }
2109
2110 pid_t machine__get_current_tid(struct machine *machine, int cpu)
2111 {
2112         if (cpu < 0 || cpu >= MAX_NR_CPUS || !machine->current_tid)
2113                 return -1;
2114
2115         return machine->current_tid[cpu];
2116 }
2117
2118 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
2119                              pid_t tid)
2120 {
2121         struct thread *thread;
2122
2123         if (cpu < 0)
2124                 return -EINVAL;
2125
2126         if (!machine->current_tid) {
2127                 int i;
2128
2129                 machine->current_tid = calloc(MAX_NR_CPUS, sizeof(pid_t));
2130                 if (!machine->current_tid)
2131                         return -ENOMEM;
2132                 for (i = 0; i < MAX_NR_CPUS; i++)
2133                         machine->current_tid[i] = -1;
2134         }
2135
2136         if (cpu >= MAX_NR_CPUS) {
2137                 pr_err("Requested CPU %d too large. ", cpu);
2138                 pr_err("Consider raising MAX_NR_CPUS\n");
2139                 return -EINVAL;
2140         }
2141
2142         machine->current_tid[cpu] = tid;
2143
2144         thread = machine__findnew_thread(machine, pid, tid);
2145         if (!thread)
2146                 return -ENOMEM;
2147
2148         thread->cpu = cpu;
2149         thread__put(thread);
2150
2151         return 0;
2152 }
2153
2154 int machine__get_kernel_start(struct machine *machine)
2155 {
2156         struct map *map = machine__kernel_map(machine);
2157         int err = 0;
2158
2159         /*
2160          * The only addresses above 2^63 are kernel addresses of a 64-bit
2161          * kernel.  Note that addresses are unsigned so that on a 32-bit system
2162          * all addresses including kernel addresses are less than 2^32.  In
2163          * that case (32-bit system), if the kernel mapping is unknown, all
2164          * addresses will be assumed to be in user space - see
2165          * machine__kernel_ip().
2166          */
2167         machine->kernel_start = 1ULL << 63;
2168         if (map) {
2169                 err = map__load(map);
2170                 if (map->start)
2171                         machine->kernel_start = map->start;
2172         }
2173         return err;
2174 }
2175
2176 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
2177 {
2178         return dsos__findnew(&machine->dsos, filename);
2179 }
2180
2181 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
2182 {
2183         struct machine *machine = vmachine;
2184         struct map *map;
2185         struct symbol *sym = map_groups__find_symbol(&machine->kmaps, MAP__FUNCTION, *addrp, &map);
2186
2187         if (sym == NULL)
2188                 return NULL;
2189
2190         *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
2191         *addrp = map->unmap_ip(map, sym->start);
2192         return sym->name;
2193 }