tools/perf/util/machine.c

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