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