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