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 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
 373 {
 374         struct machine *machine = machines__find(machines, pid);
 375
 376         if (!machine)
 377                 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
 378         return machine;
 379 }
 380
 381 void machines__process_guests(struct machines *machines,
 382                               machine__process_t process, void *data)
 383 {
 384         struct rb_node *nd;
 385
 386         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 387                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 388                 process(pos, data);
 389         }
 390 }
 391
 392 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
 393 {
 394         struct rb_node *node;
 395         struct machine *machine;
 396
 397         machines->host.id_hdr_size = id_hdr_size;
 398
 399         for (node = rb_first_cached(&machines->guests); node;
 400              node = rb_next(node)) {
 401                 machine = rb_entry(node, struct machine, rb_node);
 402                 machine->id_hdr_size = id_hdr_size;
 403         }
 404
 405         return;
 406 }
 407
 408 static void machine__update_thread_pid(struct machine *machine,
 409                                        struct thread *th, pid_t pid)
 410 {
 411         struct thread *leader;
 412
 413         if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
 414                 return;
 415
 416         th->pid_ = pid;
 417
 418         if (th->pid_ == th->tid)
 419                 return;
 420
 421         leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
 422         if (!leader)
 423                 goto out_err;
 424
 425         if (!leader->maps)
 426                 leader->maps = maps__new(machine);
 427
 428         if (!leader->maps)
 429                 goto out_err;
 430
 431         if (th->maps == leader->maps)
 432                 return;
 433
 434         if (th->maps) {
 435                 /*
 436                  * Maps are created from MMAP events which provide the pid and
 437                  * tid.  Consequently there never should be any maps on a thread
 438                  * with an unknown pid.  Just print an error if there are.
 439                  */
 440                 if (!maps__empty(th->maps))
 441                         pr_err("Discarding thread maps for %d:%d\n",
 442                                th->pid_, th->tid);
 443                 maps__put(th->maps);
 444         }
 445
 446         th->maps = maps__get(leader->maps);
 447 out_put:
 448         thread__put(leader);
 449         return;
 450 out_err:
 451         pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
 452         goto out_put;
 453 }
 454
 455 /*
 456  * Front-end cache - TID lookups come in blocks,
 457  * so most of the time we dont have to look up
 458  * the full rbtree:
 459  */
 460 static struct thread*
 461 __threads__get_last_match(struct threads *threads, struct machine *machine,
 462                           int pid, int tid)
 463 {
 464         struct thread *th;
 465
 466         th = threads->last_match;
 467         if (th != NULL) {
 468                 if (th->tid == tid) {
 469                         machine__update_thread_pid(machine, th, pid);
 470                         return thread__get(th);
 471                 }
 472
 473                 threads->last_match = NULL;
 474         }
 475
 476         return NULL;
 477 }
 478
 479 static struct thread*
 480 threads__get_last_match(struct threads *threads, struct machine *machine,
 481                         int pid, int tid)
 482 {
 483         struct thread *th = NULL;
 484
 485         if (perf_singlethreaded)
 486                 th = __threads__get_last_match(threads, machine, pid, tid);
 487
 488         return th;
 489 }
 490
 491 static void
 492 __threads__set_last_match(struct threads *threads, struct thread *th)
 493 {
 494         threads->last_match = th;
 495 }
 496
 497 static void
 498 threads__set_last_match(struct threads *threads, struct thread *th)
 499 {
 500         if (perf_singlethreaded)
 501                 __threads__set_last_match(threads, th);
 502 }
 503
 504 /*
 505  * Caller must eventually drop thread->refcnt returned with a successful
 506  * lookup/new thread inserted.
 507  */
 508 static struct thread *____machine__findnew_thread(struct machine *machine,
 509                                                   struct threads *threads,
 510                                                   pid_t pid, pid_t tid,
 511                                                   bool create)
 512 {
 513         struct rb_node **p = &threads->entries.rb_root.rb_node;
 514         struct rb_node *parent = NULL;
 515         struct thread *th;
 516         bool leftmost = true;
 517
 518         th = threads__get_last_match(threads, machine, pid, tid);
 519         if (th)
 520                 return th;
 521
 522         while (*p != NULL) {
 523                 parent = *p;
 524                 th = rb_entry(parent, struct thread, rb_node);
 525
 526                 if (th->tid == tid) {
 527                         threads__set_last_match(threads, th);
 528                         machine__update_thread_pid(machine, th, pid);
 529                         return thread__get(th);
 530                 }
 531
 532                 if (tid < th->tid)
 533                         p = &(*p)->rb_left;
 534                 else {
 535                         p = &(*p)->rb_right;
 536                         leftmost = false;
 537                 }
 538         }
 539
 540         if (!create)
 541                 return NULL;
 542
 543         th = thread__new(pid, tid);
 544         if (th != NULL) {
 545                 rb_link_node(&th->rb_node, parent, p);
 546                 rb_insert_color_cached(&th->rb_node, &threads->entries, leftmost);
 547
 548                 /*
 549                  * We have to initialize maps separately after rb tree is updated.
 550                  *
 551                  * The reason is that we call machine__findnew_thread
 552                  * within thread__init_maps to find the thread
 553                  * leader and that would screwed the rb tree.
 554                  */
 555                 if (thread__init_maps(th, machine)) {
 556                         rb_erase_cached(&th->rb_node, &threads->entries);
 557                         RB_CLEAR_NODE(&th->rb_node);
 558                         thread__put(th);
 559                         return NULL;
 560                 }
 561                 /*
 562                  * It is now in the rbtree, get a ref
 563                  */
 564                 thread__get(th);
 565                 threads__set_last_match(threads, th);
 566                 ++threads->nr;
 567         }
 568
 569         return th;
 570 }
 571
 572 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
 573 {
 574         return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
 575 }
 576
 577 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
 578                                        pid_t tid)
 579 {
 580         struct threads *threads = machine__threads(machine, tid);
 581         struct thread *th;
 582
 583         down_write(&threads->lock);
 584         th = __machine__findnew_thread(machine, pid, tid);
 585         up_write(&threads->lock);
 586         return th;
 587 }
 588
 589 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
 590                                     pid_t tid)
 591 {
 592         struct threads *threads = machine__threads(machine, tid);
 593         struct thread *th;
 594
 595         down_read(&threads->lock);
 596         th =  ____machine__findnew_thread(machine, threads, pid, tid, false);
 597         up_read(&threads->lock);
 598         return th;
 599 }
 600
 601 /*
 602  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
 603  * So here a single thread is created for that, but actually there is a separate
 604  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
 605  * is only 1. That causes problems for some tools, requiring workarounds. For
 606  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
 607  */
 608 struct thread *machine__idle_thread(struct machine *machine)
 609 {
 610         struct thread *thread = machine__findnew_thread(machine, 0, 0);
 611
 612         if (!thread || thread__set_comm(thread, "swapper", 0) ||
 613             thread__set_namespaces(thread, 0, NULL))
 614                 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
 615
 616         return thread;
 617 }
 618
 619 struct comm *machine__thread_exec_comm(struct machine *machine,
 620                                        struct thread *thread)
 621 {
 622         if (machine->comm_exec)
 623                 return thread__exec_comm(thread);
 624         else
 625                 return thread__comm(thread);
 626 }
 627
 628 int machine__process_comm_event(struct machine *machine, union perf_event *event,
 629                                 struct perf_sample *sample)
 630 {
 631         struct thread *thread = machine__findnew_thread(machine,
 632                                                         event->comm.pid,
 633                                                         event->comm.tid);
 634         bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
 635         int err = 0;
 636
 637         if (exec)
 638                 machine->comm_exec = true;
 639
 640         if (dump_trace)
 641                 perf_event__fprintf_comm(event, stdout);
 642
 643         if (thread == NULL ||
 644             __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 645                 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
 646                 err = -1;
 647         }
 648
 649         thread__put(thread);
 650
 651         return err;
 652 }
 653
 654 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
 655                                       union perf_event *event,
 656                                       struct perf_sample *sample __maybe_unused)
 657 {
 658         struct thread *thread = machine__findnew_thread(machine,
 659                                                         event->namespaces.pid,
 660                                                         event->namespaces.tid);
 661         int err = 0;
 662
 663         WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
 664                   "\nWARNING: kernel seems to support more namespaces than perf"
 665                   " tool.\nTry updating the perf tool..\n\n");
 666
 667         WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
 668                   "\nWARNING: perf tool seems to support more namespaces than"
 669                   " the kernel.\nTry updating the kernel..\n\n");
 670
 671         if (dump_trace)
 672                 perf_event__fprintf_namespaces(event, stdout);
 673
 674         if (thread == NULL ||
 675             thread__set_namespaces(thread, sample->time, &event->namespaces)) {
 676                 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
 677                 err = -1;
 678         }
 679
 680         thread__put(thread);
 681
 682         return err;
 683 }
 684
 685 int machine__process_cgroup_event(struct machine *machine,
 686                                   union perf_event *event,
 687                                   struct perf_sample *sample __maybe_unused)
 688 {
 689         struct cgroup *cgrp;
 690
 691         if (dump_trace)
 692                 perf_event__fprintf_cgroup(event, stdout);
 693
 694         cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
 695         if (cgrp == NULL)
 696                 return -ENOMEM;
 697
 698         return 0;
 699 }
 700
 701 int machine__process_lost_event(struct machine *machine __maybe_unused,
 702                                 union perf_event *event, struct perf_sample *sample __maybe_unused)
 703 {
 704         dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
 705                     event->lost.id, event->lost.lost);
 706         return 0;
 707 }
 708
 709 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
 710                                         union perf_event *event, struct perf_sample *sample)
 711 {
 712         dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
 713                     sample->id, event->lost_samples.lost);
 714         return 0;
 715 }
 716
 717 static struct dso *machine__findnew_module_dso(struct machine *machine,
 718                                                struct kmod_path *m,
 719                                                const char *filename)
 720 {
 721         struct dso *dso;
 722
 723         down_write(&machine->dsos.lock);
 724
 725         dso = __dsos__find(&machine->dsos, m->name, true);
 726         if (!dso) {
 727                 dso = __dsos__addnew(&machine->dsos, m->name);
 728                 if (dso == NULL)
 729                         goto out_unlock;
 730
 731                 dso__set_module_info(dso, m, machine);
 732                 dso__set_long_name(dso, strdup(filename), true);
 733                 dso->kernel = DSO_SPACE__KERNEL;
 734         }
 735
 736         dso__get(dso);
 737 out_unlock:
 738         up_write(&machine->dsos.lock);
 739         return dso;
 740 }
 741
 742 int machine__process_aux_event(struct machine *machine __maybe_unused,
 743                                union perf_event *event)
 744 {
 745         if (dump_trace)
 746                 perf_event__fprintf_aux(event, stdout);
 747         return 0;
 748 }
 749
 750 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
 751                                         union perf_event *event)
 752 {
 753         if (dump_trace)
 754                 perf_event__fprintf_itrace_start(event, stdout);
 755         return 0;
 756 }
 757
 758 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
 759                                             union perf_event *event)
 760 {
 761         if (dump_trace)
 762                 perf_event__fprintf_aux_output_hw_id(event, stdout);
 763         return 0;
 764 }
 765
 766 int machine__process_switch_event(struct machine *machine __maybe_unused,
 767                                   union perf_event *event)
 768 {
 769         if (dump_trace)
 770                 perf_event__fprintf_switch(event, stdout);
 771         return 0;
 772 }
 773
 774 static int machine__process_ksymbol_register(struct machine *machine,
 775                                              union perf_event *event,
 776                                              struct perf_sample *sample __maybe_unused)
 777 {
 778         struct symbol *sym;
 779         struct map *map = maps__find(&machine->kmaps, event->ksymbol.addr);
 780
 781         if (!map) {
 782                 struct dso *dso = dso__new(event->ksymbol.name);
 783
 784                 if (dso) {
 785                         dso->kernel = DSO_SPACE__KERNEL;
 786                         map = map__new2(0, dso);
 787                         dso__put(dso);
 788                 }
 789
 790                 if (!dso || !map) {
 791                         return -ENOMEM;
 792                 }
 793
 794                 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 795                         map->dso->binary_type = DSO_BINARY_TYPE__OOL;
 796                         map->dso->data.file_size = event->ksymbol.len;
 797                         dso__set_loaded(map->dso);
 798                 }
 799
 800                 map->start = event->ksymbol.addr;
 801                 map->end = map->start + event->ksymbol.len;
 802                 maps__insert(&machine->kmaps, map);
 803                 map__put(map);
 804                 dso__set_loaded(dso);
 805
 806                 if (is_bpf_image(event->ksymbol.name)) {
 807                         dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
 808                         dso__set_long_name(dso, "", false);
 809                 }
 810         }
 811
 812         sym = symbol__new(map->map_ip(map, map->start),
 813                           event->ksymbol.len,
 814                           0, 0, event->ksymbol.name);
 815         if (!sym)
 816                 return -ENOMEM;
 817         dso__insert_symbol(map->dso, sym);
 818         return 0;
 819 }
 820
 821 static int machine__process_ksymbol_unregister(struct machine *machine,
 822                                                union perf_event *event,
 823                                                struct perf_sample *sample __maybe_unused)
 824 {
 825         struct symbol *sym;
 826         struct map *map;
 827
 828         map = maps__find(&machine->kmaps, event->ksymbol.addr);
 829         if (!map)
 830                 return 0;
 831
 832         if (map != machine->vmlinux_map)
 833                 maps__remove(&machine->kmaps, map);
 834         else {
 835                 sym = dso__find_symbol(map->dso, map->map_ip(map, map->start));
 836                 if (sym)
 837                         dso__delete_symbol(map->dso, sym);
 838         }
 839
 840         return 0;
 841 }
 842
 843 int machine__process_ksymbol(struct machine *machine __maybe_unused,
 844                              union perf_event *event,
 845                              struct perf_sample *sample)
 846 {
 847         if (dump_trace)
 848                 perf_event__fprintf_ksymbol(event, stdout);
 849
 850         if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 851                 return machine__process_ksymbol_unregister(machine, event,
 852                                                            sample);
 853         return machine__process_ksymbol_register(machine, event, sample);
 854 }
 855
 856 int machine__process_text_poke(struct machine *machine, union perf_event *event,
 857                                struct perf_sample *sample __maybe_unused)
 858 {
 859         struct map *map = maps__find(&machine->kmaps, event->text_poke.addr);
 860         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 861
 862         if (dump_trace)
 863                 perf_event__fprintf_text_poke(event, machine, stdout);
 864
 865         if (!event->text_poke.new_len)
 866                 return 0;
 867
 868         if (cpumode != PERF_RECORD_MISC_KERNEL) {
 869                 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
 870                 return 0;
 871         }
 872
 873         if (map && map->dso) {
 874                 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
 875                 int ret;
 876
 877                 /*
 878                  * Kernel maps might be changed when loading symbols so loading
 879                  * must be done prior to using kernel maps.
 880                  */
 881                 map__load(map);
 882                 ret = dso__data_write_cache_addr(map->dso, map, machine,
 883                                                  event->text_poke.addr,
 884                                                  new_bytes,
 885                                                  event->text_poke.new_len);
 886                 if (ret != event->text_poke.new_len)
 887                         pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
 888                                  event->text_poke.addr);
 889         } else {
 890                 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
 891                          event->text_poke.addr);
 892         }
 893
 894         return 0;
 895 }
 896
 897 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
 898                                               const char *filename)
 899 {
 900         struct map *map = NULL;
 901         struct kmod_path m;
 902         struct dso *dso;
 903
 904         if (kmod_path__parse_name(&m, filename))
 905                 return NULL;
 906
 907         dso = machine__findnew_module_dso(machine, &m, filename);
 908         if (dso == NULL)
 909                 goto out;
 910
 911         map = map__new2(start, dso);
 912         if (map == NULL)
 913                 goto out;
 914
 915         maps__insert(&machine->kmaps, map);
 916
 917         /* Put the map here because maps__insert already got it */
 918         map__put(map);
 919 out:
 920         /* put the dso here, corresponding to  machine__findnew_module_dso */
 921         dso__put(dso);
 922         zfree(&m.name);
 923         return map;
 924 }
 925
 926 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
 927 {
 928         struct rb_node *nd;
 929         size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
 930
 931         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 932                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 933                 ret += __dsos__fprintf(&pos->dsos.head, fp);
 934         }
 935
 936         return ret;
 937 }
 938
 939 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
 940                                      bool (skip)(struct dso *dso, int parm), int parm)
 941 {
 942         return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
 943 }
 944
 945 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
 946                                      bool (skip)(struct dso *dso, int parm), int parm)
 947 {
 948         struct rb_node *nd;
 949         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
 950
 951         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 952                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 953                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
 954         }
 955         return ret;
 956 }
 957
 958 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
 959 {
 960         int i;
 961         size_t printed = 0;
 962         struct dso *kdso = machine__kernel_dso(machine);
 963
 964         if (kdso->has_build_id) {
 965                 char filename[PATH_MAX];
 966                 if (dso__build_id_filename(kdso, filename, sizeof(filename),
 967                                            false))
 968                         printed += fprintf(fp, "[0] %s\n", filename);
 969         }
 970
 971         for (i = 0; i < vmlinux_path__nr_entries; ++i)
 972                 printed += fprintf(fp, "[%d] %s\n",
 973                                    i + kdso->has_build_id, vmlinux_path[i]);
 974
 975         return printed;
 976 }
 977
 978 size_t machine__fprintf(struct machine *machine, FILE *fp)
 979 {
 980         struct rb_node *nd;
 981         size_t ret;
 982         int i;
 983
 984         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 985                 struct threads *threads = &machine->threads[i];
 986
 987                 down_read(&threads->lock);
 988
 989                 ret = fprintf(fp, "Threads: %u\n", threads->nr);
 990
 991                 for (nd = rb_first_cached(&threads->entries); nd;
 992                      nd = rb_next(nd)) {
 993                         struct thread *pos = rb_entry(nd, struct thread, rb_node);
 994
 995                         ret += thread__fprintf(pos, fp);
 996                 }
 997
 998                 up_read(&threads->lock);
 999         }
1000         return ret;
1001 }
1002
1003 static struct dso *machine__get_kernel(struct machine *machine)
1004 {
1005         const char *vmlinux_name = machine->mmap_name;
1006         struct dso *kernel;
1007
1008         if (machine__is_host(machine)) {
1009                 if (symbol_conf.vmlinux_name)
1010                         vmlinux_name = symbol_conf.vmlinux_name;
1011
1012                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1013                                                  "[kernel]", DSO_SPACE__KERNEL);
1014         } else {
1015                 if (symbol_conf.default_guest_vmlinux_name)
1016                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1017
1018                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1019                                                  "[guest.kernel]",
1020                                                  DSO_SPACE__KERNEL_GUEST);
1021         }
1022
1023         if (kernel != NULL && (!kernel->has_build_id))
1024                 dso__read_running_kernel_build_id(kernel, machine);
1025
1026         return kernel;
1027 }
1028
1029 struct process_args {
1030         u64 start;
1031 };
1032
1033 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1034                                     size_t bufsz)
1035 {
1036         if (machine__is_default_guest(machine))
1037                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1038         else
1039                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1040 }
1041
1042 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1043
1044 /* Figure out the start address of kernel map from /proc/kallsyms.
1045  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1046  * symbol_name if it's not that important.
1047  */
1048 static int machine__get_running_kernel_start(struct machine *machine,
1049                                              const char **symbol_name,
1050                                              u64 *start, u64 *end)
1051 {
1052         char filename[PATH_MAX];
1053         int i, err = -1;
1054         const char *name;
1055         u64 addr = 0;
1056
1057         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1058
1059         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1060                 return 0;
1061
1062         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1063                 err = kallsyms__get_function_start(filename, name, &addr);
1064                 if (!err)
1065                         break;
1066         }
1067
1068         if (err)
1069                 return -1;
1070
1071         if (symbol_name)
1072                 *symbol_name = name;
1073
1074         *start = addr;
1075
1076         err = kallsyms__get_function_start(filename, "_etext", &addr);
1077         if (!err)
1078                 *end = addr;
1079
1080         return 0;
1081 }
1082
1083 int machine__create_extra_kernel_map(struct machine *machine,
1084                                      struct dso *kernel,
1085                                      struct extra_kernel_map *xm)
1086 {
1087         struct kmap *kmap;
1088         struct map *map;
1089
1090         map = map__new2(xm->start, kernel);
1091         if (!map)
1092                 return -1;
1093
1094         map->end   = xm->end;
1095         map->pgoff = xm->pgoff;
1096
1097         kmap = map__kmap(map);
1098
1099         strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1100
1101         maps__insert(&machine->kmaps, map);
1102
1103         pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1104                   kmap->name, map->start, map->end);
1105
1106         map__put(map);
1107
1108         return 0;
1109 }
1110
1111 static u64 find_entry_trampoline(struct dso *dso)
1112 {
1113         /* Duplicates are removed so lookup all aliases */
1114         const char *syms[] = {
1115                 "_entry_trampoline",
1116                 "__entry_trampoline_start",
1117                 "entry_SYSCALL_64_trampoline",
1118         };
1119         struct symbol *sym = dso__first_symbol(dso);
1120         unsigned int i;
1121
1122         for (; sym; sym = dso__next_symbol(sym)) {
1123                 if (sym->binding != STB_GLOBAL)
1124                         continue;
1125                 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1126                         if (!strcmp(sym->name, syms[i]))
1127                                 return sym->start;
1128                 }
1129         }
1130
1131         return 0;
1132 }
1133
1134 /*
1135  * These values can be used for kernels that do not have symbols for the entry
1136  * trampolines in kallsyms.
1137  */
1138 #define X86_64_CPU_ENTRY_AREA_PER_CPU   0xfffffe0000000000ULL
1139 #define X86_64_CPU_ENTRY_AREA_SIZE      0x2c000
1140 #define X86_64_ENTRY_TRAMPOLINE         0x6000
1141
1142 /* Map x86_64 PTI entry trampolines */
1143 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1144                                           struct dso *kernel)
1145 {
1146         struct maps *kmaps = &machine->kmaps;
1147         int nr_cpus_avail, cpu;
1148         bool found = false;
1149         struct map *map;
1150         u64 pgoff;
1151
1152         /*
1153          * In the vmlinux case, pgoff is a virtual address which must now be
1154          * mapped to a vmlinux offset.
1155          */
1156         maps__for_each_entry(kmaps, map) {
1157                 struct kmap *kmap = __map__kmap(map);
1158                 struct map *dest_map;
1159
1160                 if (!kmap || !is_entry_trampoline(kmap->name))
1161                         continue;
1162
1163                 dest_map = maps__find(kmaps, map->pgoff);
1164                 if (dest_map != map)
1165                         map->pgoff = dest_map->map_ip(dest_map, map->pgoff);
1166                 found = true;
1167         }
1168         if (found || machine->trampolines_mapped)
1169                 return 0;
1170
1171         pgoff = find_entry_trampoline(kernel);
1172         if (!pgoff)
1173                 return 0;
1174
1175         nr_cpus_avail = machine__nr_cpus_avail(machine);
1176
1177         /* Add a 1 page map for each CPU's entry trampoline */
1178         for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1179                 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1180                          cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1181                          X86_64_ENTRY_TRAMPOLINE;
1182                 struct extra_kernel_map xm = {
1183                         .start = va,
1184                         .end   = va + page_size,
1185                         .pgoff = pgoff,
1186                 };
1187
1188                 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1189
1190                 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1191                         return -1;
1192         }
1193
1194         machine->trampolines_mapped = nr_cpus_avail;
1195
1196         return 0;
1197 }
1198
1199 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1200                                              struct dso *kernel __maybe_unused)
1201 {
1202         return 0;
1203 }
1204
1205 static int
1206 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1207 {
1208         /* In case of renewal the kernel map, destroy previous one */
1209         machine__destroy_kernel_maps(machine);
1210
1211         machine->vmlinux_map = map__new2(0, kernel);
1212         if (machine->vmlinux_map == NULL)
1213                 return -1;
1214
1215         machine->vmlinux_map->map_ip = machine->vmlinux_map->unmap_ip = identity__map_ip;
1216         maps__insert(&machine->kmaps, machine->vmlinux_map);
1217         return 0;
1218 }
1219
1220 void machine__destroy_kernel_maps(struct machine *machine)
1221 {
1222         struct kmap *kmap;
1223         struct map *map = machine__kernel_map(machine);
1224
1225         if (map == NULL)
1226                 return;
1227
1228         kmap = map__kmap(map);
1229         maps__remove(&machine->kmaps, map);
1230         if (kmap && kmap->ref_reloc_sym) {
1231                 zfree((char **)&kmap->ref_reloc_sym->name);
1232                 zfree(&kmap->ref_reloc_sym);
1233         }
1234
1235         map__zput(machine->vmlinux_map);
1236 }
1237
1238 int machines__create_guest_kernel_maps(struct machines *machines)
1239 {
1240         int ret = 0;
1241         struct dirent **namelist = NULL;
1242         int i, items = 0;
1243         char path[PATH_MAX];
1244         pid_t pid;
1245         char *endp;
1246
1247         if (symbol_conf.default_guest_vmlinux_name ||
1248             symbol_conf.default_guest_modules ||
1249             symbol_conf.default_guest_kallsyms) {
1250                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1251         }
1252
1253         if (symbol_conf.guestmount) {
1254                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1255                 if (items <= 0)
1256                         return -ENOENT;
1257                 for (i = 0; i < items; i++) {
1258                         if (!isdigit(namelist[i]->d_name[0])) {
1259                                 /* Filter out . and .. */
1260                                 continue;
1261                         }
1262                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1263                         if ((*endp != '\0') ||
1264                             (endp == namelist[i]->d_name) ||
1265                             (errno == ERANGE)) {
1266                                 pr_debug("invalid directory (%s). Skipping.\n",
1267                                          namelist[i]->d_name);
1268                                 continue;
1269                         }
1270                         sprintf(path, "%s/%s/proc/kallsyms",
1271                                 symbol_conf.guestmount,
1272                                 namelist[i]->d_name);
1273                         ret = access(path, R_OK);
1274                         if (ret) {
1275                                 pr_debug("Can't access file %s\n", path);
1276                                 goto failure;
1277                         }
1278                         machines__create_kernel_maps(machines, pid);
1279                 }
1280 failure:
1281                 free(namelist);
1282         }
1283
1284         return ret;
1285 }
1286
1287 void machines__destroy_kernel_maps(struct machines *machines)
1288 {
1289         struct rb_node *next = rb_first_cached(&machines->guests);
1290
1291         machine__destroy_kernel_maps(&machines->host);
1292
1293         while (next) {
1294                 struct machine *pos = rb_entry(next, struct machine, rb_node);
1295
1296                 next = rb_next(&pos->rb_node);
1297                 rb_erase_cached(&pos->rb_node, &machines->guests);
1298                 machine__delete(pos);
1299         }
1300 }
1301
1302 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1303 {
1304         struct machine *machine = machines__findnew(machines, pid);
1305
1306         if (machine == NULL)
1307                 return -1;
1308
1309         return machine__create_kernel_maps(machine);
1310 }
1311
1312 int machine__load_kallsyms(struct machine *machine, const char *filename)
1313 {
1314         struct map *map = machine__kernel_map(machine);
1315         int ret = __dso__load_kallsyms(map->dso, filename, map, true);
1316
1317         if (ret > 0) {
1318                 dso__set_loaded(map->dso);
1319                 /*
1320                  * Since /proc/kallsyms will have multiple sessions for the
1321                  * kernel, with modules between them, fixup the end of all
1322                  * sections.
1323                  */
1324                 maps__fixup_end(&machine->kmaps);
1325         }
1326
1327         return ret;
1328 }
1329
1330 int machine__load_vmlinux_path(struct machine *machine)
1331 {
1332         struct map *map = machine__kernel_map(machine);
1333         int ret = dso__load_vmlinux_path(map->dso, map);
1334
1335         if (ret > 0)
1336                 dso__set_loaded(map->dso);
1337
1338         return ret;
1339 }
1340
1341 static char *get_kernel_version(const char *root_dir)
1342 {
1343         char version[PATH_MAX];
1344         FILE *file;
1345         char *name, *tmp;
1346         const char *prefix = "Linux version ";
1347
1348         sprintf(version, "%s/proc/version", root_dir);
1349         file = fopen(version, "r");
1350         if (!file)
1351                 return NULL;
1352
1353         tmp = fgets(version, sizeof(version), file);
1354         fclose(file);
1355         if (!tmp)
1356                 return NULL;
1357
1358         name = strstr(version, prefix);
1359         if (!name)
1360                 return NULL;
1361         name += strlen(prefix);
1362         tmp = strchr(name, ' ');
1363         if (tmp)
1364                 *tmp = '\0';
1365
1366         return strdup(name);
1367 }
1368
1369 static bool is_kmod_dso(struct dso *dso)
1370 {
1371         return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1372                dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1373 }
1374
1375 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1376 {
1377         char *long_name;
1378         struct map *map = maps__find_by_name(maps, m->name);
1379
1380         if (map == NULL)
1381                 return 0;
1382
1383         long_name = strdup(path);
1384         if (long_name == NULL)
1385                 return -ENOMEM;
1386
1387         dso__set_long_name(map->dso, long_name, true);
1388         dso__kernel_module_get_build_id(map->dso, "");
1389
1390         /*
1391          * Full name could reveal us kmod compression, so
1392          * we need to update the symtab_type if needed.
1393          */
1394         if (m->comp && is_kmod_dso(map->dso)) {
1395                 map->dso->symtab_type++;
1396                 map->dso->comp = m->comp;
1397         }
1398
1399         return 0;
1400 }
1401
1402 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1403 {
1404         struct dirent *dent;
1405         DIR *dir = opendir(dir_name);
1406         int ret = 0;
1407
1408         if (!dir) {
1409                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1410                 return -1;
1411         }
1412
1413         while ((dent = readdir(dir)) != NULL) {
1414                 char path[PATH_MAX];
1415                 struct stat st;
1416
1417                 /*sshfs might return bad dent->d_type, so we have to stat*/
1418                 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1419                 if (stat(path, &st))
1420                         continue;
1421
1422                 if (S_ISDIR(st.st_mode)) {
1423                         if (!strcmp(dent->d_name, ".") ||
1424                             !strcmp(dent->d_name, ".."))
1425                                 continue;
1426
1427                         /* Do not follow top-level source and build symlinks */
1428                         if (depth == 0) {
1429                                 if (!strcmp(dent->d_name, "source") ||
1430                                     !strcmp(dent->d_name, "build"))
1431                                         continue;
1432                         }
1433
1434                         ret = maps__set_modules_path_dir(maps, path, depth + 1);
1435                         if (ret < 0)
1436                                 goto out;
1437                 } else {
1438                         struct kmod_path m;
1439
1440                         ret = kmod_path__parse_name(&m, dent->d_name);
1441                         if (ret)
1442                                 goto out;
1443
1444                         if (m.kmod)
1445                                 ret = maps__set_module_path(maps, path, &m);
1446
1447                         zfree(&m.name);
1448
1449                         if (ret)
1450                                 goto out;
1451                 }
1452         }
1453
1454 out:
1455         closedir(dir);
1456         return ret;
1457 }
1458
1459 static int machine__set_modules_path(struct machine *machine)
1460 {
1461         char *version;
1462         char modules_path[PATH_MAX];
1463
1464         version = get_kernel_version(machine->root_dir);
1465         if (!version)
1466                 return -1;
1467
1468         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1469                  machine->root_dir, version);
1470         free(version);
1471
1472         return maps__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1473 }
1474 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1475                                 u64 *size __maybe_unused,
1476                                 const char *name __maybe_unused)
1477 {
1478         return 0;
1479 }
1480
1481 static int machine__create_module(void *arg, const char *name, u64 start,
1482                                   u64 size)
1483 {
1484         struct machine *machine = arg;
1485         struct map *map;
1486
1487         if (arch__fix_module_text_start(&start, &size, name) < 0)
1488                 return -1;
1489
1490         map = machine__addnew_module_map(machine, start, name);
1491         if (map == NULL)
1492                 return -1;
1493         map->end = start + size;
1494
1495         dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1496
1497         return 0;
1498 }
1499
1500 static int machine__create_modules(struct machine *machine)
1501 {
1502         const char *modules;
1503         char path[PATH_MAX];
1504
1505         if (machine__is_default_guest(machine)) {
1506                 modules = symbol_conf.default_guest_modules;
1507         } else {
1508                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1509                 modules = path;
1510         }
1511
1512         if (symbol__restricted_filename(modules, "/proc/modules"))
1513                 return -1;
1514
1515         if (modules__parse(modules, machine, machine__create_module))
1516                 return -1;
1517
1518         if (!machine__set_modules_path(machine))
1519                 return 0;
1520
1521         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1522
1523         return 0;
1524 }
1525
1526 static void machine__set_kernel_mmap(struct machine *machine,
1527                                      u64 start, u64 end)
1528 {
1529         machine->vmlinux_map->start = start;
1530         machine->vmlinux_map->end   = end;
1531         /*
1532          * Be a bit paranoid here, some perf.data file came with
1533          * a zero sized synthesized MMAP event for the kernel.
1534          */
1535         if (start == 0 && end == 0)
1536                 machine->vmlinux_map->end = ~0ULL;
1537 }
1538
1539 static void machine__update_kernel_mmap(struct machine *machine,
1540                                      u64 start, u64 end)
1541 {
1542         struct map *map = machine__kernel_map(machine);
1543
1544         map__get(map);
1545         maps__remove(&machine->kmaps, map);
1546
1547         machine__set_kernel_mmap(machine, start, end);
1548
1549         maps__insert(&machine->kmaps, map);
1550         map__put(map);
1551 }
1552
1553 int machine__create_kernel_maps(struct machine *machine)
1554 {
1555         struct dso *kernel = machine__get_kernel(machine);
1556         const char *name = NULL;
1557         struct map *map;
1558         u64 start = 0, end = ~0ULL;
1559         int ret;
1560
1561         if (kernel == NULL)
1562                 return -1;
1563
1564         ret = __machine__create_kernel_maps(machine, kernel);
1565         if (ret < 0)
1566                 goto out_put;
1567
1568         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1569                 if (machine__is_host(machine))
1570                         pr_debug("Problems creating module maps, "
1571                                  "continuing anyway...\n");
1572                 else
1573                         pr_debug("Problems creating module maps for guest %d, "
1574                                  "continuing anyway...\n", machine->pid);
1575         }
1576
1577         if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1578                 if (name &&
1579                     map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1580                         machine__destroy_kernel_maps(machine);
1581                         ret = -1;
1582                         goto out_put;
1583                 }
1584
1585                 /*
1586                  * we have a real start address now, so re-order the kmaps
1587                  * assume it's the last in the kmaps
1588                  */
1589                 machine__update_kernel_mmap(machine, start, end);
1590         }
1591
1592         if (machine__create_extra_kernel_maps(machine, kernel))
1593                 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1594
1595         if (end == ~0ULL) {
1596                 /* update end address of the kernel map using adjacent module address */
1597                 map = map__next(machine__kernel_map(machine));
1598                 if (map)
1599                         machine__set_kernel_mmap(machine, start, map->start);
1600         }
1601
1602 out_put:
1603         dso__put(kernel);
1604         return ret;
1605 }
1606
1607 static bool machine__uses_kcore(struct machine *machine)
1608 {
1609         struct dso *dso;
1610
1611         list_for_each_entry(dso, &machine->dsos.head, node) {
1612                 if (dso__is_kcore(dso))
1613                         return true;
1614         }
1615
1616         return false;
1617 }
1618
1619 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1620                                              struct extra_kernel_map *xm)
1621 {
1622         return machine__is(machine, "x86_64") &&
1623                is_entry_trampoline(xm->name);
1624 }
1625
1626 static int machine__process_extra_kernel_map(struct machine *machine,
1627                                              struct extra_kernel_map *xm)
1628 {
1629         struct dso *kernel = machine__kernel_dso(machine);
1630
1631         if (kernel == NULL)
1632                 return -1;
1633
1634         return machine__create_extra_kernel_map(machine, kernel, xm);
1635 }
1636
1637 static int machine__process_kernel_mmap_event(struct machine *machine,
1638                                               struct extra_kernel_map *xm,
1639                                               struct build_id *bid)
1640 {
1641         struct map *map;
1642         enum dso_space_type dso_space;
1643         bool is_kernel_mmap;
1644
1645         /* If we have maps from kcore then we do not need or want any others */
1646         if (machine__uses_kcore(machine))
1647                 return 0;
1648
1649         if (machine__is_host(machine))
1650                 dso_space = DSO_SPACE__KERNEL;
1651         else
1652                 dso_space = DSO_SPACE__KERNEL_GUEST;
1653
1654         is_kernel_mmap = memcmp(xm->name, machine->mmap_name,
1655                                 strlen(machine->mmap_name) - 1) == 0;
1656         if (xm->name[0] == '/' ||
1657             (!is_kernel_mmap && xm->name[0] == '[')) {
1658                 map = machine__addnew_module_map(machine, xm->start,
1659                                                  xm->name);
1660                 if (map == NULL)
1661                         goto out_problem;
1662
1663                 map->end = map->start + xm->end - xm->start;
1664
1665                 if (build_id__is_defined(bid))
1666                         dso__set_build_id(map->dso, bid);
1667
1668         } else if (is_kernel_mmap) {
1669                 const char *symbol_name = (xm->name + strlen(machine->mmap_name));
1670                 /*
1671                  * Should be there already, from the build-id table in
1672                  * the header.
1673                  */
1674                 struct dso *kernel = NULL;
1675                 struct dso *dso;
1676
1677                 down_read(&machine->dsos.lock);
1678
1679                 list_for_each_entry(dso, &machine->dsos.head, node) {
1680
1681                         /*
1682                          * The cpumode passed to is_kernel_module is not the
1683                          * cpumode of *this* event. If we insist on passing
1684                          * correct cpumode to is_kernel_module, we should
1685                          * record the cpumode when we adding this dso to the
1686                          * linked list.
1687                          *
1688                          * However we don't really need passing correct
1689                          * cpumode.  We know the correct cpumode must be kernel
1690                          * mode (if not, we should not link it onto kernel_dsos
1691                          * list).
1692                          *
1693                          * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1694                          * is_kernel_module() treats it as a kernel cpumode.
1695                          */
1696
1697                         if (!dso->kernel ||
1698                             is_kernel_module(dso->long_name,
1699                                              PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1700                                 continue;
1701
1702
1703                         kernel = dso;
1704                         break;
1705                 }
1706
1707                 up_read(&machine->dsos.lock);
1708
1709                 if (kernel == NULL)
1710                         kernel = machine__findnew_dso(machine, machine->mmap_name);
1711                 if (kernel == NULL)
1712                         goto out_problem;
1713
1714                 kernel->kernel = dso_space;
1715                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1716                         dso__put(kernel);
1717                         goto out_problem;
1718                 }
1719
1720                 if (strstr(kernel->long_name, "vmlinux"))
1721                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1722
1723                 machine__update_kernel_mmap(machine, xm->start, xm->end);
1724
1725                 if (build_id__is_defined(bid))
1726                         dso__set_build_id(kernel, bid);
1727
1728                 /*
1729                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1730                  * symbol. Effectively having zero here means that at record
1731                  * time /proc/sys/kernel/kptr_restrict was non zero.
1732                  */
1733                 if (xm->pgoff != 0) {
1734                         map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1735                                                         symbol_name,
1736                                                         xm->pgoff);
1737                 }
1738
1739                 if (machine__is_default_guest(machine)) {
1740                         /*
1741                          * preload dso of guest kernel and modules
1742                          */
1743                         dso__load(kernel, machine__kernel_map(machine));
1744                 }
1745         } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1746                 return machine__process_extra_kernel_map(machine, xm);
1747         }
1748         return 0;
1749 out_problem:
1750         return -1;
1751 }
1752
1753 int machine__process_mmap2_event(struct machine *machine,
1754                                  union perf_event *event,
1755                                  struct perf_sample *sample)
1756 {
1757         struct thread *thread;
1758         struct map *map;
1759         struct dso_id dso_id = {
1760                 .maj = event->mmap2.maj,
1761                 .min = event->mmap2.min,
1762                 .ino = event->mmap2.ino,
1763                 .ino_generation = event->mmap2.ino_generation,
1764         };
1765         struct build_id __bid, *bid = NULL;
1766         int ret = 0;
1767
1768         if (dump_trace)
1769                 perf_event__fprintf_mmap2(event, stdout);
1770
1771         if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1772                 bid = &__bid;
1773                 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1774         }
1775
1776         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1777             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1778                 struct extra_kernel_map xm = {
1779                         .start = event->mmap2.start,
1780                         .end   = event->mmap2.start + event->mmap2.len,
1781                         .pgoff = event->mmap2.pgoff,
1782                 };
1783
1784                 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1785                 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1786                 if (ret < 0)
1787                         goto out_problem;
1788                 return 0;
1789         }
1790
1791         thread = machine__findnew_thread(machine, event->mmap2.pid,
1792                                         event->mmap2.tid);
1793         if (thread == NULL)
1794                 goto out_problem;
1795
1796         map = map__new(machine, event->mmap2.start,
1797                         event->mmap2.len, event->mmap2.pgoff,
1798                         &dso_id, event->mmap2.prot,
1799                         event->mmap2.flags, bid,
1800                         event->mmap2.filename, thread);
1801
1802         if (map == NULL)
1803                 goto out_problem_map;
1804
1805         ret = thread__insert_map(thread, map);
1806         if (ret)
1807                 goto out_problem_insert;
1808
1809         thread__put(thread);
1810         map__put(map);
1811         return 0;
1812
1813 out_problem_insert:
1814         map__put(map);
1815 out_problem_map:
1816         thread__put(thread);
1817 out_problem:
1818         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1819         return 0;
1820 }
1821
1822 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1823                                 struct perf_sample *sample)
1824 {
1825         struct thread *thread;
1826         struct map *map;
1827         u32 prot = 0;
1828         int ret = 0;
1829
1830         if (dump_trace)
1831                 perf_event__fprintf_mmap(event, stdout);
1832
1833         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1834             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1835                 struct extra_kernel_map xm = {
1836                         .start = event->mmap.start,
1837                         .end   = event->mmap.start + event->mmap.len,
1838                         .pgoff = event->mmap.pgoff,
1839                 };
1840
1841                 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1842                 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1843                 if (ret < 0)
1844                         goto out_problem;
1845                 return 0;
1846         }
1847
1848         thread = machine__findnew_thread(machine, event->mmap.pid,
1849                                          event->mmap.tid);
1850         if (thread == NULL)
1851                 goto out_problem;
1852
1853         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1854                 prot = PROT_EXEC;
1855
1856         map = map__new(machine, event->mmap.start,
1857                         event->mmap.len, event->mmap.pgoff,
1858                         NULL, prot, 0, NULL, event->mmap.filename, thread);
1859
1860         if (map == NULL)
1861                 goto out_problem_map;
1862
1863         ret = thread__insert_map(thread, map);
1864         if (ret)
1865                 goto out_problem_insert;
1866
1867         thread__put(thread);
1868         map__put(map);
1869         return 0;
1870
1871 out_problem_insert:
1872         map__put(map);
1873 out_problem_map:
1874         thread__put(thread);
1875 out_problem:
1876         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1877         return 0;
1878 }
1879
1880 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1881 {
1882         struct threads *threads = machine__threads(machine, th->tid);
1883
1884         if (threads->last_match == th)
1885                 threads__set_last_match(threads, NULL);
1886
1887         if (lock)
1888                 down_write(&threads->lock);
1889
1890         BUG_ON(refcount_read(&th->refcnt) == 0);
1891
1892         rb_erase_cached(&th->rb_node, &threads->entries);
1893         RB_CLEAR_NODE(&th->rb_node);
1894         --threads->nr;
1895         /*
1896          * Move it first to the dead_threads list, then drop the reference,
1897          * if this is the last reference, then the thread__delete destructor
1898          * will be called and we will remove it from the dead_threads list.
1899          */
1900         list_add_tail(&th->node, &threads->dead);
1901
1902         /*
1903          * We need to do the put here because if this is the last refcount,
1904          * then we will be touching the threads->dead head when removing the
1905          * thread.
1906          */
1907         thread__put(th);
1908
1909         if (lock)
1910                 up_write(&threads->lock);
1911 }
1912
1913 void machine__remove_thread(struct machine *machine, struct thread *th)
1914 {
1915         return __machine__remove_thread(machine, th, true);
1916 }
1917
1918 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1919                                 struct perf_sample *sample)
1920 {
1921         struct thread *thread = machine__find_thread(machine,
1922                                                      event->fork.pid,
1923                                                      event->fork.tid);
1924         struct thread *parent = machine__findnew_thread(machine,
1925                                                         event->fork.ppid,
1926                                                         event->fork.ptid);
1927         bool do_maps_clone = true;
1928         int err = 0;
1929
1930         if (dump_trace)
1931                 perf_event__fprintf_task(event, stdout);
1932
1933         /*
1934          * There may be an existing thread that is not actually the parent,
1935          * either because we are processing events out of order, or because the
1936          * (fork) event that would have removed the thread was lost. Assume the
1937          * latter case and continue on as best we can.
1938          */
1939         if (parent->pid_ != (pid_t)event->fork.ppid) {
1940                 dump_printf("removing erroneous parent thread %d/%d\n",
1941                             parent->pid_, parent->tid);
1942                 machine__remove_thread(machine, parent);
1943                 thread__put(parent);
1944                 parent = machine__findnew_thread(machine, event->fork.ppid,
1945                                                  event->fork.ptid);
1946         }
1947
1948         /* if a thread currently exists for the thread id remove it */
1949         if (thread != NULL) {
1950                 machine__remove_thread(machine, thread);
1951                 thread__put(thread);
1952         }
1953
1954         thread = machine__findnew_thread(machine, event->fork.pid,
1955                                          event->fork.tid);
1956         /*
1957          * When synthesizing FORK events, we are trying to create thread
1958          * objects for the already running tasks on the machine.
1959          *
1960          * Normally, for a kernel FORK event, we want to clone the parent's
1961          * maps because that is what the kernel just did.
1962          *
1963          * But when synthesizing, this should not be done.  If we do, we end up
1964          * with overlapping maps as we process the synthesized MMAP2 events that
1965          * get delivered shortly thereafter.
1966          *
1967          * Use the FORK event misc flags in an internal way to signal this
1968          * situation, so we can elide the map clone when appropriate.
1969          */
1970         if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1971                 do_maps_clone = false;
1972
1973         if (thread == NULL || parent == NULL ||
1974             thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1975                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1976                 err = -1;
1977         }
1978         thread__put(thread);
1979         thread__put(parent);
1980
1981         return err;
1982 }
1983
1984 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1985                                 struct perf_sample *sample __maybe_unused)
1986 {
1987         struct thread *thread = machine__find_thread(machine,
1988                                                      event->fork.pid,
1989                                                      event->fork.tid);
1990
1991         if (dump_trace)
1992                 perf_event__fprintf_task(event, stdout);
1993
1994         if (thread != NULL) {
1995                 thread__exited(thread);
1996                 thread__put(thread);
1997         }
1998
1999         return 0;
2000 }
2001
2002 int machine__process_event(struct machine *machine, union perf_event *event,
2003                            struct perf_sample *sample)
2004 {
2005         int ret;
2006
2007         switch (event->header.type) {
2008         case PERF_RECORD_COMM:
2009                 ret = machine__process_comm_event(machine, event, sample); break;
2010         case PERF_RECORD_MMAP:
2011                 ret = machine__process_mmap_event(machine, event, sample); break;
2012         case PERF_RECORD_NAMESPACES:
2013                 ret = machine__process_namespaces_event(machine, event, sample); break;
2014         case PERF_RECORD_CGROUP:
2015                 ret = machine__process_cgroup_event(machine, event, sample); break;
2016         case PERF_RECORD_MMAP2:
2017                 ret = machine__process_mmap2_event(machine, event, sample); break;
2018         case PERF_RECORD_FORK:
2019                 ret = machine__process_fork_event(machine, event, sample); break;
2020         case PERF_RECORD_EXIT:
2021                 ret = machine__process_exit_event(machine, event, sample); break;
2022         case PERF_RECORD_LOST:
2023                 ret = machine__process_lost_event(machine, event, sample); break;
2024         case PERF_RECORD_AUX:
2025                 ret = machine__process_aux_event(machine, event); break;
2026         case PERF_RECORD_ITRACE_START:
2027                 ret = machine__process_itrace_start_event(machine, event); break;
2028         case PERF_RECORD_LOST_SAMPLES:
2029                 ret = machine__process_lost_samples_event(machine, event, sample); break;
2030         case PERF_RECORD_SWITCH:
2031         case PERF_RECORD_SWITCH_CPU_WIDE:
2032                 ret = machine__process_switch_event(machine, event); break;
2033         case PERF_RECORD_KSYMBOL:
2034                 ret = machine__process_ksymbol(machine, event, sample); break;
2035         case PERF_RECORD_BPF_EVENT:
2036                 ret = machine__process_bpf(machine, event, sample); break;
2037         case PERF_RECORD_TEXT_POKE:
2038                 ret = machine__process_text_poke(machine, event, sample); break;
2039         case PERF_RECORD_AUX_OUTPUT_HW_ID:
2040                 ret = machine__process_aux_output_hw_id_event(machine, event); break;
2041         default:
2042                 ret = -1;
2043                 break;
2044         }
2045
2046         return ret;
2047 }
2048
2049 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2050 {
2051         if (!regexec(regex, sym->name, 0, NULL, 0))
2052                 return true;
2053         return false;
2054 }
2055
2056 static void ip__resolve_ams(struct thread *thread,
2057                             struct addr_map_symbol *ams,
2058                             u64 ip)
2059 {
2060         struct addr_location al;
2061
2062         memset(&al, 0, sizeof(al));
2063         /*
2064          * We cannot use the header.misc hint to determine whether a
2065          * branch stack address is user, kernel, guest, hypervisor.
2066          * Branches may straddle the kernel/user/hypervisor boundaries.
2067          * Thus, we have to try consecutively until we find a match
2068          * or else, the symbol is unknown
2069          */
2070         thread__find_cpumode_addr_location(thread, ip, &al);
2071
2072         ams->addr = ip;
2073         ams->al_addr = al.addr;
2074         ams->ms.maps = al.maps;
2075         ams->ms.sym = al.sym;
2076         ams->ms.map = al.map;
2077         ams->phys_addr = 0;
2078         ams->data_page_size = 0;
2079 }
2080
2081 static void ip__resolve_data(struct thread *thread,
2082                              u8 m, struct addr_map_symbol *ams,
2083                              u64 addr, u64 phys_addr, u64 daddr_page_size)
2084 {
2085         struct addr_location al;
2086
2087         memset(&al, 0, sizeof(al));
2088
2089         thread__find_symbol(thread, m, addr, &al);
2090
2091         ams->addr = addr;
2092         ams->al_addr = al.addr;
2093         ams->ms.maps = al.maps;
2094         ams->ms.sym = al.sym;
2095         ams->ms.map = al.map;
2096         ams->phys_addr = phys_addr;
2097         ams->data_page_size = daddr_page_size;
2098 }
2099
2100 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2101                                      struct addr_location *al)
2102 {
2103         struct mem_info *mi = mem_info__new();
2104
2105         if (!mi)
2106                 return NULL;
2107
2108         ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2109         ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2110                          sample->addr, sample->phys_addr,
2111                          sample->data_page_size);
2112         mi->data_src.val = sample->data_src;
2113
2114         return mi;
2115 }
2116
2117 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2118 {
2119         struct map *map = ms->map;
2120         char *srcline = NULL;
2121
2122         if (!map || callchain_param.key == CCKEY_FUNCTION)
2123                 return srcline;
2124
2125         srcline = srcline__tree_find(&map->dso->srclines, ip);
2126         if (!srcline) {
2127                 bool show_sym = false;
2128                 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2129
2130                 srcline = get_srcline(map->dso, map__rip_2objdump(map, ip),
2131                                       ms->sym, show_sym, show_addr, ip);
2132                 srcline__tree_insert(&map->dso->srclines, ip, srcline);
2133         }
2134
2135         return srcline;
2136 }
2137
2138 struct iterations {
2139         int nr_loop_iter;
2140         u64 cycles;
2141 };
2142
2143 static int add_callchain_ip(struct thread *thread,
2144                             struct callchain_cursor *cursor,
2145                             struct symbol **parent,
2146                             struct addr_location *root_al,
2147                             u8 *cpumode,
2148                             u64 ip,
2149                             bool branch,
2150                             struct branch_flags *flags,
2151                             struct iterations *iter,
2152                             u64 branch_from)
2153 {
2154         struct map_symbol ms;
2155         struct addr_location al;
2156         int nr_loop_iter = 0;
2157         u64 iter_cycles = 0;
2158         const char *srcline = NULL;
2159
2160         al.filtered = 0;
2161         al.sym = NULL;
2162         al.srcline = NULL;
2163         if (!cpumode) {
2164                 thread__find_cpumode_addr_location(thread, ip, &al);
2165         } else {
2166                 if (ip >= PERF_CONTEXT_MAX) {
2167                         switch (ip) {
2168                         case PERF_CONTEXT_HV:
2169                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2170                                 break;
2171                         case PERF_CONTEXT_KERNEL:
2172                                 *cpumode = PERF_RECORD_MISC_KERNEL;
2173                                 break;
2174                         case PERF_CONTEXT_USER:
2175                                 *cpumode = PERF_RECORD_MISC_USER;
2176                                 break;
2177                         default:
2178                                 pr_debug("invalid callchain context: "
2179                                          "%"PRId64"\n", (s64) ip);
2180                                 /*
2181                                  * It seems the callchain is corrupted.
2182                                  * Discard all.
2183                                  */
2184                                 callchain_cursor_reset(cursor);
2185                                 return 1;
2186                         }
2187                         return 0;
2188                 }
2189                 thread__find_symbol(thread, *cpumode, ip, &al);
2190         }
2191
2192         if (al.sym != NULL) {
2193                 if (perf_hpp_list.parent && !*parent &&
2194                     symbol__match_regex(al.sym, &parent_regex))
2195                         *parent = al.sym;
2196                 else if (have_ignore_callees && root_al &&
2197                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2198                         /* Treat this symbol as the root,
2199                            forgetting its callees. */
2200                         *root_al = al;
2201                         callchain_cursor_reset(cursor);
2202                 }
2203         }
2204
2205         if (symbol_conf.hide_unresolved && al.sym == NULL)
2206                 return 0;
2207
2208         if (iter) {
2209                 nr_loop_iter = iter->nr_loop_iter;
2210                 iter_cycles = iter->cycles;
2211         }
2212
2213         ms.maps = al.maps;
2214         ms.map = al.map;
2215         ms.sym = al.sym;
2216         srcline = callchain_srcline(&ms, al.addr);
2217         return callchain_cursor_append(cursor, ip, &ms,
2218                                        branch, flags, nr_loop_iter,
2219                                        iter_cycles, branch_from, srcline);
2220 }
2221
2222 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2223                                            struct addr_location *al)
2224 {
2225         unsigned int i;
2226         const struct branch_stack *bs = sample->branch_stack;
2227         struct branch_entry *entries = perf_sample__branch_entries(sample);
2228         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2229
2230         if (!bi)
2231                 return NULL;
2232
2233         for (i = 0; i < bs->nr; i++) {
2234                 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2235                 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2236                 bi[i].flags = entries[i].flags;
2237         }
2238         return bi;
2239 }
2240
2241 static void save_iterations(struct iterations *iter,
2242                             struct branch_entry *be, int nr)
2243 {
2244         int i;
2245
2246         iter->nr_loop_iter++;
2247         iter->cycles = 0;
2248
2249         for (i = 0; i < nr; i++)
2250                 iter->cycles += be[i].flags.cycles;
2251 }
2252
2253 #define CHASHSZ 127
2254 #define CHASHBITS 7
2255 #define NO_ENTRY 0xff
2256
2257 #define PERF_MAX_BRANCH_DEPTH 127
2258
2259 /* Remove loops. */
2260 static int remove_loops(struct branch_entry *l, int nr,
2261                         struct iterations *iter)
2262 {
2263         int i, j, off;
2264         unsigned char chash[CHASHSZ];
2265
2266         memset(chash, NO_ENTRY, sizeof(chash));
2267
2268         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2269
2270         for (i = 0; i < nr; i++) {
2271                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2272
2273                 /* no collision handling for now */
2274                 if (chash[h] == NO_ENTRY) {
2275                         chash[h] = i;
2276                 } else if (l[chash[h]].from == l[i].from) {
2277                         bool is_loop = true;
2278                         /* check if it is a real loop */
2279                         off = 0;
2280                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
2281                                 if (l[j].from != l[i + off].from) {
2282                                         is_loop = false;
2283                                         break;
2284                                 }
2285                         if (is_loop) {
2286                                 j = nr - (i + off);
2287                                 if (j > 0) {
2288                                         save_iterations(iter + i + off,
2289                                                 l + i, off);
2290
2291                                         memmove(iter + i, iter + i + off,
2292                                                 j * sizeof(*iter));
2293
2294                                         memmove(l + i, l + i + off,
2295                                                 j * sizeof(*l));
2296                                 }
2297
2298                                 nr -= off;
2299                         }
2300                 }
2301         }
2302         return nr;
2303 }
2304
2305 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2306                                        struct callchain_cursor *cursor,
2307                                        struct perf_sample *sample,
2308                                        struct symbol **parent,
2309                                        struct addr_location *root_al,
2310                                        u64 branch_from,
2311                                        bool callee, int end)
2312 {
2313         struct ip_callchain *chain = sample->callchain;
2314         u8 cpumode = PERF_RECORD_MISC_USER;
2315         int err, i;
2316
2317         if (callee) {
2318                 for (i = 0; i < end + 1; i++) {
2319                         err = add_callchain_ip(thread, cursor, parent,
2320                                                root_al, &cpumode, chain->ips[i],
2321                                                false, NULL, NULL, branch_from);
2322                         if (err)
2323                                 return err;
2324                 }
2325                 return 0;
2326         }
2327
2328         for (i = end; i >= 0; i--) {
2329                 err = add_callchain_ip(thread, cursor, parent,
2330                                        root_al, &cpumode, chain->ips[i],
2331                                        false, NULL, NULL, branch_from);
2332                 if (err)
2333                         return err;
2334         }
2335
2336         return 0;
2337 }
2338
2339 static void save_lbr_cursor_node(struct thread *thread,
2340                                  struct callchain_cursor *cursor,
2341                                  int idx)
2342 {
2343         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2344
2345         if (!lbr_stitch)
2346                 return;
2347
2348         if (cursor->pos == cursor->nr) {
2349                 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2350                 return;
2351         }
2352
2353         if (!cursor->curr)
2354                 cursor->curr = cursor->first;
2355         else
2356                 cursor->curr = cursor->curr->next;
2357         memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2358                sizeof(struct callchain_cursor_node));
2359
2360         lbr_stitch->prev_lbr_cursor[idx].valid = true;
2361         cursor->pos++;
2362 }
2363
2364 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2365                                     struct callchain_cursor *cursor,
2366                                     struct perf_sample *sample,
2367                                     struct symbol **parent,
2368                                     struct addr_location *root_al,
2369                                     u64 *branch_from,
2370                                     bool callee)
2371 {
2372         struct branch_stack *lbr_stack = sample->branch_stack;
2373         struct branch_entry *entries = perf_sample__branch_entries(sample);
2374         u8 cpumode = PERF_RECORD_MISC_USER;
2375         int lbr_nr = lbr_stack->nr;
2376         struct branch_flags *flags;
2377         int err, i;
2378         u64 ip;
2379
2380         /*
2381          * The curr and pos are not used in writing session. They are cleared
2382          * in callchain_cursor_commit() when the writing session is closed.
2383          * Using curr and pos to track the current cursor node.
2384          */
2385         if (thread->lbr_stitch) {
2386                 cursor->curr = NULL;
2387                 cursor->pos = cursor->nr;
2388                 if (cursor->nr) {
2389                         cursor->curr = cursor->first;
2390                         for (i = 0; i < (int)(cursor->nr - 1); i++)
2391                                 cursor->curr = cursor->curr->next;
2392                 }
2393         }
2394
2395         if (callee) {
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                 /*
2408                  * The number of cursor node increases.
2409                  * Move the current cursor node.
2410                  * But does not need to save current cursor node for entry 0.
2411                  * It's impossible to stitch the whole LBRs of previous sample.
2412                  */
2413                 if (thread->lbr_stitch && (cursor->pos != cursor->nr)) {
2414                         if (!cursor->curr)
2415                                 cursor->curr = cursor->first;
2416                         else
2417                                 cursor->curr = cursor->curr->next;
2418                         cursor->pos++;
2419                 }
2420
2421                 /* Add LBR ip from entries.from one by one. */
2422                 for (i = 0; i < lbr_nr; i++) {
2423                         ip = entries[i].from;
2424                         flags = &entries[i].flags;
2425                         err = add_callchain_ip(thread, cursor, parent,
2426                                                root_al, &cpumode, ip,
2427                                                true, flags, NULL,
2428                                                *branch_from);
2429                         if (err)
2430                                 return err;
2431                         save_lbr_cursor_node(thread, cursor, i);
2432                 }
2433                 return 0;
2434         }
2435
2436         /* Add LBR ip from entries.from one by one. */
2437         for (i = lbr_nr - 1; i >= 0; i--) {
2438                 ip = entries[i].from;
2439                 flags = &entries[i].flags;
2440                 err = add_callchain_ip(thread, cursor, parent,
2441                                        root_al, &cpumode, ip,
2442                                        true, flags, NULL,
2443                                        *branch_from);
2444                 if (err)
2445                         return err;
2446                 save_lbr_cursor_node(thread, cursor, i);
2447         }
2448
2449         /* Add LBR ip from first entries.to */
2450         ip = entries[0].to;
2451         flags = &entries[0].flags;
2452         *branch_from = entries[0].from;
2453         err = add_callchain_ip(thread, cursor, parent,
2454                                root_al, &cpumode, ip,
2455                                true, flags, NULL,
2456                                *branch_from);
2457         if (err)
2458                 return err;
2459
2460         return 0;
2461 }
2462
2463 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2464                                              struct callchain_cursor *cursor)
2465 {
2466         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2467         struct callchain_cursor_node *cnode;
2468         struct stitch_list *stitch_node;
2469         int err;
2470
2471         list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2472                 cnode = &stitch_node->cursor;
2473
2474                 err = callchain_cursor_append(cursor, cnode->ip,
2475                                               &cnode->ms,
2476                                               cnode->branch,
2477                                               &cnode->branch_flags,
2478                                               cnode->nr_loop_iter,
2479                                               cnode->iter_cycles,
2480                                               cnode->branch_from,
2481                                               cnode->srcline);
2482                 if (err)
2483                         return err;
2484         }
2485         return 0;
2486 }
2487
2488 static struct stitch_list *get_stitch_node(struct thread *thread)
2489 {
2490         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2491         struct stitch_list *stitch_node;
2492
2493         if (!list_empty(&lbr_stitch->free_lists)) {
2494                 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2495                                                struct stitch_list, node);
2496                 list_del(&stitch_node->node);
2497
2498                 return stitch_node;
2499         }
2500
2501         return malloc(sizeof(struct stitch_list));
2502 }
2503
2504 static bool has_stitched_lbr(struct thread *thread,
2505                              struct perf_sample *cur,
2506                              struct perf_sample *prev,
2507                              unsigned int max_lbr,
2508                              bool callee)
2509 {
2510         struct branch_stack *cur_stack = cur->branch_stack;
2511         struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2512         struct branch_stack *prev_stack = prev->branch_stack;
2513         struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2514         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2515         int i, j, nr_identical_branches = 0;
2516         struct stitch_list *stitch_node;
2517         u64 cur_base, distance;
2518
2519         if (!cur_stack || !prev_stack)
2520                 return false;
2521
2522         /* Find the physical index of the base-of-stack for current sample. */
2523         cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2524
2525         distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2526                                                      (max_lbr + prev_stack->hw_idx - cur_base);
2527         /* Previous sample has shorter stack. Nothing can be stitched. */
2528         if (distance + 1 > prev_stack->nr)
2529                 return false;
2530
2531         /*
2532          * Check if there are identical LBRs between two samples.
2533          * Identical LBRs must have same from, to and flags values. Also,
2534          * they have to be saved in the same LBR registers (same physical
2535          * index).
2536          *
2537          * Starts from the base-of-stack of current sample.
2538          */
2539         for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2540                 if ((prev_entries[i].from != cur_entries[j].from) ||
2541                     (prev_entries[i].to != cur_entries[j].to) ||
2542                     (prev_entries[i].flags.value != cur_entries[j].flags.value))
2543                         break;
2544                 nr_identical_branches++;
2545         }
2546
2547         if (!nr_identical_branches)
2548                 return false;
2549
2550         /*
2551          * Save the LBRs between the base-of-stack of previous sample
2552          * and the base-of-stack of current sample into lbr_stitch->lists.
2553          * These LBRs will be stitched later.
2554          */
2555         for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2556
2557                 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2558                         continue;
2559
2560                 stitch_node = get_stitch_node(thread);
2561                 if (!stitch_node)
2562                         return false;
2563
2564                 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2565                        sizeof(struct callchain_cursor_node));
2566
2567                 if (callee)
2568                         list_add(&stitch_node->node, &lbr_stitch->lists);
2569                 else
2570                         list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2571         }
2572
2573         return true;
2574 }
2575
2576 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2577 {
2578         if (thread->lbr_stitch)
2579                 return true;
2580
2581         thread->lbr_stitch = zalloc(sizeof(*thread->lbr_stitch));
2582         if (!thread->lbr_stitch)
2583                 goto err;
2584
2585         thread->lbr_stitch->prev_lbr_cursor = calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2586         if (!thread->lbr_stitch->prev_lbr_cursor)
2587                 goto free_lbr_stitch;
2588
2589         INIT_LIST_HEAD(&thread->lbr_stitch->lists);
2590         INIT_LIST_HEAD(&thread->lbr_stitch->free_lists);
2591
2592         return true;
2593
2594 free_lbr_stitch:
2595         zfree(&thread->lbr_stitch);
2596 err:
2597         pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2598         thread->lbr_stitch_enable = false;
2599         return false;
2600 }
2601
2602 /*
2603  * Resolve LBR callstack chain sample
2604  * Return:
2605  * 1 on success get LBR callchain information
2606  * 0 no available LBR callchain information, should try fp
2607  * negative error code on other errors.
2608  */
2609 static int resolve_lbr_callchain_sample(struct thread *thread,
2610                                         struct callchain_cursor *cursor,
2611                                         struct perf_sample *sample,
2612                                         struct symbol **parent,
2613                                         struct addr_location *root_al,
2614                                         int max_stack,
2615                                         unsigned int max_lbr)
2616 {
2617         bool callee = (callchain_param.order == ORDER_CALLEE);
2618         struct ip_callchain *chain = sample->callchain;
2619         int chain_nr = min(max_stack, (int)chain->nr), i;
2620         struct lbr_stitch *lbr_stitch;
2621         bool stitched_lbr = false;
2622         u64 branch_from = 0;
2623         int err;
2624
2625         for (i = 0; i < chain_nr; i++) {
2626                 if (chain->ips[i] == PERF_CONTEXT_USER)
2627                         break;
2628         }
2629
2630         /* LBR only affects the user callchain */
2631         if (i == chain_nr)
2632                 return 0;
2633
2634         if (thread->lbr_stitch_enable && !sample->no_hw_idx &&
2635             (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2636                 lbr_stitch = thread->lbr_stitch;
2637
2638                 stitched_lbr = has_stitched_lbr(thread, sample,
2639                                                 &lbr_stitch->prev_sample,
2640                                                 max_lbr, callee);
2641
2642                 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2643                         list_replace_init(&lbr_stitch->lists,
2644                                           &lbr_stitch->free_lists);
2645                 }
2646                 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2647         }
2648
2649         if (callee) {
2650                 /* Add kernel ip */
2651                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2652                                                   parent, root_al, branch_from,
2653                                                   true, i);
2654                 if (err)
2655                         goto error;
2656
2657                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2658                                                root_al, &branch_from, true);
2659                 if (err)
2660                         goto error;
2661
2662                 if (stitched_lbr) {
2663                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2664                         if (err)
2665                                 goto error;
2666                 }
2667
2668         } else {
2669                 if (stitched_lbr) {
2670                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2671                         if (err)
2672                                 goto error;
2673                 }
2674                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2675                                                root_al, &branch_from, false);
2676                 if (err)
2677                         goto error;
2678
2679                 /* Add kernel ip */
2680                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2681                                                   parent, root_al, branch_from,
2682                                                   false, i);
2683                 if (err)
2684                         goto error;
2685         }
2686         return 1;
2687
2688 error:
2689         return (err < 0) ? err : 0;
2690 }
2691
2692 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2693                              struct callchain_cursor *cursor,
2694                              struct symbol **parent,
2695                              struct addr_location *root_al,
2696                              u8 *cpumode, int ent)
2697 {
2698         int err = 0;
2699
2700         while (--ent >= 0) {
2701                 u64 ip = chain->ips[ent];
2702
2703                 if (ip >= PERF_CONTEXT_MAX) {
2704                         err = add_callchain_ip(thread, cursor, parent,
2705                                                root_al, cpumode, ip,
2706                                                false, NULL, NULL, 0);
2707                         break;
2708                 }
2709         }
2710         return err;
2711 }
2712
2713 static int thread__resolve_callchain_sample(struct thread *thread,
2714                                             struct callchain_cursor *cursor,
2715                                             struct evsel *evsel,
2716                                             struct perf_sample *sample,
2717                                             struct symbol **parent,
2718                                             struct addr_location *root_al,
2719                                             int max_stack)
2720 {
2721         struct branch_stack *branch = sample->branch_stack;
2722         struct branch_entry *entries = perf_sample__branch_entries(sample);
2723         struct ip_callchain *chain = sample->callchain;
2724         int chain_nr = 0;
2725         u8 cpumode = PERF_RECORD_MISC_USER;
2726         int i, j, err, nr_entries;
2727         int skip_idx = -1;
2728         int first_call = 0;
2729
2730         if (chain)
2731                 chain_nr = chain->nr;
2732
2733         if (evsel__has_branch_callstack(evsel)) {
2734                 struct perf_env *env = evsel__env(evsel);
2735
2736                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2737                                                    root_al, max_stack,
2738                                                    !env ? 0 : env->max_branches);
2739                 if (err)
2740                         return (err < 0) ? err : 0;
2741         }
2742
2743         /*
2744          * Based on DWARF debug information, some architectures skip
2745          * a callchain entry saved by the kernel.
2746          */
2747         skip_idx = arch_skip_callchain_idx(thread, chain);
2748
2749         /*
2750          * Add branches to call stack for easier browsing. This gives
2751          * more context for a sample than just the callers.
2752          *
2753          * This uses individual histograms of paths compared to the
2754          * aggregated histograms the normal LBR mode uses.
2755          *
2756          * Limitations for now:
2757          * - No extra filters
2758          * - No annotations (should annotate somehow)
2759          */
2760
2761         if (branch && callchain_param.branch_callstack) {
2762                 int nr = min(max_stack, (int)branch->nr);
2763                 struct branch_entry be[nr];
2764                 struct iterations iter[nr];
2765
2766                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2767                         pr_warning("corrupted branch chain. skipping...\n");
2768                         goto check_calls;
2769                 }
2770
2771                 for (i = 0; i < nr; i++) {
2772                         if (callchain_param.order == ORDER_CALLEE) {
2773                                 be[i] = entries[i];
2774
2775                                 if (chain == NULL)
2776                                         continue;
2777
2778                                 /*
2779                                  * Check for overlap into the callchain.
2780                                  * The return address is one off compared to
2781                                  * the branch entry. To adjust for this
2782                                  * assume the calling instruction is not longer
2783                                  * than 8 bytes.
2784                                  */
2785                                 if (i == skip_idx ||
2786                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
2787                                         first_call++;
2788                                 else if (be[i].from < chain->ips[first_call] &&
2789                                     be[i].from >= chain->ips[first_call] - 8)
2790                                         first_call++;
2791                         } else
2792                                 be[i] = entries[branch->nr - i - 1];
2793                 }
2794
2795                 memset(iter, 0, sizeof(struct iterations) * nr);
2796                 nr = remove_loops(be, nr, iter);
2797
2798                 for (i = 0; i < nr; i++) {
2799                         err = add_callchain_ip(thread, cursor, parent,
2800                                                root_al,
2801                                                NULL, be[i].to,
2802                                                true, &be[i].flags,
2803                                                NULL, be[i].from);
2804
2805                         if (!err)
2806                                 err = add_callchain_ip(thread, cursor, parent, root_al,
2807                                                        NULL, be[i].from,
2808                                                        true, &be[i].flags,
2809                                                        &iter[i], 0);
2810                         if (err == -EINVAL)
2811                                 break;
2812                         if (err)
2813                                 return err;
2814                 }
2815
2816                 if (chain_nr == 0)
2817                         return 0;
2818
2819                 chain_nr -= nr;
2820         }
2821
2822 check_calls:
2823         if (chain && callchain_param.order != ORDER_CALLEE) {
2824                 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2825                                         &cpumode, chain->nr - first_call);
2826                 if (err)
2827                         return (err < 0) ? err : 0;
2828         }
2829         for (i = first_call, nr_entries = 0;
2830              i < chain_nr && nr_entries < max_stack; i++) {
2831                 u64 ip;
2832
2833                 if (callchain_param.order == ORDER_CALLEE)
2834                         j = i;
2835                 else
2836                         j = chain->nr - i - 1;
2837
2838 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2839                 if (j == skip_idx)
2840                         continue;
2841 #endif
2842                 ip = chain->ips[j];
2843                 if (ip < PERF_CONTEXT_MAX)
2844                        ++nr_entries;
2845                 else if (callchain_param.order != ORDER_CALLEE) {
2846                         err = find_prev_cpumode(chain, thread, cursor, parent,
2847                                                 root_al, &cpumode, j);
2848                         if (err)
2849                                 return (err < 0) ? err : 0;
2850                         continue;
2851                 }
2852
2853                 err = add_callchain_ip(thread, cursor, parent,
2854                                        root_al, &cpumode, ip,
2855                                        false, NULL, NULL, 0);
2856
2857                 if (err)
2858                         return (err < 0) ? err : 0;
2859         }
2860
2861         return 0;
2862 }
2863
2864 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2865 {
2866         struct symbol *sym = ms->sym;
2867         struct map *map = ms->map;
2868         struct inline_node *inline_node;
2869         struct inline_list *ilist;
2870         u64 addr;
2871         int ret = 1;
2872
2873         if (!symbol_conf.inline_name || !map || !sym)
2874                 return ret;
2875
2876         addr = map__map_ip(map, ip);
2877         addr = map__rip_2objdump(map, addr);
2878
2879         inline_node = inlines__tree_find(&map->dso->inlined_nodes, addr);
2880         if (!inline_node) {
2881                 inline_node = dso__parse_addr_inlines(map->dso, addr, sym);
2882                 if (!inline_node)
2883                         return ret;
2884                 inlines__tree_insert(&map->dso->inlined_nodes, inline_node);
2885         }
2886
2887         list_for_each_entry(ilist, &inline_node->val, list) {
2888                 struct map_symbol ilist_ms = {
2889                         .maps = ms->maps,
2890                         .map = map,
2891                         .sym = ilist->symbol,
2892                 };
2893                 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2894                                               NULL, 0, 0, 0, ilist->srcline);
2895
2896                 if (ret != 0)
2897                         return ret;
2898         }
2899
2900         return ret;
2901 }
2902
2903 static int unwind_entry(struct unwind_entry *entry, void *arg)
2904 {
2905         struct callchain_cursor *cursor = arg;
2906         const char *srcline = NULL;
2907         u64 addr = entry->ip;
2908
2909         if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2910                 return 0;
2911
2912         if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2913                 return 0;
2914
2915         /*
2916          * Convert entry->ip from a virtual address to an offset in
2917          * its corresponding binary.
2918          */
2919         if (entry->ms.map)
2920                 addr = map__map_ip(entry->ms.map, entry->ip);
2921
2922         srcline = callchain_srcline(&entry->ms, addr);
2923         return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2924                                        false, NULL, 0, 0, 0, srcline);
2925 }
2926
2927 static int thread__resolve_callchain_unwind(struct thread *thread,
2928                                             struct callchain_cursor *cursor,
2929                                             struct evsel *evsel,
2930                                             struct perf_sample *sample,
2931                                             int max_stack)
2932 {
2933         /* Can we do dwarf post unwind? */
2934         if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2935               (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2936                 return 0;
2937
2938         /* Bail out if nothing was captured. */
2939         if ((!sample->user_regs.regs) ||
2940             (!sample->user_stack.size))
2941                 return 0;
2942
2943         return unwind__get_entries(unwind_entry, cursor,
2944                                    thread, sample, max_stack);
2945 }
2946
2947 int thread__resolve_callchain(struct thread *thread,
2948                               struct callchain_cursor *cursor,
2949                               struct evsel *evsel,
2950                               struct perf_sample *sample,
2951                               struct symbol **parent,
2952                               struct addr_location *root_al,
2953                               int max_stack)
2954 {
2955         int ret = 0;
2956
2957         callchain_cursor_reset(cursor);
2958
2959         if (callchain_param.order == ORDER_CALLEE) {
2960                 ret = thread__resolve_callchain_sample(thread, cursor,
2961                                                        evsel, sample,
2962                                                        parent, root_al,
2963                                                        max_stack);
2964                 if (ret)
2965                         return ret;
2966                 ret = thread__resolve_callchain_unwind(thread, cursor,
2967                                                        evsel, sample,
2968                                                        max_stack);
2969         } else {
2970                 ret = thread__resolve_callchain_unwind(thread, cursor,
2971                                                        evsel, sample,
2972                                                        max_stack);
2973                 if (ret)
2974                         return ret;
2975                 ret = thread__resolve_callchain_sample(thread, cursor,
2976                                                        evsel, sample,
2977                                                        parent, root_al,
2978                                                        max_stack);
2979         }
2980
2981         return ret;
2982 }
2983
2984 int machine__for_each_thread(struct machine *machine,
2985                              int (*fn)(struct thread *thread, void *p),
2986                              void *priv)
2987 {
2988         struct threads *threads;
2989         struct rb_node *nd;
2990         struct thread *thread;
2991         int rc = 0;
2992         int i;
2993
2994         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
2995                 threads = &machine->threads[i];
2996                 for (nd = rb_first_cached(&threads->entries); nd;
2997                      nd = rb_next(nd)) {
2998                         thread = rb_entry(nd, struct thread, rb_node);
2999                         rc = fn(thread, priv);
3000                         if (rc != 0)
3001                                 return rc;
3002                 }
3003
3004                 list_for_each_entry(thread, &threads->dead, node) {
3005                         rc = fn(thread, priv);
3006                         if (rc != 0)
3007                                 return rc;
3008                 }
3009         }
3010         return rc;
3011 }
3012
3013 int machines__for_each_thread(struct machines *machines,
3014                               int (*fn)(struct thread *thread, void *p),
3015                               void *priv)
3016 {
3017         struct rb_node *nd;
3018         int rc = 0;
3019
3020         rc = machine__for_each_thread(&machines->host, fn, priv);
3021         if (rc != 0)
3022                 return rc;
3023
3024         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3025                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3026
3027                 rc = machine__for_each_thread(machine, fn, priv);
3028                 if (rc != 0)
3029                         return rc;
3030         }
3031         return rc;
3032 }
3033
3034 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3035 {
3036         int nr_cpus = min(machine->env->nr_cpus_avail, MAX_NR_CPUS);
3037
3038         if (cpu < 0 || cpu >= nr_cpus || !machine->current_tid)
3039                 return -1;
3040
3041         return machine->current_tid[cpu];
3042 }
3043
3044 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3045                              pid_t tid)
3046 {
3047         struct thread *thread;
3048         int nr_cpus = min(machine->env->nr_cpus_avail, MAX_NR_CPUS);
3049
3050         if (cpu < 0)
3051                 return -EINVAL;
3052
3053         if (!machine->current_tid) {
3054                 int i;
3055
3056                 machine->current_tid = calloc(nr_cpus, sizeof(pid_t));
3057                 if (!machine->current_tid)
3058                         return -ENOMEM;
3059                 for (i = 0; i < nr_cpus; i++)
3060                         machine->current_tid[i] = -1;
3061         }
3062
3063         if (cpu >= nr_cpus) {
3064                 pr_err("Requested CPU %d too large. ", cpu);
3065                 pr_err("Consider raising MAX_NR_CPUS\n");
3066                 return -EINVAL;
3067         }
3068
3069         machine->current_tid[cpu] = tid;
3070
3071         thread = machine__findnew_thread(machine, pid, tid);
3072         if (!thread)
3073                 return -ENOMEM;
3074
3075         thread->cpu = cpu;
3076         thread__put(thread);
3077
3078         return 0;
3079 }
3080
3081 /*
3082  * Compares the raw arch string. N.B. see instead perf_env__arch() if a
3083  * normalized arch is needed.
3084  */
3085 bool machine__is(struct machine *machine, const char *arch)
3086 {
3087         return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3088 }
3089
3090 int machine__nr_cpus_avail(struct machine *machine)
3091 {
3092         return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3093 }
3094
3095 int machine__get_kernel_start(struct machine *machine)
3096 {
3097         struct map *map = machine__kernel_map(machine);
3098         int err = 0;
3099
3100         /*
3101          * The only addresses above 2^63 are kernel addresses of a 64-bit
3102          * kernel.  Note that addresses are unsigned so that on a 32-bit system
3103          * all addresses including kernel addresses are less than 2^32.  In
3104          * that case (32-bit system), if the kernel mapping is unknown, all
3105          * addresses will be assumed to be in user space - see
3106          * machine__kernel_ip().
3107          */
3108         machine->kernel_start = 1ULL << 63;
3109         if (map) {
3110                 err = map__load(map);
3111                 /*
3112                  * On x86_64, PTI entry trampolines are less than the
3113                  * start of kernel text, but still above 2^63. So leave
3114                  * kernel_start = 1ULL << 63 for x86_64.
3115                  */
3116                 if (!err && !machine__is(machine, "x86_64"))
3117                         machine->kernel_start = map->start;
3118         }
3119         return err;
3120 }
3121
3122 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3123 {
3124         u8 addr_cpumode = cpumode;
3125         bool kernel_ip;
3126
3127         if (!machine->single_address_space)
3128                 goto out;
3129
3130         kernel_ip = machine__kernel_ip(machine, addr);
3131         switch (cpumode) {
3132         case PERF_RECORD_MISC_KERNEL:
3133         case PERF_RECORD_MISC_USER:
3134                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3135                                            PERF_RECORD_MISC_USER;
3136                 break;
3137         case PERF_RECORD_MISC_GUEST_KERNEL:
3138         case PERF_RECORD_MISC_GUEST_USER:
3139                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3140                                            PERF_RECORD_MISC_GUEST_USER;
3141                 break;
3142         default:
3143                 break;
3144         }
3145 out:
3146         return addr_cpumode;
3147 }
3148
3149 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3150 {
3151         return dsos__findnew_id(&machine->dsos, filename, id);
3152 }
3153
3154 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3155 {
3156         return machine__findnew_dso_id(machine, filename, NULL);
3157 }
3158
3159 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3160 {
3161         struct machine *machine = vmachine;
3162         struct map *map;
3163         struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3164
3165         if (sym == NULL)
3166                 return NULL;
3167
3168         *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
3169         *addrp = map->unmap_ip(map, sym->start);
3170         return sym->name;
3171 }
3172
3173 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3174 {
3175         struct dso *pos;
3176         int err = 0;
3177
3178         list_for_each_entry(pos, &machine->dsos.head, node) {
3179                 if (fn(pos, machine, priv))
3180                         err = -1;
3181         }
3182         return err;
3183 }