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_switch_event(struct machine *machine __maybe_unused,
 759                                   union perf_event *event)
 760 {
 761         if (dump_trace)
 762                 perf_event__fprintf_switch(event, stdout);
 763         return 0;
 764 }
 765
 766 static int machine__process_ksymbol_register(struct machine *machine,
 767                                              union perf_event *event,
 768                                              struct perf_sample *sample __maybe_unused)
 769 {
 770         struct symbol *sym;
 771         struct map *map = maps__find(&machine->kmaps, event->ksymbol.addr);
 772
 773         if (!map) {
 774                 struct dso *dso = dso__new(event->ksymbol.name);
 775
 776                 if (dso) {
 777                         dso->kernel = DSO_SPACE__KERNEL;
 778                         map = map__new2(0, dso);
 779                         dso__put(dso);
 780                 }
 781
 782                 if (!dso || !map) {
 783                         return -ENOMEM;
 784                 }
 785
 786                 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 787                         map->dso->binary_type = DSO_BINARY_TYPE__OOL;
 788                         map->dso->data.file_size = event->ksymbol.len;
 789                         dso__set_loaded(map->dso);
 790                 }
 791
 792                 map->start = event->ksymbol.addr;
 793                 map->end = map->start + event->ksymbol.len;
 794                 maps__insert(&machine->kmaps, map);
 795                 map__put(map);
 796                 dso__set_loaded(dso);
 797
 798                 if (is_bpf_image(event->ksymbol.name)) {
 799                         dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
 800                         dso__set_long_name(dso, "", false);
 801                 }
 802         }
 803
 804         sym = symbol__new(map->map_ip(map, map->start),
 805                           event->ksymbol.len,
 806                           0, 0, event->ksymbol.name);
 807         if (!sym)
 808                 return -ENOMEM;
 809         dso__insert_symbol(map->dso, sym);
 810         return 0;
 811 }
 812
 813 static int machine__process_ksymbol_unregister(struct machine *machine,
 814                                                union perf_event *event,
 815                                                struct perf_sample *sample __maybe_unused)
 816 {
 817         struct symbol *sym;
 818         struct map *map;
 819
 820         map = maps__find(&machine->kmaps, event->ksymbol.addr);
 821         if (!map)
 822                 return 0;
 823
 824         if (map != machine->vmlinux_map)
 825                 maps__remove(&machine->kmaps, map);
 826         else {
 827                 sym = dso__find_symbol(map->dso, map->map_ip(map, map->start));
 828                 if (sym)
 829                         dso__delete_symbol(map->dso, sym);
 830         }
 831
 832         return 0;
 833 }
 834
 835 int machine__process_ksymbol(struct machine *machine __maybe_unused,
 836                              union perf_event *event,
 837                              struct perf_sample *sample)
 838 {
 839         if (dump_trace)
 840                 perf_event__fprintf_ksymbol(event, stdout);
 841
 842         if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 843                 return machine__process_ksymbol_unregister(machine, event,
 844                                                            sample);
 845         return machine__process_ksymbol_register(machine, event, sample);
 846 }
 847
 848 int machine__process_text_poke(struct machine *machine, union perf_event *event,
 849                                struct perf_sample *sample __maybe_unused)
 850 {
 851         struct map *map = maps__find(&machine->kmaps, event->text_poke.addr);
 852         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 853
 854         if (dump_trace)
 855                 perf_event__fprintf_text_poke(event, machine, stdout);
 856
 857         if (!event->text_poke.new_len)
 858                 return 0;
 859
 860         if (cpumode != PERF_RECORD_MISC_KERNEL) {
 861                 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
 862                 return 0;
 863         }
 864
 865         if (map && map->dso) {
 866                 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
 867                 int ret;
 868
 869                 /*
 870                  * Kernel maps might be changed when loading symbols so loading
 871                  * must be done prior to using kernel maps.
 872                  */
 873                 map__load(map);
 874                 ret = dso__data_write_cache_addr(map->dso, map, machine,
 875                                                  event->text_poke.addr,
 876                                                  new_bytes,
 877                                                  event->text_poke.new_len);
 878                 if (ret != event->text_poke.new_len)
 879                         pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
 880                                  event->text_poke.addr);
 881         } else {
 882                 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
 883                          event->text_poke.addr);
 884         }
 885
 886         return 0;
 887 }
 888
 889 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
 890                                               const char *filename)
 891 {
 892         struct map *map = NULL;
 893         struct kmod_path m;
 894         struct dso *dso;
 895
 896         if (kmod_path__parse_name(&m, filename))
 897                 return NULL;
 898
 899         dso = machine__findnew_module_dso(machine, &m, filename);
 900         if (dso == NULL)
 901                 goto out;
 902
 903         map = map__new2(start, dso);
 904         if (map == NULL)
 905                 goto out;
 906
 907         maps__insert(&machine->kmaps, map);
 908
 909         /* Put the map here because maps__insert already got it */
 910         map__put(map);
 911 out:
 912         /* put the dso here, corresponding to  machine__findnew_module_dso */
 913         dso__put(dso);
 914         zfree(&m.name);
 915         return map;
 916 }
 917
 918 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
 919 {
 920         struct rb_node *nd;
 921         size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
 922
 923         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 924                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 925                 ret += __dsos__fprintf(&pos->dsos.head, fp);
 926         }
 927
 928         return ret;
 929 }
 930
 931 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
 932                                      bool (skip)(struct dso *dso, int parm), int parm)
 933 {
 934         return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
 935 }
 936
 937 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
 938                                      bool (skip)(struct dso *dso, int parm), int parm)
 939 {
 940         struct rb_node *nd;
 941         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
 942
 943         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 944                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 945                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
 946         }
 947         return ret;
 948 }
 949
 950 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
 951 {
 952         int i;
 953         size_t printed = 0;
 954         struct dso *kdso = machine__kernel_dso(machine);
 955
 956         if (kdso->has_build_id) {
 957                 char filename[PATH_MAX];
 958                 if (dso__build_id_filename(kdso, filename, sizeof(filename),
 959                                            false))
 960                         printed += fprintf(fp, "[0] %s\n", filename);
 961         }
 962
 963         for (i = 0; i < vmlinux_path__nr_entries; ++i)
 964                 printed += fprintf(fp, "[%d] %s\n",
 965                                    i + kdso->has_build_id, vmlinux_path[i]);
 966
 967         return printed;
 968 }
 969
 970 size_t machine__fprintf(struct machine *machine, FILE *fp)
 971 {
 972         struct rb_node *nd;
 973         size_t ret;
 974         int i;
 975
 976         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 977                 struct threads *threads = &machine->threads[i];
 978
 979                 down_read(&threads->lock);
 980
 981                 ret = fprintf(fp, "Threads: %u\n", threads->nr);
 982
 983                 for (nd = rb_first_cached(&threads->entries); nd;
 984                      nd = rb_next(nd)) {
 985                         struct thread *pos = rb_entry(nd, struct thread, rb_node);
 986
 987                         ret += thread__fprintf(pos, fp);
 988                 }
 989
 990                 up_read(&threads->lock);
 991         }
 992         return ret;
 993 }
 994
 995 static struct dso *machine__get_kernel(struct machine *machine)
 996 {
 997         const char *vmlinux_name = machine->mmap_name;
 998         struct dso *kernel;
 999
1000         if (machine__is_host(machine)) {
1001                 if (symbol_conf.vmlinux_name)
1002                         vmlinux_name = symbol_conf.vmlinux_name;
1003
1004                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1005                                                  "[kernel]", DSO_SPACE__KERNEL);
1006         } else {
1007                 if (symbol_conf.default_guest_vmlinux_name)
1008                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1009
1010                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1011                                                  "[guest.kernel]",
1012                                                  DSO_SPACE__KERNEL_GUEST);
1013         }
1014
1015         if (kernel != NULL && (!kernel->has_build_id))
1016                 dso__read_running_kernel_build_id(kernel, machine);
1017
1018         return kernel;
1019 }
1020
1021 struct process_args {
1022         u64 start;
1023 };
1024
1025 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1026                                     size_t bufsz)
1027 {
1028         if (machine__is_default_guest(machine))
1029                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1030         else
1031                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1032 }
1033
1034 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1035
1036 /* Figure out the start address of kernel map from /proc/kallsyms.
1037  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1038  * symbol_name if it's not that important.
1039  */
1040 static int machine__get_running_kernel_start(struct machine *machine,
1041                                              const char **symbol_name,
1042                                              u64 *start, u64 *end)
1043 {
1044         char filename[PATH_MAX];
1045         int i, err = -1;
1046         const char *name;
1047         u64 addr = 0;
1048
1049         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1050
1051         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1052                 return 0;
1053
1054         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1055                 err = kallsyms__get_function_start(filename, name, &addr);
1056                 if (!err)
1057                         break;
1058         }
1059
1060         if (err)
1061                 return -1;
1062
1063         if (symbol_name)
1064                 *symbol_name = name;
1065
1066         *start = addr;
1067
1068         err = kallsyms__get_function_start(filename, "_etext", &addr);
1069         if (!err)
1070                 *end = addr;
1071
1072         return 0;
1073 }
1074
1075 int machine__create_extra_kernel_map(struct machine *machine,
1076                                      struct dso *kernel,
1077                                      struct extra_kernel_map *xm)
1078 {
1079         struct kmap *kmap;
1080         struct map *map;
1081
1082         map = map__new2(xm->start, kernel);
1083         if (!map)
1084                 return -1;
1085
1086         map->end   = xm->end;
1087         map->pgoff = xm->pgoff;
1088
1089         kmap = map__kmap(map);
1090
1091         strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1092
1093         maps__insert(&machine->kmaps, map);
1094
1095         pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1096                   kmap->name, map->start, map->end);
1097
1098         map__put(map);
1099
1100         return 0;
1101 }
1102
1103 static u64 find_entry_trampoline(struct dso *dso)
1104 {
1105         /* Duplicates are removed so lookup all aliases */
1106         const char *syms[] = {
1107                 "_entry_trampoline",
1108                 "__entry_trampoline_start",
1109                 "entry_SYSCALL_64_trampoline",
1110         };
1111         struct symbol *sym = dso__first_symbol(dso);
1112         unsigned int i;
1113
1114         for (; sym; sym = dso__next_symbol(sym)) {
1115                 if (sym->binding != STB_GLOBAL)
1116                         continue;
1117                 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1118                         if (!strcmp(sym->name, syms[i]))
1119                                 return sym->start;
1120                 }
1121         }
1122
1123         return 0;
1124 }
1125
1126 /*
1127  * These values can be used for kernels that do not have symbols for the entry
1128  * trampolines in kallsyms.
1129  */
1130 #define X86_64_CPU_ENTRY_AREA_PER_CPU   0xfffffe0000000000ULL
1131 #define X86_64_CPU_ENTRY_AREA_SIZE      0x2c000
1132 #define X86_64_ENTRY_TRAMPOLINE         0x6000
1133
1134 /* Map x86_64 PTI entry trampolines */
1135 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1136                                           struct dso *kernel)
1137 {
1138         struct maps *kmaps = &machine->kmaps;
1139         int nr_cpus_avail, cpu;
1140         bool found = false;
1141         struct map *map;
1142         u64 pgoff;
1143
1144         /*
1145          * In the vmlinux case, pgoff is a virtual address which must now be
1146          * mapped to a vmlinux offset.
1147          */
1148         maps__for_each_entry(kmaps, map) {
1149                 struct kmap *kmap = __map__kmap(map);
1150                 struct map *dest_map;
1151
1152                 if (!kmap || !is_entry_trampoline(kmap->name))
1153                         continue;
1154
1155                 dest_map = maps__find(kmaps, map->pgoff);
1156                 if (dest_map != map)
1157                         map->pgoff = dest_map->map_ip(dest_map, map->pgoff);
1158                 found = true;
1159         }
1160         if (found || machine->trampolines_mapped)
1161                 return 0;
1162
1163         pgoff = find_entry_trampoline(kernel);
1164         if (!pgoff)
1165                 return 0;
1166
1167         nr_cpus_avail = machine__nr_cpus_avail(machine);
1168
1169         /* Add a 1 page map for each CPU's entry trampoline */
1170         for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1171                 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1172                          cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1173                          X86_64_ENTRY_TRAMPOLINE;
1174                 struct extra_kernel_map xm = {
1175                         .start = va,
1176                         .end   = va + page_size,
1177                         .pgoff = pgoff,
1178                 };
1179
1180                 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1181
1182                 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1183                         return -1;
1184         }
1185
1186         machine->trampolines_mapped = nr_cpus_avail;
1187
1188         return 0;
1189 }
1190
1191 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1192                                              struct dso *kernel __maybe_unused)
1193 {
1194         return 0;
1195 }
1196
1197 static int
1198 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1199 {
1200         /* In case of renewal the kernel map, destroy previous one */
1201         machine__destroy_kernel_maps(machine);
1202
1203         machine->vmlinux_map = map__new2(0, kernel);
1204         if (machine->vmlinux_map == NULL)
1205                 return -1;
1206
1207         machine->vmlinux_map->map_ip = machine->vmlinux_map->unmap_ip = identity__map_ip;
1208         maps__insert(&machine->kmaps, machine->vmlinux_map);
1209         return 0;
1210 }
1211
1212 void machine__destroy_kernel_maps(struct machine *machine)
1213 {
1214         struct kmap *kmap;
1215         struct map *map = machine__kernel_map(machine);
1216
1217         if (map == NULL)
1218                 return;
1219
1220         kmap = map__kmap(map);
1221         maps__remove(&machine->kmaps, map);
1222         if (kmap && kmap->ref_reloc_sym) {
1223                 zfree((char **)&kmap->ref_reloc_sym->name);
1224                 zfree(&kmap->ref_reloc_sym);
1225         }
1226
1227         map__zput(machine->vmlinux_map);
1228 }
1229
1230 int machines__create_guest_kernel_maps(struct machines *machines)
1231 {
1232         int ret = 0;
1233         struct dirent **namelist = NULL;
1234         int i, items = 0;
1235         char path[PATH_MAX];
1236         pid_t pid;
1237         char *endp;
1238
1239         if (symbol_conf.default_guest_vmlinux_name ||
1240             symbol_conf.default_guest_modules ||
1241             symbol_conf.default_guest_kallsyms) {
1242                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1243         }
1244
1245         if (symbol_conf.guestmount) {
1246                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1247                 if (items <= 0)
1248                         return -ENOENT;
1249                 for (i = 0; i < items; i++) {
1250                         if (!isdigit(namelist[i]->d_name[0])) {
1251                                 /* Filter out . and .. */
1252                                 continue;
1253                         }
1254                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1255                         if ((*endp != '\0') ||
1256                             (endp == namelist[i]->d_name) ||
1257                             (errno == ERANGE)) {
1258                                 pr_debug("invalid directory (%s). Skipping.\n",
1259                                          namelist[i]->d_name);
1260                                 continue;
1261                         }
1262                         sprintf(path, "%s/%s/proc/kallsyms",
1263                                 symbol_conf.guestmount,
1264                                 namelist[i]->d_name);
1265                         ret = access(path, R_OK);
1266                         if (ret) {
1267                                 pr_debug("Can't access file %s\n", path);
1268                                 goto failure;
1269                         }
1270                         machines__create_kernel_maps(machines, pid);
1271                 }
1272 failure:
1273                 free(namelist);
1274         }
1275
1276         return ret;
1277 }
1278
1279 void machines__destroy_kernel_maps(struct machines *machines)
1280 {
1281         struct rb_node *next = rb_first_cached(&machines->guests);
1282
1283         machine__destroy_kernel_maps(&machines->host);
1284
1285         while (next) {
1286                 struct machine *pos = rb_entry(next, struct machine, rb_node);
1287
1288                 next = rb_next(&pos->rb_node);
1289                 rb_erase_cached(&pos->rb_node, &machines->guests);
1290                 machine__delete(pos);
1291         }
1292 }
1293
1294 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1295 {
1296         struct machine *machine = machines__findnew(machines, pid);
1297
1298         if (machine == NULL)
1299                 return -1;
1300
1301         return machine__create_kernel_maps(machine);
1302 }
1303
1304 int machine__load_kallsyms(struct machine *machine, const char *filename)
1305 {
1306         struct map *map = machine__kernel_map(machine);
1307         int ret = __dso__load_kallsyms(map->dso, filename, map, true);
1308
1309         if (ret > 0) {
1310                 dso__set_loaded(map->dso);
1311                 /*
1312                  * Since /proc/kallsyms will have multiple sessions for the
1313                  * kernel, with modules between them, fixup the end of all
1314                  * sections.
1315                  */
1316                 maps__fixup_end(&machine->kmaps);
1317         }
1318
1319         return ret;
1320 }
1321
1322 int machine__load_vmlinux_path(struct machine *machine)
1323 {
1324         struct map *map = machine__kernel_map(machine);
1325         int ret = dso__load_vmlinux_path(map->dso, map);
1326
1327         if (ret > 0)
1328                 dso__set_loaded(map->dso);
1329
1330         return ret;
1331 }
1332
1333 static char *get_kernel_version(const char *root_dir)
1334 {
1335         char version[PATH_MAX];
1336         FILE *file;
1337         char *name, *tmp;
1338         const char *prefix = "Linux version ";
1339
1340         sprintf(version, "%s/proc/version", root_dir);
1341         file = fopen(version, "r");
1342         if (!file)
1343                 return NULL;
1344
1345         tmp = fgets(version, sizeof(version), file);
1346         fclose(file);
1347         if (!tmp)
1348                 return NULL;
1349
1350         name = strstr(version, prefix);
1351         if (!name)
1352                 return NULL;
1353         name += strlen(prefix);
1354         tmp = strchr(name, ' ');
1355         if (tmp)
1356                 *tmp = '\0';
1357
1358         return strdup(name);
1359 }
1360
1361 static bool is_kmod_dso(struct dso *dso)
1362 {
1363         return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1364                dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1365 }
1366
1367 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1368 {
1369         char *long_name;
1370         struct map *map = maps__find_by_name(maps, m->name);
1371
1372         if (map == NULL)
1373                 return 0;
1374
1375         long_name = strdup(path);
1376         if (long_name == NULL)
1377                 return -ENOMEM;
1378
1379         dso__set_long_name(map->dso, long_name, true);
1380         dso__kernel_module_get_build_id(map->dso, "");
1381
1382         /*
1383          * Full name could reveal us kmod compression, so
1384          * we need to update the symtab_type if needed.
1385          */
1386         if (m->comp && is_kmod_dso(map->dso)) {
1387                 map->dso->symtab_type++;
1388                 map->dso->comp = m->comp;
1389         }
1390
1391         return 0;
1392 }
1393
1394 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1395 {
1396         struct dirent *dent;
1397         DIR *dir = opendir(dir_name);
1398         int ret = 0;
1399
1400         if (!dir) {
1401                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1402                 return -1;
1403         }
1404
1405         while ((dent = readdir(dir)) != NULL) {
1406                 char path[PATH_MAX];
1407                 struct stat st;
1408
1409                 /*sshfs might return bad dent->d_type, so we have to stat*/
1410                 snprintf(path, sizeof(path), "%s/%s", dir_name, dent->d_name);
1411                 if (stat(path, &st))
1412                         continue;
1413
1414                 if (S_ISDIR(st.st_mode)) {
1415                         if (!strcmp(dent->d_name, ".") ||
1416                             !strcmp(dent->d_name, ".."))
1417                                 continue;
1418
1419                         /* Do not follow top-level source and build symlinks */
1420                         if (depth == 0) {
1421                                 if (!strcmp(dent->d_name, "source") ||
1422                                     !strcmp(dent->d_name, "build"))
1423                                         continue;
1424                         }
1425
1426                         ret = maps__set_modules_path_dir(maps, path, depth + 1);
1427                         if (ret < 0)
1428                                 goto out;
1429                 } else {
1430                         struct kmod_path m;
1431
1432                         ret = kmod_path__parse_name(&m, dent->d_name);
1433                         if (ret)
1434                                 goto out;
1435
1436                         if (m.kmod)
1437                                 ret = maps__set_module_path(maps, path, &m);
1438
1439                         zfree(&m.name);
1440
1441                         if (ret)
1442                                 goto out;
1443                 }
1444         }
1445
1446 out:
1447         closedir(dir);
1448         return ret;
1449 }
1450
1451 static int machine__set_modules_path(struct machine *machine)
1452 {
1453         char *version;
1454         char modules_path[PATH_MAX];
1455
1456         version = get_kernel_version(machine->root_dir);
1457         if (!version)
1458                 return -1;
1459
1460         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1461                  machine->root_dir, version);
1462         free(version);
1463
1464         return maps__set_modules_path_dir(&machine->kmaps, modules_path, 0);
1465 }
1466 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1467                                 u64 *size __maybe_unused,
1468                                 const char *name __maybe_unused)
1469 {
1470         return 0;
1471 }
1472
1473 static int machine__create_module(void *arg, const char *name, u64 start,
1474                                   u64 size)
1475 {
1476         struct machine *machine = arg;
1477         struct map *map;
1478
1479         if (arch__fix_module_text_start(&start, &size, name) < 0)
1480                 return -1;
1481
1482         map = machine__addnew_module_map(machine, start, name);
1483         if (map == NULL)
1484                 return -1;
1485         map->end = start + size;
1486
1487         dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1488
1489         return 0;
1490 }
1491
1492 static int machine__create_modules(struct machine *machine)
1493 {
1494         const char *modules;
1495         char path[PATH_MAX];
1496
1497         if (machine__is_default_guest(machine)) {
1498                 modules = symbol_conf.default_guest_modules;
1499         } else {
1500                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1501                 modules = path;
1502         }
1503
1504         if (symbol__restricted_filename(modules, "/proc/modules"))
1505                 return -1;
1506
1507         if (modules__parse(modules, machine, machine__create_module))
1508                 return -1;
1509
1510         if (!machine__set_modules_path(machine))
1511                 return 0;
1512
1513         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1514
1515         return 0;
1516 }
1517
1518 static void machine__set_kernel_mmap(struct machine *machine,
1519                                      u64 start, u64 end)
1520 {
1521         machine->vmlinux_map->start = start;
1522         machine->vmlinux_map->end   = end;
1523         /*
1524          * Be a bit paranoid here, some perf.data file came with
1525          * a zero sized synthesized MMAP event for the kernel.
1526          */
1527         if (start == 0 && end == 0)
1528                 machine->vmlinux_map->end = ~0ULL;
1529 }
1530
1531 static void machine__update_kernel_mmap(struct machine *machine,
1532                                      u64 start, u64 end)
1533 {
1534         struct map *map = machine__kernel_map(machine);
1535
1536         map__get(map);
1537         maps__remove(&machine->kmaps, map);
1538
1539         machine__set_kernel_mmap(machine, start, end);
1540
1541         maps__insert(&machine->kmaps, map);
1542         map__put(map);
1543 }
1544
1545 int machine__create_kernel_maps(struct machine *machine)
1546 {
1547         struct dso *kernel = machine__get_kernel(machine);
1548         const char *name = NULL;
1549         struct map *map;
1550         u64 start = 0, end = ~0ULL;
1551         int ret;
1552
1553         if (kernel == NULL)
1554                 return -1;
1555
1556         ret = __machine__create_kernel_maps(machine, kernel);
1557         if (ret < 0)
1558                 goto out_put;
1559
1560         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1561                 if (machine__is_host(machine))
1562                         pr_debug("Problems creating module maps, "
1563                                  "continuing anyway...\n");
1564                 else
1565                         pr_debug("Problems creating module maps for guest %d, "
1566                                  "continuing anyway...\n", machine->pid);
1567         }
1568
1569         if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1570                 if (name &&
1571                     map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1572                         machine__destroy_kernel_maps(machine);
1573                         ret = -1;
1574                         goto out_put;
1575                 }
1576
1577                 /*
1578                  * we have a real start address now, so re-order the kmaps
1579                  * assume it's the last in the kmaps
1580                  */
1581                 machine__update_kernel_mmap(machine, start, end);
1582         }
1583
1584         if (machine__create_extra_kernel_maps(machine, kernel))
1585                 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1586
1587         if (end == ~0ULL) {
1588                 /* update end address of the kernel map using adjacent module address */
1589                 map = map__next(machine__kernel_map(machine));
1590                 if (map)
1591                         machine__set_kernel_mmap(machine, start, map->start);
1592         }
1593
1594 out_put:
1595         dso__put(kernel);
1596         return ret;
1597 }
1598
1599 static bool machine__uses_kcore(struct machine *machine)
1600 {
1601         struct dso *dso;
1602
1603         list_for_each_entry(dso, &machine->dsos.head, node) {
1604                 if (dso__is_kcore(dso))
1605                         return true;
1606         }
1607
1608         return false;
1609 }
1610
1611 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1612                                              struct extra_kernel_map *xm)
1613 {
1614         return machine__is(machine, "x86_64") &&
1615                is_entry_trampoline(xm->name);
1616 }
1617
1618 static int machine__process_extra_kernel_map(struct machine *machine,
1619                                              struct extra_kernel_map *xm)
1620 {
1621         struct dso *kernel = machine__kernel_dso(machine);
1622
1623         if (kernel == NULL)
1624                 return -1;
1625
1626         return machine__create_extra_kernel_map(machine, kernel, xm);
1627 }
1628
1629 static int machine__process_kernel_mmap_event(struct machine *machine,
1630                                               struct extra_kernel_map *xm,
1631                                               struct build_id *bid)
1632 {
1633         struct map *map;
1634         enum dso_space_type dso_space;
1635         bool is_kernel_mmap;
1636
1637         /* If we have maps from kcore then we do not need or want any others */
1638         if (machine__uses_kcore(machine))
1639                 return 0;
1640
1641         if (machine__is_host(machine))
1642                 dso_space = DSO_SPACE__KERNEL;
1643         else
1644                 dso_space = DSO_SPACE__KERNEL_GUEST;
1645
1646         is_kernel_mmap = memcmp(xm->name, machine->mmap_name,
1647                                 strlen(machine->mmap_name) - 1) == 0;
1648         if (xm->name[0] == '/' ||
1649             (!is_kernel_mmap && xm->name[0] == '[')) {
1650                 map = machine__addnew_module_map(machine, xm->start,
1651                                                  xm->name);
1652                 if (map == NULL)
1653                         goto out_problem;
1654
1655                 map->end = map->start + xm->end - xm->start;
1656
1657                 if (build_id__is_defined(bid))
1658                         dso__set_build_id(map->dso, bid);
1659
1660         } else if (is_kernel_mmap) {
1661                 const char *symbol_name = (xm->name + strlen(machine->mmap_name));
1662                 /*
1663                  * Should be there already, from the build-id table in
1664                  * the header.
1665                  */
1666                 struct dso *kernel = NULL;
1667                 struct dso *dso;
1668
1669                 down_read(&machine->dsos.lock);
1670
1671                 list_for_each_entry(dso, &machine->dsos.head, node) {
1672
1673                         /*
1674                          * The cpumode passed to is_kernel_module is not the
1675                          * cpumode of *this* event. If we insist on passing
1676                          * correct cpumode to is_kernel_module, we should
1677                          * record the cpumode when we adding this dso to the
1678                          * linked list.
1679                          *
1680                          * However we don't really need passing correct
1681                          * cpumode.  We know the correct cpumode must be kernel
1682                          * mode (if not, we should not link it onto kernel_dsos
1683                          * list).
1684                          *
1685                          * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1686                          * is_kernel_module() treats it as a kernel cpumode.
1687                          */
1688
1689                         if (!dso->kernel ||
1690                             is_kernel_module(dso->long_name,
1691                                              PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1692                                 continue;
1693
1694
1695                         kernel = dso;
1696                         break;
1697                 }
1698
1699                 up_read(&machine->dsos.lock);
1700
1701                 if (kernel == NULL)
1702                         kernel = machine__findnew_dso(machine, machine->mmap_name);
1703                 if (kernel == NULL)
1704                         goto out_problem;
1705
1706                 kernel->kernel = dso_space;
1707                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1708                         dso__put(kernel);
1709                         goto out_problem;
1710                 }
1711
1712                 if (strstr(kernel->long_name, "vmlinux"))
1713                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1714
1715                 machine__update_kernel_mmap(machine, xm->start, xm->end);
1716
1717                 if (build_id__is_defined(bid))
1718                         dso__set_build_id(kernel, bid);
1719
1720                 /*
1721                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1722                  * symbol. Effectively having zero here means that at record
1723                  * time /proc/sys/kernel/kptr_restrict was non zero.
1724                  */
1725                 if (xm->pgoff != 0) {
1726                         map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1727                                                         symbol_name,
1728                                                         xm->pgoff);
1729                 }
1730
1731                 if (machine__is_default_guest(machine)) {
1732                         /*
1733                          * preload dso of guest kernel and modules
1734                          */
1735                         dso__load(kernel, machine__kernel_map(machine));
1736                 }
1737         } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1738                 return machine__process_extra_kernel_map(machine, xm);
1739         }
1740         return 0;
1741 out_problem:
1742         return -1;
1743 }
1744
1745 int machine__process_mmap2_event(struct machine *machine,
1746                                  union perf_event *event,
1747                                  struct perf_sample *sample)
1748 {
1749         struct thread *thread;
1750         struct map *map;
1751         struct dso_id dso_id = {
1752                 .maj = event->mmap2.maj,
1753                 .min = event->mmap2.min,
1754                 .ino = event->mmap2.ino,
1755                 .ino_generation = event->mmap2.ino_generation,
1756         };
1757         struct build_id __bid, *bid = NULL;
1758         int ret = 0;
1759
1760         if (dump_trace)
1761                 perf_event__fprintf_mmap2(event, stdout);
1762
1763         if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1764                 bid = &__bid;
1765                 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1766         }
1767
1768         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1769             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1770                 struct extra_kernel_map xm = {
1771                         .start = event->mmap2.start,
1772                         .end   = event->mmap2.start + event->mmap2.len,
1773                         .pgoff = event->mmap2.pgoff,
1774                 };
1775
1776                 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1777                 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1778                 if (ret < 0)
1779                         goto out_problem;
1780                 return 0;
1781         }
1782
1783         thread = machine__findnew_thread(machine, event->mmap2.pid,
1784                                         event->mmap2.tid);
1785         if (thread == NULL)
1786                 goto out_problem;
1787
1788         map = map__new(machine, event->mmap2.start,
1789                         event->mmap2.len, event->mmap2.pgoff,
1790                         &dso_id, event->mmap2.prot,
1791                         event->mmap2.flags, bid,
1792                         event->mmap2.filename, thread);
1793
1794         if (map == NULL)
1795                 goto out_problem_map;
1796
1797         ret = thread__insert_map(thread, map);
1798         if (ret)
1799                 goto out_problem_insert;
1800
1801         thread__put(thread);
1802         map__put(map);
1803         return 0;
1804
1805 out_problem_insert:
1806         map__put(map);
1807 out_problem_map:
1808         thread__put(thread);
1809 out_problem:
1810         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1811         return 0;
1812 }
1813
1814 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1815                                 struct perf_sample *sample)
1816 {
1817         struct thread *thread;
1818         struct map *map;
1819         u32 prot = 0;
1820         int ret = 0;
1821
1822         if (dump_trace)
1823                 perf_event__fprintf_mmap(event, stdout);
1824
1825         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1826             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1827                 struct extra_kernel_map xm = {
1828                         .start = event->mmap.start,
1829                         .end   = event->mmap.start + event->mmap.len,
1830                         .pgoff = event->mmap.pgoff,
1831                 };
1832
1833                 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1834                 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1835                 if (ret < 0)
1836                         goto out_problem;
1837                 return 0;
1838         }
1839
1840         thread = machine__findnew_thread(machine, event->mmap.pid,
1841                                          event->mmap.tid);
1842         if (thread == NULL)
1843                 goto out_problem;
1844
1845         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1846                 prot = PROT_EXEC;
1847
1848         map = map__new(machine, event->mmap.start,
1849                         event->mmap.len, event->mmap.pgoff,
1850                         NULL, prot, 0, NULL, event->mmap.filename, thread);
1851
1852         if (map == NULL)
1853                 goto out_problem_map;
1854
1855         ret = thread__insert_map(thread, map);
1856         if (ret)
1857                 goto out_problem_insert;
1858
1859         thread__put(thread);
1860         map__put(map);
1861         return 0;
1862
1863 out_problem_insert:
1864         map__put(map);
1865 out_problem_map:
1866         thread__put(thread);
1867 out_problem:
1868         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1869         return 0;
1870 }
1871
1872 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1873 {
1874         struct threads *threads = machine__threads(machine, th->tid);
1875
1876         if (threads->last_match == th)
1877                 threads__set_last_match(threads, NULL);
1878
1879         if (lock)
1880                 down_write(&threads->lock);
1881
1882         BUG_ON(refcount_read(&th->refcnt) == 0);
1883
1884         rb_erase_cached(&th->rb_node, &threads->entries);
1885         RB_CLEAR_NODE(&th->rb_node);
1886         --threads->nr;
1887         /*
1888          * Move it first to the dead_threads list, then drop the reference,
1889          * if this is the last reference, then the thread__delete destructor
1890          * will be called and we will remove it from the dead_threads list.
1891          */
1892         list_add_tail(&th->node, &threads->dead);
1893
1894         /*
1895          * We need to do the put here because if this is the last refcount,
1896          * then we will be touching the threads->dead head when removing the
1897          * thread.
1898          */
1899         thread__put(th);
1900
1901         if (lock)
1902                 up_write(&threads->lock);
1903 }
1904
1905 void machine__remove_thread(struct machine *machine, struct thread *th)
1906 {
1907         return __machine__remove_thread(machine, th, true);
1908 }
1909
1910 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1911                                 struct perf_sample *sample)
1912 {
1913         struct thread *thread = machine__find_thread(machine,
1914                                                      event->fork.pid,
1915                                                      event->fork.tid);
1916         struct thread *parent = machine__findnew_thread(machine,
1917                                                         event->fork.ppid,
1918                                                         event->fork.ptid);
1919         bool do_maps_clone = true;
1920         int err = 0;
1921
1922         if (dump_trace)
1923                 perf_event__fprintf_task(event, stdout);
1924
1925         /*
1926          * There may be an existing thread that is not actually the parent,
1927          * either because we are processing events out of order, or because the
1928          * (fork) event that would have removed the thread was lost. Assume the
1929          * latter case and continue on as best we can.
1930          */
1931         if (parent->pid_ != (pid_t)event->fork.ppid) {
1932                 dump_printf("removing erroneous parent thread %d/%d\n",
1933                             parent->pid_, parent->tid);
1934                 machine__remove_thread(machine, parent);
1935                 thread__put(parent);
1936                 parent = machine__findnew_thread(machine, event->fork.ppid,
1937                                                  event->fork.ptid);
1938         }
1939
1940         /* if a thread currently exists for the thread id remove it */
1941         if (thread != NULL) {
1942                 machine__remove_thread(machine, thread);
1943                 thread__put(thread);
1944         }
1945
1946         thread = machine__findnew_thread(machine, event->fork.pid,
1947                                          event->fork.tid);
1948         /*
1949          * When synthesizing FORK events, we are trying to create thread
1950          * objects for the already running tasks on the machine.
1951          *
1952          * Normally, for a kernel FORK event, we want to clone the parent's
1953          * maps because that is what the kernel just did.
1954          *
1955          * But when synthesizing, this should not be done.  If we do, we end up
1956          * with overlapping maps as we process the synthesized MMAP2 events that
1957          * get delivered shortly thereafter.
1958          *
1959          * Use the FORK event misc flags in an internal way to signal this
1960          * situation, so we can elide the map clone when appropriate.
1961          */
1962         if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1963                 do_maps_clone = false;
1964
1965         if (thread == NULL || parent == NULL ||
1966             thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1967                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1968                 err = -1;
1969         }
1970         thread__put(thread);
1971         thread__put(parent);
1972
1973         return err;
1974 }
1975
1976 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1977                                 struct perf_sample *sample __maybe_unused)
1978 {
1979         struct thread *thread = machine__find_thread(machine,
1980                                                      event->fork.pid,
1981                                                      event->fork.tid);
1982
1983         if (dump_trace)
1984                 perf_event__fprintf_task(event, stdout);
1985
1986         if (thread != NULL) {
1987                 thread__exited(thread);
1988                 thread__put(thread);
1989         }
1990
1991         return 0;
1992 }
1993
1994 int machine__process_event(struct machine *machine, union perf_event *event,
1995                            struct perf_sample *sample)
1996 {
1997         int ret;
1998
1999         switch (event->header.type) {
2000         case PERF_RECORD_COMM:
2001                 ret = machine__process_comm_event(machine, event, sample); break;
2002         case PERF_RECORD_MMAP:
2003                 ret = machine__process_mmap_event(machine, event, sample); break;
2004         case PERF_RECORD_NAMESPACES:
2005                 ret = machine__process_namespaces_event(machine, event, sample); break;
2006         case PERF_RECORD_CGROUP:
2007                 ret = machine__process_cgroup_event(machine, event, sample); break;
2008         case PERF_RECORD_MMAP2:
2009                 ret = machine__process_mmap2_event(machine, event, sample); break;
2010         case PERF_RECORD_FORK:
2011                 ret = machine__process_fork_event(machine, event, sample); break;
2012         case PERF_RECORD_EXIT:
2013                 ret = machine__process_exit_event(machine, event, sample); break;
2014         case PERF_RECORD_LOST:
2015                 ret = machine__process_lost_event(machine, event, sample); break;
2016         case PERF_RECORD_AUX:
2017                 ret = machine__process_aux_event(machine, event); break;
2018         case PERF_RECORD_ITRACE_START:
2019                 ret = machine__process_itrace_start_event(machine, event); break;
2020         case PERF_RECORD_LOST_SAMPLES:
2021                 ret = machine__process_lost_samples_event(machine, event, sample); break;
2022         case PERF_RECORD_SWITCH:
2023         case PERF_RECORD_SWITCH_CPU_WIDE:
2024                 ret = machine__process_switch_event(machine, event); break;
2025         case PERF_RECORD_KSYMBOL:
2026                 ret = machine__process_ksymbol(machine, event, sample); break;
2027         case PERF_RECORD_BPF_EVENT:
2028                 ret = machine__process_bpf(machine, event, sample); break;
2029         case PERF_RECORD_TEXT_POKE:
2030                 ret = machine__process_text_poke(machine, event, sample); break;
2031         default:
2032                 ret = -1;
2033                 break;
2034         }
2035
2036         return ret;
2037 }
2038
2039 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2040 {
2041         if (!regexec(regex, sym->name, 0, NULL, 0))
2042                 return true;
2043         return false;
2044 }
2045
2046 static void ip__resolve_ams(struct thread *thread,
2047                             struct addr_map_symbol *ams,
2048                             u64 ip)
2049 {
2050         struct addr_location al;
2051
2052         memset(&al, 0, sizeof(al));
2053         /*
2054          * We cannot use the header.misc hint to determine whether a
2055          * branch stack address is user, kernel, guest, hypervisor.
2056          * Branches may straddle the kernel/user/hypervisor boundaries.
2057          * Thus, we have to try consecutively until we find a match
2058          * or else, the symbol is unknown
2059          */
2060         thread__find_cpumode_addr_location(thread, ip, &al);
2061
2062         ams->addr = ip;
2063         ams->al_addr = al.addr;
2064         ams->ms.maps = al.maps;
2065         ams->ms.sym = al.sym;
2066         ams->ms.map = al.map;
2067         ams->phys_addr = 0;
2068         ams->data_page_size = 0;
2069 }
2070
2071 static void ip__resolve_data(struct thread *thread,
2072                              u8 m, struct addr_map_symbol *ams,
2073                              u64 addr, u64 phys_addr, u64 daddr_page_size)
2074 {
2075         struct addr_location al;
2076
2077         memset(&al, 0, sizeof(al));
2078
2079         thread__find_symbol(thread, m, addr, &al);
2080
2081         ams->addr = addr;
2082         ams->al_addr = al.addr;
2083         ams->ms.maps = al.maps;
2084         ams->ms.sym = al.sym;
2085         ams->ms.map = al.map;
2086         ams->phys_addr = phys_addr;
2087         ams->data_page_size = daddr_page_size;
2088 }
2089
2090 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2091                                      struct addr_location *al)
2092 {
2093         struct mem_info *mi = mem_info__new();
2094
2095         if (!mi)
2096                 return NULL;
2097
2098         ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2099         ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2100                          sample->addr, sample->phys_addr,
2101                          sample->data_page_size);
2102         mi->data_src.val = sample->data_src;
2103
2104         return mi;
2105 }
2106
2107 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2108 {
2109         struct map *map = ms->map;
2110         char *srcline = NULL;
2111
2112         if (!map || callchain_param.key == CCKEY_FUNCTION)
2113                 return srcline;
2114
2115         srcline = srcline__tree_find(&map->dso->srclines, ip);
2116         if (!srcline) {
2117                 bool show_sym = false;
2118                 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2119
2120                 srcline = get_srcline(map->dso, map__rip_2objdump(map, ip),
2121                                       ms->sym, show_sym, show_addr, ip);
2122                 srcline__tree_insert(&map->dso->srclines, ip, srcline);
2123         }
2124
2125         return srcline;
2126 }
2127
2128 struct iterations {
2129         int nr_loop_iter;
2130         u64 cycles;
2131 };
2132
2133 static int add_callchain_ip(struct thread *thread,
2134                             struct callchain_cursor *cursor,
2135                             struct symbol **parent,
2136                             struct addr_location *root_al,
2137                             u8 *cpumode,
2138                             u64 ip,
2139                             bool branch,
2140                             struct branch_flags *flags,
2141                             struct iterations *iter,
2142                             u64 branch_from)
2143 {
2144         struct map_symbol ms;
2145         struct addr_location al;
2146         int nr_loop_iter = 0;
2147         u64 iter_cycles = 0;
2148         const char *srcline = NULL;
2149
2150         al.filtered = 0;
2151         al.sym = NULL;
2152         if (!cpumode) {
2153                 thread__find_cpumode_addr_location(thread, ip, &al);
2154         } else {
2155                 if (ip >= PERF_CONTEXT_MAX) {
2156                         switch (ip) {
2157                         case PERF_CONTEXT_HV:
2158                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2159                                 break;
2160                         case PERF_CONTEXT_KERNEL:
2161                                 *cpumode = PERF_RECORD_MISC_KERNEL;
2162                                 break;
2163                         case PERF_CONTEXT_USER:
2164                                 *cpumode = PERF_RECORD_MISC_USER;
2165                                 break;
2166                         default:
2167                                 pr_debug("invalid callchain context: "
2168                                          "%"PRId64"\n", (s64) ip);
2169                                 /*
2170                                  * It seems the callchain is corrupted.
2171                                  * Discard all.
2172                                  */
2173                                 callchain_cursor_reset(cursor);
2174                                 return 1;
2175                         }
2176                         return 0;
2177                 }
2178                 thread__find_symbol(thread, *cpumode, ip, &al);
2179         }
2180
2181         if (al.sym != NULL) {
2182                 if (perf_hpp_list.parent && !*parent &&
2183                     symbol__match_regex(al.sym, &parent_regex))
2184                         *parent = al.sym;
2185                 else if (have_ignore_callees && root_al &&
2186                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2187                         /* Treat this symbol as the root,
2188                            forgetting its callees. */
2189                         *root_al = al;
2190                         callchain_cursor_reset(cursor);
2191                 }
2192         }
2193
2194         if (symbol_conf.hide_unresolved && al.sym == NULL)
2195                 return 0;
2196
2197         if (iter) {
2198                 nr_loop_iter = iter->nr_loop_iter;
2199                 iter_cycles = iter->cycles;
2200         }
2201
2202         ms.maps = al.maps;
2203         ms.map = al.map;
2204         ms.sym = al.sym;
2205         srcline = callchain_srcline(&ms, al.addr);
2206         return callchain_cursor_append(cursor, ip, &ms,
2207                                        branch, flags, nr_loop_iter,
2208                                        iter_cycles, branch_from, srcline);
2209 }
2210
2211 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2212                                            struct addr_location *al)
2213 {
2214         unsigned int i;
2215         const struct branch_stack *bs = sample->branch_stack;
2216         struct branch_entry *entries = perf_sample__branch_entries(sample);
2217         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2218
2219         if (!bi)
2220                 return NULL;
2221
2222         for (i = 0; i < bs->nr; i++) {
2223                 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2224                 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2225                 bi[i].flags = entries[i].flags;
2226         }
2227         return bi;
2228 }
2229
2230 static void save_iterations(struct iterations *iter,
2231                             struct branch_entry *be, int nr)
2232 {
2233         int i;
2234
2235         iter->nr_loop_iter++;
2236         iter->cycles = 0;
2237
2238         for (i = 0; i < nr; i++)
2239                 iter->cycles += be[i].flags.cycles;
2240 }
2241
2242 #define CHASHSZ 127
2243 #define CHASHBITS 7
2244 #define NO_ENTRY 0xff
2245
2246 #define PERF_MAX_BRANCH_DEPTH 127
2247
2248 /* Remove loops. */
2249 static int remove_loops(struct branch_entry *l, int nr,
2250                         struct iterations *iter)
2251 {
2252         int i, j, off;
2253         unsigned char chash[CHASHSZ];
2254
2255         memset(chash, NO_ENTRY, sizeof(chash));
2256
2257         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2258
2259         for (i = 0; i < nr; i++) {
2260                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2261
2262                 /* no collision handling for now */
2263                 if (chash[h] == NO_ENTRY) {
2264                         chash[h] = i;
2265                 } else if (l[chash[h]].from == l[i].from) {
2266                         bool is_loop = true;
2267                         /* check if it is a real loop */
2268                         off = 0;
2269                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
2270                                 if (l[j].from != l[i + off].from) {
2271                                         is_loop = false;
2272                                         break;
2273                                 }
2274                         if (is_loop) {
2275                                 j = nr - (i + off);
2276                                 if (j > 0) {
2277                                         save_iterations(iter + i + off,
2278                                                 l + i, off);
2279
2280                                         memmove(iter + i, iter + i + off,
2281                                                 j * sizeof(*iter));
2282
2283                                         memmove(l + i, l + i + off,
2284                                                 j * sizeof(*l));
2285                                 }
2286
2287                                 nr -= off;
2288                         }
2289                 }
2290         }
2291         return nr;
2292 }
2293
2294 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2295                                        struct callchain_cursor *cursor,
2296                                        struct perf_sample *sample,
2297                                        struct symbol **parent,
2298                                        struct addr_location *root_al,
2299                                        u64 branch_from,
2300                                        bool callee, int end)
2301 {
2302         struct ip_callchain *chain = sample->callchain;
2303         u8 cpumode = PERF_RECORD_MISC_USER;
2304         int err, i;
2305
2306         if (callee) {
2307                 for (i = 0; i < end + 1; i++) {
2308                         err = add_callchain_ip(thread, cursor, parent,
2309                                                root_al, &cpumode, chain->ips[i],
2310                                                false, NULL, NULL, branch_from);
2311                         if (err)
2312                                 return err;
2313                 }
2314                 return 0;
2315         }
2316
2317         for (i = end; i >= 0; i--) {
2318                 err = add_callchain_ip(thread, cursor, parent,
2319                                        root_al, &cpumode, chain->ips[i],
2320                                        false, NULL, NULL, branch_from);
2321                 if (err)
2322                         return err;
2323         }
2324
2325         return 0;
2326 }
2327
2328 static void save_lbr_cursor_node(struct thread *thread,
2329                                  struct callchain_cursor *cursor,
2330                                  int idx)
2331 {
2332         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2333
2334         if (!lbr_stitch)
2335                 return;
2336
2337         if (cursor->pos == cursor->nr) {
2338                 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2339                 return;
2340         }
2341
2342         if (!cursor->curr)
2343                 cursor->curr = cursor->first;
2344         else
2345                 cursor->curr = cursor->curr->next;
2346         memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2347                sizeof(struct callchain_cursor_node));
2348
2349         lbr_stitch->prev_lbr_cursor[idx].valid = true;
2350         cursor->pos++;
2351 }
2352
2353 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2354                                     struct callchain_cursor *cursor,
2355                                     struct perf_sample *sample,
2356                                     struct symbol **parent,
2357                                     struct addr_location *root_al,
2358                                     u64 *branch_from,
2359                                     bool callee)
2360 {
2361         struct branch_stack *lbr_stack = sample->branch_stack;
2362         struct branch_entry *entries = perf_sample__branch_entries(sample);
2363         u8 cpumode = PERF_RECORD_MISC_USER;
2364         int lbr_nr = lbr_stack->nr;
2365         struct branch_flags *flags;
2366         int err, i;
2367         u64 ip;
2368
2369         /*
2370          * The curr and pos are not used in writing session. They are cleared
2371          * in callchain_cursor_commit() when the writing session is closed.
2372          * Using curr and pos to track the current cursor node.
2373          */
2374         if (thread->lbr_stitch) {
2375                 cursor->curr = NULL;
2376                 cursor->pos = cursor->nr;
2377                 if (cursor->nr) {
2378                         cursor->curr = cursor->first;
2379                         for (i = 0; i < (int)(cursor->nr - 1); i++)
2380                                 cursor->curr = cursor->curr->next;
2381                 }
2382         }
2383
2384         if (callee) {
2385                 /* Add LBR ip from first entries.to */
2386                 ip = entries[0].to;
2387                 flags = &entries[0].flags;
2388                 *branch_from = entries[0].from;
2389                 err = add_callchain_ip(thread, cursor, parent,
2390                                        root_al, &cpumode, ip,
2391                                        true, flags, NULL,
2392                                        *branch_from);
2393                 if (err)
2394                         return err;
2395
2396                 /*
2397                  * The number of cursor node increases.
2398                  * Move the current cursor node.
2399                  * But does not need to save current cursor node for entry 0.
2400                  * It's impossible to stitch the whole LBRs of previous sample.
2401                  */
2402                 if (thread->lbr_stitch && (cursor->pos != cursor->nr)) {
2403                         if (!cursor->curr)
2404                                 cursor->curr = cursor->first;
2405                         else
2406                                 cursor->curr = cursor->curr->next;
2407                         cursor->pos++;
2408                 }
2409
2410                 /* Add LBR ip from entries.from one by one. */
2411                 for (i = 0; i < lbr_nr; i++) {
2412                         ip = entries[i].from;
2413                         flags = &entries[i].flags;
2414                         err = add_callchain_ip(thread, cursor, parent,
2415                                                root_al, &cpumode, ip,
2416                                                true, flags, NULL,
2417                                                *branch_from);
2418                         if (err)
2419                                 return err;
2420                         save_lbr_cursor_node(thread, cursor, i);
2421                 }
2422                 return 0;
2423         }
2424
2425         /* Add LBR ip from entries.from one by one. */
2426         for (i = lbr_nr - 1; i >= 0; i--) {
2427                 ip = entries[i].from;
2428                 flags = &entries[i].flags;
2429                 err = add_callchain_ip(thread, cursor, parent,
2430                                        root_al, &cpumode, ip,
2431                                        true, flags, NULL,
2432                                        *branch_from);
2433                 if (err)
2434                         return err;
2435                 save_lbr_cursor_node(thread, cursor, i);
2436         }
2437
2438         /* Add LBR ip from first entries.to */
2439         ip = entries[0].to;
2440         flags = &entries[0].flags;
2441         *branch_from = entries[0].from;
2442         err = add_callchain_ip(thread, cursor, parent,
2443                                root_al, &cpumode, ip,
2444                                true, flags, NULL,
2445                                *branch_from);
2446         if (err)
2447                 return err;
2448
2449         return 0;
2450 }
2451
2452 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2453                                              struct callchain_cursor *cursor)
2454 {
2455         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2456         struct callchain_cursor_node *cnode;
2457         struct stitch_list *stitch_node;
2458         int err;
2459
2460         list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2461                 cnode = &stitch_node->cursor;
2462
2463                 err = callchain_cursor_append(cursor, cnode->ip,
2464                                               &cnode->ms,
2465                                               cnode->branch,
2466                                               &cnode->branch_flags,
2467                                               cnode->nr_loop_iter,
2468                                               cnode->iter_cycles,
2469                                               cnode->branch_from,
2470                                               cnode->srcline);
2471                 if (err)
2472                         return err;
2473         }
2474         return 0;
2475 }
2476
2477 static struct stitch_list *get_stitch_node(struct thread *thread)
2478 {
2479         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2480         struct stitch_list *stitch_node;
2481
2482         if (!list_empty(&lbr_stitch->free_lists)) {
2483                 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2484                                                struct stitch_list, node);
2485                 list_del(&stitch_node->node);
2486
2487                 return stitch_node;
2488         }
2489
2490         return malloc(sizeof(struct stitch_list));
2491 }
2492
2493 static bool has_stitched_lbr(struct thread *thread,
2494                              struct perf_sample *cur,
2495                              struct perf_sample *prev,
2496                              unsigned int max_lbr,
2497                              bool callee)
2498 {
2499         struct branch_stack *cur_stack = cur->branch_stack;
2500         struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2501         struct branch_stack *prev_stack = prev->branch_stack;
2502         struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2503         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2504         int i, j, nr_identical_branches = 0;
2505         struct stitch_list *stitch_node;
2506         u64 cur_base, distance;
2507
2508         if (!cur_stack || !prev_stack)
2509                 return false;
2510
2511         /* Find the physical index of the base-of-stack for current sample. */
2512         cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2513
2514         distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2515                                                      (max_lbr + prev_stack->hw_idx - cur_base);
2516         /* Previous sample has shorter stack. Nothing can be stitched. */
2517         if (distance + 1 > prev_stack->nr)
2518                 return false;
2519
2520         /*
2521          * Check if there are identical LBRs between two samples.
2522          * Identical LBRs must have same from, to and flags values. Also,
2523          * they have to be saved in the same LBR registers (same physical
2524          * index).
2525          *
2526          * Starts from the base-of-stack of current sample.
2527          */
2528         for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2529                 if ((prev_entries[i].from != cur_entries[j].from) ||
2530                     (prev_entries[i].to != cur_entries[j].to) ||
2531                     (prev_entries[i].flags.value != cur_entries[j].flags.value))
2532                         break;
2533                 nr_identical_branches++;
2534         }
2535
2536         if (!nr_identical_branches)
2537                 return false;
2538
2539         /*
2540          * Save the LBRs between the base-of-stack of previous sample
2541          * and the base-of-stack of current sample into lbr_stitch->lists.
2542          * These LBRs will be stitched later.
2543          */
2544         for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2545
2546                 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2547                         continue;
2548
2549                 stitch_node = get_stitch_node(thread);
2550                 if (!stitch_node)
2551                         return false;
2552
2553                 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2554                        sizeof(struct callchain_cursor_node));
2555
2556                 if (callee)
2557                         list_add(&stitch_node->node, &lbr_stitch->lists);
2558                 else
2559                         list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2560         }
2561
2562         return true;
2563 }
2564
2565 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2566 {
2567         if (thread->lbr_stitch)
2568                 return true;
2569
2570         thread->lbr_stitch = zalloc(sizeof(*thread->lbr_stitch));
2571         if (!thread->lbr_stitch)
2572                 goto err;
2573
2574         thread->lbr_stitch->prev_lbr_cursor = calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2575         if (!thread->lbr_stitch->prev_lbr_cursor)
2576                 goto free_lbr_stitch;
2577
2578         INIT_LIST_HEAD(&thread->lbr_stitch->lists);
2579         INIT_LIST_HEAD(&thread->lbr_stitch->free_lists);
2580
2581         return true;
2582
2583 free_lbr_stitch:
2584         zfree(&thread->lbr_stitch);
2585 err:
2586         pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2587         thread->lbr_stitch_enable = false;
2588         return false;
2589 }
2590
2591 /*
2592  * Resolve LBR callstack chain sample
2593  * Return:
2594  * 1 on success get LBR callchain information
2595  * 0 no available LBR callchain information, should try fp
2596  * negative error code on other errors.
2597  */
2598 static int resolve_lbr_callchain_sample(struct thread *thread,
2599                                         struct callchain_cursor *cursor,
2600                                         struct perf_sample *sample,
2601                                         struct symbol **parent,
2602                                         struct addr_location *root_al,
2603                                         int max_stack,
2604                                         unsigned int max_lbr)
2605 {
2606         bool callee = (callchain_param.order == ORDER_CALLEE);
2607         struct ip_callchain *chain = sample->callchain;
2608         int chain_nr = min(max_stack, (int)chain->nr), i;
2609         struct lbr_stitch *lbr_stitch;
2610         bool stitched_lbr = false;
2611         u64 branch_from = 0;
2612         int err;
2613
2614         for (i = 0; i < chain_nr; i++) {
2615                 if (chain->ips[i] == PERF_CONTEXT_USER)
2616                         break;
2617         }
2618
2619         /* LBR only affects the user callchain */
2620         if (i == chain_nr)
2621                 return 0;
2622
2623         if (thread->lbr_stitch_enable && !sample->no_hw_idx &&
2624             (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2625                 lbr_stitch = thread->lbr_stitch;
2626
2627                 stitched_lbr = has_stitched_lbr(thread, sample,
2628                                                 &lbr_stitch->prev_sample,
2629                                                 max_lbr, callee);
2630
2631                 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2632                         list_replace_init(&lbr_stitch->lists,
2633                                           &lbr_stitch->free_lists);
2634                 }
2635                 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2636         }
2637
2638         if (callee) {
2639                 /* Add kernel ip */
2640                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2641                                                   parent, root_al, branch_from,
2642                                                   true, i);
2643                 if (err)
2644                         goto error;
2645
2646                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2647                                                root_al, &branch_from, true);
2648                 if (err)
2649                         goto error;
2650
2651                 if (stitched_lbr) {
2652                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2653                         if (err)
2654                                 goto error;
2655                 }
2656
2657         } else {
2658                 if (stitched_lbr) {
2659                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2660                         if (err)
2661                                 goto error;
2662                 }
2663                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2664                                                root_al, &branch_from, false);
2665                 if (err)
2666                         goto error;
2667
2668                 /* Add kernel ip */
2669                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2670                                                   parent, root_al, branch_from,
2671                                                   false, i);
2672                 if (err)
2673                         goto error;
2674         }
2675         return 1;
2676
2677 error:
2678         return (err < 0) ? err : 0;
2679 }
2680
2681 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2682                              struct callchain_cursor *cursor,
2683                              struct symbol **parent,
2684                              struct addr_location *root_al,
2685                              u8 *cpumode, int ent)
2686 {
2687         int err = 0;
2688
2689         while (--ent >= 0) {
2690                 u64 ip = chain->ips[ent];
2691
2692                 if (ip >= PERF_CONTEXT_MAX) {
2693                         err = add_callchain_ip(thread, cursor, parent,
2694                                                root_al, cpumode, ip,
2695                                                false, NULL, NULL, 0);
2696                         break;
2697                 }
2698         }
2699         return err;
2700 }
2701
2702 static int thread__resolve_callchain_sample(struct thread *thread,
2703                                             struct callchain_cursor *cursor,
2704                                             struct evsel *evsel,
2705                                             struct perf_sample *sample,
2706                                             struct symbol **parent,
2707                                             struct addr_location *root_al,
2708                                             int max_stack)
2709 {
2710         struct branch_stack *branch = sample->branch_stack;
2711         struct branch_entry *entries = perf_sample__branch_entries(sample);
2712         struct ip_callchain *chain = sample->callchain;
2713         int chain_nr = 0;
2714         u8 cpumode = PERF_RECORD_MISC_USER;
2715         int i, j, err, nr_entries;
2716         int skip_idx = -1;
2717         int first_call = 0;
2718
2719         if (chain)
2720                 chain_nr = chain->nr;
2721
2722         if (evsel__has_branch_callstack(evsel)) {
2723                 struct perf_env *env = evsel__env(evsel);
2724
2725                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2726                                                    root_al, max_stack,
2727                                                    !env ? 0 : env->max_branches);
2728                 if (err)
2729                         return (err < 0) ? err : 0;
2730         }
2731
2732         /*
2733          * Based on DWARF debug information, some architectures skip
2734          * a callchain entry saved by the kernel.
2735          */
2736         skip_idx = arch_skip_callchain_idx(thread, chain);
2737
2738         /*
2739          * Add branches to call stack for easier browsing. This gives
2740          * more context for a sample than just the callers.
2741          *
2742          * This uses individual histograms of paths compared to the
2743          * aggregated histograms the normal LBR mode uses.
2744          *
2745          * Limitations for now:
2746          * - No extra filters
2747          * - No annotations (should annotate somehow)
2748          */
2749
2750         if (branch && callchain_param.branch_callstack) {
2751                 int nr = min(max_stack, (int)branch->nr);
2752                 struct branch_entry be[nr];
2753                 struct iterations iter[nr];
2754
2755                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2756                         pr_warning("corrupted branch chain. skipping...\n");
2757                         goto check_calls;
2758                 }
2759
2760                 for (i = 0; i < nr; i++) {
2761                         if (callchain_param.order == ORDER_CALLEE) {
2762                                 be[i] = entries[i];
2763
2764                                 if (chain == NULL)
2765                                         continue;
2766
2767                                 /*
2768                                  * Check for overlap into the callchain.
2769                                  * The return address is one off compared to
2770                                  * the branch entry. To adjust for this
2771                                  * assume the calling instruction is not longer
2772                                  * than 8 bytes.
2773                                  */
2774                                 if (i == skip_idx ||
2775                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
2776                                         first_call++;
2777                                 else if (be[i].from < chain->ips[first_call] &&
2778                                     be[i].from >= chain->ips[first_call] - 8)
2779                                         first_call++;
2780                         } else
2781                                 be[i] = entries[branch->nr - i - 1];
2782                 }
2783
2784                 memset(iter, 0, sizeof(struct iterations) * nr);
2785                 nr = remove_loops(be, nr, iter);
2786
2787                 for (i = 0; i < nr; i++) {
2788                         err = add_callchain_ip(thread, cursor, parent,
2789                                                root_al,
2790                                                NULL, be[i].to,
2791                                                true, &be[i].flags,
2792                                                NULL, be[i].from);
2793
2794                         if (!err)
2795                                 err = add_callchain_ip(thread, cursor, parent, root_al,
2796                                                        NULL, be[i].from,
2797                                                        true, &be[i].flags,
2798                                                        &iter[i], 0);
2799                         if (err == -EINVAL)
2800                                 break;
2801                         if (err)
2802                                 return err;
2803                 }
2804
2805                 if (chain_nr == 0)
2806                         return 0;
2807
2808                 chain_nr -= nr;
2809         }
2810
2811 check_calls:
2812         if (chain && callchain_param.order != ORDER_CALLEE) {
2813                 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2814                                         &cpumode, chain->nr - first_call);
2815                 if (err)
2816                         return (err < 0) ? err : 0;
2817         }
2818         for (i = first_call, nr_entries = 0;
2819              i < chain_nr && nr_entries < max_stack; i++) {
2820                 u64 ip;
2821
2822                 if (callchain_param.order == ORDER_CALLEE)
2823                         j = i;
2824                 else
2825                         j = chain->nr - i - 1;
2826
2827 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2828                 if (j == skip_idx)
2829                         continue;
2830 #endif
2831                 ip = chain->ips[j];
2832                 if (ip < PERF_CONTEXT_MAX)
2833                        ++nr_entries;
2834                 else if (callchain_param.order != ORDER_CALLEE) {
2835                         err = find_prev_cpumode(chain, thread, cursor, parent,
2836                                                 root_al, &cpumode, j);
2837                         if (err)
2838                                 return (err < 0) ? err : 0;
2839                         continue;
2840                 }
2841
2842                 err = add_callchain_ip(thread, cursor, parent,
2843                                        root_al, &cpumode, ip,
2844                                        false, NULL, NULL, 0);
2845
2846                 if (err)
2847                         return (err < 0) ? err : 0;
2848         }
2849
2850         return 0;
2851 }
2852
2853 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2854 {
2855         struct symbol *sym = ms->sym;
2856         struct map *map = ms->map;
2857         struct inline_node *inline_node;
2858         struct inline_list *ilist;
2859         u64 addr;
2860         int ret = 1;
2861
2862         if (!symbol_conf.inline_name || !map || !sym)
2863                 return ret;
2864
2865         addr = map__map_ip(map, ip);
2866         addr = map__rip_2objdump(map, addr);
2867
2868         inline_node = inlines__tree_find(&map->dso->inlined_nodes, addr);
2869         if (!inline_node) {
2870                 inline_node = dso__parse_addr_inlines(map->dso, addr, sym);
2871                 if (!inline_node)
2872                         return ret;
2873                 inlines__tree_insert(&map->dso->inlined_nodes, inline_node);
2874         }
2875
2876         list_for_each_entry(ilist, &inline_node->val, list) {
2877                 struct map_symbol ilist_ms = {
2878                         .maps = ms->maps,
2879                         .map = map,
2880                         .sym = ilist->symbol,
2881                 };
2882                 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2883                                               NULL, 0, 0, 0, ilist->srcline);
2884
2885                 if (ret != 0)
2886                         return ret;
2887         }
2888
2889         return ret;
2890 }
2891
2892 static int unwind_entry(struct unwind_entry *entry, void *arg)
2893 {
2894         struct callchain_cursor *cursor = arg;
2895         const char *srcline = NULL;
2896         u64 addr = entry->ip;
2897
2898         if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2899                 return 0;
2900
2901         if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2902                 return 0;
2903
2904         /*
2905          * Convert entry->ip from a virtual address to an offset in
2906          * its corresponding binary.
2907          */
2908         if (entry->ms.map)
2909                 addr = map__map_ip(entry->ms.map, entry->ip);
2910
2911         srcline = callchain_srcline(&entry->ms, addr);
2912         return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2913                                        false, NULL, 0, 0, 0, srcline);
2914 }
2915
2916 static int thread__resolve_callchain_unwind(struct thread *thread,
2917                                             struct callchain_cursor *cursor,
2918                                             struct evsel *evsel,
2919                                             struct perf_sample *sample,
2920                                             int max_stack)
2921 {
2922         /* Can we do dwarf post unwind? */
2923         if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2924               (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2925                 return 0;
2926
2927         /* Bail out if nothing was captured. */
2928         if ((!sample->user_regs.regs) ||
2929             (!sample->user_stack.size))
2930                 return 0;
2931
2932         return unwind__get_entries(unwind_entry, cursor,
2933                                    thread, sample, max_stack);
2934 }
2935
2936 int thread__resolve_callchain(struct thread *thread,
2937                               struct callchain_cursor *cursor,
2938                               struct evsel *evsel,
2939                               struct perf_sample *sample,
2940                               struct symbol **parent,
2941                               struct addr_location *root_al,
2942                               int max_stack)
2943 {
2944         int ret = 0;
2945
2946         callchain_cursor_reset(cursor);
2947
2948         if (callchain_param.order == ORDER_CALLEE) {
2949                 ret = thread__resolve_callchain_sample(thread, cursor,
2950                                                        evsel, sample,
2951                                                        parent, root_al,
2952                                                        max_stack);
2953                 if (ret)
2954                         return ret;
2955                 ret = thread__resolve_callchain_unwind(thread, cursor,
2956                                                        evsel, sample,
2957                                                        max_stack);
2958         } else {
2959                 ret = thread__resolve_callchain_unwind(thread, cursor,
2960                                                        evsel, sample,
2961                                                        max_stack);
2962                 if (ret)
2963                         return ret;
2964                 ret = thread__resolve_callchain_sample(thread, cursor,
2965                                                        evsel, sample,
2966                                                        parent, root_al,
2967                                                        max_stack);
2968         }
2969
2970         return ret;
2971 }
2972
2973 int machine__for_each_thread(struct machine *machine,
2974                              int (*fn)(struct thread *thread, void *p),
2975                              void *priv)
2976 {
2977         struct threads *threads;
2978         struct rb_node *nd;
2979         struct thread *thread;
2980         int rc = 0;
2981         int i;
2982
2983         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
2984                 threads = &machine->threads[i];
2985                 for (nd = rb_first_cached(&threads->entries); nd;
2986                      nd = rb_next(nd)) {
2987                         thread = rb_entry(nd, struct thread, rb_node);
2988                         rc = fn(thread, priv);
2989                         if (rc != 0)
2990                                 return rc;
2991                 }
2992
2993                 list_for_each_entry(thread, &threads->dead, node) {
2994                         rc = fn(thread, priv);
2995                         if (rc != 0)
2996                                 return rc;
2997                 }
2998         }
2999         return rc;
3000 }
3001
3002 int machines__for_each_thread(struct machines *machines,
3003                               int (*fn)(struct thread *thread, void *p),
3004                               void *priv)
3005 {
3006         struct rb_node *nd;
3007         int rc = 0;
3008
3009         rc = machine__for_each_thread(&machines->host, fn, priv);
3010         if (rc != 0)
3011                 return rc;
3012
3013         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3014                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3015
3016                 rc = machine__for_each_thread(machine, fn, priv);
3017                 if (rc != 0)
3018                         return rc;
3019         }
3020         return rc;
3021 }
3022
3023 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3024 {
3025         int nr_cpus = min(machine->env->nr_cpus_avail, MAX_NR_CPUS);
3026
3027         if (cpu < 0 || cpu >= nr_cpus || !machine->current_tid)
3028                 return -1;
3029
3030         return machine->current_tid[cpu];
3031 }
3032
3033 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3034                              pid_t tid)
3035 {
3036         struct thread *thread;
3037         int nr_cpus = min(machine->env->nr_cpus_avail, MAX_NR_CPUS);
3038
3039         if (cpu < 0)
3040                 return -EINVAL;
3041
3042         if (!machine->current_tid) {
3043                 int i;
3044
3045                 machine->current_tid = calloc(nr_cpus, sizeof(pid_t));
3046                 if (!machine->current_tid)
3047                         return -ENOMEM;
3048                 for (i = 0; i < nr_cpus; i++)
3049                         machine->current_tid[i] = -1;
3050         }
3051
3052         if (cpu >= nr_cpus) {
3053                 pr_err("Requested CPU %d too large. ", cpu);
3054                 pr_err("Consider raising MAX_NR_CPUS\n");
3055                 return -EINVAL;
3056         }
3057
3058         machine->current_tid[cpu] = tid;
3059
3060         thread = machine__findnew_thread(machine, pid, tid);
3061         if (!thread)
3062                 return -ENOMEM;
3063
3064         thread->cpu = cpu;
3065         thread__put(thread);
3066
3067         return 0;
3068 }
3069
3070 /*
3071  * Compares the raw arch string. N.B. see instead perf_env__arch() if a
3072  * normalized arch is needed.
3073  */
3074 bool machine__is(struct machine *machine, const char *arch)
3075 {
3076         return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3077 }
3078
3079 int machine__nr_cpus_avail(struct machine *machine)
3080 {
3081         return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3082 }
3083
3084 int machine__get_kernel_start(struct machine *machine)
3085 {
3086         struct map *map = machine__kernel_map(machine);
3087         int err = 0;
3088
3089         /*
3090          * The only addresses above 2^63 are kernel addresses of a 64-bit
3091          * kernel.  Note that addresses are unsigned so that on a 32-bit system
3092          * all addresses including kernel addresses are less than 2^32.  In
3093          * that case (32-bit system), if the kernel mapping is unknown, all
3094          * addresses will be assumed to be in user space - see
3095          * machine__kernel_ip().
3096          */
3097         machine->kernel_start = 1ULL << 63;
3098         if (map) {
3099                 err = map__load(map);
3100                 /*
3101                  * On x86_64, PTI entry trampolines are less than the
3102                  * start of kernel text, but still above 2^63. So leave
3103                  * kernel_start = 1ULL << 63 for x86_64.
3104                  */
3105                 if (!err && !machine__is(machine, "x86_64"))
3106                         machine->kernel_start = map->start;
3107         }
3108         return err;
3109 }
3110
3111 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3112 {
3113         u8 addr_cpumode = cpumode;
3114         bool kernel_ip;
3115
3116         if (!machine->single_address_space)
3117                 goto out;
3118
3119         kernel_ip = machine__kernel_ip(machine, addr);
3120         switch (cpumode) {
3121         case PERF_RECORD_MISC_KERNEL:
3122         case PERF_RECORD_MISC_USER:
3123                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3124                                            PERF_RECORD_MISC_USER;
3125                 break;
3126         case PERF_RECORD_MISC_GUEST_KERNEL:
3127         case PERF_RECORD_MISC_GUEST_USER:
3128                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3129                                            PERF_RECORD_MISC_GUEST_USER;
3130                 break;
3131         default:
3132                 break;
3133         }
3134 out:
3135         return addr_cpumode;
3136 }
3137
3138 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3139 {
3140         return dsos__findnew_id(&machine->dsos, filename, id);
3141 }
3142
3143 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3144 {
3145         return machine__findnew_dso_id(machine, filename, NULL);
3146 }
3147
3148 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3149 {
3150         struct machine *machine = vmachine;
3151         struct map *map;
3152         struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3153
3154         if (sym == NULL)
3155                 return NULL;
3156
3157         *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
3158         *addrp = map->unmap_ip(map, sym->start);
3159         return sym->name;
3160 }
3161
3162 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3163 {
3164         struct dso *pos;
3165         int err = 0;
3166
3167         list_for_each_entry(pos, &machine->dsos.head, node) {
3168                 if (fn(pos, machine, priv))
3169                         err = -1;
3170         }
3171         return err;
3172 }