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 "mem-info.h"
  20 #include "path.h"
  21 #include "srcline.h"
  22 #include "symbol.h"
  23 #include "sort.h"
  24 #include "strlist.h"
  25 #include "target.h"
  26 #include "thread.h"
  27 #include "util.h"
  28 #include "vdso.h"
  29 #include <stdbool.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <unistd.h>
  33 #include "unwind.h"
  34 #include "linux/hash.h"
  35 #include "asm/bug.h"
  36 #include "bpf-event.h"
  37 #include <internal/lib.h> // page_size
  38 #include "cgroup.h"
  39 #include "arm64-frame-pointer-unwind-support.h"
  40
  41 #include <linux/ctype.h>
  42 #include <symbol/kallsyms.h>
  43 #include <linux/mman.h>
  44 #include <linux/string.h>
  45 #include <linux/zalloc.h>
  46
  47 static struct dso *machine__kernel_dso(struct machine *machine)
  48 {
  49         return map__dso(machine->vmlinux_map);
  50 }
  51
  52 static int machine__set_mmap_name(struct machine *machine)
  53 {
  54         if (machine__is_host(machine))
  55                 machine->mmap_name = strdup("[kernel.kallsyms]");
  56         else if (machine__is_default_guest(machine))
  57                 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
  58         else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
  59                           machine->pid) < 0)
  60                 machine->mmap_name = NULL;
  61
  62         return machine->mmap_name ? 0 : -ENOMEM;
  63 }
  64
  65 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
  66 {
  67         char comm[64];
  68
  69         snprintf(comm, sizeof(comm), "[guest/%d]", pid);
  70         thread__set_comm(thread, comm, 0);
  71 }
  72
  73 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
  74 {
  75         int err = -ENOMEM;
  76
  77         memset(machine, 0, sizeof(*machine));
  78         machine->kmaps = maps__new(machine);
  79         if (machine->kmaps == NULL)
  80                 return -ENOMEM;
  81
  82         RB_CLEAR_NODE(&machine->rb_node);
  83         dsos__init(&machine->dsos);
  84
  85         threads__init(&machine->threads);
  86
  87         machine->vdso_info = NULL;
  88         machine->env = NULL;
  89
  90         machine->pid = pid;
  91
  92         machine->id_hdr_size = 0;
  93         machine->kptr_restrict_warned = false;
  94         machine->comm_exec = false;
  95         machine->kernel_start = 0;
  96         machine->vmlinux_map = NULL;
  97
  98         machine->root_dir = strdup(root_dir);
  99         if (machine->root_dir == NULL)
 100                 goto out;
 101
 102         if (machine__set_mmap_name(machine))
 103                 goto out;
 104
 105         if (pid != HOST_KERNEL_ID) {
 106                 struct thread *thread = machine__findnew_thread(machine, -1,
 107                                                                 pid);
 108
 109                 if (thread == NULL)
 110                         goto out;
 111
 112                 thread__set_guest_comm(thread, pid);
 113                 thread__put(thread);
 114         }
 115
 116         machine->current_tid = NULL;
 117         err = 0;
 118
 119 out:
 120         if (err) {
 121                 zfree(&machine->kmaps);
 122                 zfree(&machine->root_dir);
 123                 zfree(&machine->mmap_name);
 124         }
 125         return 0;
 126 }
 127
 128 struct machine *machine__new_host(void)
 129 {
 130         struct machine *machine = malloc(sizeof(*machine));
 131
 132         if (machine != NULL) {
 133                 machine__init(machine, "", HOST_KERNEL_ID);
 134
 135                 if (machine__create_kernel_maps(machine) < 0)
 136                         goto out_delete;
 137
 138                 machine->env = &perf_env;
 139         }
 140
 141         return machine;
 142 out_delete:
 143         free(machine);
 144         return NULL;
 145 }
 146
 147 struct machine *machine__new_kallsyms(void)
 148 {
 149         struct machine *machine = machine__new_host();
 150         /*
 151          * FIXME:
 152          * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
 153          *    ask for not using the kcore parsing code, once this one is fixed
 154          *    to create a map per module.
 155          */
 156         if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
 157                 machine__delete(machine);
 158                 machine = NULL;
 159         }
 160
 161         return machine;
 162 }
 163
 164 void machine__delete_threads(struct machine *machine)
 165 {
 166         threads__remove_all_threads(&machine->threads);
 167 }
 168
 169 void machine__exit(struct machine *machine)
 170 {
 171         if (machine == NULL)
 172                 return;
 173
 174         machine__destroy_kernel_maps(machine);
 175         maps__zput(machine->kmaps);
 176         dsos__exit(&machine->dsos);
 177         machine__exit_vdso(machine);
 178         zfree(&machine->root_dir);
 179         zfree(&machine->mmap_name);
 180         zfree(&machine->current_tid);
 181         zfree(&machine->kallsyms_filename);
 182
 183         threads__exit(&machine->threads);
 184 }
 185
 186 void machine__delete(struct machine *machine)
 187 {
 188         if (machine) {
 189                 machine__exit(machine);
 190                 free(machine);
 191         }
 192 }
 193
 194 void machines__init(struct machines *machines)
 195 {
 196         machine__init(&machines->host, "", HOST_KERNEL_ID);
 197         machines->guests = RB_ROOT_CACHED;
 198 }
 199
 200 void machines__exit(struct machines *machines)
 201 {
 202         machine__exit(&machines->host);
 203         /* XXX exit guest */
 204 }
 205
 206 struct machine *machines__add(struct machines *machines, pid_t pid,
 207                               const char *root_dir)
 208 {
 209         struct rb_node **p = &machines->guests.rb_root.rb_node;
 210         struct rb_node *parent = NULL;
 211         struct machine *pos, *machine = malloc(sizeof(*machine));
 212         bool leftmost = true;
 213
 214         if (machine == NULL)
 215                 return NULL;
 216
 217         if (machine__init(machine, root_dir, pid) != 0) {
 218                 free(machine);
 219                 return NULL;
 220         }
 221
 222         while (*p != NULL) {
 223                 parent = *p;
 224                 pos = rb_entry(parent, struct machine, rb_node);
 225                 if (pid < pos->pid)
 226                         p = &(*p)->rb_left;
 227                 else {
 228                         p = &(*p)->rb_right;
 229                         leftmost = false;
 230                 }
 231         }
 232
 233         rb_link_node(&machine->rb_node, parent, p);
 234         rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
 235
 236         machine->machines = machines;
 237
 238         return machine;
 239 }
 240
 241 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
 242 {
 243         struct rb_node *nd;
 244
 245         machines->host.comm_exec = comm_exec;
 246
 247         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 248                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
 249
 250                 machine->comm_exec = comm_exec;
 251         }
 252 }
 253
 254 struct machine *machines__find(struct machines *machines, pid_t pid)
 255 {
 256         struct rb_node **p = &machines->guests.rb_root.rb_node;
 257         struct rb_node *parent = NULL;
 258         struct machine *machine;
 259         struct machine *default_machine = NULL;
 260
 261         if (pid == HOST_KERNEL_ID)
 262                 return &machines->host;
 263
 264         while (*p != NULL) {
 265                 parent = *p;
 266                 machine = rb_entry(parent, struct machine, rb_node);
 267                 if (pid < machine->pid)
 268                         p = &(*p)->rb_left;
 269                 else if (pid > machine->pid)
 270                         p = &(*p)->rb_right;
 271                 else
 272                         return machine;
 273                 if (!machine->pid)
 274                         default_machine = machine;
 275         }
 276
 277         return default_machine;
 278 }
 279
 280 struct machine *machines__findnew(struct machines *machines, pid_t pid)
 281 {
 282         char path[PATH_MAX];
 283         const char *root_dir = "";
 284         struct machine *machine = machines__find(machines, pid);
 285
 286         if (machine && (machine->pid == pid))
 287                 goto out;
 288
 289         if ((pid != HOST_KERNEL_ID) &&
 290             (pid != DEFAULT_GUEST_KERNEL_ID) &&
 291             (symbol_conf.guestmount)) {
 292                 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
 293                 if (access(path, R_OK)) {
 294                         static struct strlist *seen;
 295
 296                         if (!seen)
 297                                 seen = strlist__new(NULL, NULL);
 298
 299                         if (!strlist__has_entry(seen, path)) {
 300                                 pr_err("Can't access file %s\n", path);
 301                                 strlist__add(seen, path);
 302                         }
 303                         machine = NULL;
 304                         goto out;
 305                 }
 306                 root_dir = path;
 307         }
 308
 309         machine = machines__add(machines, pid, root_dir);
 310 out:
 311         return machine;
 312 }
 313
 314 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
 315 {
 316         struct machine *machine = machines__find(machines, pid);
 317
 318         if (!machine)
 319                 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
 320         return machine;
 321 }
 322
 323 /*
 324  * A common case for KVM test programs is that the test program acts as the
 325  * hypervisor, creating, running and destroying the virtual machine, and
 326  * providing the guest object code from its own object code. In this case,
 327  * the VM is not running an OS, but only the functions loaded into it by the
 328  * hypervisor test program, and conveniently, loaded at the same virtual
 329  * addresses.
 330  *
 331  * Normally to resolve addresses, MMAP events are needed to map addresses
 332  * back to the object code and debug symbols for that object code.
 333  *
 334  * Currently, there is no way to get such mapping information from guests
 335  * but, in the scenario described above, the guest has the same mappings
 336  * as the hypervisor, so support for that scenario can be achieved.
 337  *
 338  * To support that, copy the host thread's maps to the guest thread's maps.
 339  * Note, we do not discover the guest until we encounter a guest event,
 340  * which works well because it is not until then that we know that the host
 341  * thread's maps have been set up.
 342  *
 343  * This function returns the guest thread. Apart from keeping the data
 344  * structures sane, using a thread belonging to the guest machine, instead
 345  * of the host thread, allows it to have its own comm (refer
 346  * thread__set_guest_comm()).
 347  */
 348 static struct thread *findnew_guest_code(struct machine *machine,
 349                                          struct machine *host_machine,
 350                                          pid_t pid)
 351 {
 352         struct thread *host_thread;
 353         struct thread *thread;
 354         int err;
 355
 356         if (!machine)
 357                 return NULL;
 358
 359         thread = machine__findnew_thread(machine, -1, pid);
 360         if (!thread)
 361                 return NULL;
 362
 363         /* Assume maps are set up if there are any */
 364         if (!maps__empty(thread__maps(thread)))
 365                 return thread;
 366
 367         host_thread = machine__find_thread(host_machine, -1, pid);
 368         if (!host_thread)
 369                 goto out_err;
 370
 371         thread__set_guest_comm(thread, pid);
 372
 373         /*
 374          * Guest code can be found in hypervisor process at the same address
 375          * so copy host maps.
 376          */
 377         err = maps__copy_from(thread__maps(thread), thread__maps(host_thread));
 378         thread__put(host_thread);
 379         if (err)
 380                 goto out_err;
 381
 382         return thread;
 383
 384 out_err:
 385         thread__zput(thread);
 386         return NULL;
 387 }
 388
 389 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
 390 {
 391         struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
 392         struct machine *machine = machines__findnew(machines, pid);
 393
 394         return findnew_guest_code(machine, host_machine, pid);
 395 }
 396
 397 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
 398 {
 399         struct machines *machines = machine->machines;
 400         struct machine *host_machine;
 401
 402         if (!machines)
 403                 return NULL;
 404
 405         host_machine = machines__find(machines, HOST_KERNEL_ID);
 406
 407         return findnew_guest_code(machine, host_machine, pid);
 408 }
 409
 410 void machines__process_guests(struct machines *machines,
 411                               machine__process_t process, void *data)
 412 {
 413         struct rb_node *nd;
 414
 415         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 416                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 417                 process(pos, data);
 418         }
 419 }
 420
 421 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
 422 {
 423         struct rb_node *node;
 424         struct machine *machine;
 425
 426         machines->host.id_hdr_size = id_hdr_size;
 427
 428         for (node = rb_first_cached(&machines->guests); node;
 429              node = rb_next(node)) {
 430                 machine = rb_entry(node, struct machine, rb_node);
 431                 machine->id_hdr_size = id_hdr_size;
 432         }
 433
 434         return;
 435 }
 436
 437 static void machine__update_thread_pid(struct machine *machine,
 438                                        struct thread *th, pid_t pid)
 439 {
 440         struct thread *leader;
 441
 442         if (pid == thread__pid(th) || pid == -1 || thread__pid(th) != -1)
 443                 return;
 444
 445         thread__set_pid(th, pid);
 446
 447         if (thread__pid(th) == thread__tid(th))
 448                 return;
 449
 450         leader = machine__findnew_thread(machine, thread__pid(th), thread__pid(th));
 451         if (!leader)
 452                 goto out_err;
 453
 454         if (!thread__maps(leader))
 455                 thread__set_maps(leader, maps__new(machine));
 456
 457         if (!thread__maps(leader))
 458                 goto out_err;
 459
 460         if (thread__maps(th) == thread__maps(leader))
 461                 goto out_put;
 462
 463         if (thread__maps(th)) {
 464                 /*
 465                  * Maps are created from MMAP events which provide the pid and
 466                  * tid.  Consequently there never should be any maps on a thread
 467                  * with an unknown pid.  Just print an error if there are.
 468                  */
 469                 if (!maps__empty(thread__maps(th)))
 470                         pr_err("Discarding thread maps for %d:%d\n",
 471                                 thread__pid(th), thread__tid(th));
 472                 maps__put(thread__maps(th));
 473         }
 474
 475         thread__set_maps(th, maps__get(thread__maps(leader)));
 476 out_put:
 477         thread__put(leader);
 478         return;
 479 out_err:
 480         pr_err("Failed to join map groups for %d:%d\n", thread__pid(th), thread__tid(th));
 481         goto out_put;
 482 }
 483
 484 /*
 485  * Caller must eventually drop thread->refcnt returned with a successful
 486  * lookup/new thread inserted.
 487  */
 488 static struct thread *__machine__findnew_thread(struct machine *machine,
 489                                                 pid_t pid,
 490                                                 pid_t tid,
 491                                                 bool create)
 492 {
 493         struct thread *th = threads__find(&machine->threads, tid);
 494         bool created;
 495
 496         if (th) {
 497                 machine__update_thread_pid(machine, th, pid);
 498                 return th;
 499         }
 500         if (!create)
 501                 return NULL;
 502
 503         th = threads__findnew(&machine->threads, pid, tid, &created);
 504         if (created) {
 505                 /*
 506                  * We have to initialize maps separately after rb tree is
 507                  * updated.
 508                  *
 509                  * The reason is that we call machine__findnew_thread within
 510                  * thread__init_maps to find the thread leader and that would
 511                  * screwed the rb tree.
 512                  */
 513                 if (thread__init_maps(th, machine)) {
 514                         pr_err("Thread init failed thread %d\n", pid);
 515                         threads__remove(&machine->threads, th);
 516                         thread__put(th);
 517                         return NULL;
 518                 }
 519         } else
 520                 machine__update_thread_pid(machine, th, pid);
 521
 522         return th;
 523 }
 524
 525 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
 526 {
 527         return __machine__findnew_thread(machine, pid, tid, /*create=*/true);
 528 }
 529
 530 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
 531                                     pid_t tid)
 532 {
 533         return __machine__findnew_thread(machine, pid, tid, /*create=*/false);
 534 }
 535
 536 /*
 537  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
 538  * So here a single thread is created for that, but actually there is a separate
 539  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
 540  * is only 1. That causes problems for some tools, requiring workarounds. For
 541  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
 542  */
 543 struct thread *machine__idle_thread(struct machine *machine)
 544 {
 545         struct thread *thread = machine__findnew_thread(machine, 0, 0);
 546
 547         if (!thread || thread__set_comm(thread, "swapper", 0) ||
 548             thread__set_namespaces(thread, 0, NULL))
 549                 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
 550
 551         return thread;
 552 }
 553
 554 struct comm *machine__thread_exec_comm(struct machine *machine,
 555                                        struct thread *thread)
 556 {
 557         if (machine->comm_exec)
 558                 return thread__exec_comm(thread);
 559         else
 560                 return thread__comm(thread);
 561 }
 562
 563 int machine__process_comm_event(struct machine *machine, union perf_event *event,
 564                                 struct perf_sample *sample)
 565 {
 566         struct thread *thread = machine__findnew_thread(machine,
 567                                                         event->comm.pid,
 568                                                         event->comm.tid);
 569         bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
 570         int err = 0;
 571
 572         if (exec)
 573                 machine->comm_exec = true;
 574
 575         if (dump_trace)
 576                 perf_event__fprintf_comm(event, stdout);
 577
 578         if (thread == NULL ||
 579             __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 580                 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
 581                 err = -1;
 582         }
 583
 584         thread__put(thread);
 585
 586         return err;
 587 }
 588
 589 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
 590                                       union perf_event *event,
 591                                       struct perf_sample *sample __maybe_unused)
 592 {
 593         struct thread *thread = machine__findnew_thread(machine,
 594                                                         event->namespaces.pid,
 595                                                         event->namespaces.tid);
 596         int err = 0;
 597
 598         WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
 599                   "\nWARNING: kernel seems to support more namespaces than perf"
 600                   " tool.\nTry updating the perf tool..\n\n");
 601
 602         WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
 603                   "\nWARNING: perf tool seems to support more namespaces than"
 604                   " the kernel.\nTry updating the kernel..\n\n");
 605
 606         if (dump_trace)
 607                 perf_event__fprintf_namespaces(event, stdout);
 608
 609         if (thread == NULL ||
 610             thread__set_namespaces(thread, sample->time, &event->namespaces)) {
 611                 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
 612                 err = -1;
 613         }
 614
 615         thread__put(thread);
 616
 617         return err;
 618 }
 619
 620 int machine__process_cgroup_event(struct machine *machine,
 621                                   union perf_event *event,
 622                                   struct perf_sample *sample __maybe_unused)
 623 {
 624         struct cgroup *cgrp;
 625
 626         if (dump_trace)
 627                 perf_event__fprintf_cgroup(event, stdout);
 628
 629         cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
 630         if (cgrp == NULL)
 631                 return -ENOMEM;
 632
 633         return 0;
 634 }
 635
 636 int machine__process_lost_event(struct machine *machine __maybe_unused,
 637                                 union perf_event *event, struct perf_sample *sample __maybe_unused)
 638 {
 639         dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
 640                     event->lost.id, event->lost.lost);
 641         return 0;
 642 }
 643
 644 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
 645                                         union perf_event *event, struct perf_sample *sample)
 646 {
 647         dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "%s\n",
 648                     sample->id, event->lost_samples.lost,
 649                     event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF ? " (BPF)" : "");
 650         return 0;
 651 }
 652
 653 int machine__process_aux_event(struct machine *machine __maybe_unused,
 654                                union perf_event *event)
 655 {
 656         if (dump_trace)
 657                 perf_event__fprintf_aux(event, stdout);
 658         return 0;
 659 }
 660
 661 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
 662                                         union perf_event *event)
 663 {
 664         if (dump_trace)
 665                 perf_event__fprintf_itrace_start(event, stdout);
 666         return 0;
 667 }
 668
 669 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
 670                                             union perf_event *event)
 671 {
 672         if (dump_trace)
 673                 perf_event__fprintf_aux_output_hw_id(event, stdout);
 674         return 0;
 675 }
 676
 677 int machine__process_switch_event(struct machine *machine __maybe_unused,
 678                                   union perf_event *event)
 679 {
 680         if (dump_trace)
 681                 perf_event__fprintf_switch(event, stdout);
 682         return 0;
 683 }
 684
 685 static int machine__process_ksymbol_register(struct machine *machine,
 686                                              union perf_event *event,
 687                                              struct perf_sample *sample __maybe_unused)
 688 {
 689         struct symbol *sym;
 690         struct dso *dso = NULL;
 691         struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 692         int err = 0;
 693
 694         if (!map) {
 695                 dso = dso__new(event->ksymbol.name);
 696
 697                 if (!dso) {
 698                         err = -ENOMEM;
 699                         goto out;
 700                 }
 701                 dso__set_kernel(dso, DSO_SPACE__KERNEL);
 702                 map = map__new2(0, dso);
 703                 if (!map) {
 704                         err = -ENOMEM;
 705                         goto out;
 706                 }
 707                 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 708                         dso__set_binary_type(dso, DSO_BINARY_TYPE__OOL);
 709                         dso__data(dso)->file_size = event->ksymbol.len;
 710                         dso__set_loaded(dso);
 711                 }
 712
 713                 map__set_start(map, event->ksymbol.addr);
 714                 map__set_end(map, map__start(map) + event->ksymbol.len);
 715                 err = maps__insert(machine__kernel_maps(machine), map);
 716                 if (err) {
 717                         err = -ENOMEM;
 718                         goto out;
 719                 }
 720
 721                 dso__set_loaded(dso);
 722
 723                 if (is_bpf_image(event->ksymbol.name)) {
 724                         dso__set_binary_type(dso, DSO_BINARY_TYPE__BPF_IMAGE);
 725                         dso__set_long_name(dso, "", false);
 726                 }
 727         } else {
 728                 dso = dso__get(map__dso(map));
 729         }
 730
 731         sym = symbol__new(map__map_ip(map, map__start(map)),
 732                           event->ksymbol.len,
 733                           0, 0, event->ksymbol.name);
 734         if (!sym) {
 735                 err = -ENOMEM;
 736                 goto out;
 737         }
 738         dso__insert_symbol(dso, sym);
 739 out:
 740         map__put(map);
 741         dso__put(dso);
 742         return err;
 743 }
 744
 745 static int machine__process_ksymbol_unregister(struct machine *machine,
 746                                                union perf_event *event,
 747                                                struct perf_sample *sample __maybe_unused)
 748 {
 749         struct symbol *sym;
 750         struct map *map;
 751
 752         map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 753         if (!map)
 754                 return 0;
 755
 756         if (!RC_CHK_EQUAL(map, machine->vmlinux_map))
 757                 maps__remove(machine__kernel_maps(machine), map);
 758         else {
 759                 struct dso *dso = map__dso(map);
 760
 761                 sym = dso__find_symbol(dso, map__map_ip(map, map__start(map)));
 762                 if (sym)
 763                         dso__delete_symbol(dso, sym);
 764         }
 765         map__put(map);
 766         return 0;
 767 }
 768
 769 int machine__process_ksymbol(struct machine *machine __maybe_unused,
 770                              union perf_event *event,
 771                              struct perf_sample *sample)
 772 {
 773         if (dump_trace)
 774                 perf_event__fprintf_ksymbol(event, stdout);
 775
 776         if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 777                 return machine__process_ksymbol_unregister(machine, event,
 778                                                            sample);
 779         return machine__process_ksymbol_register(machine, event, sample);
 780 }
 781
 782 int machine__process_text_poke(struct machine *machine, union perf_event *event,
 783                                struct perf_sample *sample __maybe_unused)
 784 {
 785         struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
 786         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 787         struct dso *dso = map ? map__dso(map) : NULL;
 788
 789         if (dump_trace)
 790                 perf_event__fprintf_text_poke(event, machine, stdout);
 791
 792         if (!event->text_poke.new_len)
 793                 goto out;
 794
 795         if (cpumode != PERF_RECORD_MISC_KERNEL) {
 796                 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
 797                 goto out;
 798         }
 799
 800         if (dso) {
 801                 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
 802                 int ret;
 803
 804                 /*
 805                  * Kernel maps might be changed when loading symbols so loading
 806                  * must be done prior to using kernel maps.
 807                  */
 808                 map__load(map);
 809                 ret = dso__data_write_cache_addr(dso, map, machine,
 810                                                  event->text_poke.addr,
 811                                                  new_bytes,
 812                                                  event->text_poke.new_len);
 813                 if (ret != event->text_poke.new_len)
 814                         pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
 815                                  event->text_poke.addr);
 816         } else {
 817                 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
 818                          event->text_poke.addr);
 819         }
 820 out:
 821         map__put(map);
 822         return 0;
 823 }
 824
 825 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
 826                                               const char *filename)
 827 {
 828         struct map *map = NULL;
 829         struct kmod_path m;
 830         struct dso *dso;
 831         int err;
 832
 833         if (kmod_path__parse_name(&m, filename))
 834                 return NULL;
 835
 836         dso = dsos__findnew_module_dso(&machine->dsos, machine, &m, filename);
 837         if (dso == NULL)
 838                 goto out;
 839
 840         map = map__new2(start, dso);
 841         if (map == NULL)
 842                 goto out;
 843
 844         err = maps__insert(machine__kernel_maps(machine), map);
 845         /* If maps__insert failed, return NULL. */
 846         if (err) {
 847                 map__put(map);
 848                 map = NULL;
 849         }
 850 out:
 851         /* put the dso here, corresponding to  machine__findnew_module_dso */
 852         dso__put(dso);
 853         zfree(&m.name);
 854         return map;
 855 }
 856
 857 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
 858 {
 859         struct rb_node *nd;
 860         size_t ret = dsos__fprintf(&machines->host.dsos, fp);
 861
 862         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 863                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 864                 ret += dsos__fprintf(&pos->dsos, fp);
 865         }
 866
 867         return ret;
 868 }
 869
 870 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
 871                                      bool (skip)(struct dso *dso, int parm), int parm)
 872 {
 873         return dsos__fprintf_buildid(&m->dsos, fp, skip, parm);
 874 }
 875
 876 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
 877                                      bool (skip)(struct dso *dso, int parm), int parm)
 878 {
 879         struct rb_node *nd;
 880         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
 881
 882         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 883                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 884                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
 885         }
 886         return ret;
 887 }
 888
 889 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
 890 {
 891         int i;
 892         size_t printed = 0;
 893         struct dso *kdso = machine__kernel_dso(machine);
 894
 895         if (dso__has_build_id(kdso)) {
 896                 char filename[PATH_MAX];
 897
 898                 if (dso__build_id_filename(kdso, filename, sizeof(filename), false))
 899                         printed += fprintf(fp, "[0] %s\n", filename);
 900         }
 901
 902         for (i = 0; i < vmlinux_path__nr_entries; ++i) {
 903                 printed += fprintf(fp, "[%d] %s\n", i + dso__has_build_id(kdso),
 904                                    vmlinux_path[i]);
 905         }
 906         return printed;
 907 }
 908
 909 struct machine_fprintf_cb_args {
 910         FILE *fp;
 911         size_t printed;
 912 };
 913
 914 static int machine_fprintf_cb(struct thread *thread, void *data)
 915 {
 916         struct machine_fprintf_cb_args *args = data;
 917
 918         /* TODO: handle fprintf errors. */
 919         args->printed += thread__fprintf(thread, args->fp);
 920         return 0;
 921 }
 922
 923 size_t machine__fprintf(struct machine *machine, FILE *fp)
 924 {
 925         struct machine_fprintf_cb_args args = {
 926                 .fp = fp,
 927                 .printed = 0,
 928         };
 929         size_t ret = fprintf(fp, "Threads: %zu\n", threads__nr(&machine->threads));
 930
 931         machine__for_each_thread(machine, machine_fprintf_cb, &args);
 932         return ret + args.printed;
 933 }
 934
 935 static struct dso *machine__get_kernel(struct machine *machine)
 936 {
 937         const char *vmlinux_name = machine->mmap_name;
 938         struct dso *kernel;
 939
 940         if (machine__is_host(machine)) {
 941                 if (symbol_conf.vmlinux_name)
 942                         vmlinux_name = symbol_conf.vmlinux_name;
 943
 944                 kernel = machine__findnew_kernel(machine, vmlinux_name,
 945                                                  "[kernel]", DSO_SPACE__KERNEL);
 946         } else {
 947                 if (symbol_conf.default_guest_vmlinux_name)
 948                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
 949
 950                 kernel = machine__findnew_kernel(machine, vmlinux_name,
 951                                                  "[guest.kernel]",
 952                                                  DSO_SPACE__KERNEL_GUEST);
 953         }
 954
 955         if (kernel != NULL && (!dso__has_build_id(kernel)))
 956                 dso__read_running_kernel_build_id(kernel, machine);
 957
 958         return kernel;
 959 }
 960
 961 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
 962                                     size_t bufsz)
 963 {
 964         if (machine__is_default_guest(machine))
 965                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
 966         else
 967                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
 968 }
 969
 970 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
 971
 972 /* Figure out the start address of kernel map from /proc/kallsyms.
 973  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
 974  * symbol_name if it's not that important.
 975  */
 976 static int machine__get_running_kernel_start(struct machine *machine,
 977                                              const char **symbol_name,
 978                                              u64 *start, u64 *end)
 979 {
 980         char filename[PATH_MAX];
 981         int i, err = -1;
 982         const char *name;
 983         u64 addr = 0;
 984
 985         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
 986
 987         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
 988                 return 0;
 989
 990         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
 991                 err = kallsyms__get_function_start(filename, name, &addr);
 992                 if (!err)
 993                         break;
 994         }
 995
 996         if (err)
 997                 return -1;
 998
 999         if (symbol_name)
1000                 *symbol_name = name;
1001
1002         *start = addr;
1003
1004         err = kallsyms__get_symbol_start(filename, "_edata", &addr);
1005         if (err)
1006                 err = kallsyms__get_function_start(filename, "_etext", &addr);
1007         if (!err)
1008                 *end = addr;
1009
1010         return 0;
1011 }
1012
1013 int machine__create_extra_kernel_map(struct machine *machine,
1014                                      struct dso *kernel,
1015                                      struct extra_kernel_map *xm)
1016 {
1017         struct kmap *kmap;
1018         struct map *map;
1019         int err;
1020
1021         map = map__new2(xm->start, kernel);
1022         if (!map)
1023                 return -ENOMEM;
1024
1025         map__set_end(map, xm->end);
1026         map__set_pgoff(map, xm->pgoff);
1027
1028         kmap = map__kmap(map);
1029
1030         strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1031
1032         err = maps__insert(machine__kernel_maps(machine), map);
1033
1034         if (!err) {
1035                 pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1036                         kmap->name, map__start(map), map__end(map));
1037         }
1038
1039         map__put(map);
1040
1041         return err;
1042 }
1043
1044 static u64 find_entry_trampoline(struct dso *dso)
1045 {
1046         /* Duplicates are removed so lookup all aliases */
1047         const char *syms[] = {
1048                 "_entry_trampoline",
1049                 "__entry_trampoline_start",
1050                 "entry_SYSCALL_64_trampoline",
1051         };
1052         struct symbol *sym = dso__first_symbol(dso);
1053         unsigned int i;
1054
1055         for (; sym; sym = dso__next_symbol(sym)) {
1056                 if (sym->binding != STB_GLOBAL)
1057                         continue;
1058                 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1059                         if (!strcmp(sym->name, syms[i]))
1060                                 return sym->start;
1061                 }
1062         }
1063
1064         return 0;
1065 }
1066
1067 /*
1068  * These values can be used for kernels that do not have symbols for the entry
1069  * trampolines in kallsyms.
1070  */
1071 #define X86_64_CPU_ENTRY_AREA_PER_CPU   0xfffffe0000000000ULL
1072 #define X86_64_CPU_ENTRY_AREA_SIZE      0x2c000
1073 #define X86_64_ENTRY_TRAMPOLINE         0x6000
1074
1075 struct machine__map_x86_64_entry_trampolines_args {
1076         struct maps *kmaps;
1077         bool found;
1078 };
1079
1080 static int machine__map_x86_64_entry_trampolines_cb(struct map *map, void *data)
1081 {
1082         struct machine__map_x86_64_entry_trampolines_args *args = data;
1083         struct map *dest_map;
1084         struct kmap *kmap = __map__kmap(map);
1085
1086         if (!kmap || !is_entry_trampoline(kmap->name))
1087                 return 0;
1088
1089         dest_map = maps__find(args->kmaps, map__pgoff(map));
1090         if (RC_CHK_ACCESS(dest_map) != RC_CHK_ACCESS(map))
1091                 map__set_pgoff(map, map__map_ip(dest_map, map__pgoff(map)));
1092
1093         map__put(dest_map);
1094         args->found = true;
1095         return 0;
1096 }
1097
1098 /* Map x86_64 PTI entry trampolines */
1099 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1100                                           struct dso *kernel)
1101 {
1102         struct machine__map_x86_64_entry_trampolines_args args = {
1103                 .kmaps = machine__kernel_maps(machine),
1104                 .found = false,
1105         };
1106         int nr_cpus_avail, cpu;
1107         u64 pgoff;
1108
1109         /*
1110          * In the vmlinux case, pgoff is a virtual address which must now be
1111          * mapped to a vmlinux offset.
1112          */
1113         maps__for_each_map(args.kmaps, machine__map_x86_64_entry_trampolines_cb, &args);
1114
1115         if (args.found || machine->trampolines_mapped)
1116                 return 0;
1117
1118         pgoff = find_entry_trampoline(kernel);
1119         if (!pgoff)
1120                 return 0;
1121
1122         nr_cpus_avail = machine__nr_cpus_avail(machine);
1123
1124         /* Add a 1 page map for each CPU's entry trampoline */
1125         for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1126                 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1127                          cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1128                          X86_64_ENTRY_TRAMPOLINE;
1129                 struct extra_kernel_map xm = {
1130                         .start = va,
1131                         .end   = va + page_size,
1132                         .pgoff = pgoff,
1133                 };
1134
1135                 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1136
1137                 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1138                         return -1;
1139         }
1140
1141         machine->trampolines_mapped = nr_cpus_avail;
1142
1143         return 0;
1144 }
1145
1146 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1147                                              struct dso *kernel __maybe_unused)
1148 {
1149         return 0;
1150 }
1151
1152 static int
1153 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1154 {
1155         /* In case of renewal the kernel map, destroy previous one */
1156         machine__destroy_kernel_maps(machine);
1157
1158         map__put(machine->vmlinux_map);
1159         machine->vmlinux_map = map__new2(0, kernel);
1160         if (machine->vmlinux_map == NULL)
1161                 return -ENOMEM;
1162
1163         map__set_mapping_type(machine->vmlinux_map, MAPPING_TYPE__IDENTITY);
1164         return maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1165 }
1166
1167 void machine__destroy_kernel_maps(struct machine *machine)
1168 {
1169         struct kmap *kmap;
1170         struct map *map = machine__kernel_map(machine);
1171
1172         if (map == NULL)
1173                 return;
1174
1175         kmap = map__kmap(map);
1176         maps__remove(machine__kernel_maps(machine), map);
1177         if (kmap && kmap->ref_reloc_sym) {
1178                 zfree((char **)&kmap->ref_reloc_sym->name);
1179                 zfree(&kmap->ref_reloc_sym);
1180         }
1181
1182         map__zput(machine->vmlinux_map);
1183 }
1184
1185 int machines__create_guest_kernel_maps(struct machines *machines)
1186 {
1187         int ret = 0;
1188         struct dirent **namelist = NULL;
1189         int i, items = 0;
1190         char path[PATH_MAX];
1191         pid_t pid;
1192         char *endp;
1193
1194         if (symbol_conf.default_guest_vmlinux_name ||
1195             symbol_conf.default_guest_modules ||
1196             symbol_conf.default_guest_kallsyms) {
1197                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1198         }
1199
1200         if (symbol_conf.guestmount) {
1201                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1202                 if (items <= 0)
1203                         return -ENOENT;
1204                 for (i = 0; i < items; i++) {
1205                         if (!isdigit(namelist[i]->d_name[0])) {
1206                                 /* Filter out . and .. */
1207                                 continue;
1208                         }
1209                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1210                         if ((*endp != '\0') ||
1211                             (endp == namelist[i]->d_name) ||
1212                             (errno == ERANGE)) {
1213                                 pr_debug("invalid directory (%s). Skipping.\n",
1214                                          namelist[i]->d_name);
1215                                 continue;
1216                         }
1217                         sprintf(path, "%s/%s/proc/kallsyms",
1218                                 symbol_conf.guestmount,
1219                                 namelist[i]->d_name);
1220                         ret = access(path, R_OK);
1221                         if (ret) {
1222                                 pr_debug("Can't access file %s\n", path);
1223                                 goto failure;
1224                         }
1225                         machines__create_kernel_maps(machines, pid);
1226                 }
1227 failure:
1228                 free(namelist);
1229         }
1230
1231         return ret;
1232 }
1233
1234 void machines__destroy_kernel_maps(struct machines *machines)
1235 {
1236         struct rb_node *next = rb_first_cached(&machines->guests);
1237
1238         machine__destroy_kernel_maps(&machines->host);
1239
1240         while (next) {
1241                 struct machine *pos = rb_entry(next, struct machine, rb_node);
1242
1243                 next = rb_next(&pos->rb_node);
1244                 rb_erase_cached(&pos->rb_node, &machines->guests);
1245                 machine__delete(pos);
1246         }
1247 }
1248
1249 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1250 {
1251         struct machine *machine = machines__findnew(machines, pid);
1252
1253         if (machine == NULL)
1254                 return -1;
1255
1256         return machine__create_kernel_maps(machine);
1257 }
1258
1259 int machine__load_kallsyms(struct machine *machine, const char *filename)
1260 {
1261         struct map *map = machine__kernel_map(machine);
1262         struct dso *dso = map__dso(map);
1263         int ret = __dso__load_kallsyms(dso, filename, map, true);
1264
1265         if (ret > 0) {
1266                 dso__set_loaded(dso);
1267                 /*
1268                  * Since /proc/kallsyms will have multiple sessions for the
1269                  * kernel, with modules between them, fixup the end of all
1270                  * sections.
1271                  */
1272                 maps__fixup_end(machine__kernel_maps(machine));
1273         }
1274
1275         return ret;
1276 }
1277
1278 int machine__load_vmlinux_path(struct machine *machine)
1279 {
1280         struct map *map = machine__kernel_map(machine);
1281         struct dso *dso = map__dso(map);
1282         int ret = dso__load_vmlinux_path(dso, map);
1283
1284         if (ret > 0)
1285                 dso__set_loaded(dso);
1286
1287         return ret;
1288 }
1289
1290 static char *get_kernel_version(const char *root_dir)
1291 {
1292         char version[PATH_MAX];
1293         FILE *file;
1294         char *name, *tmp;
1295         const char *prefix = "Linux version ";
1296
1297         sprintf(version, "%s/proc/version", root_dir);
1298         file = fopen(version, "r");
1299         if (!file)
1300                 return NULL;
1301
1302         tmp = fgets(version, sizeof(version), file);
1303         fclose(file);
1304         if (!tmp)
1305                 return NULL;
1306
1307         name = strstr(version, prefix);
1308         if (!name)
1309                 return NULL;
1310         name += strlen(prefix);
1311         tmp = strchr(name, ' ');
1312         if (tmp)
1313                 *tmp = '\0';
1314
1315         return strdup(name);
1316 }
1317
1318 static bool is_kmod_dso(struct dso *dso)
1319 {
1320         return dso__symtab_type(dso) == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1321                dso__symtab_type(dso) == DSO_BINARY_TYPE__GUEST_KMODULE;
1322 }
1323
1324 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1325 {
1326         char *long_name;
1327         struct dso *dso;
1328         struct map *map = maps__find_by_name(maps, m->name);
1329
1330         if (map == NULL)
1331                 return 0;
1332
1333         long_name = strdup(path);
1334         if (long_name == NULL) {
1335                 map__put(map);
1336                 return -ENOMEM;
1337         }
1338
1339         dso = map__dso(map);
1340         dso__set_long_name(dso, long_name, true);
1341         dso__kernel_module_get_build_id(dso, "");
1342
1343         /*
1344          * Full name could reveal us kmod compression, so
1345          * we need to update the symtab_type if needed.
1346          */
1347         if (m->comp && is_kmod_dso(dso)) {
1348                 dso__set_symtab_type(dso, dso__symtab_type(dso)+1);
1349                 dso__set_comp(dso, m->comp);
1350         }
1351         map__put(map);
1352         return 0;
1353 }
1354
1355 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1356 {
1357         struct dirent *dent;
1358         DIR *dir = opendir(dir_name);
1359         int ret = 0;
1360
1361         if (!dir) {
1362                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1363                 return -1;
1364         }
1365
1366         while ((dent = readdir(dir)) != NULL) {
1367                 char path[PATH_MAX];
1368                 struct stat st;
1369
1370                 /*sshfs might return bad dent->d_type, so we have to stat*/
1371                 path__join(path, sizeof(path), dir_name, dent->d_name);
1372                 if (stat(path, &st))
1373                         continue;
1374
1375                 if (S_ISDIR(st.st_mode)) {
1376                         if (!strcmp(dent->d_name, ".") ||
1377                             !strcmp(dent->d_name, ".."))
1378                                 continue;
1379
1380                         /* Do not follow top-level source and build symlinks */
1381                         if (depth == 0) {
1382                                 if (!strcmp(dent->d_name, "source") ||
1383                                     !strcmp(dent->d_name, "build"))
1384                                         continue;
1385                         }
1386
1387                         ret = maps__set_modules_path_dir(maps, path, depth + 1);
1388                         if (ret < 0)
1389                                 goto out;
1390                 } else {
1391                         struct kmod_path m;
1392
1393                         ret = kmod_path__parse_name(&m, dent->d_name);
1394                         if (ret)
1395                                 goto out;
1396
1397                         if (m.kmod)
1398                                 ret = maps__set_module_path(maps, path, &m);
1399
1400                         zfree(&m.name);
1401
1402                         if (ret)
1403                                 goto out;
1404                 }
1405         }
1406
1407 out:
1408         closedir(dir);
1409         return ret;
1410 }
1411
1412 static int machine__set_modules_path(struct machine *machine)
1413 {
1414         char *version;
1415         char modules_path[PATH_MAX];
1416
1417         version = get_kernel_version(machine->root_dir);
1418         if (!version)
1419                 return -1;
1420
1421         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1422                  machine->root_dir, version);
1423         free(version);
1424
1425         return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1426 }
1427 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1428                                 u64 *size __maybe_unused,
1429                                 const char *name __maybe_unused)
1430 {
1431         return 0;
1432 }
1433
1434 static int machine__create_module(void *arg, const char *name, u64 start,
1435                                   u64 size)
1436 {
1437         struct machine *machine = arg;
1438         struct map *map;
1439
1440         if (arch__fix_module_text_start(&start, &size, name) < 0)
1441                 return -1;
1442
1443         map = machine__addnew_module_map(machine, start, name);
1444         if (map == NULL)
1445                 return -1;
1446         map__set_end(map, start + size);
1447
1448         dso__kernel_module_get_build_id(map__dso(map), machine->root_dir);
1449         map__put(map);
1450         return 0;
1451 }
1452
1453 static int machine__create_modules(struct machine *machine)
1454 {
1455         const char *modules;
1456         char path[PATH_MAX];
1457
1458         if (machine__is_default_guest(machine)) {
1459                 modules = symbol_conf.default_guest_modules;
1460         } else {
1461                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1462                 modules = path;
1463         }
1464
1465         if (symbol__restricted_filename(modules, "/proc/modules"))
1466                 return -1;
1467
1468         if (modules__parse(modules, machine, machine__create_module))
1469                 return -1;
1470
1471         if (!machine__set_modules_path(machine))
1472                 return 0;
1473
1474         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1475
1476         return 0;
1477 }
1478
1479 static void machine__set_kernel_mmap(struct machine *machine,
1480                                      u64 start, u64 end)
1481 {
1482         map__set_start(machine->vmlinux_map, start);
1483         map__set_end(machine->vmlinux_map, end);
1484         /*
1485          * Be a bit paranoid here, some perf.data file came with
1486          * a zero sized synthesized MMAP event for the kernel.
1487          */
1488         if (start == 0 && end == 0)
1489                 map__set_end(machine->vmlinux_map, ~0ULL);
1490 }
1491
1492 static int machine__update_kernel_mmap(struct machine *machine,
1493                                      u64 start, u64 end)
1494 {
1495         struct map *orig, *updated;
1496         int err;
1497
1498         orig = machine->vmlinux_map;
1499         updated = map__get(orig);
1500
1501         machine->vmlinux_map = updated;
1502         maps__remove(machine__kernel_maps(machine), orig);
1503         machine__set_kernel_mmap(machine, start, end);
1504         err = maps__insert(machine__kernel_maps(machine), updated);
1505         map__put(orig);
1506
1507         return err;
1508 }
1509
1510 int machine__create_kernel_maps(struct machine *machine)
1511 {
1512         struct dso *kernel = machine__get_kernel(machine);
1513         const char *name = NULL;
1514         u64 start = 0, end = ~0ULL;
1515         int ret;
1516
1517         if (kernel == NULL)
1518                 return -1;
1519
1520         ret = __machine__create_kernel_maps(machine, kernel);
1521         if (ret < 0)
1522                 goto out_put;
1523
1524         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1525                 if (machine__is_host(machine))
1526                         pr_debug("Problems creating module maps, "
1527                                  "continuing anyway...\n");
1528                 else
1529                         pr_debug("Problems creating module maps for guest %d, "
1530                                  "continuing anyway...\n", machine->pid);
1531         }
1532
1533         if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1534                 if (name &&
1535                     map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1536                         machine__destroy_kernel_maps(machine);
1537                         ret = -1;
1538                         goto out_put;
1539                 }
1540
1541                 /*
1542                  * we have a real start address now, so re-order the kmaps
1543                  * assume it's the last in the kmaps
1544                  */
1545                 ret = machine__update_kernel_mmap(machine, start, end);
1546                 if (ret < 0)
1547                         goto out_put;
1548         }
1549
1550         if (machine__create_extra_kernel_maps(machine, kernel))
1551                 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1552
1553         if (end == ~0ULL) {
1554                 /* update end address of the kernel map using adjacent module address */
1555                 struct map *next = maps__find_next_entry(machine__kernel_maps(machine),
1556                                                          machine__kernel_map(machine));
1557
1558                 if (next) {
1559                         machine__set_kernel_mmap(machine, start, map__start(next));
1560                         map__put(next);
1561                 }
1562         }
1563
1564 out_put:
1565         dso__put(kernel);
1566         return ret;
1567 }
1568
1569 static int machine__uses_kcore_cb(struct dso *dso, void *data __maybe_unused)
1570 {
1571         return dso__is_kcore(dso) ? 1 : 0;
1572 }
1573
1574 static bool machine__uses_kcore(struct machine *machine)
1575 {
1576         return dsos__for_each_dso(&machine->dsos, machine__uses_kcore_cb, NULL) != 0 ? true : false;
1577 }
1578
1579 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1580                                              struct extra_kernel_map *xm)
1581 {
1582         return machine__is(machine, "x86_64") &&
1583                is_entry_trampoline(xm->name);
1584 }
1585
1586 static int machine__process_extra_kernel_map(struct machine *machine,
1587                                              struct extra_kernel_map *xm)
1588 {
1589         struct dso *kernel = machine__kernel_dso(machine);
1590
1591         if (kernel == NULL)
1592                 return -1;
1593
1594         return machine__create_extra_kernel_map(machine, kernel, xm);
1595 }
1596
1597 static int machine__process_kernel_mmap_event(struct machine *machine,
1598                                               struct extra_kernel_map *xm,
1599                                               struct build_id *bid)
1600 {
1601         enum dso_space_type dso_space;
1602         bool is_kernel_mmap;
1603         const char *mmap_name = machine->mmap_name;
1604
1605         /* If we have maps from kcore then we do not need or want any others */
1606         if (machine__uses_kcore(machine))
1607                 return 0;
1608
1609         if (machine__is_host(machine))
1610                 dso_space = DSO_SPACE__KERNEL;
1611         else
1612                 dso_space = DSO_SPACE__KERNEL_GUEST;
1613
1614         is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1615         if (!is_kernel_mmap && !machine__is_host(machine)) {
1616                 /*
1617                  * If the event was recorded inside the guest and injected into
1618                  * the host perf.data file, then it will match a host mmap_name,
1619                  * so try that - see machine__set_mmap_name().
1620                  */
1621                 mmap_name = "[kernel.kallsyms]";
1622                 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1623         }
1624         if (xm->name[0] == '/' ||
1625             (!is_kernel_mmap && xm->name[0] == '[')) {
1626                 struct map *map = machine__addnew_module_map(machine, xm->start, xm->name);
1627
1628                 if (map == NULL)
1629                         goto out_problem;
1630
1631                 map__set_end(map, map__start(map) + xm->end - xm->start);
1632
1633                 if (build_id__is_defined(bid))
1634                         dso__set_build_id(map__dso(map), bid);
1635
1636                 map__put(map);
1637         } else if (is_kernel_mmap) {
1638                 const char *symbol_name = xm->name + strlen(mmap_name);
1639                 /*
1640                  * Should be there already, from the build-id table in
1641                  * the header.
1642                  */
1643                 struct dso *kernel = dsos__find_kernel_dso(&machine->dsos);
1644
1645                 if (kernel == NULL)
1646                         kernel = machine__findnew_dso(machine, machine->mmap_name);
1647                 if (kernel == NULL)
1648                         goto out_problem;
1649
1650                 dso__set_kernel(kernel, dso_space);
1651                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1652                         dso__put(kernel);
1653                         goto out_problem;
1654                 }
1655
1656                 if (strstr(dso__long_name(kernel), "vmlinux"))
1657                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1658
1659                 if (machine__update_kernel_mmap(machine, xm->start, xm->end) < 0) {
1660                         dso__put(kernel);
1661                         goto out_problem;
1662                 }
1663
1664                 if (build_id__is_defined(bid))
1665                         dso__set_build_id(kernel, bid);
1666
1667                 /*
1668                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1669                  * symbol. Effectively having zero here means that at record
1670                  * time /proc/sys/kernel/kptr_restrict was non zero.
1671                  */
1672                 if (xm->pgoff != 0) {
1673                         map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1674                                                         symbol_name,
1675                                                         xm->pgoff);
1676                 }
1677
1678                 if (machine__is_default_guest(machine)) {
1679                         /*
1680                          * preload dso of guest kernel and modules
1681                          */
1682                         dso__load(kernel, machine__kernel_map(machine));
1683                 }
1684                 dso__put(kernel);
1685         } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1686                 return machine__process_extra_kernel_map(machine, xm);
1687         }
1688         return 0;
1689 out_problem:
1690         return -1;
1691 }
1692
1693 int machine__process_mmap2_event(struct machine *machine,
1694                                  union perf_event *event,
1695                                  struct perf_sample *sample)
1696 {
1697         struct thread *thread;
1698         struct map *map;
1699         struct dso_id dso_id = {
1700                 .maj = event->mmap2.maj,
1701                 .min = event->mmap2.min,
1702                 .ino = event->mmap2.ino,
1703                 .ino_generation = event->mmap2.ino_generation,
1704         };
1705         struct build_id __bid, *bid = NULL;
1706         int ret = 0;
1707
1708         if (dump_trace)
1709                 perf_event__fprintf_mmap2(event, stdout);
1710
1711         if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1712                 bid = &__bid;
1713                 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1714         }
1715
1716         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1717             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1718                 struct extra_kernel_map xm = {
1719                         .start = event->mmap2.start,
1720                         .end   = event->mmap2.start + event->mmap2.len,
1721                         .pgoff = event->mmap2.pgoff,
1722                 };
1723
1724                 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1725                 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1726                 if (ret < 0)
1727                         goto out_problem;
1728                 return 0;
1729         }
1730
1731         thread = machine__findnew_thread(machine, event->mmap2.pid,
1732                                         event->mmap2.tid);
1733         if (thread == NULL)
1734                 goto out_problem;
1735
1736         map = map__new(machine, event->mmap2.start,
1737                         event->mmap2.len, event->mmap2.pgoff,
1738                         &dso_id, event->mmap2.prot,
1739                         event->mmap2.flags, bid,
1740                         event->mmap2.filename, thread);
1741
1742         if (map == NULL)
1743                 goto out_problem_map;
1744
1745         ret = thread__insert_map(thread, map);
1746         if (ret)
1747                 goto out_problem_insert;
1748
1749         thread__put(thread);
1750         map__put(map);
1751         return 0;
1752
1753 out_problem_insert:
1754         map__put(map);
1755 out_problem_map:
1756         thread__put(thread);
1757 out_problem:
1758         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1759         return 0;
1760 }
1761
1762 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1763                                 struct perf_sample *sample)
1764 {
1765         struct thread *thread;
1766         struct map *map;
1767         u32 prot = 0;
1768         int ret = 0;
1769
1770         if (dump_trace)
1771                 perf_event__fprintf_mmap(event, stdout);
1772
1773         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1774             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1775                 struct extra_kernel_map xm = {
1776                         .start = event->mmap.start,
1777                         .end   = event->mmap.start + event->mmap.len,
1778                         .pgoff = event->mmap.pgoff,
1779                 };
1780
1781                 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1782                 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1783                 if (ret < 0)
1784                         goto out_problem;
1785                 return 0;
1786         }
1787
1788         thread = machine__findnew_thread(machine, event->mmap.pid,
1789                                          event->mmap.tid);
1790         if (thread == NULL)
1791                 goto out_problem;
1792
1793         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1794                 prot = PROT_EXEC;
1795
1796         map = map__new(machine, event->mmap.start,
1797                         event->mmap.len, event->mmap.pgoff,
1798                         NULL, prot, 0, NULL, event->mmap.filename, thread);
1799
1800         if (map == NULL)
1801                 goto out_problem_map;
1802
1803         ret = thread__insert_map(thread, map);
1804         if (ret)
1805                 goto out_problem_insert;
1806
1807         thread__put(thread);
1808         map__put(map);
1809         return 0;
1810
1811 out_problem_insert:
1812         map__put(map);
1813 out_problem_map:
1814         thread__put(thread);
1815 out_problem:
1816         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1817         return 0;
1818 }
1819
1820 void machine__remove_thread(struct machine *machine, struct thread *th)
1821 {
1822         return threads__remove(&machine->threads, th);
1823 }
1824
1825 int machine__process_fork_event(struct machine *machine, union perf_event *event,
1826                                 struct perf_sample *sample)
1827 {
1828         struct thread *thread = machine__find_thread(machine,
1829                                                      event->fork.pid,
1830                                                      event->fork.tid);
1831         struct thread *parent = machine__findnew_thread(machine,
1832                                                         event->fork.ppid,
1833                                                         event->fork.ptid);
1834         bool do_maps_clone = true;
1835         int err = 0;
1836
1837         if (dump_trace)
1838                 perf_event__fprintf_task(event, stdout);
1839
1840         /*
1841          * There may be an existing thread that is not actually the parent,
1842          * either because we are processing events out of order, or because the
1843          * (fork) event that would have removed the thread was lost. Assume the
1844          * latter case and continue on as best we can.
1845          */
1846         if (thread__pid(parent) != (pid_t)event->fork.ppid) {
1847                 dump_printf("removing erroneous parent thread %d/%d\n",
1848                             thread__pid(parent), thread__tid(parent));
1849                 machine__remove_thread(machine, parent);
1850                 thread__put(parent);
1851                 parent = machine__findnew_thread(machine, event->fork.ppid,
1852                                                  event->fork.ptid);
1853         }
1854
1855         /* if a thread currently exists for the thread id remove it */
1856         if (thread != NULL) {
1857                 machine__remove_thread(machine, thread);
1858                 thread__put(thread);
1859         }
1860
1861         thread = machine__findnew_thread(machine, event->fork.pid,
1862                                          event->fork.tid);
1863         /*
1864          * When synthesizing FORK events, we are trying to create thread
1865          * objects for the already running tasks on the machine.
1866          *
1867          * Normally, for a kernel FORK event, we want to clone the parent's
1868          * maps because that is what the kernel just did.
1869          *
1870          * But when synthesizing, this should not be done.  If we do, we end up
1871          * with overlapping maps as we process the synthesized MMAP2 events that
1872          * get delivered shortly thereafter.
1873          *
1874          * Use the FORK event misc flags in an internal way to signal this
1875          * situation, so we can elide the map clone when appropriate.
1876          */
1877         if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
1878                 do_maps_clone = false;
1879
1880         if (thread == NULL || parent == NULL ||
1881             thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
1882                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
1883                 err = -1;
1884         }
1885         thread__put(thread);
1886         thread__put(parent);
1887
1888         return err;
1889 }
1890
1891 int machine__process_exit_event(struct machine *machine, union perf_event *event,
1892                                 struct perf_sample *sample __maybe_unused)
1893 {
1894         struct thread *thread = machine__find_thread(machine,
1895                                                      event->fork.pid,
1896                                                      event->fork.tid);
1897
1898         if (dump_trace)
1899                 perf_event__fprintf_task(event, stdout);
1900
1901         if (thread != NULL) {
1902                 if (symbol_conf.keep_exited_threads)
1903                         thread__set_exited(thread, /*exited=*/true);
1904                 else
1905                         machine__remove_thread(machine, thread);
1906         }
1907         thread__put(thread);
1908         return 0;
1909 }
1910
1911 int machine__process_event(struct machine *machine, union perf_event *event,
1912                            struct perf_sample *sample)
1913 {
1914         int ret;
1915
1916         switch (event->header.type) {
1917         case PERF_RECORD_COMM:
1918                 ret = machine__process_comm_event(machine, event, sample); break;
1919         case PERF_RECORD_MMAP:
1920                 ret = machine__process_mmap_event(machine, event, sample); break;
1921         case PERF_RECORD_NAMESPACES:
1922                 ret = machine__process_namespaces_event(machine, event, sample); break;
1923         case PERF_RECORD_CGROUP:
1924                 ret = machine__process_cgroup_event(machine, event, sample); break;
1925         case PERF_RECORD_MMAP2:
1926                 ret = machine__process_mmap2_event(machine, event, sample); break;
1927         case PERF_RECORD_FORK:
1928                 ret = machine__process_fork_event(machine, event, sample); break;
1929         case PERF_RECORD_EXIT:
1930                 ret = machine__process_exit_event(machine, event, sample); break;
1931         case PERF_RECORD_LOST:
1932                 ret = machine__process_lost_event(machine, event, sample); break;
1933         case PERF_RECORD_AUX:
1934                 ret = machine__process_aux_event(machine, event); break;
1935         case PERF_RECORD_ITRACE_START:
1936                 ret = machine__process_itrace_start_event(machine, event); break;
1937         case PERF_RECORD_LOST_SAMPLES:
1938                 ret = machine__process_lost_samples_event(machine, event, sample); break;
1939         case PERF_RECORD_SWITCH:
1940         case PERF_RECORD_SWITCH_CPU_WIDE:
1941                 ret = machine__process_switch_event(machine, event); break;
1942         case PERF_RECORD_KSYMBOL:
1943                 ret = machine__process_ksymbol(machine, event, sample); break;
1944         case PERF_RECORD_BPF_EVENT:
1945                 ret = machine__process_bpf(machine, event, sample); break;
1946         case PERF_RECORD_TEXT_POKE:
1947                 ret = machine__process_text_poke(machine, event, sample); break;
1948         case PERF_RECORD_AUX_OUTPUT_HW_ID:
1949                 ret = machine__process_aux_output_hw_id_event(machine, event); break;
1950         default:
1951                 ret = -1;
1952                 break;
1953         }
1954
1955         return ret;
1956 }
1957
1958 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
1959 {
1960         return regexec(regex, sym->name, 0, NULL, 0) == 0;
1961 }
1962
1963 static void ip__resolve_ams(struct thread *thread,
1964                             struct addr_map_symbol *ams,
1965                             u64 ip)
1966 {
1967         struct addr_location al;
1968
1969         addr_location__init(&al);
1970         /*
1971          * We cannot use the header.misc hint to determine whether a
1972          * branch stack address is user, kernel, guest, hypervisor.
1973          * Branches may straddle the kernel/user/hypervisor boundaries.
1974          * Thus, we have to try consecutively until we find a match
1975          * or else, the symbol is unknown
1976          */
1977         thread__find_cpumode_addr_location(thread, ip, &al);
1978
1979         ams->addr = ip;
1980         ams->al_addr = al.addr;
1981         ams->al_level = al.level;
1982         ams->ms.maps = maps__get(al.maps);
1983         ams->ms.sym = al.sym;
1984         ams->ms.map = map__get(al.map);
1985         ams->phys_addr = 0;
1986         ams->data_page_size = 0;
1987         addr_location__exit(&al);
1988 }
1989
1990 static void ip__resolve_data(struct thread *thread,
1991                              u8 m, struct addr_map_symbol *ams,
1992                              u64 addr, u64 phys_addr, u64 daddr_page_size)
1993 {
1994         struct addr_location al;
1995
1996         addr_location__init(&al);
1997
1998         thread__find_symbol(thread, m, addr, &al);
1999
2000         ams->addr = addr;
2001         ams->al_addr = al.addr;
2002         ams->al_level = al.level;
2003         ams->ms.maps = maps__get(al.maps);
2004         ams->ms.sym = al.sym;
2005         ams->ms.map = map__get(al.map);
2006         ams->phys_addr = phys_addr;
2007         ams->data_page_size = daddr_page_size;
2008         addr_location__exit(&al);
2009 }
2010
2011 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2012                                      struct addr_location *al)
2013 {
2014         struct mem_info *mi = mem_info__new();
2015
2016         if (!mi)
2017                 return NULL;
2018
2019         ip__resolve_ams(al->thread, mem_info__iaddr(mi), sample->ip);
2020         ip__resolve_data(al->thread, al->cpumode, mem_info__daddr(mi),
2021                          sample->addr, sample->phys_addr,
2022                          sample->data_page_size);
2023         mem_info__data_src(mi)->val = sample->data_src;
2024
2025         return mi;
2026 }
2027
2028 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2029 {
2030         struct map *map = ms->map;
2031         char *srcline = NULL;
2032         struct dso *dso;
2033
2034         if (!map || callchain_param.key == CCKEY_FUNCTION)
2035                 return srcline;
2036
2037         dso = map__dso(map);
2038         srcline = srcline__tree_find(dso__srclines(dso), ip);
2039         if (!srcline) {
2040                 bool show_sym = false;
2041                 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2042
2043                 srcline = get_srcline(dso, map__rip_2objdump(map, ip),
2044                                       ms->sym, show_sym, show_addr, ip);
2045                 srcline__tree_insert(dso__srclines(dso), ip, srcline);
2046         }
2047
2048         return srcline;
2049 }
2050
2051 struct iterations {
2052         int nr_loop_iter;
2053         u64 cycles;
2054 };
2055
2056 static int add_callchain_ip(struct thread *thread,
2057                             struct callchain_cursor *cursor,
2058                             struct symbol **parent,
2059                             struct addr_location *root_al,
2060                             u8 *cpumode,
2061                             u64 ip,
2062                             bool branch,
2063                             struct branch_flags *flags,
2064                             struct iterations *iter,
2065                             u64 branch_from,
2066                             bool symbols)
2067 {
2068         struct map_symbol ms = {};
2069         struct addr_location al;
2070         int nr_loop_iter = 0, err = 0;
2071         u64 iter_cycles = 0;
2072         const char *srcline = NULL;
2073
2074         addr_location__init(&al);
2075         al.filtered = 0;
2076         al.sym = NULL;
2077         al.srcline = NULL;
2078         if (!cpumode) {
2079                 thread__find_cpumode_addr_location(thread, ip, &al);
2080         } else {
2081                 if (ip >= PERF_CONTEXT_MAX) {
2082                         switch (ip) {
2083                         case PERF_CONTEXT_HV:
2084                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2085                                 break;
2086                         case PERF_CONTEXT_KERNEL:
2087                                 *cpumode = PERF_RECORD_MISC_KERNEL;
2088                                 break;
2089                         case PERF_CONTEXT_USER:
2090                                 *cpumode = PERF_RECORD_MISC_USER;
2091                                 break;
2092                         default:
2093                                 pr_debug("invalid callchain context: "
2094                                          "%"PRId64"\n", (s64) ip);
2095                                 /*
2096                                  * It seems the callchain is corrupted.
2097                                  * Discard all.
2098                                  */
2099                                 callchain_cursor_reset(cursor);
2100                                 err = 1;
2101                                 goto out;
2102                         }
2103                         goto out;
2104                 }
2105                 if (symbols)
2106                         thread__find_symbol(thread, *cpumode, ip, &al);
2107         }
2108
2109         if (al.sym != NULL) {
2110                 if (perf_hpp_list.parent && !*parent &&
2111                     symbol__match_regex(al.sym, &parent_regex))
2112                         *parent = al.sym;
2113                 else if (have_ignore_callees && root_al &&
2114                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2115                         /* Treat this symbol as the root,
2116                            forgetting its callees. */
2117                         addr_location__copy(root_al, &al);
2118                         callchain_cursor_reset(cursor);
2119                 }
2120         }
2121
2122         if (symbol_conf.hide_unresolved && al.sym == NULL)
2123                 goto out;
2124
2125         if (iter) {
2126                 nr_loop_iter = iter->nr_loop_iter;
2127                 iter_cycles = iter->cycles;
2128         }
2129
2130         ms.maps = maps__get(al.maps);
2131         ms.map = map__get(al.map);
2132         ms.sym = al.sym;
2133         srcline = callchain_srcline(&ms, al.addr);
2134         err = callchain_cursor_append(cursor, ip, &ms,
2135                                       branch, flags, nr_loop_iter,
2136                                       iter_cycles, branch_from, srcline);
2137 out:
2138         addr_location__exit(&al);
2139         map_symbol__exit(&ms);
2140         return err;
2141 }
2142
2143 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2144                                            struct addr_location *al)
2145 {
2146         unsigned int i;
2147         const struct branch_stack *bs = sample->branch_stack;
2148         struct branch_entry *entries = perf_sample__branch_entries(sample);
2149         u64 *branch_stack_cntr = sample->branch_stack_cntr;
2150         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2151
2152         if (!bi)
2153                 return NULL;
2154
2155         for (i = 0; i < bs->nr; i++) {
2156                 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2157                 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2158                 bi[i].flags = entries[i].flags;
2159                 if (branch_stack_cntr)
2160                         bi[i].branch_stack_cntr  = branch_stack_cntr[i];
2161         }
2162         return bi;
2163 }
2164
2165 static void save_iterations(struct iterations *iter,
2166                             struct branch_entry *be, int nr)
2167 {
2168         int i;
2169
2170         iter->nr_loop_iter++;
2171         iter->cycles = 0;
2172
2173         for (i = 0; i < nr; i++)
2174                 iter->cycles += be[i].flags.cycles;
2175 }
2176
2177 #define CHASHSZ 127
2178 #define CHASHBITS 7
2179 #define NO_ENTRY 0xff
2180
2181 #define PERF_MAX_BRANCH_DEPTH 127
2182
2183 /* Remove loops. */
2184 static int remove_loops(struct branch_entry *l, int nr,
2185                         struct iterations *iter)
2186 {
2187         int i, j, off;
2188         unsigned char chash[CHASHSZ];
2189
2190         memset(chash, NO_ENTRY, sizeof(chash));
2191
2192         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2193
2194         for (i = 0; i < nr; i++) {
2195                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2196
2197                 /* no collision handling for now */
2198                 if (chash[h] == NO_ENTRY) {
2199                         chash[h] = i;
2200                 } else if (l[chash[h]].from == l[i].from) {
2201                         bool is_loop = true;
2202                         /* check if it is a real loop */
2203                         off = 0;
2204                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
2205                                 if (l[j].from != l[i + off].from) {
2206                                         is_loop = false;
2207                                         break;
2208                                 }
2209                         if (is_loop) {
2210                                 j = nr - (i + off);
2211                                 if (j > 0) {
2212                                         save_iterations(iter + i + off,
2213                                                 l + i, off);
2214
2215                                         memmove(iter + i, iter + i + off,
2216                                                 j * sizeof(*iter));
2217
2218                                         memmove(l + i, l + i + off,
2219                                                 j * sizeof(*l));
2220                                 }
2221
2222                                 nr -= off;
2223                         }
2224                 }
2225         }
2226         return nr;
2227 }
2228
2229 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2230                                        struct callchain_cursor *cursor,
2231                                        struct perf_sample *sample,
2232                                        struct symbol **parent,
2233                                        struct addr_location *root_al,
2234                                        u64 branch_from,
2235                                        bool callee, int end,
2236                                        bool symbols)
2237 {
2238         struct ip_callchain *chain = sample->callchain;
2239         u8 cpumode = PERF_RECORD_MISC_USER;
2240         int err, i;
2241
2242         if (callee) {
2243                 for (i = 0; i < end + 1; i++) {
2244                         err = add_callchain_ip(thread, cursor, parent,
2245                                                root_al, &cpumode, chain->ips[i],
2246                                                false, NULL, NULL, branch_from,
2247                                                symbols);
2248                         if (err)
2249                                 return err;
2250                 }
2251                 return 0;
2252         }
2253
2254         for (i = end; i >= 0; i--) {
2255                 err = add_callchain_ip(thread, cursor, parent,
2256                                        root_al, &cpumode, chain->ips[i],
2257                                        false, NULL, NULL, branch_from,
2258                                        symbols);
2259                 if (err)
2260                         return err;
2261         }
2262
2263         return 0;
2264 }
2265
2266 static void save_lbr_cursor_node(struct thread *thread,
2267                                  struct callchain_cursor *cursor,
2268                                  int idx)
2269 {
2270         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2271
2272         if (!lbr_stitch)
2273                 return;
2274
2275         if (cursor->pos == cursor->nr) {
2276                 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2277                 return;
2278         }
2279
2280         if (!cursor->curr)
2281                 cursor->curr = cursor->first;
2282         else
2283                 cursor->curr = cursor->curr->next;
2284
2285         map_symbol__exit(&lbr_stitch->prev_lbr_cursor[idx].ms);
2286         memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2287                sizeof(struct callchain_cursor_node));
2288         lbr_stitch->prev_lbr_cursor[idx].ms.maps = maps__get(cursor->curr->ms.maps);
2289         lbr_stitch->prev_lbr_cursor[idx].ms.map = map__get(cursor->curr->ms.map);
2290
2291         lbr_stitch->prev_lbr_cursor[idx].valid = true;
2292         cursor->pos++;
2293 }
2294
2295 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2296                                     struct callchain_cursor *cursor,
2297                                     struct perf_sample *sample,
2298                                     struct symbol **parent,
2299                                     struct addr_location *root_al,
2300                                     u64 *branch_from,
2301                                     bool callee,
2302                                     bool symbols)
2303 {
2304         struct branch_stack *lbr_stack = sample->branch_stack;
2305         struct branch_entry *entries = perf_sample__branch_entries(sample);
2306         u8 cpumode = PERF_RECORD_MISC_USER;
2307         int lbr_nr = lbr_stack->nr;
2308         struct branch_flags *flags;
2309         int err, i;
2310         u64 ip;
2311
2312         /*
2313          * The curr and pos are not used in writing session. They are cleared
2314          * in callchain_cursor_commit() when the writing session is closed.
2315          * Using curr and pos to track the current cursor node.
2316          */
2317         if (thread__lbr_stitch(thread)) {
2318                 cursor->curr = NULL;
2319                 cursor->pos = cursor->nr;
2320                 if (cursor->nr) {
2321                         cursor->curr = cursor->first;
2322                         for (i = 0; i < (int)(cursor->nr - 1); i++)
2323                                 cursor->curr = cursor->curr->next;
2324                 }
2325         }
2326
2327         if (callee) {
2328                 /* Add LBR ip from first entries.to */
2329                 ip = entries[0].to;
2330                 flags = &entries[0].flags;
2331                 *branch_from = entries[0].from;
2332                 err = add_callchain_ip(thread, cursor, parent,
2333                                        root_al, &cpumode, ip,
2334                                        true, flags, NULL,
2335                                        *branch_from, symbols);
2336                 if (err)
2337                         return err;
2338
2339                 /*
2340                  * The number of cursor node increases.
2341                  * Move the current cursor node.
2342                  * But does not need to save current cursor node for entry 0.
2343                  * It's impossible to stitch the whole LBRs of previous sample.
2344                  */
2345                 if (thread__lbr_stitch(thread) && (cursor->pos != cursor->nr)) {
2346                         if (!cursor->curr)
2347                                 cursor->curr = cursor->first;
2348                         else
2349                                 cursor->curr = cursor->curr->next;
2350                         cursor->pos++;
2351                 }
2352
2353                 /* Add LBR ip from entries.from one by one. */
2354                 for (i = 0; i < lbr_nr; i++) {
2355                         ip = entries[i].from;
2356                         flags = &entries[i].flags;
2357                         err = add_callchain_ip(thread, cursor, parent,
2358                                                root_al, &cpumode, ip,
2359                                                true, flags, NULL,
2360                                                *branch_from, symbols);
2361                         if (err)
2362                                 return err;
2363                         save_lbr_cursor_node(thread, cursor, i);
2364                 }
2365                 return 0;
2366         }
2367
2368         /* Add LBR ip from entries.from one by one. */
2369         for (i = lbr_nr - 1; i >= 0; i--) {
2370                 ip = entries[i].from;
2371                 flags = &entries[i].flags;
2372                 err = add_callchain_ip(thread, cursor, parent,
2373                                        root_al, &cpumode, ip,
2374                                        true, flags, NULL,
2375                                        *branch_from, symbols);
2376                 if (err)
2377                         return err;
2378                 save_lbr_cursor_node(thread, cursor, i);
2379         }
2380
2381         if (lbr_nr > 0) {
2382                 /* Add LBR ip from first entries.to */
2383                 ip = entries[0].to;
2384                 flags = &entries[0].flags;
2385                 *branch_from = entries[0].from;
2386                 err = add_callchain_ip(thread, cursor, parent,
2387                                 root_al, &cpumode, ip,
2388                                 true, flags, NULL,
2389                                 *branch_from, symbols);
2390                 if (err)
2391                         return err;
2392         }
2393
2394         return 0;
2395 }
2396
2397 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2398                                              struct callchain_cursor *cursor)
2399 {
2400         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2401         struct callchain_cursor_node *cnode;
2402         struct stitch_list *stitch_node;
2403         int err;
2404
2405         list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2406                 cnode = &stitch_node->cursor;
2407
2408                 err = callchain_cursor_append(cursor, cnode->ip,
2409                                               &cnode->ms,
2410                                               cnode->branch,
2411                                               &cnode->branch_flags,
2412                                               cnode->nr_loop_iter,
2413                                               cnode->iter_cycles,
2414                                               cnode->branch_from,
2415                                               cnode->srcline);
2416                 if (err)
2417                         return err;
2418         }
2419         return 0;
2420 }
2421
2422 static struct stitch_list *get_stitch_node(struct thread *thread)
2423 {
2424         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2425         struct stitch_list *stitch_node;
2426
2427         if (!list_empty(&lbr_stitch->free_lists)) {
2428                 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2429                                                struct stitch_list, node);
2430                 list_del(&stitch_node->node);
2431
2432                 return stitch_node;
2433         }
2434
2435         return malloc(sizeof(struct stitch_list));
2436 }
2437
2438 static bool has_stitched_lbr(struct thread *thread,
2439                              struct perf_sample *cur,
2440                              struct perf_sample *prev,
2441                              unsigned int max_lbr,
2442                              bool callee)
2443 {
2444         struct branch_stack *cur_stack = cur->branch_stack;
2445         struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2446         struct branch_stack *prev_stack = prev->branch_stack;
2447         struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2448         struct lbr_stitch *lbr_stitch = thread__lbr_stitch(thread);
2449         int i, j, nr_identical_branches = 0;
2450         struct stitch_list *stitch_node;
2451         u64 cur_base, distance;
2452
2453         if (!cur_stack || !prev_stack)
2454                 return false;
2455
2456         /* Find the physical index of the base-of-stack for current sample. */
2457         cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2458
2459         distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2460                                                      (max_lbr + prev_stack->hw_idx - cur_base);
2461         /* Previous sample has shorter stack. Nothing can be stitched. */
2462         if (distance + 1 > prev_stack->nr)
2463                 return false;
2464
2465         /*
2466          * Check if there are identical LBRs between two samples.
2467          * Identical LBRs must have same from, to and flags values. Also,
2468          * they have to be saved in the same LBR registers (same physical
2469          * index).
2470          *
2471          * Starts from the base-of-stack of current sample.
2472          */
2473         for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2474                 if ((prev_entries[i].from != cur_entries[j].from) ||
2475                     (prev_entries[i].to != cur_entries[j].to) ||
2476                     (prev_entries[i].flags.value != cur_entries[j].flags.value))
2477                         break;
2478                 nr_identical_branches++;
2479         }
2480
2481         if (!nr_identical_branches)
2482                 return false;
2483
2484         /*
2485          * Save the LBRs between the base-of-stack of previous sample
2486          * and the base-of-stack of current sample into lbr_stitch->lists.
2487          * These LBRs will be stitched later.
2488          */
2489         for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2490
2491                 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2492                         continue;
2493
2494                 stitch_node = get_stitch_node(thread);
2495                 if (!stitch_node)
2496                         return false;
2497
2498                 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2499                        sizeof(struct callchain_cursor_node));
2500
2501                 stitch_node->cursor.ms.maps = maps__get(lbr_stitch->prev_lbr_cursor[i].ms.maps);
2502                 stitch_node->cursor.ms.map = map__get(lbr_stitch->prev_lbr_cursor[i].ms.map);
2503
2504                 if (callee)
2505                         list_add(&stitch_node->node, &lbr_stitch->lists);
2506                 else
2507                         list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2508         }
2509
2510         return true;
2511 }
2512
2513 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2514 {
2515         if (thread__lbr_stitch(thread))
2516                 return true;
2517
2518         thread__set_lbr_stitch(thread, zalloc(sizeof(struct lbr_stitch)));
2519         if (!thread__lbr_stitch(thread))
2520                 goto err;
2521
2522         thread__lbr_stitch(thread)->prev_lbr_cursor =
2523                 calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2524         if (!thread__lbr_stitch(thread)->prev_lbr_cursor)
2525                 goto free_lbr_stitch;
2526
2527         thread__lbr_stitch(thread)->prev_lbr_cursor_size = max_lbr + 1;
2528
2529         INIT_LIST_HEAD(&thread__lbr_stitch(thread)->lists);
2530         INIT_LIST_HEAD(&thread__lbr_stitch(thread)->free_lists);
2531
2532         return true;
2533
2534 free_lbr_stitch:
2535         free(thread__lbr_stitch(thread));
2536         thread__set_lbr_stitch(thread, NULL);
2537 err:
2538         pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2539         thread__set_lbr_stitch_enable(thread, false);
2540         return false;
2541 }
2542
2543 /*
2544  * Resolve LBR callstack chain sample
2545  * Return:
2546  * 1 on success get LBR callchain information
2547  * 0 no available LBR callchain information, should try fp
2548  * negative error code on other errors.
2549  */
2550 static int resolve_lbr_callchain_sample(struct thread *thread,
2551                                         struct callchain_cursor *cursor,
2552                                         struct perf_sample *sample,
2553                                         struct symbol **parent,
2554                                         struct addr_location *root_al,
2555                                         int max_stack,
2556                                         unsigned int max_lbr,
2557                                         bool symbols)
2558 {
2559         bool callee = (callchain_param.order == ORDER_CALLEE);
2560         struct ip_callchain *chain = sample->callchain;
2561         int chain_nr = min(max_stack, (int)chain->nr), i;
2562         struct lbr_stitch *lbr_stitch;
2563         bool stitched_lbr = false;
2564         u64 branch_from = 0;
2565         int err;
2566
2567         for (i = 0; i < chain_nr; i++) {
2568                 if (chain->ips[i] == PERF_CONTEXT_USER)
2569                         break;
2570         }
2571
2572         /* LBR only affects the user callchain */
2573         if (i == chain_nr)
2574                 return 0;
2575
2576         if (thread__lbr_stitch_enable(thread) && !sample->no_hw_idx &&
2577             (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2578                 lbr_stitch = thread__lbr_stitch(thread);
2579
2580                 stitched_lbr = has_stitched_lbr(thread, sample,
2581                                                 &lbr_stitch->prev_sample,
2582                                                 max_lbr, callee);
2583
2584                 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2585                         struct stitch_list *stitch_node;
2586
2587                         list_for_each_entry(stitch_node, &lbr_stitch->lists, node)
2588                                 map_symbol__exit(&stitch_node->cursor.ms);
2589
2590                         list_splice_init(&lbr_stitch->lists, &lbr_stitch->free_lists);
2591                 }
2592                 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2593         }
2594
2595         if (callee) {
2596                 /* Add kernel ip */
2597                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2598                                                   parent, root_al, branch_from,
2599                                                   true, i, symbols);
2600                 if (err)
2601                         goto error;
2602
2603                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2604                                                root_al, &branch_from, true, symbols);
2605                 if (err)
2606                         goto error;
2607
2608                 if (stitched_lbr) {
2609                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2610                         if (err)
2611                                 goto error;
2612                 }
2613
2614         } else {
2615                 if (stitched_lbr) {
2616                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2617                         if (err)
2618                                 goto error;
2619                 }
2620                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2621                                                root_al, &branch_from, false, symbols);
2622                 if (err)
2623                         goto error;
2624
2625                 /* Add kernel ip */
2626                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2627                                                   parent, root_al, branch_from,
2628                                                   false, i, symbols);
2629                 if (err)
2630                         goto error;
2631         }
2632         return 1;
2633
2634 error:
2635         return (err < 0) ? err : 0;
2636 }
2637
2638 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2639                              struct callchain_cursor *cursor,
2640                              struct symbol **parent,
2641                              struct addr_location *root_al,
2642                              u8 *cpumode, int ent, bool symbols)
2643 {
2644         int err = 0;
2645
2646         while (--ent >= 0) {
2647                 u64 ip = chain->ips[ent];
2648
2649                 if (ip >= PERF_CONTEXT_MAX) {
2650                         err = add_callchain_ip(thread, cursor, parent,
2651                                                root_al, cpumode, ip,
2652                                                false, NULL, NULL, 0, symbols);
2653                         break;
2654                 }
2655         }
2656         return err;
2657 }
2658
2659 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2660                 struct thread *thread, int usr_idx)
2661 {
2662         if (machine__normalized_is(maps__machine(thread__maps(thread)), "arm64"))
2663                 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2664         else
2665                 return 0;
2666 }
2667
2668 static int thread__resolve_callchain_sample(struct thread *thread,
2669                                             struct callchain_cursor *cursor,
2670                                             struct evsel *evsel,
2671                                             struct perf_sample *sample,
2672                                             struct symbol **parent,
2673                                             struct addr_location *root_al,
2674                                             int max_stack,
2675                                             bool symbols)
2676 {
2677         struct branch_stack *branch = sample->branch_stack;
2678         struct branch_entry *entries = perf_sample__branch_entries(sample);
2679         struct ip_callchain *chain = sample->callchain;
2680         int chain_nr = 0;
2681         u8 cpumode = PERF_RECORD_MISC_USER;
2682         int i, j, err, nr_entries, usr_idx;
2683         int skip_idx = -1;
2684         int first_call = 0;
2685         u64 leaf_frame_caller;
2686
2687         if (chain)
2688                 chain_nr = chain->nr;
2689
2690         if (evsel__has_branch_callstack(evsel)) {
2691                 struct perf_env *env = evsel__env(evsel);
2692
2693                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2694                                                    root_al, max_stack,
2695                                                    !env ? 0 : env->max_branches,
2696                                                    symbols);
2697                 if (err)
2698                         return (err < 0) ? err : 0;
2699         }
2700
2701         /*
2702          * Based on DWARF debug information, some architectures skip
2703          * a callchain entry saved by the kernel.
2704          */
2705         skip_idx = arch_skip_callchain_idx(thread, chain);
2706
2707         /*
2708          * Add branches to call stack for easier browsing. This gives
2709          * more context for a sample than just the callers.
2710          *
2711          * This uses individual histograms of paths compared to the
2712          * aggregated histograms the normal LBR mode uses.
2713          *
2714          * Limitations for now:
2715          * - No extra filters
2716          * - No annotations (should annotate somehow)
2717          */
2718
2719         if (branch && callchain_param.branch_callstack) {
2720                 int nr = min(max_stack, (int)branch->nr);
2721                 struct branch_entry be[nr];
2722                 struct iterations iter[nr];
2723
2724                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2725                         pr_warning("corrupted branch chain. skipping...\n");
2726                         goto check_calls;
2727                 }
2728
2729                 for (i = 0; i < nr; i++) {
2730                         if (callchain_param.order == ORDER_CALLEE) {
2731                                 be[i] = entries[i];
2732
2733                                 if (chain == NULL)
2734                                         continue;
2735
2736                                 /*
2737                                  * Check for overlap into the callchain.
2738                                  * The return address is one off compared to
2739                                  * the branch entry. To adjust for this
2740                                  * assume the calling instruction is not longer
2741                                  * than 8 bytes.
2742                                  */
2743                                 if (i == skip_idx ||
2744                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
2745                                         first_call++;
2746                                 else if (be[i].from < chain->ips[first_call] &&
2747                                     be[i].from >= chain->ips[first_call] - 8)
2748                                         first_call++;
2749                         } else
2750                                 be[i] = entries[branch->nr - i - 1];
2751                 }
2752
2753                 memset(iter, 0, sizeof(struct iterations) * nr);
2754                 nr = remove_loops(be, nr, iter);
2755
2756                 for (i = 0; i < nr; i++) {
2757                         err = add_callchain_ip(thread, cursor, parent,
2758                                                root_al,
2759                                                NULL, be[i].to,
2760                                                true, &be[i].flags,
2761                                                NULL, be[i].from, symbols);
2762
2763                         if (!err) {
2764                                 err = add_callchain_ip(thread, cursor, parent, root_al,
2765                                                        NULL, be[i].from,
2766                                                        true, &be[i].flags,
2767                                                        &iter[i], 0, symbols);
2768                         }
2769                         if (err == -EINVAL)
2770                                 break;
2771                         if (err)
2772                                 return err;
2773                 }
2774
2775                 if (chain_nr == 0)
2776                         return 0;
2777
2778                 chain_nr -= nr;
2779         }
2780
2781 check_calls:
2782         if (chain && callchain_param.order != ORDER_CALLEE) {
2783                 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2784                                         &cpumode, chain->nr - first_call, symbols);
2785                 if (err)
2786                         return (err < 0) ? err : 0;
2787         }
2788         for (i = first_call, nr_entries = 0;
2789              i < chain_nr && nr_entries < max_stack; i++) {
2790                 u64 ip;
2791
2792                 if (callchain_param.order == ORDER_CALLEE)
2793                         j = i;
2794                 else
2795                         j = chain->nr - i - 1;
2796
2797 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2798                 if (j == skip_idx)
2799                         continue;
2800 #endif
2801                 ip = chain->ips[j];
2802                 if (ip < PERF_CONTEXT_MAX)
2803                        ++nr_entries;
2804                 else if (callchain_param.order != ORDER_CALLEE) {
2805                         err = find_prev_cpumode(chain, thread, cursor, parent,
2806                                                 root_al, &cpumode, j, symbols);
2807                         if (err)
2808                                 return (err < 0) ? err : 0;
2809                         continue;
2810                 }
2811
2812                 /*
2813                  * PERF_CONTEXT_USER allows us to locate where the user stack ends.
2814                  * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
2815                  * the index will be different in order to add the missing frame
2816                  * at the right place.
2817                  */
2818
2819                 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
2820
2821                 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
2822
2823                         leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
2824
2825                         /*
2826                          * check if leaf_frame_Caller != ip to not add the same
2827                          * value twice.
2828                          */
2829
2830                         if (leaf_frame_caller && leaf_frame_caller != ip) {
2831
2832                                 err = add_callchain_ip(thread, cursor, parent,
2833                                                 root_al, &cpumode, leaf_frame_caller,
2834                                                 false, NULL, NULL, 0, symbols);
2835                                 if (err)
2836                                         return (err < 0) ? err : 0;
2837                         }
2838                 }
2839
2840                 err = add_callchain_ip(thread, cursor, parent,
2841                                        root_al, &cpumode, ip,
2842                                        false, NULL, NULL, 0, symbols);
2843
2844                 if (err)
2845                         return (err < 0) ? err : 0;
2846         }
2847
2848         return 0;
2849 }
2850
2851 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
2852 {
2853         struct symbol *sym = ms->sym;
2854         struct map *map = ms->map;
2855         struct inline_node *inline_node;
2856         struct inline_list *ilist;
2857         struct dso *dso;
2858         u64 addr;
2859         int ret = 1;
2860         struct map_symbol ilist_ms;
2861
2862         if (!symbol_conf.inline_name || !map || !sym)
2863                 return ret;
2864
2865         addr = map__dso_map_ip(map, ip);
2866         addr = map__rip_2objdump(map, addr);
2867         dso = map__dso(map);
2868
2869         inline_node = inlines__tree_find(dso__inlined_nodes(dso), addr);
2870         if (!inline_node) {
2871                 inline_node = dso__parse_addr_inlines(dso, addr, sym);
2872                 if (!inline_node)
2873                         return ret;
2874                 inlines__tree_insert(dso__inlined_nodes(dso), inline_node);
2875         }
2876
2877         ilist_ms = (struct map_symbol) {
2878                 .maps = maps__get(ms->maps),
2879                 .map = map__get(map),
2880         };
2881         list_for_each_entry(ilist, &inline_node->val, list) {
2882                 ilist_ms.sym = ilist->symbol;
2883                 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
2884                                               NULL, 0, 0, 0, ilist->srcline);
2885
2886                 if (ret != 0)
2887                         return ret;
2888         }
2889         map_symbol__exit(&ilist_ms);
2890
2891         return ret;
2892 }
2893
2894 static int unwind_entry(struct unwind_entry *entry, void *arg)
2895 {
2896         struct callchain_cursor *cursor = arg;
2897         const char *srcline = NULL;
2898         u64 addr = entry->ip;
2899
2900         if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
2901                 return 0;
2902
2903         if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
2904                 return 0;
2905
2906         /*
2907          * Convert entry->ip from a virtual address to an offset in
2908          * its corresponding binary.
2909          */
2910         if (entry->ms.map)
2911                 addr = map__dso_map_ip(entry->ms.map, entry->ip);
2912
2913         srcline = callchain_srcline(&entry->ms, addr);
2914         return callchain_cursor_append(cursor, entry->ip, &entry->ms,
2915                                        false, NULL, 0, 0, 0, srcline);
2916 }
2917
2918 static int thread__resolve_callchain_unwind(struct thread *thread,
2919                                             struct callchain_cursor *cursor,
2920                                             struct evsel *evsel,
2921                                             struct perf_sample *sample,
2922                                             int max_stack, bool symbols)
2923 {
2924         /* Can we do dwarf post unwind? */
2925         if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
2926               (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
2927                 return 0;
2928
2929         /* Bail out if nothing was captured. */
2930         if ((!sample->user_regs.regs) ||
2931             (!sample->user_stack.size))
2932                 return 0;
2933
2934         if (!symbols)
2935                 pr_debug("Not resolving symbols with an unwinder isn't currently supported\n");
2936
2937         return unwind__get_entries(unwind_entry, cursor,
2938                                    thread, sample, max_stack, false);
2939 }
2940
2941 int __thread__resolve_callchain(struct thread *thread,
2942                                 struct callchain_cursor *cursor,
2943                                 struct evsel *evsel,
2944                                 struct perf_sample *sample,
2945                                 struct symbol **parent,
2946                                 struct addr_location *root_al,
2947                                 int max_stack,
2948                                 bool symbols)
2949 {
2950         int ret = 0;
2951
2952         if (cursor == NULL)
2953                 return -ENOMEM;
2954
2955         callchain_cursor_reset(cursor);
2956
2957         if (callchain_param.order == ORDER_CALLEE) {
2958                 ret = thread__resolve_callchain_sample(thread, cursor,
2959                                                        evsel, sample,
2960                                                        parent, root_al,
2961                                                        max_stack, symbols);
2962                 if (ret)
2963                         return ret;
2964                 ret = thread__resolve_callchain_unwind(thread, cursor,
2965                                                        evsel, sample,
2966                                                        max_stack, symbols);
2967         } else {
2968                 ret = thread__resolve_callchain_unwind(thread, cursor,
2969                                                        evsel, sample,
2970                                                        max_stack, symbols);
2971                 if (ret)
2972                         return ret;
2973                 ret = thread__resolve_callchain_sample(thread, cursor,
2974                                                        evsel, sample,
2975                                                        parent, root_al,
2976                                                        max_stack, symbols);
2977         }
2978
2979         return ret;
2980 }
2981
2982 int machine__for_each_thread(struct machine *machine,
2983                              int (*fn)(struct thread *thread, void *p),
2984                              void *priv)
2985 {
2986         return threads__for_each_thread(&machine->threads, fn, priv);
2987 }
2988
2989 int machines__for_each_thread(struct machines *machines,
2990                               int (*fn)(struct thread *thread, void *p),
2991                               void *priv)
2992 {
2993         struct rb_node *nd;
2994         int rc = 0;
2995
2996         rc = machine__for_each_thread(&machines->host, fn, priv);
2997         if (rc != 0)
2998                 return rc;
2999
3000         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3001                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3002
3003                 rc = machine__for_each_thread(machine, fn, priv);
3004                 if (rc != 0)
3005                         return rc;
3006         }
3007         return rc;
3008 }
3009
3010
3011 static int thread_list_cb(struct thread *thread, void *data)
3012 {
3013         struct list_head *list = data;
3014         struct thread_list *entry = malloc(sizeof(*entry));
3015
3016         if (!entry)
3017                 return -ENOMEM;
3018
3019         entry->thread = thread__get(thread);
3020         list_add_tail(&entry->list, list);
3021         return 0;
3022 }
3023
3024 int machine__thread_list(struct machine *machine, struct list_head *list)
3025 {
3026         return machine__for_each_thread(machine, thread_list_cb, list);
3027 }
3028
3029 void thread_list__delete(struct list_head *list)
3030 {
3031         struct thread_list *pos, *next;
3032
3033         list_for_each_entry_safe(pos, next, list, list) {
3034                 thread__zput(pos->thread);
3035                 list_del(&pos->list);
3036                 free(pos);
3037         }
3038 }
3039
3040 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3041 {
3042         if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3043                 return -1;
3044
3045         return machine->current_tid[cpu];
3046 }
3047
3048 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3049                              pid_t tid)
3050 {
3051         struct thread *thread;
3052         const pid_t init_val = -1;
3053
3054         if (cpu < 0)
3055                 return -EINVAL;
3056
3057         if (realloc_array_as_needed(machine->current_tid,
3058                                     machine->current_tid_sz,
3059                                     (unsigned int)cpu,
3060                                     &init_val))
3061                 return -ENOMEM;
3062
3063         machine->current_tid[cpu] = tid;
3064
3065         thread = machine__findnew_thread(machine, pid, tid);
3066         if (!thread)
3067                 return -ENOMEM;
3068
3069         thread__set_cpu(thread, cpu);
3070         thread__put(thread);
3071
3072         return 0;
3073 }
3074
3075 /*
3076  * Compares the raw arch string. N.B. see instead perf_env__arch() or
3077  * machine__normalized_is() if a normalized arch is needed.
3078  */
3079 bool machine__is(struct machine *machine, const char *arch)
3080 {
3081         return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3082 }
3083
3084 bool machine__normalized_is(struct machine *machine, const char *arch)
3085 {
3086         return machine && !strcmp(perf_env__arch(machine->env), arch);
3087 }
3088
3089 int machine__nr_cpus_avail(struct machine *machine)
3090 {
3091         return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3092 }
3093
3094 int machine__get_kernel_start(struct machine *machine)
3095 {
3096         struct map *map = machine__kernel_map(machine);
3097         int err = 0;
3098
3099         /*
3100          * The only addresses above 2^63 are kernel addresses of a 64-bit
3101          * kernel.  Note that addresses are unsigned so that on a 32-bit system
3102          * all addresses including kernel addresses are less than 2^32.  In
3103          * that case (32-bit system), if the kernel mapping is unknown, all
3104          * addresses will be assumed to be in user space - see
3105          * machine__kernel_ip().
3106          */
3107         machine->kernel_start = 1ULL << 63;
3108         if (map) {
3109                 err = map__load(map);
3110                 /*
3111                  * On x86_64, PTI entry trampolines are less than the
3112                  * start of kernel text, but still above 2^63. So leave
3113                  * kernel_start = 1ULL << 63 for x86_64.
3114                  */
3115                 if (!err && !machine__is(machine, "x86_64"))
3116                         machine->kernel_start = map__start(map);
3117         }
3118         return err;
3119 }
3120
3121 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3122 {
3123         u8 addr_cpumode = cpumode;
3124         bool kernel_ip;
3125
3126         if (!machine->single_address_space)
3127                 goto out;
3128
3129         kernel_ip = machine__kernel_ip(machine, addr);
3130         switch (cpumode) {
3131         case PERF_RECORD_MISC_KERNEL:
3132         case PERF_RECORD_MISC_USER:
3133                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3134                                            PERF_RECORD_MISC_USER;
3135                 break;
3136         case PERF_RECORD_MISC_GUEST_KERNEL:
3137         case PERF_RECORD_MISC_GUEST_USER:
3138                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3139                                            PERF_RECORD_MISC_GUEST_USER;
3140                 break;
3141         default:
3142                 break;
3143         }
3144 out:
3145         return addr_cpumode;
3146 }
3147
3148 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename,
3149                                     const struct dso_id *id)
3150 {
3151         return dsos__findnew_id(&machine->dsos, filename, id);
3152 }
3153
3154 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3155 {
3156         return machine__findnew_dso_id(machine, filename, NULL);
3157 }
3158
3159 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3160 {
3161         struct machine *machine = vmachine;
3162         struct map *map;
3163         struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3164
3165         if (sym == NULL)
3166                 return NULL;
3167
3168         *modp = __map__is_kmodule(map) ? (char *)dso__short_name(map__dso(map)) : NULL;
3169         *addrp = map__unmap_ip(map, sym->start);
3170         return sym->name;
3171 }
3172
3173 struct machine__for_each_dso_cb_args {
3174         struct machine *machine;
3175         machine__dso_t fn;
3176         void *priv;
3177 };
3178
3179 static int machine__for_each_dso_cb(struct dso *dso, void *data)
3180 {
3181         struct machine__for_each_dso_cb_args *args = data;
3182
3183         return args->fn(dso, args->machine, args->priv);
3184 }
3185
3186 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3187 {
3188         struct machine__for_each_dso_cb_args args = {
3189                 .machine = machine,
3190                 .fn = fn,
3191                 .priv = priv,
3192         };
3193
3194         return dsos__for_each_dso(&machine->dsos, machine__for_each_dso_cb, &args);
3195 }
3196
3197 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3198 {
3199         struct maps *maps = machine__kernel_maps(machine);
3200
3201         return maps__for_each_map(maps, fn, priv);
3202 }
3203
3204 bool machine__is_lock_function(struct machine *machine, u64 addr)
3205 {
3206         if (!machine->sched.text_start) {
3207                 struct map *kmap;
3208                 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3209
3210                 if (!sym) {
3211                         /* to avoid retry */
3212                         machine->sched.text_start = 1;
3213                         return false;
3214                 }
3215
3216                 machine->sched.text_start = map__unmap_ip(kmap, sym->start);
3217
3218                 /* should not fail from here */
3219                 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3220                 machine->sched.text_end = map__unmap_ip(kmap, sym->start);
3221
3222                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3223                 machine->lock.text_start = map__unmap_ip(kmap, sym->start);
3224
3225                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3226                 machine->lock.text_end = map__unmap_ip(kmap, sym->start);
3227
3228                 sym = machine__find_kernel_symbol_by_name(machine, "__traceiter_contention_begin", &kmap);
3229                 if (sym) {
3230                         machine->traceiter.text_start = map__unmap_ip(kmap, sym->start);
3231                         machine->traceiter.text_end = map__unmap_ip(kmap, sym->end);
3232                 }
3233                 sym = machine__find_kernel_symbol_by_name(machine, "trace_contention_begin", &kmap);
3234                 if (sym) {
3235                         machine->trace.text_start = map__unmap_ip(kmap, sym->start);
3236                         machine->trace.text_end = map__unmap_ip(kmap, sym->end);
3237                 }
3238         }
3239
3240         /* failed to get kernel symbols */
3241         if (machine->sched.text_start == 1)
3242                 return false;
3243
3244         /* mutex and rwsem functions are in sched text section */
3245         if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3246                 return true;
3247
3248         /* spinlock functions are in lock text section */
3249         if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3250                 return true;
3251
3252         /* traceiter functions currently don't have their own section
3253          * but we consider them lock functions
3254          */
3255         if (machine->traceiter.text_start != 0) {
3256                 if (machine->traceiter.text_start <= addr && addr < machine->traceiter.text_end)
3257                         return true;
3258         }
3259
3260         if (machine->trace.text_start != 0) {
3261                 if (machine->trace.text_start <= addr && addr < machine->trace.text_end)
3262                         return true;
3263         }
3264
3265         return false;
3266 }
3267
3268 int machine__hit_all_dsos(struct machine *machine)
3269 {
3270         return dsos__hit_all(&machine->dsos);
3271 }