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 "path.h"
  20 #include "srcline.h"
  21 #include "symbol.h"
  22 #include "sort.h"
  23 #include "strlist.h"
  24 #include "target.h"
  25 #include "thread.h"
  26 #include "util.h"
  27 #include "vdso.h"
  28 #include <stdbool.h>
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <unistd.h>
  32 #include "unwind.h"
  33 #include "linux/hash.h"
  34 #include "asm/bug.h"
  35 #include "bpf-event.h"
  36 #include <internal/lib.h> // page_size
  37 #include "cgroup.h"
  38 #include "arm64-frame-pointer-unwind-support.h"
  39
  40 #include <linux/ctype.h>
  41 #include <symbol/kallsyms.h>
  42 #include <linux/mman.h>
  43 #include <linux/string.h>
  44 #include <linux/zalloc.h>
  45
  46 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock);
  47
  48 static struct dso *machine__kernel_dso(struct machine *machine)
  49 {
  50         return machine->vmlinux_map->dso;
  51 }
  52
  53 static void dsos__init(struct dsos *dsos)
  54 {
  55         INIT_LIST_HEAD(&dsos->head);
  56         dsos->root = RB_ROOT;
  57         init_rwsem(&dsos->lock);
  58 }
  59
  60 static void machine__threads_init(struct machine *machine)
  61 {
  62         int i;
  63
  64         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
  65                 struct threads *threads = &machine->threads[i];
  66                 threads->entries = RB_ROOT_CACHED;
  67                 init_rwsem(&threads->lock);
  68                 threads->nr = 0;
  69                 INIT_LIST_HEAD(&threads->dead);
  70                 threads->last_match = NULL;
  71         }
  72 }
  73
  74 static int machine__set_mmap_name(struct machine *machine)
  75 {
  76         if (machine__is_host(machine))
  77                 machine->mmap_name = strdup("[kernel.kallsyms]");
  78         else if (machine__is_default_guest(machine))
  79                 machine->mmap_name = strdup("[guest.kernel.kallsyms]");
  80         else if (asprintf(&machine->mmap_name, "[guest.kernel.kallsyms.%d]",
  81                           machine->pid) < 0)
  82                 machine->mmap_name = NULL;
  83
  84         return machine->mmap_name ? 0 : -ENOMEM;
  85 }
  86
  87 static void thread__set_guest_comm(struct thread *thread, pid_t pid)
  88 {
  89         char comm[64];
  90
  91         snprintf(comm, sizeof(comm), "[guest/%d]", pid);
  92         thread__set_comm(thread, comm, 0);
  93 }
  94
  95 int machine__init(struct machine *machine, const char *root_dir, pid_t pid)
  96 {
  97         int err = -ENOMEM;
  98
  99         memset(machine, 0, sizeof(*machine));
 100         machine->kmaps = maps__new(machine);
 101         if (machine->kmaps == NULL)
 102                 return -ENOMEM;
 103
 104         RB_CLEAR_NODE(&machine->rb_node);
 105         dsos__init(&machine->dsos);
 106
 107         machine__threads_init(machine);
 108
 109         machine->vdso_info = NULL;
 110         machine->env = NULL;
 111
 112         machine->pid = pid;
 113
 114         machine->id_hdr_size = 0;
 115         machine->kptr_restrict_warned = false;
 116         machine->comm_exec = false;
 117         machine->kernel_start = 0;
 118         machine->vmlinux_map = NULL;
 119
 120         machine->root_dir = strdup(root_dir);
 121         if (machine->root_dir == NULL)
 122                 goto out;
 123
 124         if (machine__set_mmap_name(machine))
 125                 goto out;
 126
 127         if (pid != HOST_KERNEL_ID) {
 128                 struct thread *thread = machine__findnew_thread(machine, -1,
 129                                                                 pid);
 130
 131                 if (thread == NULL)
 132                         goto out;
 133
 134                 thread__set_guest_comm(thread, pid);
 135                 thread__put(thread);
 136         }
 137
 138         machine->current_tid = NULL;
 139         err = 0;
 140
 141 out:
 142         if (err) {
 143                 zfree(&machine->kmaps);
 144                 zfree(&machine->root_dir);
 145                 zfree(&machine->mmap_name);
 146         }
 147         return 0;
 148 }
 149
 150 struct machine *machine__new_host(void)
 151 {
 152         struct machine *machine = malloc(sizeof(*machine));
 153
 154         if (machine != NULL) {
 155                 machine__init(machine, "", HOST_KERNEL_ID);
 156
 157                 if (machine__create_kernel_maps(machine) < 0)
 158                         goto out_delete;
 159         }
 160
 161         return machine;
 162 out_delete:
 163         free(machine);
 164         return NULL;
 165 }
 166
 167 struct machine *machine__new_kallsyms(void)
 168 {
 169         struct machine *machine = machine__new_host();
 170         /*
 171          * FIXME:
 172          * 1) We should switch to machine__load_kallsyms(), i.e. not explicitly
 173          *    ask for not using the kcore parsing code, once this one is fixed
 174          *    to create a map per module.
 175          */
 176         if (machine && machine__load_kallsyms(machine, "/proc/kallsyms") <= 0) {
 177                 machine__delete(machine);
 178                 machine = NULL;
 179         }
 180
 181         return machine;
 182 }
 183
 184 static void dsos__purge(struct dsos *dsos)
 185 {
 186         struct dso *pos, *n;
 187
 188         down_write(&dsos->lock);
 189
 190         list_for_each_entry_safe(pos, n, &dsos->head, node) {
 191                 RB_CLEAR_NODE(&pos->rb_node);
 192                 pos->root = NULL;
 193                 list_del_init(&pos->node);
 194                 dso__put(pos);
 195         }
 196
 197         up_write(&dsos->lock);
 198 }
 199
 200 static void dsos__exit(struct dsos *dsos)
 201 {
 202         dsos__purge(dsos);
 203         exit_rwsem(&dsos->lock);
 204 }
 205
 206 void machine__delete_threads(struct machine *machine)
 207 {
 208         struct rb_node *nd;
 209         int i;
 210
 211         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 212                 struct threads *threads = &machine->threads[i];
 213                 down_write(&threads->lock);
 214                 nd = rb_first_cached(&threads->entries);
 215                 while (nd) {
 216                         struct thread *t = rb_entry(nd, struct thread, rb_node);
 217
 218                         nd = rb_next(nd);
 219                         __machine__remove_thread(machine, t, false);
 220                 }
 221                 up_write(&threads->lock);
 222         }
 223 }
 224
 225 void machine__exit(struct machine *machine)
 226 {
 227         int i;
 228
 229         if (machine == NULL)
 230                 return;
 231
 232         machine__destroy_kernel_maps(machine);
 233         maps__delete(machine->kmaps);
 234         dsos__exit(&machine->dsos);
 235         machine__exit_vdso(machine);
 236         zfree(&machine->root_dir);
 237         zfree(&machine->mmap_name);
 238         zfree(&machine->current_tid);
 239         zfree(&machine->kallsyms_filename);
 240
 241         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
 242                 struct threads *threads = &machine->threads[i];
 243                 struct thread *thread, *n;
 244                 /*
 245                  * Forget about the dead, at this point whatever threads were
 246                  * left in the dead lists better have a reference count taken
 247                  * by who is using them, and then, when they drop those references
 248                  * and it finally hits zero, thread__put() will check and see that
 249                  * its not in the dead threads list and will not try to remove it
 250                  * from there, just calling thread__delete() straight away.
 251                  */
 252                 list_for_each_entry_safe(thread, n, &threads->dead, node)
 253                         list_del_init(&thread->node);
 254
 255                 exit_rwsem(&threads->lock);
 256         }
 257 }
 258
 259 void machine__delete(struct machine *machine)
 260 {
 261         if (machine) {
 262                 machine__exit(machine);
 263                 free(machine);
 264         }
 265 }
 266
 267 void machines__init(struct machines *machines)
 268 {
 269         machine__init(&machines->host, "", HOST_KERNEL_ID);
 270         machines->guests = RB_ROOT_CACHED;
 271 }
 272
 273 void machines__exit(struct machines *machines)
 274 {
 275         machine__exit(&machines->host);
 276         /* XXX exit guest */
 277 }
 278
 279 struct machine *machines__add(struct machines *machines, pid_t pid,
 280                               const char *root_dir)
 281 {
 282         struct rb_node **p = &machines->guests.rb_root.rb_node;
 283         struct rb_node *parent = NULL;
 284         struct machine *pos, *machine = malloc(sizeof(*machine));
 285         bool leftmost = true;
 286
 287         if (machine == NULL)
 288                 return NULL;
 289
 290         if (machine__init(machine, root_dir, pid) != 0) {
 291                 free(machine);
 292                 return NULL;
 293         }
 294
 295         while (*p != NULL) {
 296                 parent = *p;
 297                 pos = rb_entry(parent, struct machine, rb_node);
 298                 if (pid < pos->pid)
 299                         p = &(*p)->rb_left;
 300                 else {
 301                         p = &(*p)->rb_right;
 302                         leftmost = false;
 303                 }
 304         }
 305
 306         rb_link_node(&machine->rb_node, parent, p);
 307         rb_insert_color_cached(&machine->rb_node, &machines->guests, leftmost);
 308
 309         machine->machines = machines;
 310
 311         return machine;
 312 }
 313
 314 void machines__set_comm_exec(struct machines *machines, bool comm_exec)
 315 {
 316         struct rb_node *nd;
 317
 318         machines->host.comm_exec = comm_exec;
 319
 320         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 321                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
 322
 323                 machine->comm_exec = comm_exec;
 324         }
 325 }
 326
 327 struct machine *machines__find(struct machines *machines, pid_t pid)
 328 {
 329         struct rb_node **p = &machines->guests.rb_root.rb_node;
 330         struct rb_node *parent = NULL;
 331         struct machine *machine;
 332         struct machine *default_machine = NULL;
 333
 334         if (pid == HOST_KERNEL_ID)
 335                 return &machines->host;
 336
 337         while (*p != NULL) {
 338                 parent = *p;
 339                 machine = rb_entry(parent, struct machine, rb_node);
 340                 if (pid < machine->pid)
 341                         p = &(*p)->rb_left;
 342                 else if (pid > machine->pid)
 343                         p = &(*p)->rb_right;
 344                 else
 345                         return machine;
 346                 if (!machine->pid)
 347                         default_machine = machine;
 348         }
 349
 350         return default_machine;
 351 }
 352
 353 struct machine *machines__findnew(struct machines *machines, pid_t pid)
 354 {
 355         char path[PATH_MAX];
 356         const char *root_dir = "";
 357         struct machine *machine = machines__find(machines, pid);
 358
 359         if (machine && (machine->pid == pid))
 360                 goto out;
 361
 362         if ((pid != HOST_KERNEL_ID) &&
 363             (pid != DEFAULT_GUEST_KERNEL_ID) &&
 364             (symbol_conf.guestmount)) {
 365                 sprintf(path, "%s/%d", symbol_conf.guestmount, pid);
 366                 if (access(path, R_OK)) {
 367                         static struct strlist *seen;
 368
 369                         if (!seen)
 370                                 seen = strlist__new(NULL, NULL);
 371
 372                         if (!strlist__has_entry(seen, path)) {
 373                                 pr_err("Can't access file %s\n", path);
 374                                 strlist__add(seen, path);
 375                         }
 376                         machine = NULL;
 377                         goto out;
 378                 }
 379                 root_dir = path;
 380         }
 381
 382         machine = machines__add(machines, pid, root_dir);
 383 out:
 384         return machine;
 385 }
 386
 387 struct machine *machines__find_guest(struct machines *machines, pid_t pid)
 388 {
 389         struct machine *machine = machines__find(machines, pid);
 390
 391         if (!machine)
 392                 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
 393         return machine;
 394 }
 395
 396 /*
 397  * A common case for KVM test programs is that the test program acts as the
 398  * hypervisor, creating, running and destroying the virtual machine, and
 399  * providing the guest object code from its own object code. In this case,
 400  * the VM is not running an OS, but only the functions loaded into it by the
 401  * hypervisor test program, and conveniently, loaded at the same virtual
 402  * addresses.
 403  *
 404  * Normally to resolve addresses, MMAP events are needed to map addresses
 405  * back to the object code and debug symbols for that object code.
 406  *
 407  * Currently, there is no way to get such mapping information from guests
 408  * but, in the scenario described above, the guest has the same mappings
 409  * as the hypervisor, so support for that scenario can be achieved.
 410  *
 411  * To support that, copy the host thread's maps to the guest thread's maps.
 412  * Note, we do not discover the guest until we encounter a guest event,
 413  * which works well because it is not until then that we know that the host
 414  * thread's maps have been set up.
 415  *
 416  * This function returns the guest thread. Apart from keeping the data
 417  * structures sane, using a thread belonging to the guest machine, instead
 418  * of the host thread, allows it to have its own comm (refer
 419  * thread__set_guest_comm()).
 420  */
 421 static struct thread *findnew_guest_code(struct machine *machine,
 422                                          struct machine *host_machine,
 423                                          pid_t pid)
 424 {
 425         struct thread *host_thread;
 426         struct thread *thread;
 427         int err;
 428
 429         if (!machine)
 430                 return NULL;
 431
 432         thread = machine__findnew_thread(machine, -1, pid);
 433         if (!thread)
 434                 return NULL;
 435
 436         /* Assume maps are set up if there are any */
 437         if (thread->maps->nr_maps)
 438                 return thread;
 439
 440         host_thread = machine__find_thread(host_machine, -1, pid);
 441         if (!host_thread)
 442                 goto out_err;
 443
 444         thread__set_guest_comm(thread, pid);
 445
 446         /*
 447          * Guest code can be found in hypervisor process at the same address
 448          * so copy host maps.
 449          */
 450         err = maps__clone(thread, host_thread->maps);
 451         thread__put(host_thread);
 452         if (err)
 453                 goto out_err;
 454
 455         return thread;
 456
 457 out_err:
 458         thread__zput(thread);
 459         return NULL;
 460 }
 461
 462 struct thread *machines__findnew_guest_code(struct machines *machines, pid_t pid)
 463 {
 464         struct machine *host_machine = machines__find(machines, HOST_KERNEL_ID);
 465         struct machine *machine = machines__findnew(machines, pid);
 466
 467         return findnew_guest_code(machine, host_machine, pid);
 468 }
 469
 470 struct thread *machine__findnew_guest_code(struct machine *machine, pid_t pid)
 471 {
 472         struct machines *machines = machine->machines;
 473         struct machine *host_machine;
 474
 475         if (!machines)
 476                 return NULL;
 477
 478         host_machine = machines__find(machines, HOST_KERNEL_ID);
 479
 480         return findnew_guest_code(machine, host_machine, pid);
 481 }
 482
 483 void machines__process_guests(struct machines *machines,
 484                               machine__process_t process, void *data)
 485 {
 486         struct rb_node *nd;
 487
 488         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
 489                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
 490                 process(pos, data);
 491         }
 492 }
 493
 494 void machines__set_id_hdr_size(struct machines *machines, u16 id_hdr_size)
 495 {
 496         struct rb_node *node;
 497         struct machine *machine;
 498
 499         machines->host.id_hdr_size = id_hdr_size;
 500
 501         for (node = rb_first_cached(&machines->guests); node;
 502              node = rb_next(node)) {
 503                 machine = rb_entry(node, struct machine, rb_node);
 504                 machine->id_hdr_size = id_hdr_size;
 505         }
 506
 507         return;
 508 }
 509
 510 static void machine__update_thread_pid(struct machine *machine,
 511                                        struct thread *th, pid_t pid)
 512 {
 513         struct thread *leader;
 514
 515         if (pid == th->pid_ || pid == -1 || th->pid_ != -1)
 516                 return;
 517
 518         th->pid_ = pid;
 519
 520         if (th->pid_ == th->tid)
 521                 return;
 522
 523         leader = __machine__findnew_thread(machine, th->pid_, th->pid_);
 524         if (!leader)
 525                 goto out_err;
 526
 527         if (!leader->maps)
 528                 leader->maps = maps__new(machine);
 529
 530         if (!leader->maps)
 531                 goto out_err;
 532
 533         if (th->maps == leader->maps)
 534                 return;
 535
 536         if (th->maps) {
 537                 /*
 538                  * Maps are created from MMAP events which provide the pid and
 539                  * tid.  Consequently there never should be any maps on a thread
 540                  * with an unknown pid.  Just print an error if there are.
 541                  */
 542                 if (!maps__empty(th->maps))
 543                         pr_err("Discarding thread maps for %d:%d\n",
 544                                th->pid_, th->tid);
 545                 maps__put(th->maps);
 546         }
 547
 548         th->maps = maps__get(leader->maps);
 549 out_put:
 550         thread__put(leader);
 551         return;
 552 out_err:
 553         pr_err("Failed to join map groups for %d:%d\n", th->pid_, th->tid);
 554         goto out_put;
 555 }
 556
 557 /*
 558  * Front-end cache - TID lookups come in blocks,
 559  * so most of the time we dont have to look up
 560  * the full rbtree:
 561  */
 562 static struct thread*
 563 __threads__get_last_match(struct threads *threads, struct machine *machine,
 564                           int pid, int tid)
 565 {
 566         struct thread *th;
 567
 568         th = threads->last_match;
 569         if (th != NULL) {
 570                 if (th->tid == tid) {
 571                         machine__update_thread_pid(machine, th, pid);
 572                         return thread__get(th);
 573                 }
 574
 575                 threads->last_match = NULL;
 576         }
 577
 578         return NULL;
 579 }
 580
 581 static struct thread*
 582 threads__get_last_match(struct threads *threads, struct machine *machine,
 583                         int pid, int tid)
 584 {
 585         struct thread *th = NULL;
 586
 587         if (perf_singlethreaded)
 588                 th = __threads__get_last_match(threads, machine, pid, tid);
 589
 590         return th;
 591 }
 592
 593 static void
 594 __threads__set_last_match(struct threads *threads, struct thread *th)
 595 {
 596         threads->last_match = th;
 597 }
 598
 599 static void
 600 threads__set_last_match(struct threads *threads, struct thread *th)
 601 {
 602         if (perf_singlethreaded)
 603                 __threads__set_last_match(threads, th);
 604 }
 605
 606 /*
 607  * Caller must eventually drop thread->refcnt returned with a successful
 608  * lookup/new thread inserted.
 609  */
 610 static struct thread *____machine__findnew_thread(struct machine *machine,
 611                                                   struct threads *threads,
 612                                                   pid_t pid, pid_t tid,
 613                                                   bool create)
 614 {
 615         struct rb_node **p = &threads->entries.rb_root.rb_node;
 616         struct rb_node *parent = NULL;
 617         struct thread *th;
 618         bool leftmost = true;
 619
 620         th = threads__get_last_match(threads, machine, pid, tid);
 621         if (th)
 622                 return th;
 623
 624         while (*p != NULL) {
 625                 parent = *p;
 626                 th = rb_entry(parent, struct thread, rb_node);
 627
 628                 if (th->tid == tid) {
 629                         threads__set_last_match(threads, th);
 630                         machine__update_thread_pid(machine, th, pid);
 631                         return thread__get(th);
 632                 }
 633
 634                 if (tid < th->tid)
 635                         p = &(*p)->rb_left;
 636                 else {
 637                         p = &(*p)->rb_right;
 638                         leftmost = false;
 639                 }
 640         }
 641
 642         if (!create)
 643                 return NULL;
 644
 645         th = thread__new(pid, tid);
 646         if (th != NULL) {
 647                 rb_link_node(&th->rb_node, parent, p);
 648                 rb_insert_color_cached(&th->rb_node, &threads->entries, leftmost);
 649
 650                 /*
 651                  * We have to initialize maps separately after rb tree is updated.
 652                  *
 653                  * The reason is that we call machine__findnew_thread
 654                  * within thread__init_maps to find the thread
 655                  * leader and that would screwed the rb tree.
 656                  */
 657                 if (thread__init_maps(th, machine)) {
 658                         rb_erase_cached(&th->rb_node, &threads->entries);
 659                         RB_CLEAR_NODE(&th->rb_node);
 660                         thread__put(th);
 661                         return NULL;
 662                 }
 663                 /*
 664                  * It is now in the rbtree, get a ref
 665                  */
 666                 thread__get(th);
 667                 threads__set_last_match(threads, th);
 668                 ++threads->nr;
 669         }
 670
 671         return th;
 672 }
 673
 674 struct thread *__machine__findnew_thread(struct machine *machine, pid_t pid, pid_t tid)
 675 {
 676         return ____machine__findnew_thread(machine, machine__threads(machine, tid), pid, tid, true);
 677 }
 678
 679 struct thread *machine__findnew_thread(struct machine *machine, pid_t pid,
 680                                        pid_t tid)
 681 {
 682         struct threads *threads = machine__threads(machine, tid);
 683         struct thread *th;
 684
 685         down_write(&threads->lock);
 686         th = __machine__findnew_thread(machine, pid, tid);
 687         up_write(&threads->lock);
 688         return th;
 689 }
 690
 691 struct thread *machine__find_thread(struct machine *machine, pid_t pid,
 692                                     pid_t tid)
 693 {
 694         struct threads *threads = machine__threads(machine, tid);
 695         struct thread *th;
 696
 697         down_read(&threads->lock);
 698         th =  ____machine__findnew_thread(machine, threads, pid, tid, false);
 699         up_read(&threads->lock);
 700         return th;
 701 }
 702
 703 /*
 704  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
 705  * So here a single thread is created for that, but actually there is a separate
 706  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
 707  * is only 1. That causes problems for some tools, requiring workarounds. For
 708  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
 709  */
 710 struct thread *machine__idle_thread(struct machine *machine)
 711 {
 712         struct thread *thread = machine__findnew_thread(machine, 0, 0);
 713
 714         if (!thread || thread__set_comm(thread, "swapper", 0) ||
 715             thread__set_namespaces(thread, 0, NULL))
 716                 pr_err("problem inserting idle task for machine pid %d\n", machine->pid);
 717
 718         return thread;
 719 }
 720
 721 struct comm *machine__thread_exec_comm(struct machine *machine,
 722                                        struct thread *thread)
 723 {
 724         if (machine->comm_exec)
 725                 return thread__exec_comm(thread);
 726         else
 727                 return thread__comm(thread);
 728 }
 729
 730 int machine__process_comm_event(struct machine *machine, union perf_event *event,
 731                                 struct perf_sample *sample)
 732 {
 733         struct thread *thread = machine__findnew_thread(machine,
 734                                                         event->comm.pid,
 735                                                         event->comm.tid);
 736         bool exec = event->header.misc & PERF_RECORD_MISC_COMM_EXEC;
 737         int err = 0;
 738
 739         if (exec)
 740                 machine->comm_exec = true;
 741
 742         if (dump_trace)
 743                 perf_event__fprintf_comm(event, stdout);
 744
 745         if (thread == NULL ||
 746             __thread__set_comm(thread, event->comm.comm, sample->time, exec)) {
 747                 dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n");
 748                 err = -1;
 749         }
 750
 751         thread__put(thread);
 752
 753         return err;
 754 }
 755
 756 int machine__process_namespaces_event(struct machine *machine __maybe_unused,
 757                                       union perf_event *event,
 758                                       struct perf_sample *sample __maybe_unused)
 759 {
 760         struct thread *thread = machine__findnew_thread(machine,
 761                                                         event->namespaces.pid,
 762                                                         event->namespaces.tid);
 763         int err = 0;
 764
 765         WARN_ONCE(event->namespaces.nr_namespaces > NR_NAMESPACES,
 766                   "\nWARNING: kernel seems to support more namespaces than perf"
 767                   " tool.\nTry updating the perf tool..\n\n");
 768
 769         WARN_ONCE(event->namespaces.nr_namespaces < NR_NAMESPACES,
 770                   "\nWARNING: perf tool seems to support more namespaces than"
 771                   " the kernel.\nTry updating the kernel..\n\n");
 772
 773         if (dump_trace)
 774                 perf_event__fprintf_namespaces(event, stdout);
 775
 776         if (thread == NULL ||
 777             thread__set_namespaces(thread, sample->time, &event->namespaces)) {
 778                 dump_printf("problem processing PERF_RECORD_NAMESPACES, skipping event.\n");
 779                 err = -1;
 780         }
 781
 782         thread__put(thread);
 783
 784         return err;
 785 }
 786
 787 int machine__process_cgroup_event(struct machine *machine,
 788                                   union perf_event *event,
 789                                   struct perf_sample *sample __maybe_unused)
 790 {
 791         struct cgroup *cgrp;
 792
 793         if (dump_trace)
 794                 perf_event__fprintf_cgroup(event, stdout);
 795
 796         cgrp = cgroup__findnew(machine->env, event->cgroup.id, event->cgroup.path);
 797         if (cgrp == NULL)
 798                 return -ENOMEM;
 799
 800         return 0;
 801 }
 802
 803 int machine__process_lost_event(struct machine *machine __maybe_unused,
 804                                 union perf_event *event, struct perf_sample *sample __maybe_unused)
 805 {
 806         dump_printf(": id:%" PRI_lu64 ": lost:%" PRI_lu64 "\n",
 807                     event->lost.id, event->lost.lost);
 808         return 0;
 809 }
 810
 811 int machine__process_lost_samples_event(struct machine *machine __maybe_unused,
 812                                         union perf_event *event, struct perf_sample *sample)
 813 {
 814         dump_printf(": id:%" PRIu64 ": lost samples :%" PRI_lu64 "\n",
 815                     sample->id, event->lost_samples.lost);
 816         return 0;
 817 }
 818
 819 static struct dso *machine__findnew_module_dso(struct machine *machine,
 820                                                struct kmod_path *m,
 821                                                const char *filename)
 822 {
 823         struct dso *dso;
 824
 825         down_write(&machine->dsos.lock);
 826
 827         dso = __dsos__find(&machine->dsos, m->name, true);
 828         if (!dso) {
 829                 dso = __dsos__addnew(&machine->dsos, m->name);
 830                 if (dso == NULL)
 831                         goto out_unlock;
 832
 833                 dso__set_module_info(dso, m, machine);
 834                 dso__set_long_name(dso, strdup(filename), true);
 835                 dso->kernel = DSO_SPACE__KERNEL;
 836         }
 837
 838         dso__get(dso);
 839 out_unlock:
 840         up_write(&machine->dsos.lock);
 841         return dso;
 842 }
 843
 844 int machine__process_aux_event(struct machine *machine __maybe_unused,
 845                                union perf_event *event)
 846 {
 847         if (dump_trace)
 848                 perf_event__fprintf_aux(event, stdout);
 849         return 0;
 850 }
 851
 852 int machine__process_itrace_start_event(struct machine *machine __maybe_unused,
 853                                         union perf_event *event)
 854 {
 855         if (dump_trace)
 856                 perf_event__fprintf_itrace_start(event, stdout);
 857         return 0;
 858 }
 859
 860 int machine__process_aux_output_hw_id_event(struct machine *machine __maybe_unused,
 861                                             union perf_event *event)
 862 {
 863         if (dump_trace)
 864                 perf_event__fprintf_aux_output_hw_id(event, stdout);
 865         return 0;
 866 }
 867
 868 int machine__process_switch_event(struct machine *machine __maybe_unused,
 869                                   union perf_event *event)
 870 {
 871         if (dump_trace)
 872                 perf_event__fprintf_switch(event, stdout);
 873         return 0;
 874 }
 875
 876 static int machine__process_ksymbol_register(struct machine *machine,
 877                                              union perf_event *event,
 878                                              struct perf_sample *sample __maybe_unused)
 879 {
 880         struct symbol *sym;
 881         struct map *map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 882
 883         if (!map) {
 884                 struct dso *dso = dso__new(event->ksymbol.name);
 885
 886                 if (dso) {
 887                         dso->kernel = DSO_SPACE__KERNEL;
 888                         map = map__new2(0, dso);
 889                         dso__put(dso);
 890                 }
 891
 892                 if (!dso || !map) {
 893                         return -ENOMEM;
 894                 }
 895
 896                 if (event->ksymbol.ksym_type == PERF_RECORD_KSYMBOL_TYPE_OOL) {
 897                         map->dso->binary_type = DSO_BINARY_TYPE__OOL;
 898                         map->dso->data.file_size = event->ksymbol.len;
 899                         dso__set_loaded(map->dso);
 900                 }
 901
 902                 map->start = event->ksymbol.addr;
 903                 map->end = map->start + event->ksymbol.len;
 904                 maps__insert(machine__kernel_maps(machine), map);
 905                 map__put(map);
 906                 dso__set_loaded(dso);
 907
 908                 if (is_bpf_image(event->ksymbol.name)) {
 909                         dso->binary_type = DSO_BINARY_TYPE__BPF_IMAGE;
 910                         dso__set_long_name(dso, "", false);
 911                 }
 912         }
 913
 914         sym = symbol__new(map->map_ip(map, map->start),
 915                           event->ksymbol.len,
 916                           0, 0, event->ksymbol.name);
 917         if (!sym)
 918                 return -ENOMEM;
 919         dso__insert_symbol(map->dso, sym);
 920         return 0;
 921 }
 922
 923 static int machine__process_ksymbol_unregister(struct machine *machine,
 924                                                union perf_event *event,
 925                                                struct perf_sample *sample __maybe_unused)
 926 {
 927         struct symbol *sym;
 928         struct map *map;
 929
 930         map = maps__find(machine__kernel_maps(machine), event->ksymbol.addr);
 931         if (!map)
 932                 return 0;
 933
 934         if (map != machine->vmlinux_map)
 935                 maps__remove(machine__kernel_maps(machine), map);
 936         else {
 937                 sym = dso__find_symbol(map->dso, map->map_ip(map, map->start));
 938                 if (sym)
 939                         dso__delete_symbol(map->dso, sym);
 940         }
 941
 942         return 0;
 943 }
 944
 945 int machine__process_ksymbol(struct machine *machine __maybe_unused,
 946                              union perf_event *event,
 947                              struct perf_sample *sample)
 948 {
 949         if (dump_trace)
 950                 perf_event__fprintf_ksymbol(event, stdout);
 951
 952         if (event->ksymbol.flags & PERF_RECORD_KSYMBOL_FLAGS_UNREGISTER)
 953                 return machine__process_ksymbol_unregister(machine, event,
 954                                                            sample);
 955         return machine__process_ksymbol_register(machine, event, sample);
 956 }
 957
 958 int machine__process_text_poke(struct machine *machine, union perf_event *event,
 959                                struct perf_sample *sample __maybe_unused)
 960 {
 961         struct map *map = maps__find(machine__kernel_maps(machine), event->text_poke.addr);
 962         u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 963
 964         if (dump_trace)
 965                 perf_event__fprintf_text_poke(event, machine, stdout);
 966
 967         if (!event->text_poke.new_len)
 968                 return 0;
 969
 970         if (cpumode != PERF_RECORD_MISC_KERNEL) {
 971                 pr_debug("%s: unsupported cpumode - ignoring\n", __func__);
 972                 return 0;
 973         }
 974
 975         if (map && map->dso) {
 976                 u8 *new_bytes = event->text_poke.bytes + event->text_poke.old_len;
 977                 int ret;
 978
 979                 /*
 980                  * Kernel maps might be changed when loading symbols so loading
 981                  * must be done prior to using kernel maps.
 982                  */
 983                 map__load(map);
 984                 ret = dso__data_write_cache_addr(map->dso, map, machine,
 985                                                  event->text_poke.addr,
 986                                                  new_bytes,
 987                                                  event->text_poke.new_len);
 988                 if (ret != event->text_poke.new_len)
 989                         pr_debug("Failed to write kernel text poke at %#" PRI_lx64 "\n",
 990                                  event->text_poke.addr);
 991         } else {
 992                 pr_debug("Failed to find kernel text poke address map for %#" PRI_lx64 "\n",
 993                          event->text_poke.addr);
 994         }
 995
 996         return 0;
 997 }
 998
 999 static struct map *machine__addnew_module_map(struct machine *machine, u64 start,
1000                                               const char *filename)
1001 {
1002         struct map *map = NULL;
1003         struct kmod_path m;
1004         struct dso *dso;
1005
1006         if (kmod_path__parse_name(&m, filename))
1007                 return NULL;
1008
1009         dso = machine__findnew_module_dso(machine, &m, filename);
1010         if (dso == NULL)
1011                 goto out;
1012
1013         map = map__new2(start, dso);
1014         if (map == NULL)
1015                 goto out;
1016
1017         maps__insert(machine__kernel_maps(machine), map);
1018
1019         /* Put the map here because maps__insert already got it */
1020         map__put(map);
1021 out:
1022         /* put the dso here, corresponding to  machine__findnew_module_dso */
1023         dso__put(dso);
1024         zfree(&m.name);
1025         return map;
1026 }
1027
1028 size_t machines__fprintf_dsos(struct machines *machines, FILE *fp)
1029 {
1030         struct rb_node *nd;
1031         size_t ret = __dsos__fprintf(&machines->host.dsos.head, fp);
1032
1033         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1034                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1035                 ret += __dsos__fprintf(&pos->dsos.head, fp);
1036         }
1037
1038         return ret;
1039 }
1040
1041 size_t machine__fprintf_dsos_buildid(struct machine *m, FILE *fp,
1042                                      bool (skip)(struct dso *dso, int parm), int parm)
1043 {
1044         return __dsos__fprintf_buildid(&m->dsos.head, fp, skip, parm);
1045 }
1046
1047 size_t machines__fprintf_dsos_buildid(struct machines *machines, FILE *fp,
1048                                      bool (skip)(struct dso *dso, int parm), int parm)
1049 {
1050         struct rb_node *nd;
1051         size_t ret = machine__fprintf_dsos_buildid(&machines->host, fp, skip, parm);
1052
1053         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
1054                 struct machine *pos = rb_entry(nd, struct machine, rb_node);
1055                 ret += machine__fprintf_dsos_buildid(pos, fp, skip, parm);
1056         }
1057         return ret;
1058 }
1059
1060 size_t machine__fprintf_vmlinux_path(struct machine *machine, FILE *fp)
1061 {
1062         int i;
1063         size_t printed = 0;
1064         struct dso *kdso = machine__kernel_dso(machine);
1065
1066         if (kdso->has_build_id) {
1067                 char filename[PATH_MAX];
1068                 if (dso__build_id_filename(kdso, filename, sizeof(filename),
1069                                            false))
1070                         printed += fprintf(fp, "[0] %s\n", filename);
1071         }
1072
1073         for (i = 0; i < vmlinux_path__nr_entries; ++i)
1074                 printed += fprintf(fp, "[%d] %s\n",
1075                                    i + kdso->has_build_id, vmlinux_path[i]);
1076
1077         return printed;
1078 }
1079
1080 size_t machine__fprintf(struct machine *machine, FILE *fp)
1081 {
1082         struct rb_node *nd;
1083         size_t ret;
1084         int i;
1085
1086         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
1087                 struct threads *threads = &machine->threads[i];
1088
1089                 down_read(&threads->lock);
1090
1091                 ret = fprintf(fp, "Threads: %u\n", threads->nr);
1092
1093                 for (nd = rb_first_cached(&threads->entries); nd;
1094                      nd = rb_next(nd)) {
1095                         struct thread *pos = rb_entry(nd, struct thread, rb_node);
1096
1097                         ret += thread__fprintf(pos, fp);
1098                 }
1099
1100                 up_read(&threads->lock);
1101         }
1102         return ret;
1103 }
1104
1105 static struct dso *machine__get_kernel(struct machine *machine)
1106 {
1107         const char *vmlinux_name = machine->mmap_name;
1108         struct dso *kernel;
1109
1110         if (machine__is_host(machine)) {
1111                 if (symbol_conf.vmlinux_name)
1112                         vmlinux_name = symbol_conf.vmlinux_name;
1113
1114                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1115                                                  "[kernel]", DSO_SPACE__KERNEL);
1116         } else {
1117                 if (symbol_conf.default_guest_vmlinux_name)
1118                         vmlinux_name = symbol_conf.default_guest_vmlinux_name;
1119
1120                 kernel = machine__findnew_kernel(machine, vmlinux_name,
1121                                                  "[guest.kernel]",
1122                                                  DSO_SPACE__KERNEL_GUEST);
1123         }
1124
1125         if (kernel != NULL && (!kernel->has_build_id))
1126                 dso__read_running_kernel_build_id(kernel, machine);
1127
1128         return kernel;
1129 }
1130
1131 void machine__get_kallsyms_filename(struct machine *machine, char *buf,
1132                                     size_t bufsz)
1133 {
1134         if (machine__is_default_guest(machine))
1135                 scnprintf(buf, bufsz, "%s", symbol_conf.default_guest_kallsyms);
1136         else
1137                 scnprintf(buf, bufsz, "%s/proc/kallsyms", machine->root_dir);
1138 }
1139
1140 const char *ref_reloc_sym_names[] = {"_text", "_stext", NULL};
1141
1142 /* Figure out the start address of kernel map from /proc/kallsyms.
1143  * Returns the name of the start symbol in *symbol_name. Pass in NULL as
1144  * symbol_name if it's not that important.
1145  */
1146 static int machine__get_running_kernel_start(struct machine *machine,
1147                                              const char **symbol_name,
1148                                              u64 *start, u64 *end)
1149 {
1150         char filename[PATH_MAX];
1151         int i, err = -1;
1152         const char *name;
1153         u64 addr = 0;
1154
1155         machine__get_kallsyms_filename(machine, filename, PATH_MAX);
1156
1157         if (symbol__restricted_filename(filename, "/proc/kallsyms"))
1158                 return 0;
1159
1160         for (i = 0; (name = ref_reloc_sym_names[i]) != NULL; i++) {
1161                 err = kallsyms__get_function_start(filename, name, &addr);
1162                 if (!err)
1163                         break;
1164         }
1165
1166         if (err)
1167                 return -1;
1168
1169         if (symbol_name)
1170                 *symbol_name = name;
1171
1172         *start = addr;
1173
1174         err = kallsyms__get_function_start(filename, "_etext", &addr);
1175         if (!err)
1176                 *end = addr;
1177
1178         return 0;
1179 }
1180
1181 int machine__create_extra_kernel_map(struct machine *machine,
1182                                      struct dso *kernel,
1183                                      struct extra_kernel_map *xm)
1184 {
1185         struct kmap *kmap;
1186         struct map *map;
1187
1188         map = map__new2(xm->start, kernel);
1189         if (!map)
1190                 return -1;
1191
1192         map->end   = xm->end;
1193         map->pgoff = xm->pgoff;
1194
1195         kmap = map__kmap(map);
1196
1197         strlcpy(kmap->name, xm->name, KMAP_NAME_LEN);
1198
1199         maps__insert(machine__kernel_maps(machine), map);
1200
1201         pr_debug2("Added extra kernel map %s %" PRIx64 "-%" PRIx64 "\n",
1202                   kmap->name, map->start, map->end);
1203
1204         map__put(map);
1205
1206         return 0;
1207 }
1208
1209 static u64 find_entry_trampoline(struct dso *dso)
1210 {
1211         /* Duplicates are removed so lookup all aliases */
1212         const char *syms[] = {
1213                 "_entry_trampoline",
1214                 "__entry_trampoline_start",
1215                 "entry_SYSCALL_64_trampoline",
1216         };
1217         struct symbol *sym = dso__first_symbol(dso);
1218         unsigned int i;
1219
1220         for (; sym; sym = dso__next_symbol(sym)) {
1221                 if (sym->binding != STB_GLOBAL)
1222                         continue;
1223                 for (i = 0; i < ARRAY_SIZE(syms); i++) {
1224                         if (!strcmp(sym->name, syms[i]))
1225                                 return sym->start;
1226                 }
1227         }
1228
1229         return 0;
1230 }
1231
1232 /*
1233  * These values can be used for kernels that do not have symbols for the entry
1234  * trampolines in kallsyms.
1235  */
1236 #define X86_64_CPU_ENTRY_AREA_PER_CPU   0xfffffe0000000000ULL
1237 #define X86_64_CPU_ENTRY_AREA_SIZE      0x2c000
1238 #define X86_64_ENTRY_TRAMPOLINE         0x6000
1239
1240 /* Map x86_64 PTI entry trampolines */
1241 int machine__map_x86_64_entry_trampolines(struct machine *machine,
1242                                           struct dso *kernel)
1243 {
1244         struct maps *kmaps = machine__kernel_maps(machine);
1245         int nr_cpus_avail, cpu;
1246         bool found = false;
1247         struct map *map;
1248         u64 pgoff;
1249
1250         /*
1251          * In the vmlinux case, pgoff is a virtual address which must now be
1252          * mapped to a vmlinux offset.
1253          */
1254         maps__for_each_entry(kmaps, map) {
1255                 struct kmap *kmap = __map__kmap(map);
1256                 struct map *dest_map;
1257
1258                 if (!kmap || !is_entry_trampoline(kmap->name))
1259                         continue;
1260
1261                 dest_map = maps__find(kmaps, map->pgoff);
1262                 if (dest_map != map)
1263                         map->pgoff = dest_map->map_ip(dest_map, map->pgoff);
1264                 found = true;
1265         }
1266         if (found || machine->trampolines_mapped)
1267                 return 0;
1268
1269         pgoff = find_entry_trampoline(kernel);
1270         if (!pgoff)
1271                 return 0;
1272
1273         nr_cpus_avail = machine__nr_cpus_avail(machine);
1274
1275         /* Add a 1 page map for each CPU's entry trampoline */
1276         for (cpu = 0; cpu < nr_cpus_avail; cpu++) {
1277                 u64 va = X86_64_CPU_ENTRY_AREA_PER_CPU +
1278                          cpu * X86_64_CPU_ENTRY_AREA_SIZE +
1279                          X86_64_ENTRY_TRAMPOLINE;
1280                 struct extra_kernel_map xm = {
1281                         .start = va,
1282                         .end   = va + page_size,
1283                         .pgoff = pgoff,
1284                 };
1285
1286                 strlcpy(xm.name, ENTRY_TRAMPOLINE_NAME, KMAP_NAME_LEN);
1287
1288                 if (machine__create_extra_kernel_map(machine, kernel, &xm) < 0)
1289                         return -1;
1290         }
1291
1292         machine->trampolines_mapped = nr_cpus_avail;
1293
1294         return 0;
1295 }
1296
1297 int __weak machine__create_extra_kernel_maps(struct machine *machine __maybe_unused,
1298                                              struct dso *kernel __maybe_unused)
1299 {
1300         return 0;
1301 }
1302
1303 static int
1304 __machine__create_kernel_maps(struct machine *machine, struct dso *kernel)
1305 {
1306         /* In case of renewal the kernel map, destroy previous one */
1307         machine__destroy_kernel_maps(machine);
1308
1309         machine->vmlinux_map = map__new2(0, kernel);
1310         if (machine->vmlinux_map == NULL)
1311                 return -1;
1312
1313         machine->vmlinux_map->map_ip = machine->vmlinux_map->unmap_ip = identity__map_ip;
1314         maps__insert(machine__kernel_maps(machine), machine->vmlinux_map);
1315         return 0;
1316 }
1317
1318 void machine__destroy_kernel_maps(struct machine *machine)
1319 {
1320         struct kmap *kmap;
1321         struct map *map = machine__kernel_map(machine);
1322
1323         if (map == NULL)
1324                 return;
1325
1326         kmap = map__kmap(map);
1327         maps__remove(machine__kernel_maps(machine), map);
1328         if (kmap && kmap->ref_reloc_sym) {
1329                 zfree((char **)&kmap->ref_reloc_sym->name);
1330                 zfree(&kmap->ref_reloc_sym);
1331         }
1332
1333         map__zput(machine->vmlinux_map);
1334 }
1335
1336 int machines__create_guest_kernel_maps(struct machines *machines)
1337 {
1338         int ret = 0;
1339         struct dirent **namelist = NULL;
1340         int i, items = 0;
1341         char path[PATH_MAX];
1342         pid_t pid;
1343         char *endp;
1344
1345         if (symbol_conf.default_guest_vmlinux_name ||
1346             symbol_conf.default_guest_modules ||
1347             symbol_conf.default_guest_kallsyms) {
1348                 machines__create_kernel_maps(machines, DEFAULT_GUEST_KERNEL_ID);
1349         }
1350
1351         if (symbol_conf.guestmount) {
1352                 items = scandir(symbol_conf.guestmount, &namelist, NULL, NULL);
1353                 if (items <= 0)
1354                         return -ENOENT;
1355                 for (i = 0; i < items; i++) {
1356                         if (!isdigit(namelist[i]->d_name[0])) {
1357                                 /* Filter out . and .. */
1358                                 continue;
1359                         }
1360                         pid = (pid_t)strtol(namelist[i]->d_name, &endp, 10);
1361                         if ((*endp != '\0') ||
1362                             (endp == namelist[i]->d_name) ||
1363                             (errno == ERANGE)) {
1364                                 pr_debug("invalid directory (%s). Skipping.\n",
1365                                          namelist[i]->d_name);
1366                                 continue;
1367                         }
1368                         sprintf(path, "%s/%s/proc/kallsyms",
1369                                 symbol_conf.guestmount,
1370                                 namelist[i]->d_name);
1371                         ret = access(path, R_OK);
1372                         if (ret) {
1373                                 pr_debug("Can't access file %s\n", path);
1374                                 goto failure;
1375                         }
1376                         machines__create_kernel_maps(machines, pid);
1377                 }
1378 failure:
1379                 free(namelist);
1380         }
1381
1382         return ret;
1383 }
1384
1385 void machines__destroy_kernel_maps(struct machines *machines)
1386 {
1387         struct rb_node *next = rb_first_cached(&machines->guests);
1388
1389         machine__destroy_kernel_maps(&machines->host);
1390
1391         while (next) {
1392                 struct machine *pos = rb_entry(next, struct machine, rb_node);
1393
1394                 next = rb_next(&pos->rb_node);
1395                 rb_erase_cached(&pos->rb_node, &machines->guests);
1396                 machine__delete(pos);
1397         }
1398 }
1399
1400 int machines__create_kernel_maps(struct machines *machines, pid_t pid)
1401 {
1402         struct machine *machine = machines__findnew(machines, pid);
1403
1404         if (machine == NULL)
1405                 return -1;
1406
1407         return machine__create_kernel_maps(machine);
1408 }
1409
1410 int machine__load_kallsyms(struct machine *machine, const char *filename)
1411 {
1412         struct map *map = machine__kernel_map(machine);
1413         int ret = __dso__load_kallsyms(map->dso, filename, map, true);
1414
1415         if (ret > 0) {
1416                 dso__set_loaded(map->dso);
1417                 /*
1418                  * Since /proc/kallsyms will have multiple sessions for the
1419                  * kernel, with modules between them, fixup the end of all
1420                  * sections.
1421                  */
1422                 maps__fixup_end(machine__kernel_maps(machine));
1423         }
1424
1425         return ret;
1426 }
1427
1428 int machine__load_vmlinux_path(struct machine *machine)
1429 {
1430         struct map *map = machine__kernel_map(machine);
1431         int ret = dso__load_vmlinux_path(map->dso, map);
1432
1433         if (ret > 0)
1434                 dso__set_loaded(map->dso);
1435
1436         return ret;
1437 }
1438
1439 static char *get_kernel_version(const char *root_dir)
1440 {
1441         char version[PATH_MAX];
1442         FILE *file;
1443         char *name, *tmp;
1444         const char *prefix = "Linux version ";
1445
1446         sprintf(version, "%s/proc/version", root_dir);
1447         file = fopen(version, "r");
1448         if (!file)
1449                 return NULL;
1450
1451         tmp = fgets(version, sizeof(version), file);
1452         fclose(file);
1453         if (!tmp)
1454                 return NULL;
1455
1456         name = strstr(version, prefix);
1457         if (!name)
1458                 return NULL;
1459         name += strlen(prefix);
1460         tmp = strchr(name, ' ');
1461         if (tmp)
1462                 *tmp = '\0';
1463
1464         return strdup(name);
1465 }
1466
1467 static bool is_kmod_dso(struct dso *dso)
1468 {
1469         return dso->symtab_type == DSO_BINARY_TYPE__SYSTEM_PATH_KMODULE ||
1470                dso->symtab_type == DSO_BINARY_TYPE__GUEST_KMODULE;
1471 }
1472
1473 static int maps__set_module_path(struct maps *maps, const char *path, struct kmod_path *m)
1474 {
1475         char *long_name;
1476         struct map *map = maps__find_by_name(maps, m->name);
1477
1478         if (map == NULL)
1479                 return 0;
1480
1481         long_name = strdup(path);
1482         if (long_name == NULL)
1483                 return -ENOMEM;
1484
1485         dso__set_long_name(map->dso, long_name, true);
1486         dso__kernel_module_get_build_id(map->dso, "");
1487
1488         /*
1489          * Full name could reveal us kmod compression, so
1490          * we need to update the symtab_type if needed.
1491          */
1492         if (m->comp && is_kmod_dso(map->dso)) {
1493                 map->dso->symtab_type++;
1494                 map->dso->comp = m->comp;
1495         }
1496
1497         return 0;
1498 }
1499
1500 static int maps__set_modules_path_dir(struct maps *maps, const char *dir_name, int depth)
1501 {
1502         struct dirent *dent;
1503         DIR *dir = opendir(dir_name);
1504         int ret = 0;
1505
1506         if (!dir) {
1507                 pr_debug("%s: cannot open %s dir\n", __func__, dir_name);
1508                 return -1;
1509         }
1510
1511         while ((dent = readdir(dir)) != NULL) {
1512                 char path[PATH_MAX];
1513                 struct stat st;
1514
1515                 /*sshfs might return bad dent->d_type, so we have to stat*/
1516                 path__join(path, sizeof(path), dir_name, dent->d_name);
1517                 if (stat(path, &st))
1518                         continue;
1519
1520                 if (S_ISDIR(st.st_mode)) {
1521                         if (!strcmp(dent->d_name, ".") ||
1522                             !strcmp(dent->d_name, ".."))
1523                                 continue;
1524
1525                         /* Do not follow top-level source and build symlinks */
1526                         if (depth == 0) {
1527                                 if (!strcmp(dent->d_name, "source") ||
1528                                     !strcmp(dent->d_name, "build"))
1529                                         continue;
1530                         }
1531
1532                         ret = maps__set_modules_path_dir(maps, path, depth + 1);
1533                         if (ret < 0)
1534                                 goto out;
1535                 } else {
1536                         struct kmod_path m;
1537
1538                         ret = kmod_path__parse_name(&m, dent->d_name);
1539                         if (ret)
1540                                 goto out;
1541
1542                         if (m.kmod)
1543                                 ret = maps__set_module_path(maps, path, &m);
1544
1545                         zfree(&m.name);
1546
1547                         if (ret)
1548                                 goto out;
1549                 }
1550         }
1551
1552 out:
1553         closedir(dir);
1554         return ret;
1555 }
1556
1557 static int machine__set_modules_path(struct machine *machine)
1558 {
1559         char *version;
1560         char modules_path[PATH_MAX];
1561
1562         version = get_kernel_version(machine->root_dir);
1563         if (!version)
1564                 return -1;
1565
1566         snprintf(modules_path, sizeof(modules_path), "%s/lib/modules/%s",
1567                  machine->root_dir, version);
1568         free(version);
1569
1570         return maps__set_modules_path_dir(machine__kernel_maps(machine), modules_path, 0);
1571 }
1572 int __weak arch__fix_module_text_start(u64 *start __maybe_unused,
1573                                 u64 *size __maybe_unused,
1574                                 const char *name __maybe_unused)
1575 {
1576         return 0;
1577 }
1578
1579 static int machine__create_module(void *arg, const char *name, u64 start,
1580                                   u64 size)
1581 {
1582         struct machine *machine = arg;
1583         struct map *map;
1584
1585         if (arch__fix_module_text_start(&start, &size, name) < 0)
1586                 return -1;
1587
1588         map = machine__addnew_module_map(machine, start, name);
1589         if (map == NULL)
1590                 return -1;
1591         map->end = start + size;
1592
1593         dso__kernel_module_get_build_id(map->dso, machine->root_dir);
1594
1595         return 0;
1596 }
1597
1598 static int machine__create_modules(struct machine *machine)
1599 {
1600         const char *modules;
1601         char path[PATH_MAX];
1602
1603         if (machine__is_default_guest(machine)) {
1604                 modules = symbol_conf.default_guest_modules;
1605         } else {
1606                 snprintf(path, PATH_MAX, "%s/proc/modules", machine->root_dir);
1607                 modules = path;
1608         }
1609
1610         if (symbol__restricted_filename(modules, "/proc/modules"))
1611                 return -1;
1612
1613         if (modules__parse(modules, machine, machine__create_module))
1614                 return -1;
1615
1616         if (!machine__set_modules_path(machine))
1617                 return 0;
1618
1619         pr_debug("Problems setting modules path maps, continuing anyway...\n");
1620
1621         return 0;
1622 }
1623
1624 static void machine__set_kernel_mmap(struct machine *machine,
1625                                      u64 start, u64 end)
1626 {
1627         machine->vmlinux_map->start = start;
1628         machine->vmlinux_map->end   = end;
1629         /*
1630          * Be a bit paranoid here, some perf.data file came with
1631          * a zero sized synthesized MMAP event for the kernel.
1632          */
1633         if (start == 0 && end == 0)
1634                 machine->vmlinux_map->end = ~0ULL;
1635 }
1636
1637 static void machine__update_kernel_mmap(struct machine *machine,
1638                                      u64 start, u64 end)
1639 {
1640         struct map *map = machine__kernel_map(machine);
1641
1642         map__get(map);
1643         maps__remove(machine__kernel_maps(machine), map);
1644
1645         machine__set_kernel_mmap(machine, start, end);
1646
1647         maps__insert(machine__kernel_maps(machine), map);
1648         map__put(map);
1649 }
1650
1651 int machine__create_kernel_maps(struct machine *machine)
1652 {
1653         struct dso *kernel = machine__get_kernel(machine);
1654         const char *name = NULL;
1655         struct map *map;
1656         u64 start = 0, end = ~0ULL;
1657         int ret;
1658
1659         if (kernel == NULL)
1660                 return -1;
1661
1662         ret = __machine__create_kernel_maps(machine, kernel);
1663         if (ret < 0)
1664                 goto out_put;
1665
1666         if (symbol_conf.use_modules && machine__create_modules(machine) < 0) {
1667                 if (machine__is_host(machine))
1668                         pr_debug("Problems creating module maps, "
1669                                  "continuing anyway...\n");
1670                 else
1671                         pr_debug("Problems creating module maps for guest %d, "
1672                                  "continuing anyway...\n", machine->pid);
1673         }
1674
1675         if (!machine__get_running_kernel_start(machine, &name, &start, &end)) {
1676                 if (name &&
1677                     map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map, name, start)) {
1678                         machine__destroy_kernel_maps(machine);
1679                         ret = -1;
1680                         goto out_put;
1681                 }
1682
1683                 /*
1684                  * we have a real start address now, so re-order the kmaps
1685                  * assume it's the last in the kmaps
1686                  */
1687                 machine__update_kernel_mmap(machine, start, end);
1688         }
1689
1690         if (machine__create_extra_kernel_maps(machine, kernel))
1691                 pr_debug("Problems creating extra kernel maps, continuing anyway...\n");
1692
1693         if (end == ~0ULL) {
1694                 /* update end address of the kernel map using adjacent module address */
1695                 map = map__next(machine__kernel_map(machine));
1696                 if (map)
1697                         machine__set_kernel_mmap(machine, start, map->start);
1698         }
1699
1700 out_put:
1701         dso__put(kernel);
1702         return ret;
1703 }
1704
1705 static bool machine__uses_kcore(struct machine *machine)
1706 {
1707         struct dso *dso;
1708
1709         list_for_each_entry(dso, &machine->dsos.head, node) {
1710                 if (dso__is_kcore(dso))
1711                         return true;
1712         }
1713
1714         return false;
1715 }
1716
1717 static bool perf_event__is_extra_kernel_mmap(struct machine *machine,
1718                                              struct extra_kernel_map *xm)
1719 {
1720         return machine__is(machine, "x86_64") &&
1721                is_entry_trampoline(xm->name);
1722 }
1723
1724 static int machine__process_extra_kernel_map(struct machine *machine,
1725                                              struct extra_kernel_map *xm)
1726 {
1727         struct dso *kernel = machine__kernel_dso(machine);
1728
1729         if (kernel == NULL)
1730                 return -1;
1731
1732         return machine__create_extra_kernel_map(machine, kernel, xm);
1733 }
1734
1735 static int machine__process_kernel_mmap_event(struct machine *machine,
1736                                               struct extra_kernel_map *xm,
1737                                               struct build_id *bid)
1738 {
1739         struct map *map;
1740         enum dso_space_type dso_space;
1741         bool is_kernel_mmap;
1742         const char *mmap_name = machine->mmap_name;
1743
1744         /* If we have maps from kcore then we do not need or want any others */
1745         if (machine__uses_kcore(machine))
1746                 return 0;
1747
1748         if (machine__is_host(machine))
1749                 dso_space = DSO_SPACE__KERNEL;
1750         else
1751                 dso_space = DSO_SPACE__KERNEL_GUEST;
1752
1753         is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1754         if (!is_kernel_mmap && !machine__is_host(machine)) {
1755                 /*
1756                  * If the event was recorded inside the guest and injected into
1757                  * the host perf.data file, then it will match a host mmap_name,
1758                  * so try that - see machine__set_mmap_name().
1759                  */
1760                 mmap_name = "[kernel.kallsyms]";
1761                 is_kernel_mmap = memcmp(xm->name, mmap_name, strlen(mmap_name) - 1) == 0;
1762         }
1763         if (xm->name[0] == '/' ||
1764             (!is_kernel_mmap && xm->name[0] == '[')) {
1765                 map = machine__addnew_module_map(machine, xm->start,
1766                                                  xm->name);
1767                 if (map == NULL)
1768                         goto out_problem;
1769
1770                 map->end = map->start + xm->end - xm->start;
1771
1772                 if (build_id__is_defined(bid))
1773                         dso__set_build_id(map->dso, bid);
1774
1775         } else if (is_kernel_mmap) {
1776                 const char *symbol_name = xm->name + strlen(mmap_name);
1777                 /*
1778                  * Should be there already, from the build-id table in
1779                  * the header.
1780                  */
1781                 struct dso *kernel = NULL;
1782                 struct dso *dso;
1783
1784                 down_read(&machine->dsos.lock);
1785
1786                 list_for_each_entry(dso, &machine->dsos.head, node) {
1787
1788                         /*
1789                          * The cpumode passed to is_kernel_module is not the
1790                          * cpumode of *this* event. If we insist on passing
1791                          * correct cpumode to is_kernel_module, we should
1792                          * record the cpumode when we adding this dso to the
1793                          * linked list.
1794                          *
1795                          * However we don't really need passing correct
1796                          * cpumode.  We know the correct cpumode must be kernel
1797                          * mode (if not, we should not link it onto kernel_dsos
1798                          * list).
1799                          *
1800                          * Therefore, we pass PERF_RECORD_MISC_CPUMODE_UNKNOWN.
1801                          * is_kernel_module() treats it as a kernel cpumode.
1802                          */
1803
1804                         if (!dso->kernel ||
1805                             is_kernel_module(dso->long_name,
1806                                              PERF_RECORD_MISC_CPUMODE_UNKNOWN))
1807                                 continue;
1808
1809
1810                         kernel = dso;
1811                         break;
1812                 }
1813
1814                 up_read(&machine->dsos.lock);
1815
1816                 if (kernel == NULL)
1817                         kernel = machine__findnew_dso(machine, machine->mmap_name);
1818                 if (kernel == NULL)
1819                         goto out_problem;
1820
1821                 kernel->kernel = dso_space;
1822                 if (__machine__create_kernel_maps(machine, kernel) < 0) {
1823                         dso__put(kernel);
1824                         goto out_problem;
1825                 }
1826
1827                 if (strstr(kernel->long_name, "vmlinux"))
1828                         dso__set_short_name(kernel, "[kernel.vmlinux]", false);
1829
1830                 machine__update_kernel_mmap(machine, xm->start, xm->end);
1831
1832                 if (build_id__is_defined(bid))
1833                         dso__set_build_id(kernel, bid);
1834
1835                 /*
1836                  * Avoid using a zero address (kptr_restrict) for the ref reloc
1837                  * symbol. Effectively having zero here means that at record
1838                  * time /proc/sys/kernel/kptr_restrict was non zero.
1839                  */
1840                 if (xm->pgoff != 0) {
1841                         map__set_kallsyms_ref_reloc_sym(machine->vmlinux_map,
1842                                                         symbol_name,
1843                                                         xm->pgoff);
1844                 }
1845
1846                 if (machine__is_default_guest(machine)) {
1847                         /*
1848                          * preload dso of guest kernel and modules
1849                          */
1850                         dso__load(kernel, machine__kernel_map(machine));
1851                 }
1852         } else if (perf_event__is_extra_kernel_mmap(machine, xm)) {
1853                 return machine__process_extra_kernel_map(machine, xm);
1854         }
1855         return 0;
1856 out_problem:
1857         return -1;
1858 }
1859
1860 int machine__process_mmap2_event(struct machine *machine,
1861                                  union perf_event *event,
1862                                  struct perf_sample *sample)
1863 {
1864         struct thread *thread;
1865         struct map *map;
1866         struct dso_id dso_id = {
1867                 .maj = event->mmap2.maj,
1868                 .min = event->mmap2.min,
1869                 .ino = event->mmap2.ino,
1870                 .ino_generation = event->mmap2.ino_generation,
1871         };
1872         struct build_id __bid, *bid = NULL;
1873         int ret = 0;
1874
1875         if (dump_trace)
1876                 perf_event__fprintf_mmap2(event, stdout);
1877
1878         if (event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID) {
1879                 bid = &__bid;
1880                 build_id__init(bid, event->mmap2.build_id, event->mmap2.build_id_size);
1881         }
1882
1883         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1884             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1885                 struct extra_kernel_map xm = {
1886                         .start = event->mmap2.start,
1887                         .end   = event->mmap2.start + event->mmap2.len,
1888                         .pgoff = event->mmap2.pgoff,
1889                 };
1890
1891                 strlcpy(xm.name, event->mmap2.filename, KMAP_NAME_LEN);
1892                 ret = machine__process_kernel_mmap_event(machine, &xm, bid);
1893                 if (ret < 0)
1894                         goto out_problem;
1895                 return 0;
1896         }
1897
1898         thread = machine__findnew_thread(machine, event->mmap2.pid,
1899                                         event->mmap2.tid);
1900         if (thread == NULL)
1901                 goto out_problem;
1902
1903         map = map__new(machine, event->mmap2.start,
1904                         event->mmap2.len, event->mmap2.pgoff,
1905                         &dso_id, event->mmap2.prot,
1906                         event->mmap2.flags, bid,
1907                         event->mmap2.filename, thread);
1908
1909         if (map == NULL)
1910                 goto out_problem_map;
1911
1912         ret = thread__insert_map(thread, map);
1913         if (ret)
1914                 goto out_problem_insert;
1915
1916         thread__put(thread);
1917         map__put(map);
1918         return 0;
1919
1920 out_problem_insert:
1921         map__put(map);
1922 out_problem_map:
1923         thread__put(thread);
1924 out_problem:
1925         dump_printf("problem processing PERF_RECORD_MMAP2, skipping event.\n");
1926         return 0;
1927 }
1928
1929 int machine__process_mmap_event(struct machine *machine, union perf_event *event,
1930                                 struct perf_sample *sample)
1931 {
1932         struct thread *thread;
1933         struct map *map;
1934         u32 prot = 0;
1935         int ret = 0;
1936
1937         if (dump_trace)
1938                 perf_event__fprintf_mmap(event, stdout);
1939
1940         if (sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL ||
1941             sample->cpumode == PERF_RECORD_MISC_KERNEL) {
1942                 struct extra_kernel_map xm = {
1943                         .start = event->mmap.start,
1944                         .end   = event->mmap.start + event->mmap.len,
1945                         .pgoff = event->mmap.pgoff,
1946                 };
1947
1948                 strlcpy(xm.name, event->mmap.filename, KMAP_NAME_LEN);
1949                 ret = machine__process_kernel_mmap_event(machine, &xm, NULL);
1950                 if (ret < 0)
1951                         goto out_problem;
1952                 return 0;
1953         }
1954
1955         thread = machine__findnew_thread(machine, event->mmap.pid,
1956                                          event->mmap.tid);
1957         if (thread == NULL)
1958                 goto out_problem;
1959
1960         if (!(event->header.misc & PERF_RECORD_MISC_MMAP_DATA))
1961                 prot = PROT_EXEC;
1962
1963         map = map__new(machine, event->mmap.start,
1964                         event->mmap.len, event->mmap.pgoff,
1965                         NULL, prot, 0, NULL, event->mmap.filename, thread);
1966
1967         if (map == NULL)
1968                 goto out_problem_map;
1969
1970         ret = thread__insert_map(thread, map);
1971         if (ret)
1972                 goto out_problem_insert;
1973
1974         thread__put(thread);
1975         map__put(map);
1976         return 0;
1977
1978 out_problem_insert:
1979         map__put(map);
1980 out_problem_map:
1981         thread__put(thread);
1982 out_problem:
1983         dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n");
1984         return 0;
1985 }
1986
1987 static void __machine__remove_thread(struct machine *machine, struct thread *th, bool lock)
1988 {
1989         struct threads *threads = machine__threads(machine, th->tid);
1990
1991         if (threads->last_match == th)
1992                 threads__set_last_match(threads, NULL);
1993
1994         if (lock)
1995                 down_write(&threads->lock);
1996
1997         BUG_ON(refcount_read(&th->refcnt) == 0);
1998
1999         rb_erase_cached(&th->rb_node, &threads->entries);
2000         RB_CLEAR_NODE(&th->rb_node);
2001         --threads->nr;
2002         /*
2003          * Move it first to the dead_threads list, then drop the reference,
2004          * if this is the last reference, then the thread__delete destructor
2005          * will be called and we will remove it from the dead_threads list.
2006          */
2007         list_add_tail(&th->node, &threads->dead);
2008
2009         /*
2010          * We need to do the put here because if this is the last refcount,
2011          * then we will be touching the threads->dead head when removing the
2012          * thread.
2013          */
2014         thread__put(th);
2015
2016         if (lock)
2017                 up_write(&threads->lock);
2018 }
2019
2020 void machine__remove_thread(struct machine *machine, struct thread *th)
2021 {
2022         return __machine__remove_thread(machine, th, true);
2023 }
2024
2025 int machine__process_fork_event(struct machine *machine, union perf_event *event,
2026                                 struct perf_sample *sample)
2027 {
2028         struct thread *thread = machine__find_thread(machine,
2029                                                      event->fork.pid,
2030                                                      event->fork.tid);
2031         struct thread *parent = machine__findnew_thread(machine,
2032                                                         event->fork.ppid,
2033                                                         event->fork.ptid);
2034         bool do_maps_clone = true;
2035         int err = 0;
2036
2037         if (dump_trace)
2038                 perf_event__fprintf_task(event, stdout);
2039
2040         /*
2041          * There may be an existing thread that is not actually the parent,
2042          * either because we are processing events out of order, or because the
2043          * (fork) event that would have removed the thread was lost. Assume the
2044          * latter case and continue on as best we can.
2045          */
2046         if (parent->pid_ != (pid_t)event->fork.ppid) {
2047                 dump_printf("removing erroneous parent thread %d/%d\n",
2048                             parent->pid_, parent->tid);
2049                 machine__remove_thread(machine, parent);
2050                 thread__put(parent);
2051                 parent = machine__findnew_thread(machine, event->fork.ppid,
2052                                                  event->fork.ptid);
2053         }
2054
2055         /* if a thread currently exists for the thread id remove it */
2056         if (thread != NULL) {
2057                 machine__remove_thread(machine, thread);
2058                 thread__put(thread);
2059         }
2060
2061         thread = machine__findnew_thread(machine, event->fork.pid,
2062                                          event->fork.tid);
2063         /*
2064          * When synthesizing FORK events, we are trying to create thread
2065          * objects for the already running tasks on the machine.
2066          *
2067          * Normally, for a kernel FORK event, we want to clone the parent's
2068          * maps because that is what the kernel just did.
2069          *
2070          * But when synthesizing, this should not be done.  If we do, we end up
2071          * with overlapping maps as we process the synthesized MMAP2 events that
2072          * get delivered shortly thereafter.
2073          *
2074          * Use the FORK event misc flags in an internal way to signal this
2075          * situation, so we can elide the map clone when appropriate.
2076          */
2077         if (event->fork.header.misc & PERF_RECORD_MISC_FORK_EXEC)
2078                 do_maps_clone = false;
2079
2080         if (thread == NULL || parent == NULL ||
2081             thread__fork(thread, parent, sample->time, do_maps_clone) < 0) {
2082                 dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n");
2083                 err = -1;
2084         }
2085         thread__put(thread);
2086         thread__put(parent);
2087
2088         return err;
2089 }
2090
2091 int machine__process_exit_event(struct machine *machine, union perf_event *event,
2092                                 struct perf_sample *sample __maybe_unused)
2093 {
2094         struct thread *thread = machine__find_thread(machine,
2095                                                      event->fork.pid,
2096                                                      event->fork.tid);
2097
2098         if (dump_trace)
2099                 perf_event__fprintf_task(event, stdout);
2100
2101         if (thread != NULL) {
2102                 thread__exited(thread);
2103                 thread__put(thread);
2104         }
2105
2106         return 0;
2107 }
2108
2109 int machine__process_event(struct machine *machine, union perf_event *event,
2110                            struct perf_sample *sample)
2111 {
2112         int ret;
2113
2114         switch (event->header.type) {
2115         case PERF_RECORD_COMM:
2116                 ret = machine__process_comm_event(machine, event, sample); break;
2117         case PERF_RECORD_MMAP:
2118                 ret = machine__process_mmap_event(machine, event, sample); break;
2119         case PERF_RECORD_NAMESPACES:
2120                 ret = machine__process_namespaces_event(machine, event, sample); break;
2121         case PERF_RECORD_CGROUP:
2122                 ret = machine__process_cgroup_event(machine, event, sample); break;
2123         case PERF_RECORD_MMAP2:
2124                 ret = machine__process_mmap2_event(machine, event, sample); break;
2125         case PERF_RECORD_FORK:
2126                 ret = machine__process_fork_event(machine, event, sample); break;
2127         case PERF_RECORD_EXIT:
2128                 ret = machine__process_exit_event(machine, event, sample); break;
2129         case PERF_RECORD_LOST:
2130                 ret = machine__process_lost_event(machine, event, sample); break;
2131         case PERF_RECORD_AUX:
2132                 ret = machine__process_aux_event(machine, event); break;
2133         case PERF_RECORD_ITRACE_START:
2134                 ret = machine__process_itrace_start_event(machine, event); break;
2135         case PERF_RECORD_LOST_SAMPLES:
2136                 ret = machine__process_lost_samples_event(machine, event, sample); break;
2137         case PERF_RECORD_SWITCH:
2138         case PERF_RECORD_SWITCH_CPU_WIDE:
2139                 ret = machine__process_switch_event(machine, event); break;
2140         case PERF_RECORD_KSYMBOL:
2141                 ret = machine__process_ksymbol(machine, event, sample); break;
2142         case PERF_RECORD_BPF_EVENT:
2143                 ret = machine__process_bpf(machine, event, sample); break;
2144         case PERF_RECORD_TEXT_POKE:
2145                 ret = machine__process_text_poke(machine, event, sample); break;
2146         case PERF_RECORD_AUX_OUTPUT_HW_ID:
2147                 ret = machine__process_aux_output_hw_id_event(machine, event); break;
2148         default:
2149                 ret = -1;
2150                 break;
2151         }
2152
2153         return ret;
2154 }
2155
2156 static bool symbol__match_regex(struct symbol *sym, regex_t *regex)
2157 {
2158         if (!regexec(regex, sym->name, 0, NULL, 0))
2159                 return true;
2160         return false;
2161 }
2162
2163 static void ip__resolve_ams(struct thread *thread,
2164                             struct addr_map_symbol *ams,
2165                             u64 ip)
2166 {
2167         struct addr_location al;
2168
2169         memset(&al, 0, sizeof(al));
2170         /*
2171          * We cannot use the header.misc hint to determine whether a
2172          * branch stack address is user, kernel, guest, hypervisor.
2173          * Branches may straddle the kernel/user/hypervisor boundaries.
2174          * Thus, we have to try consecutively until we find a match
2175          * or else, the symbol is unknown
2176          */
2177         thread__find_cpumode_addr_location(thread, ip, &al);
2178
2179         ams->addr = ip;
2180         ams->al_addr = al.addr;
2181         ams->al_level = al.level;
2182         ams->ms.maps = al.maps;
2183         ams->ms.sym = al.sym;
2184         ams->ms.map = al.map;
2185         ams->phys_addr = 0;
2186         ams->data_page_size = 0;
2187 }
2188
2189 static void ip__resolve_data(struct thread *thread,
2190                              u8 m, struct addr_map_symbol *ams,
2191                              u64 addr, u64 phys_addr, u64 daddr_page_size)
2192 {
2193         struct addr_location al;
2194
2195         memset(&al, 0, sizeof(al));
2196
2197         thread__find_symbol(thread, m, addr, &al);
2198
2199         ams->addr = addr;
2200         ams->al_addr = al.addr;
2201         ams->al_level = al.level;
2202         ams->ms.maps = al.maps;
2203         ams->ms.sym = al.sym;
2204         ams->ms.map = al.map;
2205         ams->phys_addr = phys_addr;
2206         ams->data_page_size = daddr_page_size;
2207 }
2208
2209 struct mem_info *sample__resolve_mem(struct perf_sample *sample,
2210                                      struct addr_location *al)
2211 {
2212         struct mem_info *mi = mem_info__new();
2213
2214         if (!mi)
2215                 return NULL;
2216
2217         ip__resolve_ams(al->thread, &mi->iaddr, sample->ip);
2218         ip__resolve_data(al->thread, al->cpumode, &mi->daddr,
2219                          sample->addr, sample->phys_addr,
2220                          sample->data_page_size);
2221         mi->data_src.val = sample->data_src;
2222
2223         return mi;
2224 }
2225
2226 static char *callchain_srcline(struct map_symbol *ms, u64 ip)
2227 {
2228         struct map *map = ms->map;
2229         char *srcline = NULL;
2230
2231         if (!map || callchain_param.key == CCKEY_FUNCTION)
2232                 return srcline;
2233
2234         srcline = srcline__tree_find(&map->dso->srclines, ip);
2235         if (!srcline) {
2236                 bool show_sym = false;
2237                 bool show_addr = callchain_param.key == CCKEY_ADDRESS;
2238
2239                 srcline = get_srcline(map->dso, map__rip_2objdump(map, ip),
2240                                       ms->sym, show_sym, show_addr, ip);
2241                 srcline__tree_insert(&map->dso->srclines, ip, srcline);
2242         }
2243
2244         return srcline;
2245 }
2246
2247 struct iterations {
2248         int nr_loop_iter;
2249         u64 cycles;
2250 };
2251
2252 static int add_callchain_ip(struct thread *thread,
2253                             struct callchain_cursor *cursor,
2254                             struct symbol **parent,
2255                             struct addr_location *root_al,
2256                             u8 *cpumode,
2257                             u64 ip,
2258                             bool branch,
2259                             struct branch_flags *flags,
2260                             struct iterations *iter,
2261                             u64 branch_from)
2262 {
2263         struct map_symbol ms;
2264         struct addr_location al;
2265         int nr_loop_iter = 0;
2266         u64 iter_cycles = 0;
2267         const char *srcline = NULL;
2268
2269         al.filtered = 0;
2270         al.sym = NULL;
2271         al.srcline = NULL;
2272         if (!cpumode) {
2273                 thread__find_cpumode_addr_location(thread, ip, &al);
2274         } else {
2275                 if (ip >= PERF_CONTEXT_MAX) {
2276                         switch (ip) {
2277                         case PERF_CONTEXT_HV:
2278                                 *cpumode = PERF_RECORD_MISC_HYPERVISOR;
2279                                 break;
2280                         case PERF_CONTEXT_KERNEL:
2281                                 *cpumode = PERF_RECORD_MISC_KERNEL;
2282                                 break;
2283                         case PERF_CONTEXT_USER:
2284                                 *cpumode = PERF_RECORD_MISC_USER;
2285                                 break;
2286                         default:
2287                                 pr_debug("invalid callchain context: "
2288                                          "%"PRId64"\n", (s64) ip);
2289                                 /*
2290                                  * It seems the callchain is corrupted.
2291                                  * Discard all.
2292                                  */
2293                                 callchain_cursor_reset(cursor);
2294                                 return 1;
2295                         }
2296                         return 0;
2297                 }
2298                 thread__find_symbol(thread, *cpumode, ip, &al);
2299         }
2300
2301         if (al.sym != NULL) {
2302                 if (perf_hpp_list.parent && !*parent &&
2303                     symbol__match_regex(al.sym, &parent_regex))
2304                         *parent = al.sym;
2305                 else if (have_ignore_callees && root_al &&
2306                   symbol__match_regex(al.sym, &ignore_callees_regex)) {
2307                         /* Treat this symbol as the root,
2308                            forgetting its callees. */
2309                         *root_al = al;
2310                         callchain_cursor_reset(cursor);
2311                 }
2312         }
2313
2314         if (symbol_conf.hide_unresolved && al.sym == NULL)
2315                 return 0;
2316
2317         if (iter) {
2318                 nr_loop_iter = iter->nr_loop_iter;
2319                 iter_cycles = iter->cycles;
2320         }
2321
2322         ms.maps = al.maps;
2323         ms.map = al.map;
2324         ms.sym = al.sym;
2325         srcline = callchain_srcline(&ms, al.addr);
2326         return callchain_cursor_append(cursor, ip, &ms,
2327                                        branch, flags, nr_loop_iter,
2328                                        iter_cycles, branch_from, srcline);
2329 }
2330
2331 struct branch_info *sample__resolve_bstack(struct perf_sample *sample,
2332                                            struct addr_location *al)
2333 {
2334         unsigned int i;
2335         const struct branch_stack *bs = sample->branch_stack;
2336         struct branch_entry *entries = perf_sample__branch_entries(sample);
2337         struct branch_info *bi = calloc(bs->nr, sizeof(struct branch_info));
2338
2339         if (!bi)
2340                 return NULL;
2341
2342         for (i = 0; i < bs->nr; i++) {
2343                 ip__resolve_ams(al->thread, &bi[i].to, entries[i].to);
2344                 ip__resolve_ams(al->thread, &bi[i].from, entries[i].from);
2345                 bi[i].flags = entries[i].flags;
2346         }
2347         return bi;
2348 }
2349
2350 static void save_iterations(struct iterations *iter,
2351                             struct branch_entry *be, int nr)
2352 {
2353         int i;
2354
2355         iter->nr_loop_iter++;
2356         iter->cycles = 0;
2357
2358         for (i = 0; i < nr; i++)
2359                 iter->cycles += be[i].flags.cycles;
2360 }
2361
2362 #define CHASHSZ 127
2363 #define CHASHBITS 7
2364 #define NO_ENTRY 0xff
2365
2366 #define PERF_MAX_BRANCH_DEPTH 127
2367
2368 /* Remove loops. */
2369 static int remove_loops(struct branch_entry *l, int nr,
2370                         struct iterations *iter)
2371 {
2372         int i, j, off;
2373         unsigned char chash[CHASHSZ];
2374
2375         memset(chash, NO_ENTRY, sizeof(chash));
2376
2377         BUG_ON(PERF_MAX_BRANCH_DEPTH > 255);
2378
2379         for (i = 0; i < nr; i++) {
2380                 int h = hash_64(l[i].from, CHASHBITS) % CHASHSZ;
2381
2382                 /* no collision handling for now */
2383                 if (chash[h] == NO_ENTRY) {
2384                         chash[h] = i;
2385                 } else if (l[chash[h]].from == l[i].from) {
2386                         bool is_loop = true;
2387                         /* check if it is a real loop */
2388                         off = 0;
2389                         for (j = chash[h]; j < i && i + off < nr; j++, off++)
2390                                 if (l[j].from != l[i + off].from) {
2391                                         is_loop = false;
2392                                         break;
2393                                 }
2394                         if (is_loop) {
2395                                 j = nr - (i + off);
2396                                 if (j > 0) {
2397                                         save_iterations(iter + i + off,
2398                                                 l + i, off);
2399
2400                                         memmove(iter + i, iter + i + off,
2401                                                 j * sizeof(*iter));
2402
2403                                         memmove(l + i, l + i + off,
2404                                                 j * sizeof(*l));
2405                                 }
2406
2407                                 nr -= off;
2408                         }
2409                 }
2410         }
2411         return nr;
2412 }
2413
2414 static int lbr_callchain_add_kernel_ip(struct thread *thread,
2415                                        struct callchain_cursor *cursor,
2416                                        struct perf_sample *sample,
2417                                        struct symbol **parent,
2418                                        struct addr_location *root_al,
2419                                        u64 branch_from,
2420                                        bool callee, int end)
2421 {
2422         struct ip_callchain *chain = sample->callchain;
2423         u8 cpumode = PERF_RECORD_MISC_USER;
2424         int err, i;
2425
2426         if (callee) {
2427                 for (i = 0; i < end + 1; i++) {
2428                         err = add_callchain_ip(thread, cursor, parent,
2429                                                root_al, &cpumode, chain->ips[i],
2430                                                false, NULL, NULL, branch_from);
2431                         if (err)
2432                                 return err;
2433                 }
2434                 return 0;
2435         }
2436
2437         for (i = end; i >= 0; i--) {
2438                 err = add_callchain_ip(thread, cursor, parent,
2439                                        root_al, &cpumode, chain->ips[i],
2440                                        false, NULL, NULL, branch_from);
2441                 if (err)
2442                         return err;
2443         }
2444
2445         return 0;
2446 }
2447
2448 static void save_lbr_cursor_node(struct thread *thread,
2449                                  struct callchain_cursor *cursor,
2450                                  int idx)
2451 {
2452         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2453
2454         if (!lbr_stitch)
2455                 return;
2456
2457         if (cursor->pos == cursor->nr) {
2458                 lbr_stitch->prev_lbr_cursor[idx].valid = false;
2459                 return;
2460         }
2461
2462         if (!cursor->curr)
2463                 cursor->curr = cursor->first;
2464         else
2465                 cursor->curr = cursor->curr->next;
2466         memcpy(&lbr_stitch->prev_lbr_cursor[idx], cursor->curr,
2467                sizeof(struct callchain_cursor_node));
2468
2469         lbr_stitch->prev_lbr_cursor[idx].valid = true;
2470         cursor->pos++;
2471 }
2472
2473 static int lbr_callchain_add_lbr_ip(struct thread *thread,
2474                                     struct callchain_cursor *cursor,
2475                                     struct perf_sample *sample,
2476                                     struct symbol **parent,
2477                                     struct addr_location *root_al,
2478                                     u64 *branch_from,
2479                                     bool callee)
2480 {
2481         struct branch_stack *lbr_stack = sample->branch_stack;
2482         struct branch_entry *entries = perf_sample__branch_entries(sample);
2483         u8 cpumode = PERF_RECORD_MISC_USER;
2484         int lbr_nr = lbr_stack->nr;
2485         struct branch_flags *flags;
2486         int err, i;
2487         u64 ip;
2488
2489         /*
2490          * The curr and pos are not used in writing session. They are cleared
2491          * in callchain_cursor_commit() when the writing session is closed.
2492          * Using curr and pos to track the current cursor node.
2493          */
2494         if (thread->lbr_stitch) {
2495                 cursor->curr = NULL;
2496                 cursor->pos = cursor->nr;
2497                 if (cursor->nr) {
2498                         cursor->curr = cursor->first;
2499                         for (i = 0; i < (int)(cursor->nr - 1); i++)
2500                                 cursor->curr = cursor->curr->next;
2501                 }
2502         }
2503
2504         if (callee) {
2505                 /* Add LBR ip from first entries.to */
2506                 ip = entries[0].to;
2507                 flags = &entries[0].flags;
2508                 *branch_from = entries[0].from;
2509                 err = add_callchain_ip(thread, cursor, parent,
2510                                        root_al, &cpumode, ip,
2511                                        true, flags, NULL,
2512                                        *branch_from);
2513                 if (err)
2514                         return err;
2515
2516                 /*
2517                  * The number of cursor node increases.
2518                  * Move the current cursor node.
2519                  * But does not need to save current cursor node for entry 0.
2520                  * It's impossible to stitch the whole LBRs of previous sample.
2521                  */
2522                 if (thread->lbr_stitch && (cursor->pos != cursor->nr)) {
2523                         if (!cursor->curr)
2524                                 cursor->curr = cursor->first;
2525                         else
2526                                 cursor->curr = cursor->curr->next;
2527                         cursor->pos++;
2528                 }
2529
2530                 /* Add LBR ip from entries.from one by one. */
2531                 for (i = 0; i < lbr_nr; i++) {
2532                         ip = entries[i].from;
2533                         flags = &entries[i].flags;
2534                         err = add_callchain_ip(thread, cursor, parent,
2535                                                root_al, &cpumode, ip,
2536                                                true, flags, NULL,
2537                                                *branch_from);
2538                         if (err)
2539                                 return err;
2540                         save_lbr_cursor_node(thread, cursor, i);
2541                 }
2542                 return 0;
2543         }
2544
2545         /* Add LBR ip from entries.from one by one. */
2546         for (i = lbr_nr - 1; i >= 0; i--) {
2547                 ip = entries[i].from;
2548                 flags = &entries[i].flags;
2549                 err = add_callchain_ip(thread, cursor, parent,
2550                                        root_al, &cpumode, ip,
2551                                        true, flags, NULL,
2552                                        *branch_from);
2553                 if (err)
2554                         return err;
2555                 save_lbr_cursor_node(thread, cursor, i);
2556         }
2557
2558         /* Add LBR ip from first entries.to */
2559         ip = entries[0].to;
2560         flags = &entries[0].flags;
2561         *branch_from = entries[0].from;
2562         err = add_callchain_ip(thread, cursor, parent,
2563                                root_al, &cpumode, ip,
2564                                true, flags, NULL,
2565                                *branch_from);
2566         if (err)
2567                 return err;
2568
2569         return 0;
2570 }
2571
2572 static int lbr_callchain_add_stitched_lbr_ip(struct thread *thread,
2573                                              struct callchain_cursor *cursor)
2574 {
2575         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2576         struct callchain_cursor_node *cnode;
2577         struct stitch_list *stitch_node;
2578         int err;
2579
2580         list_for_each_entry(stitch_node, &lbr_stitch->lists, node) {
2581                 cnode = &stitch_node->cursor;
2582
2583                 err = callchain_cursor_append(cursor, cnode->ip,
2584                                               &cnode->ms,
2585                                               cnode->branch,
2586                                               &cnode->branch_flags,
2587                                               cnode->nr_loop_iter,
2588                                               cnode->iter_cycles,
2589                                               cnode->branch_from,
2590                                               cnode->srcline);
2591                 if (err)
2592                         return err;
2593         }
2594         return 0;
2595 }
2596
2597 static struct stitch_list *get_stitch_node(struct thread *thread)
2598 {
2599         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2600         struct stitch_list *stitch_node;
2601
2602         if (!list_empty(&lbr_stitch->free_lists)) {
2603                 stitch_node = list_first_entry(&lbr_stitch->free_lists,
2604                                                struct stitch_list, node);
2605                 list_del(&stitch_node->node);
2606
2607                 return stitch_node;
2608         }
2609
2610         return malloc(sizeof(struct stitch_list));
2611 }
2612
2613 static bool has_stitched_lbr(struct thread *thread,
2614                              struct perf_sample *cur,
2615                              struct perf_sample *prev,
2616                              unsigned int max_lbr,
2617                              bool callee)
2618 {
2619         struct branch_stack *cur_stack = cur->branch_stack;
2620         struct branch_entry *cur_entries = perf_sample__branch_entries(cur);
2621         struct branch_stack *prev_stack = prev->branch_stack;
2622         struct branch_entry *prev_entries = perf_sample__branch_entries(prev);
2623         struct lbr_stitch *lbr_stitch = thread->lbr_stitch;
2624         int i, j, nr_identical_branches = 0;
2625         struct stitch_list *stitch_node;
2626         u64 cur_base, distance;
2627
2628         if (!cur_stack || !prev_stack)
2629                 return false;
2630
2631         /* Find the physical index of the base-of-stack for current sample. */
2632         cur_base = max_lbr - cur_stack->nr + cur_stack->hw_idx + 1;
2633
2634         distance = (prev_stack->hw_idx > cur_base) ? (prev_stack->hw_idx - cur_base) :
2635                                                      (max_lbr + prev_stack->hw_idx - cur_base);
2636         /* Previous sample has shorter stack. Nothing can be stitched. */
2637         if (distance + 1 > prev_stack->nr)
2638                 return false;
2639
2640         /*
2641          * Check if there are identical LBRs between two samples.
2642          * Identical LBRs must have same from, to and flags values. Also,
2643          * they have to be saved in the same LBR registers (same physical
2644          * index).
2645          *
2646          * Starts from the base-of-stack of current sample.
2647          */
2648         for (i = distance, j = cur_stack->nr - 1; (i >= 0) && (j >= 0); i--, j--) {
2649                 if ((prev_entries[i].from != cur_entries[j].from) ||
2650                     (prev_entries[i].to != cur_entries[j].to) ||
2651                     (prev_entries[i].flags.value != cur_entries[j].flags.value))
2652                         break;
2653                 nr_identical_branches++;
2654         }
2655
2656         if (!nr_identical_branches)
2657                 return false;
2658
2659         /*
2660          * Save the LBRs between the base-of-stack of previous sample
2661          * and the base-of-stack of current sample into lbr_stitch->lists.
2662          * These LBRs will be stitched later.
2663          */
2664         for (i = prev_stack->nr - 1; i > (int)distance; i--) {
2665
2666                 if (!lbr_stitch->prev_lbr_cursor[i].valid)
2667                         continue;
2668
2669                 stitch_node = get_stitch_node(thread);
2670                 if (!stitch_node)
2671                         return false;
2672
2673                 memcpy(&stitch_node->cursor, &lbr_stitch->prev_lbr_cursor[i],
2674                        sizeof(struct callchain_cursor_node));
2675
2676                 if (callee)
2677                         list_add(&stitch_node->node, &lbr_stitch->lists);
2678                 else
2679                         list_add_tail(&stitch_node->node, &lbr_stitch->lists);
2680         }
2681
2682         return true;
2683 }
2684
2685 static bool alloc_lbr_stitch(struct thread *thread, unsigned int max_lbr)
2686 {
2687         if (thread->lbr_stitch)
2688                 return true;
2689
2690         thread->lbr_stitch = zalloc(sizeof(*thread->lbr_stitch));
2691         if (!thread->lbr_stitch)
2692                 goto err;
2693
2694         thread->lbr_stitch->prev_lbr_cursor = calloc(max_lbr + 1, sizeof(struct callchain_cursor_node));
2695         if (!thread->lbr_stitch->prev_lbr_cursor)
2696                 goto free_lbr_stitch;
2697
2698         INIT_LIST_HEAD(&thread->lbr_stitch->lists);
2699         INIT_LIST_HEAD(&thread->lbr_stitch->free_lists);
2700
2701         return true;
2702
2703 free_lbr_stitch:
2704         zfree(&thread->lbr_stitch);
2705 err:
2706         pr_warning("Failed to allocate space for stitched LBRs. Disable LBR stitch\n");
2707         thread->lbr_stitch_enable = false;
2708         return false;
2709 }
2710
2711 /*
2712  * Resolve LBR callstack chain sample
2713  * Return:
2714  * 1 on success get LBR callchain information
2715  * 0 no available LBR callchain information, should try fp
2716  * negative error code on other errors.
2717  */
2718 static int resolve_lbr_callchain_sample(struct thread *thread,
2719                                         struct callchain_cursor *cursor,
2720                                         struct perf_sample *sample,
2721                                         struct symbol **parent,
2722                                         struct addr_location *root_al,
2723                                         int max_stack,
2724                                         unsigned int max_lbr)
2725 {
2726         bool callee = (callchain_param.order == ORDER_CALLEE);
2727         struct ip_callchain *chain = sample->callchain;
2728         int chain_nr = min(max_stack, (int)chain->nr), i;
2729         struct lbr_stitch *lbr_stitch;
2730         bool stitched_lbr = false;
2731         u64 branch_from = 0;
2732         int err;
2733
2734         for (i = 0; i < chain_nr; i++) {
2735                 if (chain->ips[i] == PERF_CONTEXT_USER)
2736                         break;
2737         }
2738
2739         /* LBR only affects the user callchain */
2740         if (i == chain_nr)
2741                 return 0;
2742
2743         if (thread->lbr_stitch_enable && !sample->no_hw_idx &&
2744             (max_lbr > 0) && alloc_lbr_stitch(thread, max_lbr)) {
2745                 lbr_stitch = thread->lbr_stitch;
2746
2747                 stitched_lbr = has_stitched_lbr(thread, sample,
2748                                                 &lbr_stitch->prev_sample,
2749                                                 max_lbr, callee);
2750
2751                 if (!stitched_lbr && !list_empty(&lbr_stitch->lists)) {
2752                         list_replace_init(&lbr_stitch->lists,
2753                                           &lbr_stitch->free_lists);
2754                 }
2755                 memcpy(&lbr_stitch->prev_sample, sample, sizeof(*sample));
2756         }
2757
2758         if (callee) {
2759                 /* Add kernel ip */
2760                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2761                                                   parent, root_al, branch_from,
2762                                                   true, i);
2763                 if (err)
2764                         goto error;
2765
2766                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2767                                                root_al, &branch_from, true);
2768                 if (err)
2769                         goto error;
2770
2771                 if (stitched_lbr) {
2772                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2773                         if (err)
2774                                 goto error;
2775                 }
2776
2777         } else {
2778                 if (stitched_lbr) {
2779                         err = lbr_callchain_add_stitched_lbr_ip(thread, cursor);
2780                         if (err)
2781                                 goto error;
2782                 }
2783                 err = lbr_callchain_add_lbr_ip(thread, cursor, sample, parent,
2784                                                root_al, &branch_from, false);
2785                 if (err)
2786                         goto error;
2787
2788                 /* Add kernel ip */
2789                 err = lbr_callchain_add_kernel_ip(thread, cursor, sample,
2790                                                   parent, root_al, branch_from,
2791                                                   false, i);
2792                 if (err)
2793                         goto error;
2794         }
2795         return 1;
2796
2797 error:
2798         return (err < 0) ? err : 0;
2799 }
2800
2801 static int find_prev_cpumode(struct ip_callchain *chain, struct thread *thread,
2802                              struct callchain_cursor *cursor,
2803                              struct symbol **parent,
2804                              struct addr_location *root_al,
2805                              u8 *cpumode, int ent)
2806 {
2807         int err = 0;
2808
2809         while (--ent >= 0) {
2810                 u64 ip = chain->ips[ent];
2811
2812                 if (ip >= PERF_CONTEXT_MAX) {
2813                         err = add_callchain_ip(thread, cursor, parent,
2814                                                root_al, cpumode, ip,
2815                                                false, NULL, NULL, 0);
2816                         break;
2817                 }
2818         }
2819         return err;
2820 }
2821
2822 static u64 get_leaf_frame_caller(struct perf_sample *sample,
2823                 struct thread *thread, int usr_idx)
2824 {
2825         if (machine__normalized_is(thread->maps->machine, "arm64"))
2826                 return get_leaf_frame_caller_aarch64(sample, thread, usr_idx);
2827         else
2828                 return 0;
2829 }
2830
2831 static int thread__resolve_callchain_sample(struct thread *thread,
2832                                             struct callchain_cursor *cursor,
2833                                             struct evsel *evsel,
2834                                             struct perf_sample *sample,
2835                                             struct symbol **parent,
2836                                             struct addr_location *root_al,
2837                                             int max_stack)
2838 {
2839         struct branch_stack *branch = sample->branch_stack;
2840         struct branch_entry *entries = perf_sample__branch_entries(sample);
2841         struct ip_callchain *chain = sample->callchain;
2842         int chain_nr = 0;
2843         u8 cpumode = PERF_RECORD_MISC_USER;
2844         int i, j, err, nr_entries, usr_idx;
2845         int skip_idx = -1;
2846         int first_call = 0;
2847         u64 leaf_frame_caller;
2848
2849         if (chain)
2850                 chain_nr = chain->nr;
2851
2852         if (evsel__has_branch_callstack(evsel)) {
2853                 struct perf_env *env = evsel__env(evsel);
2854
2855                 err = resolve_lbr_callchain_sample(thread, cursor, sample, parent,
2856                                                    root_al, max_stack,
2857                                                    !env ? 0 : env->max_branches);
2858                 if (err)
2859                         return (err < 0) ? err : 0;
2860         }
2861
2862         /*
2863          * Based on DWARF debug information, some architectures skip
2864          * a callchain entry saved by the kernel.
2865          */
2866         skip_idx = arch_skip_callchain_idx(thread, chain);
2867
2868         /*
2869          * Add branches to call stack for easier browsing. This gives
2870          * more context for a sample than just the callers.
2871          *
2872          * This uses individual histograms of paths compared to the
2873          * aggregated histograms the normal LBR mode uses.
2874          *
2875          * Limitations for now:
2876          * - No extra filters
2877          * - No annotations (should annotate somehow)
2878          */
2879
2880         if (branch && callchain_param.branch_callstack) {
2881                 int nr = min(max_stack, (int)branch->nr);
2882                 struct branch_entry be[nr];
2883                 struct iterations iter[nr];
2884
2885                 if (branch->nr > PERF_MAX_BRANCH_DEPTH) {
2886                         pr_warning("corrupted branch chain. skipping...\n");
2887                         goto check_calls;
2888                 }
2889
2890                 for (i = 0; i < nr; i++) {
2891                         if (callchain_param.order == ORDER_CALLEE) {
2892                                 be[i] = entries[i];
2893
2894                                 if (chain == NULL)
2895                                         continue;
2896
2897                                 /*
2898                                  * Check for overlap into the callchain.
2899                                  * The return address is one off compared to
2900                                  * the branch entry. To adjust for this
2901                                  * assume the calling instruction is not longer
2902                                  * than 8 bytes.
2903                                  */
2904                                 if (i == skip_idx ||
2905                                     chain->ips[first_call] >= PERF_CONTEXT_MAX)
2906                                         first_call++;
2907                                 else if (be[i].from < chain->ips[first_call] &&
2908                                     be[i].from >= chain->ips[first_call] - 8)
2909                                         first_call++;
2910                         } else
2911                                 be[i] = entries[branch->nr - i - 1];
2912                 }
2913
2914                 memset(iter, 0, sizeof(struct iterations) * nr);
2915                 nr = remove_loops(be, nr, iter);
2916
2917                 for (i = 0; i < nr; i++) {
2918                         err = add_callchain_ip(thread, cursor, parent,
2919                                                root_al,
2920                                                NULL, be[i].to,
2921                                                true, &be[i].flags,
2922                                                NULL, be[i].from);
2923
2924                         if (!err)
2925                                 err = add_callchain_ip(thread, cursor, parent, root_al,
2926                                                        NULL, be[i].from,
2927                                                        true, &be[i].flags,
2928                                                        &iter[i], 0);
2929                         if (err == -EINVAL)
2930                                 break;
2931                         if (err)
2932                                 return err;
2933                 }
2934
2935                 if (chain_nr == 0)
2936                         return 0;
2937
2938                 chain_nr -= nr;
2939         }
2940
2941 check_calls:
2942         if (chain && callchain_param.order != ORDER_CALLEE) {
2943                 err = find_prev_cpumode(chain, thread, cursor, parent, root_al,
2944                                         &cpumode, chain->nr - first_call);
2945                 if (err)
2946                         return (err < 0) ? err : 0;
2947         }
2948         for (i = first_call, nr_entries = 0;
2949              i < chain_nr && nr_entries < max_stack; i++) {
2950                 u64 ip;
2951
2952                 if (callchain_param.order == ORDER_CALLEE)
2953                         j = i;
2954                 else
2955                         j = chain->nr - i - 1;
2956
2957 #ifdef HAVE_SKIP_CALLCHAIN_IDX
2958                 if (j == skip_idx)
2959                         continue;
2960 #endif
2961                 ip = chain->ips[j];
2962                 if (ip < PERF_CONTEXT_MAX)
2963                        ++nr_entries;
2964                 else if (callchain_param.order != ORDER_CALLEE) {
2965                         err = find_prev_cpumode(chain, thread, cursor, parent,
2966                                                 root_al, &cpumode, j);
2967                         if (err)
2968                                 return (err < 0) ? err : 0;
2969                         continue;
2970                 }
2971
2972                 /*
2973                  * PERF_CONTEXT_USER allows us to locate where the user stack ends.
2974                  * Depending on callchain_param.order and the position of PERF_CONTEXT_USER,
2975                  * the index will be different in order to add the missing frame
2976                  * at the right place.
2977                  */
2978
2979                 usr_idx = callchain_param.order == ORDER_CALLEE ? j-2 : j-1;
2980
2981                 if (usr_idx >= 0 && chain->ips[usr_idx] == PERF_CONTEXT_USER) {
2982
2983                         leaf_frame_caller = get_leaf_frame_caller(sample, thread, usr_idx);
2984
2985                         /*
2986                          * check if leaf_frame_Caller != ip to not add the same
2987                          * value twice.
2988                          */
2989
2990                         if (leaf_frame_caller && leaf_frame_caller != ip) {
2991
2992                                 err = add_callchain_ip(thread, cursor, parent,
2993                                                root_al, &cpumode, leaf_frame_caller,
2994                                                false, NULL, NULL, 0);
2995                                 if (err)
2996                                         return (err < 0) ? err : 0;
2997                         }
2998                 }
2999
3000                 err = add_callchain_ip(thread, cursor, parent,
3001                                        root_al, &cpumode, ip,
3002                                        false, NULL, NULL, 0);
3003
3004                 if (err)
3005                         return (err < 0) ? err : 0;
3006         }
3007
3008         return 0;
3009 }
3010
3011 static int append_inlines(struct callchain_cursor *cursor, struct map_symbol *ms, u64 ip)
3012 {
3013         struct symbol *sym = ms->sym;
3014         struct map *map = ms->map;
3015         struct inline_node *inline_node;
3016         struct inline_list *ilist;
3017         u64 addr;
3018         int ret = 1;
3019
3020         if (!symbol_conf.inline_name || !map || !sym)
3021                 return ret;
3022
3023         addr = map__map_ip(map, ip);
3024         addr = map__rip_2objdump(map, addr);
3025
3026         inline_node = inlines__tree_find(&map->dso->inlined_nodes, addr);
3027         if (!inline_node) {
3028                 inline_node = dso__parse_addr_inlines(map->dso, addr, sym);
3029                 if (!inline_node)
3030                         return ret;
3031                 inlines__tree_insert(&map->dso->inlined_nodes, inline_node);
3032         }
3033
3034         list_for_each_entry(ilist, &inline_node->val, list) {
3035                 struct map_symbol ilist_ms = {
3036                         .maps = ms->maps,
3037                         .map = map,
3038                         .sym = ilist->symbol,
3039                 };
3040                 ret = callchain_cursor_append(cursor, ip, &ilist_ms, false,
3041                                               NULL, 0, 0, 0, ilist->srcline);
3042
3043                 if (ret != 0)
3044                         return ret;
3045         }
3046
3047         return ret;
3048 }
3049
3050 static int unwind_entry(struct unwind_entry *entry, void *arg)
3051 {
3052         struct callchain_cursor *cursor = arg;
3053         const char *srcline = NULL;
3054         u64 addr = entry->ip;
3055
3056         if (symbol_conf.hide_unresolved && entry->ms.sym == NULL)
3057                 return 0;
3058
3059         if (append_inlines(cursor, &entry->ms, entry->ip) == 0)
3060                 return 0;
3061
3062         /*
3063          * Convert entry->ip from a virtual address to an offset in
3064          * its corresponding binary.
3065          */
3066         if (entry->ms.map)
3067                 addr = map__map_ip(entry->ms.map, entry->ip);
3068
3069         srcline = callchain_srcline(&entry->ms, addr);
3070         return callchain_cursor_append(cursor, entry->ip, &entry->ms,
3071                                        false, NULL, 0, 0, 0, srcline);
3072 }
3073
3074 static int thread__resolve_callchain_unwind(struct thread *thread,
3075                                             struct callchain_cursor *cursor,
3076                                             struct evsel *evsel,
3077                                             struct perf_sample *sample,
3078                                             int max_stack)
3079 {
3080         /* Can we do dwarf post unwind? */
3081         if (!((evsel->core.attr.sample_type & PERF_SAMPLE_REGS_USER) &&
3082               (evsel->core.attr.sample_type & PERF_SAMPLE_STACK_USER)))
3083                 return 0;
3084
3085         /* Bail out if nothing was captured. */
3086         if ((!sample->user_regs.regs) ||
3087             (!sample->user_stack.size))
3088                 return 0;
3089
3090         return unwind__get_entries(unwind_entry, cursor,
3091                                    thread, sample, max_stack, false);
3092 }
3093
3094 int thread__resolve_callchain(struct thread *thread,
3095                               struct callchain_cursor *cursor,
3096                               struct evsel *evsel,
3097                               struct perf_sample *sample,
3098                               struct symbol **parent,
3099                               struct addr_location *root_al,
3100                               int max_stack)
3101 {
3102         int ret = 0;
3103
3104         callchain_cursor_reset(cursor);
3105
3106         if (callchain_param.order == ORDER_CALLEE) {
3107                 ret = thread__resolve_callchain_sample(thread, cursor,
3108                                                        evsel, sample,
3109                                                        parent, root_al,
3110                                                        max_stack);
3111                 if (ret)
3112                         return ret;
3113                 ret = thread__resolve_callchain_unwind(thread, cursor,
3114                                                        evsel, sample,
3115                                                        max_stack);
3116         } else {
3117                 ret = thread__resolve_callchain_unwind(thread, cursor,
3118                                                        evsel, sample,
3119                                                        max_stack);
3120                 if (ret)
3121                         return ret;
3122                 ret = thread__resolve_callchain_sample(thread, cursor,
3123                                                        evsel, sample,
3124                                                        parent, root_al,
3125                                                        max_stack);
3126         }
3127
3128         return ret;
3129 }
3130
3131 int machine__for_each_thread(struct machine *machine,
3132                              int (*fn)(struct thread *thread, void *p),
3133                              void *priv)
3134 {
3135         struct threads *threads;
3136         struct rb_node *nd;
3137         struct thread *thread;
3138         int rc = 0;
3139         int i;
3140
3141         for (i = 0; i < THREADS__TABLE_SIZE; i++) {
3142                 threads = &machine->threads[i];
3143                 for (nd = rb_first_cached(&threads->entries); nd;
3144                      nd = rb_next(nd)) {
3145                         thread = rb_entry(nd, struct thread, rb_node);
3146                         rc = fn(thread, priv);
3147                         if (rc != 0)
3148                                 return rc;
3149                 }
3150
3151                 list_for_each_entry(thread, &threads->dead, node) {
3152                         rc = fn(thread, priv);
3153                         if (rc != 0)
3154                                 return rc;
3155                 }
3156         }
3157         return rc;
3158 }
3159
3160 int machines__for_each_thread(struct machines *machines,
3161                               int (*fn)(struct thread *thread, void *p),
3162                               void *priv)
3163 {
3164         struct rb_node *nd;
3165         int rc = 0;
3166
3167         rc = machine__for_each_thread(&machines->host, fn, priv);
3168         if (rc != 0)
3169                 return rc;
3170
3171         for (nd = rb_first_cached(&machines->guests); nd; nd = rb_next(nd)) {
3172                 struct machine *machine = rb_entry(nd, struct machine, rb_node);
3173
3174                 rc = machine__for_each_thread(machine, fn, priv);
3175                 if (rc != 0)
3176                         return rc;
3177         }
3178         return rc;
3179 }
3180
3181 pid_t machine__get_current_tid(struct machine *machine, int cpu)
3182 {
3183         if (cpu < 0 || (size_t)cpu >= machine->current_tid_sz)
3184                 return -1;
3185
3186         return machine->current_tid[cpu];
3187 }
3188
3189 int machine__set_current_tid(struct machine *machine, int cpu, pid_t pid,
3190                              pid_t tid)
3191 {
3192         struct thread *thread;
3193         const pid_t init_val = -1;
3194
3195         if (cpu < 0)
3196                 return -EINVAL;
3197
3198         if (realloc_array_as_needed(machine->current_tid,
3199                                     machine->current_tid_sz,
3200                                     (unsigned int)cpu,
3201                                     &init_val))
3202                 return -ENOMEM;
3203
3204         machine->current_tid[cpu] = tid;
3205
3206         thread = machine__findnew_thread(machine, pid, tid);
3207         if (!thread)
3208                 return -ENOMEM;
3209
3210         thread->cpu = cpu;
3211         thread__put(thread);
3212
3213         return 0;
3214 }
3215
3216 /*
3217  * Compares the raw arch string. N.B. see instead perf_env__arch() or
3218  * machine__normalized_is() if a normalized arch is needed.
3219  */
3220 bool machine__is(struct machine *machine, const char *arch)
3221 {
3222         return machine && !strcmp(perf_env__raw_arch(machine->env), arch);
3223 }
3224
3225 bool machine__normalized_is(struct machine *machine, const char *arch)
3226 {
3227         return machine && !strcmp(perf_env__arch(machine->env), arch);
3228 }
3229
3230 int machine__nr_cpus_avail(struct machine *machine)
3231 {
3232         return machine ? perf_env__nr_cpus_avail(machine->env) : 0;
3233 }
3234
3235 int machine__get_kernel_start(struct machine *machine)
3236 {
3237         struct map *map = machine__kernel_map(machine);
3238         int err = 0;
3239
3240         /*
3241          * The only addresses above 2^63 are kernel addresses of a 64-bit
3242          * kernel.  Note that addresses are unsigned so that on a 32-bit system
3243          * all addresses including kernel addresses are less than 2^32.  In
3244          * that case (32-bit system), if the kernel mapping is unknown, all
3245          * addresses will be assumed to be in user space - see
3246          * machine__kernel_ip().
3247          */
3248         machine->kernel_start = 1ULL << 63;
3249         if (map) {
3250                 err = map__load(map);
3251                 /*
3252                  * On x86_64, PTI entry trampolines are less than the
3253                  * start of kernel text, but still above 2^63. So leave
3254                  * kernel_start = 1ULL << 63 for x86_64.
3255                  */
3256                 if (!err && !machine__is(machine, "x86_64"))
3257                         machine->kernel_start = map->start;
3258         }
3259         return err;
3260 }
3261
3262 u8 machine__addr_cpumode(struct machine *machine, u8 cpumode, u64 addr)
3263 {
3264         u8 addr_cpumode = cpumode;
3265         bool kernel_ip;
3266
3267         if (!machine->single_address_space)
3268                 goto out;
3269
3270         kernel_ip = machine__kernel_ip(machine, addr);
3271         switch (cpumode) {
3272         case PERF_RECORD_MISC_KERNEL:
3273         case PERF_RECORD_MISC_USER:
3274                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_KERNEL :
3275                                            PERF_RECORD_MISC_USER;
3276                 break;
3277         case PERF_RECORD_MISC_GUEST_KERNEL:
3278         case PERF_RECORD_MISC_GUEST_USER:
3279                 addr_cpumode = kernel_ip ? PERF_RECORD_MISC_GUEST_KERNEL :
3280                                            PERF_RECORD_MISC_GUEST_USER;
3281                 break;
3282         default:
3283                 break;
3284         }
3285 out:
3286         return addr_cpumode;
3287 }
3288
3289 struct dso *machine__findnew_dso_id(struct machine *machine, const char *filename, struct dso_id *id)
3290 {
3291         return dsos__findnew_id(&machine->dsos, filename, id);
3292 }
3293
3294 struct dso *machine__findnew_dso(struct machine *machine, const char *filename)
3295 {
3296         return machine__findnew_dso_id(machine, filename, NULL);
3297 }
3298
3299 char *machine__resolve_kernel_addr(void *vmachine, unsigned long long *addrp, char **modp)
3300 {
3301         struct machine *machine = vmachine;
3302         struct map *map;
3303         struct symbol *sym = machine__find_kernel_symbol(machine, *addrp, &map);
3304
3305         if (sym == NULL)
3306                 return NULL;
3307
3308         *modp = __map__is_kmodule(map) ? (char *)map->dso->short_name : NULL;
3309         *addrp = map->unmap_ip(map, sym->start);
3310         return sym->name;
3311 }
3312
3313 int machine__for_each_dso(struct machine *machine, machine__dso_t fn, void *priv)
3314 {
3315         struct dso *pos;
3316         int err = 0;
3317
3318         list_for_each_entry(pos, &machine->dsos.head, node) {
3319                 if (fn(pos, machine, priv))
3320                         err = -1;
3321         }
3322         return err;
3323 }
3324
3325 int machine__for_each_kernel_map(struct machine *machine, machine__map_t fn, void *priv)
3326 {
3327         struct maps *maps = machine__kernel_maps(machine);
3328         struct map *map;
3329         int err = 0;
3330
3331         for (map = maps__first(maps); map != NULL; map = map__next(map)) {
3332                 err = fn(map, priv);
3333                 if (err != 0) {
3334                         break;
3335                 }
3336         }
3337         return err;
3338 }
3339
3340 bool machine__is_lock_function(struct machine *machine, u64 addr)
3341 {
3342         if (!machine->sched.text_start) {
3343                 struct map *kmap;
3344                 struct symbol *sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_start", &kmap);
3345
3346                 if (!sym) {
3347                         /* to avoid retry */
3348                         machine->sched.text_start = 1;
3349                         return false;
3350                 }
3351
3352                 machine->sched.text_start = kmap->unmap_ip(kmap, sym->start);
3353
3354                 /* should not fail from here */
3355                 sym = machine__find_kernel_symbol_by_name(machine, "__sched_text_end", &kmap);
3356                 machine->sched.text_end = kmap->unmap_ip(kmap, sym->start);
3357
3358                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_start", &kmap);
3359                 machine->lock.text_start = kmap->unmap_ip(kmap, sym->start);
3360
3361                 sym = machine__find_kernel_symbol_by_name(machine, "__lock_text_end", &kmap);
3362                 machine->lock.text_end = kmap->unmap_ip(kmap, sym->start);
3363         }
3364
3365         /* failed to get kernel symbols */
3366         if (machine->sched.text_start == 1)
3367                 return false;
3368
3369         /* mutex and rwsem functions are in sched text section */
3370         if (machine->sched.text_start <= addr && addr < machine->sched.text_end)
3371                 return true;
3372
3373         /* spinlock functions are in lock text section */
3374         if (machine->lock.text_start <= addr && addr < machine->lock.text_end)
3375                 return true;
3376
3377         return false;
3378 }